Interannual Variability of Mean and Extreme Rainfall and Relationship with Large-Scale Circulation

Interannual variability of mean and extreme rainfall and relationship with large-scale circulation

Malcolm Haylock

A thesis submitted in fulfilment of the requirements for the degree of Doctor of Philosophy by Publication at the University of East Anglia

December 2004

© This copy of the thesis has been supplied on condition that anyone who consults it is understood to recognise that its copyright rests with the author and that no quotation from the thesis, nor any information derived therefrom, may be published without the author’s prior, written consent. Interannual variability of mean and extreme rainfall and relationship with large-scale circulation

Malcolm Haylock 2004

Abstract

Seven journal publications are presented documenting historical trends and interannual variability of mean and extreme rainfall. The seven studies cover diverse geographical regions including Australia, southeast Asia and the Pacific, Europe and South America. While two of the studies solely present trends in mean and extreme rainfall in recent decades, the remainder seek causes for the observed trends. The first paper explains the reduction in winter rainfall observed in recent decades in southwest Western Australia. Composites of sea level pressure reveal that during anomalously dry years there was a westward expansion of the continental anticyclone along with an intensification of the trough at higher latitudes, thus confining the rain bearing storms to the south of the continent. The second paper examines the seasonal dependence of the spatial coherence of Indonesian rainfall, showing that high coherence coincides with a strong relationship with the El Niño – Southern Oscillation phenomenon (ENSO). The third paper extends this work to cover the entire Maritime Continent, showing that the ENSO-affected region is strongly related to large-scale changes in the Pacific-wide ENSO related anomalies of sea surface temperature. The fourth paper uses several indices of daily extreme rainfall to document whether there has been an observed change in extreme rainfall over Australia and to determine if the results are sensitive to the choice of rainfall indices. The fifth paper presents results from a southeast Asia and Pacific regional workshop that examined trends in indices of extreme daily rainfall and temperature. While extreme temperatures have increased in recent decades, rainfall trends are more region- dependent. The sixth paper shows extreme rainfall trends from a similar workshop held in South America. A strong regional dependence of the observed trends is shown to be the result of changes in the behaviour of ENSO and a weakening of the continental trough in surface pressure. The seventh paper relates changes in European winter extreme rainfall to changes in pressure. The large-scale latitude dependent trend is the result of changes in the North Atlantic Oscillation (NAO). The seven studies show that, despite the expectation that a warmer atmosphere means an enhanced hydrological cycle, mean and extreme rainfall has shown no consistent increase. Moreover, long-term trends in rainfall are dominated by changes in large scale climate signals, such as ENSO and the NAO. Contents

Chapter 1: Introduction ...... 2

How has winter rainfall changed in southwest Western Australia and what has caused the observed change? ...... 5 How does rainfall vary over Indonesia and can we conclude anything about predictability?...... 9 How does rainfall vary over the Maritime Continent and what causes this?...... 12 How has extreme rainfall changed over Australia and are results sensitive to the choice of indices? ...... 15 How have extreme rainfall and temperature changed over Southeast Asia? ...... 19 How has extreme rainfall changed over South America and what has caused this?.. 23 How does extreme winter rainfall vary over Europe and what causes this?...... 27 Conclusions ...... 31 References ...... 32

Chapter 2: Allan and Haylock, 1993

Chapter 3: Haylock and McBride, 2001

Chapter 4: McBride et al., 2003

Chapter 5: Haylock and Nicholls, 2000

Chapter 6: Manton et al., 2001

Chapter 7: Haylock et al., 2004

Chapter 8: Haylock and Goodess, 2004

Appendix 1: Personal Contribution

Appendix 2: Citations

Appendix 3: Publications

Acknowledgements

1 Chapter 1: Introduction

There are two main reasons why we would wish to study the climate: to increase our knowledge for the sake of learning; and to be able to determine what is the most likely climate at varying timescales in the future.

Climate forecasting is particularly important for rainfall. It is one of the key climate variables that have important influence on both our lives and the health of all ecosystems. Interannual variability of mean and extreme rainfall can have devastating consequences. The El Niño-induced drought in southeast Australia in summer 2002-3 was combined with unusually warm temperatures to produce one of the worst wildfire seasons on record, with the destruction of almost 4 million Ha of forest (Nairn, 2003). In contrast, extreme and persistent rainfall in central and eastern Europe in summer 2002 led to severe flooding causing over €21 billion in economic loss and the loss of over 100 lives (Munich Re, 2003). Knowledge in advance of such events would enable the chance to take precautionary steps to reduce the impact. Longer-term forecasts are important for less dramatic, but equally important reasons, such as the ability to plan for any major changes to the spatial and temporal distribution of rainfall (e.g. building dams to provide water resources) or, in the case of human-induced climate change, to take steps now to mitigate such changes.

There are two common timescales on which much current work is involved: seasonal forecasting and climate change scenarios. Seasonal forecasts are routinely issued by several national meteorological services, which give the most probable climate for the next few months. Climate change scenarios are currently issued based on available output from climate model experiments, generally up to the end of the 21st century.

Two common methods are currently in use for operational seasonal forecasts: statistical and dynamical. Statistical methods use current conditions to predict likely future conditions. Since rainfall has very little temporal autocorrelation i.e. given that it is currently an unusually wet month, we can’t determine with sufficient statistical confidence what will probably happen in several months time, we need to use other variables with which to forecast rainfall. Slowly evolving sea surface temperature (SST) observations are usually used, provided there is sufficient predictive skill. In this method, we are not trying to forecast individual rainfall events, but rather the mean conditions, although there may be sufficient skill to also model other parts of the rainfall distribution such as the probability of observing extreme events.

2 Dynamical models are used increasingly for seasonal forecasting, which involves running usually an ensemble of climate models (the same model with different initialisation states and/or different models) to determine a probability distribution for future climate variables. Unfortunately, rainfall is very sensitive to the hydrological cycle parameterisations within dynamical models, as well as to local effects such as topography and surface processes which are not easily reproduced in the model. Therefore, statistical downscaling is often used to determine a more representative rainfall from the model. These techniques use either the model’s rainfall or, more commonly, other larger-scale variables (with higher spatial autocorrelation) such as circulation measures that are more reliably simulated by the models.

Climate change experiments, which by their nature are examining the climate many years from the present, cannot directly use current conditions to forecast. Therefore a dynamical global circulation model (GCM) is required, giving coarse resolution output on a grid of the order of several degrees. For finer scale resolution we must rely on downscaling of climate model data, achieved using statistical methods or by nesting a higher resolution regional model within the global model (dynamical downscaling).

Therefore, in statistical seasonal forecasting, as well as statistical downscaling of dynamical seasonal forecasts or climate change experiments, we need to relate rainfall to something larger scale. For downscaling we require that these variables have a higher spatial autocorrelation, for example mean sea level pressure (MSLP). For statistical seasonal forecasting we require spatial and temporal autocorrelation (e.g. SST) and use this to model rainfall. In order to find these relationships between rainfall and other variables, we therefore need to look at the past record in order to look forwards in time with any confidence.

The seven papers in this thesis are all concerned with examining historic rainfall data (or a proxy of), and determining what is happening at the regional scale. The two most important themes are, firstly, how has rainfall varied in its mean and, secondly, in its extremes. The variability of mean rainfall is examined by calculating linear trends in seasonal rainfall as well as looking at large-scale modes of variability using linear multivariate statistical techniques. The extremes have been examined using various non-parametric indices calculated seasonally or annually.

The extremes presented here could be considered “moderate” extremes, as they represent events with return periods of less than a year. The reason for this is that if we wish to examine trends and interannual variability of extremes using data of only

3 several decades, then statistical robustness limits us to this part of the rainfall distribution. Frei and Schar (2001) discuss this in detail. The important caveat is therefore that what we find for these events may not apply to higher magnitude extreme events.

The seven papers span more than ten years of research. Although they describe separate studies, there is a natural progression in the themes studied and the sophistication of the statistical tools used. In all cases, the topic of the research was provided by the desire of different government agencies to know whether we can say anything about the rainfall climatology at some time in the future. Overriding this was the desire to undertake research in areas that were outlined as deficient in knowledge by the three assessment reports of the Intergovernmental Panel on Climate Change (IPCC). The context of each paper with regards to the aims of the funding agency and IPCC will be discussed with the paper, as these are key criteria in assessing whether each study was successful in its aims.

The diverse nature of the papers is reflected in their geographical regions, which include tropical and extra tropical regions in both hemispheres. The earliest studies are concerned with smaller regions, but over time, the desire has been to determine what is happening at larger scales.

The papers add to our knowledge of changes in rainfall variability and extremes by addressing the key question:

How has mean and extreme rainfall changed in various regions in recent decades, and can we use any regional signal in these changes to determine firstly, what has caused the change and, secondly, what is likely in the future?

A small set of papers cannot hope to cover such a diverse topic for all regions with any completeness. Therefore, I address this key question by attempting to answer the following specific questions:

1. How has winter rainfall changed in southwest Western Australia and what has caused the observed change?

2. How does rainfall vary over Indonesia and can we conclude anything about predictability?

3. How does rainfall vary over the Maritime Continent and what causes this?

4 4. How has extreme rainfall changed over Australia and are the results sensitive to the choice of indices?

5. How have extreme rainfall and temperature changed over Southeast Asia?

6. How does extreme winter rainfall vary over Europe and what causes this?

7. How has extreme rainfall changed over South America and what has caused this?

These questions will be addressed in turn. In particular, for each study, I will focus on the following points:

• What was the motivation for the study?

• What was the state of knowledge before the study?

• What were the key findings?

• Was the study successful in its aims?

• How has the work progressed since the study was published?

Several other papers that I have co-authored are mentioned in the text. Although I have not included these papers as chapters, their relevance is discussed in context with the various studies. These papers are highlighted in bold and are included in the appendices.

How has winter rainfall changed in southwest Western Australia and what has caused the observed change?

Motivation and state of knowledge Since the early 1970s, there have been a number of reports detailing a strong decrease in winter rainfall over southwest Western Australia (e.g. Pittock, 1975; Pittock, 1983; Wright, 1974a; Wright, 1974b). During the early part of the 20th century, the average June-August (JJA) rainfall total was around 120 mm/month, dropping since the mid 1940s to around 100 mm/month in 1990. The rate of decline accelerated during the latter 30 years.

Although some studies had looked at circulation patterns associated with southwest Western Australia rainfall (Wright, 1974a; Wright, 1974b), no studies had examined links between the observed rainfall decrease and changes in large-scale atmospheric and oceanic circulation. Wright (1974a) and Wright (1974b) examined rainfall data from 1876-1970 and showed that there had been an increase in early winter (May-July) rainfall, due to an increase in moist inflow from lower latitudes. This coincided with a

5 decrease in late winter (August-October) rainfall, caused by a reduction in the number of frontal events from the southwest.

In the late 1980s, with the increase in public awareness of anthropogenic climate change, there was considerable concern as to whether this decrease in rainfall was part of the natural variability of the climate system, or whether it was related to anthropogenic influences. The Western Australia government were particularly concerned, as the far southwest of the state is important for agriculture as well as containing Perth, the largest city in the state. Therefore the government funded a 3-year project to assess possible changes under enhanced greenhouse conditions, and to determine the cause of the rainfall decline (Allan et al., 1992).

Key findings Chapter 2, Allan and Haylock (1993) (published in Journal of Climate) describes a key component of this project, by attempting to find large-scale circulation features associated with southwest Western Australia rainfall that could have caused the decrease in rainfall.

Allan and Haylock (1993) first created a monthly regional average rainfall series using stations back to 1870. Although there was a dry period in the late 19th century, the strong decrease for 1960-1990 appeared unusual in the context of the historical observations. They showed that Perth JJA mean sea level pressure (MSLP) back to 1870 was very strongly correlated with JJA rainfall (r=-0.80, p<0.01), and both series contained a 7-20 year spectral peak. Perth JJA MSLP also contained a strong (increasing) trend since 1960.

Gridded MSLP data were next bandpass-filtered to isolate the 7-20 year frequencies, and these were composited depending on whether bandpass-filtered rainfall showed anomalously high or low rainfall. This revealed a centre of anomalously low (high) pressure south of Western Australia during wet (dry) years. This suggests that winter cyclone tracks were forced further north (south) during wet (dry) years. Examination of the bandpass-filtered composites of MSLP showed that, during dry years, there was an intensification and westward expansion of the continental anticyclone, along with an intensification of the trough at higher latitudes. This caused stronger zonal westerly flow and confined most storms to the south of the continent. These changes in high latitude trough intensity were confirmed by examining Antarctic MSLP observations.

6 Gridded sea surface temperature (SST) observations were similarly examined using 7- 20 year bandpass-filtered data, composited by bandpass-filtered rainfall. This showed colder SSTs across the southern Indian Ocean during wet years, most likely caused by increased mixing and evaporation associated with stronger wind speeds caused by the enhanced westerly zonal flow.

Changes in atmospheric circulation over longer time scales were examined by lowpass- filtering MSLP to isolate frequencies greater than 25 years. This showed a 70 year sequence that began with anomalously low pressure over southern Australia and anomalously high over New Zealand in 1911-20 and ended with anomalies of the opposite sign during 1961-1970. These changes were superimposed on the 7-20 year oscillation. Modelling experiments using 2xCO2 equilibrium runs of slab-ocean global circulation models (GCMs) did not show such a change in MSLP, suggesting that this change may be a passing low frequency wave. Also strong negative correlations with the Southern Oscillation Index (SOI) and rainfall in 1870-1930 weakened in the latter part of the record.

Success in achieving aims and progression of work By isolating distinct multi-decadal and long wave signals in the rainfall and MSLP, and showing that correlations with the SOI had changed over time, Allan and Haylock (1993) answered many questions but posed many new problems. The study was successful in examining the drying trend in much greater detail than had been done previously, but by showing the strong links with changes to the continental anticyclone and high latitude trough, the study was not able to answer what had caused such changes to the circulation. Still it was an important step to show that regional rainfall was responding to much larger scale changes in circulation.

The study examined data up to 1990, but since this time the drying trend has continued, with 2001 being the driest winter on record for a large part of southwest Western Australia. Fig. 1 shows the total JJA rainfall at Wilgarrup, a station in the region. Therefore, the interest in the problem has only increased, so that in January 1998 a new project, the Indian Ocean Climate Initiative (IOCI – http://www.ioci.org.au) commenced. Funding of $AU1.5 million was provided to CSIRO and the Bureau of Meteorology over 5 years to promote research into the Indian Ocean, in particular to work further on the drying trend in the southwest. Part of the motivation for the initiative was the relative scarcity of research on the Indian Ocean compared to the heavily studied Pacific Ocean.

7 900

800

700

600

500

400

300

200

100

0 1910 1930 1950 1970 1990 2010

Figure 1: June-August total rainfall at Wilgarrup, southwest Western Australia, 1910-2003.

The project is now completed and the key results are available in IOCI (2001), which makes extensive reference to Allan and Haylock (1993). Key findings of the IOCI project in relation to the drying trend include:

• The reduction in rainfall is not simply due to changes in Indian Ocean SST (confirming Allan and Haylock, 1993).

• Examination of the frequency of rain-bearing synoptic patterns using a stochastic downscaling model shows that the changes around 1960 can be separated into an abrupt shift and a trend. The timing of these changes matches changes in the rainfall.

• The shift in circulation frequencies coincides with a change in the behavior of ENSO in the 1970s. The drying trend appears to be due to a mechanism different to the change in the nature of ENSO.

• The changes in the frequencies of the synoptic patterns and the reduced precipitation are due to changes in the location and intensity of low and high-pressure systems, and the moisture content of the lower troposphere.

8 IOCI has just been funded for a further period. Although explaining the rainfall decrease will still be an important aim, more effort will be placed on analysing variability and predictability of the southwest Australian climate.

How does rainfall vary over Indonesia and can we conclude anything about predictability?

Motivation and state of knowledge There has been a long history of studies into seasonal predictability of Indonesian rainfall since the early colonial work of Braak (1919), Berlage (1927) and Berlage (1934). More recent work by Nicholls (1981) and Hastenrath (1987) showed that seasonal persistence of ENSO as measured by Darwin MSLP could be used for seasonal rainfall prediction in Indonesia. However these more recent studies (see also Kirono et al., 1999) show that predictability using ENSO was limited to the dry to wet “transition season” September-November (SON). This is due to there being only small lag correlations with wet season December-February (DJF) rainfall. No study had examined whether wet season rainfall could be forecast using other predictors.

Although research into seasonal predictability of rainfall in Indonesia has a long history back to the Dutch East Indies period, routine operational forecasts are more recent. Since the mid 1990s, the organisation Badan Meteorologi dan Geofisika (BMG) has been publishing monthly seasonal outlooks for rainfall, looking up to 6 months ahead. The current subjective scheme uses the current SOI to forecast rainfall amounts for about 100 rainfall districts covering most, but not all, of the country. There have been no publications detailing the forecast methodology, nor any report of its performance skill.

BMG desired a more objective method for forecasting, similar to the SST-based probabilistic scheme operational in Australia since the mid 1990s (Drosdowsky and Chambers, 2001). The main requirements for a scheme were:

• Objective i.e. readily reproducible results

• Use of SSTs rather than SOI for greater robustness

• Probabilistic forecasts of 3-month rainfall terciles for 1-month lead times

• Assessment of performance skill

Therefore, a study was undertaken as part of a project funded by the Australian Centre for International Agricultural Research (ACIAR) to improve the seasonal forecasting

9 methodology in Indonesia. The four-year ACIAR project ran from 1999 to 2002 and was called “Capturing the benefits of seasonal climate forecasts in agricultural management”. It was aimed at linking climate scientists, agricultural scientists and farmers to determine to what extent seasonal forecasts could be effectively used by stakeholders. My role in the project, as a member of the Australian Bureau of Meteorology, was to develop an improved seasonal forecasting scheme for Indonesian rainfall.

Key findings Chapter 3, Haylock and McBride (2001) (published in Journal of Climate) used a recently compiled data set to show that DJF rainfall over Indonesia was inherently unpredictable. The new monthly rainfall data set contained 63 stations compiled by Kirono, an extension to the set used in Kirono et al. (1999).

Haylock and McBride (2001) gave an example of two neighbouring stations in Java separated by less than 100km that had a low interannual correlation of DJF rainfall of 0.18 (0.30 for non de-trended data) compared with 0.61 in SON. The conclusion was that local effects were much more important in DJF compared with SON.

The seasonal dependence of the spatial coherence was quantified across the entire region using principal component (PC) analysis. This showed that two significant PCs in DJF accounted for 22% of total rainfall variance, compared with one PC in SON accounting for 38%. The first PCs for both seasons had high correlations with the SOI of 0.76 and 0.79 for DJF and SON respectively. The difference between the two seasons was that the first PC for DJF had high loadings over only a small area in the north of the region, whereas the first PC for SON had high loadings over most of the region. The first PC loadings closely resembled the correlation between rainfall and the SOI for both seasons. The second significant PC for DJF had high loadings over a small area of central Sumatra and western Borneo.

This close relationship between correlation with the SOI and spatial coherence was further illustrated with an all-Indonesia rainfall index, calculated by measuring the rank of each season’s rainfall averaged across all stations (Fig. 2). In SON, this had a large range, showing that there were years when stations received consistently high or low rainfall. In SON, this index had a high correlation with the SOI of 0.81. In DJF, the range of the index and the correlation with the SOI were much lower.

10 Figure 2: Scatterplot of all-Indonesian rainfall index and SOI for DJF and SON. From Haylock and McBride (2001).

Haylock and McBride (2001) showed that the influence of ENSO is not totally absent in DJF, as there are stations in the north of the region with high correlations with the SOI. However, the key finding is that the low spatial coherence of DJF rainfall means that, even if ENSO is still active across the region, local effects dominate and so there can be no large-scale predictor with which to forecast rainfall.

Success in achieving aims and progression of work The key findings of Haylock and McBride (2001) were valuable in building a seasonal forecasting model for Indonesian rainfall. It meant that, apart from the two regions with high loadings for the two significant PCs (the northern region and the region covering central Sumatra and western Borneo), DJF rainfall was inherently unpredictable. Therefore, there was no point searching for possible predictors to try to explain a large proportion of variance across the entire region. Of the two regions with high DJF

11 loadings, one was very strongly related to ENSO. The study also showed the dominance of ENSO in JJA and SON. MAM had similar low spatial coherence as DJF.

My main deliverable from this project was an objective seasonal forecasting scheme. This personal computer-based scheme forecasts 3-month rainfall tercile probabilities at 1-month lag using discriminant analysis with three PCs of Indo-Pacific SST (similar to Drosdowsky and Chambers, 2001). This scheme is currently being run in parallel with the older (subjective) scheme.

Haylock and McBride (2001) raised a key question: does the lack of spatial coherence and relationship with ENSO in DJF mean that ENSO breaks down in the region in DJF, or do local effects dominate ENSO? This could not be answered in their original study, but will be addressed in the next section.

How does rainfall vary over the Maritime Continent and what causes this?

Motivation and state of knowledge The question raised by Haylock and McBride (2001) as to what extent ENSO is still active in the Indonesian region in DJF was further excited by anecdotal evidence suggesting that rainfall in the Philippines was highly correlated with the SOI in DJF. This, together with Haylock and McBride's (2001) finding that northern Indonesian rainfall was correlated with the SOI in DJF suggested that ENSO was still active in the region but had moved further north.

Key findings Chapter 4, McBride et al. (2003) (published in Journal of Climate) extended the Haylock and McBride (2001) study to cover the entire Maritime Continent. The term “Maritime Continent” was first used by Ramage (1968) to describe the domain 10°S–20°N and 90°–150°E. This unique region is made up of several large islands with complex orography and about 17,500 small islands. Containing some of the warmest ocean temperatures of the world, the rainfall in this area is dominated by convective activity. The lack of accessible, good quality rainfall data, combined with the high proportion of ocean in the region was overcome by using outgoing long wave radiation (OLR) as a proxy for rainfall (Heddinghaus and Krueger, 1981).

McBride et al. (2003) had three main conclusions. Firstly, strong resemblance between the factor loadings of the first PC of OLR and patterns of correlation between OLR and

12 SOI lent strong support to their previous hypothesis that regional predictability requires spatial coherence.

Secondly, they found that the predictable component of OLR a) remained in the Maritime Continent and b) was located in the winter hemisphere (Fig. 3). This is different to the major monsoon heat source (as measured by low OLR), which progressed annually from Indonesia in DJF to India in JJA.

Figure 3: The annual cycle of cold cloud over the maritime continent as represented by the presence of low values of OLR: the shaded area for each month being long-term mean OLR values less than 220 w m-2. Superimposed for each month is the area convection that is negatively correlated with the Southern Oscillation Index, with the hatching representing correlations of magnitude greater than 0.6. From McBride et al. (2003).

13 Finally, patterns of correlation between SST and the SOI implied a direct relationship between SST and rainfall. In the western Pacific, the “boomerang” pattern of SSTs (opposite sign of correlation to the eastern Pacific) had their highest SOI correlations in the winter hemisphere. These highly correlated SSTs were upstream of the region with “predictable” OLR. Here “predictable” referred to either as having a strong correlation with the SOI or having high spatial coherence.

Success in achieving aims and progression of work McBride et al. (2003) were successful on two accounts. Firstly, by extending the Haylock and McBride (2001) analysis to a larger region, it enabled the confirmation of the previous results as well as giving a clearer picture of why only the northern Indonesian region responded to ENSO forcing in DJF. Secondly, they found strong evidence to suggest that the seasonally varying ENSO-SST relationship in the west Pacific could explain the seasonal variation in the rainfall/OLR relationship with ENSO.

During the study by McBride et al. (2003), a similar analysis was being undertaken by Hendon (2003) using the same station rainfall data set as Haylock and McBride (2001). Hendon (2003) drew some similar conclusions to Haylock and McBride (2001) in that dry (JJA) and transition (SON) season rainfall anomalies were spatially coherent and strongly related to ENSO. However, Hendon (2003) arrived at an alternative hypothesis, suggesting that the observed SST response to ENSO in the Indonesian region was caused more by changes in ocean-atmosphere heat flux through ENSO-driven wind anomalies. He proposed a mechanism whereby during an El Niño event, anomalous easterly winds enhance the easterly monsoon circulation in JJA and so lead to cooler SSTs in the region. This in turn enhances the SST zonal gradient across the western Pacific thereby reducing rainfall and the Walker circulation. In DJF when the circulation changes to predominantly northwesterly, the anomalous easterly winds reduce the wind speed thereby reducing the cooling of the local SSTs and so reducing any ENSO response on rainfall. This is an interesting hypothesis but goes against the McBride et al. (2003) hypothesis that the seasonal SST response in the western Pacific is fundamental to ENSO and is the direct cause for the regional climate response to ENSO. The hypothesis of Hendon (2003) does not explain how ENSO SST anomalies can be so coherent in the winter hemisphere up into higher latitudes. Clearly, more work needs to be done to resolve these divergent hypotheses.

14 How has extreme rainfall changed over Australia and are results sensitive to the choice of indices?

Motivation and state of knowledge Some of the earliest climate change experiments to examine daily climate model data showed that the largest changes under enhanced greenhouse conditions would be to the frequency and magnitude of climate extremes. One of the earliest studies to identify changes to extremes was Gordon et al. (1992) who examined daily rainfall data from a slab-ocean GCM to show that a general upward shift in the rainfall histogram had a large impact on the frequency and intensity of extreme events. As the co-author on this paper that carried out most of the analyses of daily rainfall, this was my introduction to the possible change in extreme rainfall.

The importance of extremes under climate change, together with the relative paucity of studies, led the IPCC 2nd assessment (Nicholls et al., 1996) to conclude that there was a great need for further work on changes in extremes. Several studies highlighted that we needed to study the past and present as well as possible changes in the future. Karl et al. (1995) emphasised the importance of long-term climate monitoring to detect changes, through the maintenance of a consistent high quality observing network. Particular attention was given to extremes by Nicholls (1995) who discussed the use of regional climate indices for extreme events, selected on the basis of their damage potential.

Therefore several international workshops developed indices for climate extremes (Folland et al., 1999b; Nicholls and Murray, 1999). The aim was to create a set of indices that could be calculated for a variety of climates to enable inter-comparison between regions. Methodologies for calculating the indices were reported, highlighting that these indices should be designed to maximise their independence (low correlation).

During this time in Australia, the Australian State of the Environment Reporting System was established in 1996, a project aimed at monitoring the environment. The first report, released in 1996, detailed the need for a set of environmental indicators developed by experts in seven themes: Human Settlements; Biodiversity; The Atmosphere; The Land; Inland Waters; Coasts and Oceans; and Natural and Cultural Heritage. In preparation for the next report in 2001, studies using the environmental indicators on a continental scale needed to be done. A workshop was held in Melbourne in 1997 on indicators for the atmosphere, which recommended that climate extremes indicators be developed. A report was published presenting time series of various

15 indicators (Nicholls et al., 1999 - for which I calculated and analysed the rainfall indices).

There had been a few studies looking at the changes in extreme rainfall over Australia, in particular studies looking at the number of raindays and changes in percentiles and frequency of events above long-term percentiles. Hennessy et al. (1999) and Suppiah and Hennessy (1998) found a decrease in extremes in southwest Western Australia and an increase in the number of raindays over northern Australia, mainly from an increase in days with low rainfall. Both these studies used different data sets and methodologies. Hennessy et al. (1999) used 379 stations, examining total rainfall, raindays and seasonal 95th and 99th percentiles. Suppiah and Hennessy (1998) used 125 stations, examining total rainfall, raindays and 90th and 95th percentiles using 6-month seasons.

Key findings

Chapter 5, Haylock and Nicholls (2000) (published in International Journal of Climatology) was a study with two aims: to identify historical changes in extreme rainfall using an updated high quality data set; and to determine whether the changes were sensitive to the choice of extreme indices. There were problems identified with previous studies in their choice of data, in particular Hennessy et al. (1999) had used daily data from a data set that contained stations with data patched from nearby stations and had only been screened for inhomogeneities using monthly data (Lavery et al., 1997).

Lavery et al. (1992) produced a set of 191 rainfall stations that they deemed to have daily data of sufficiently high quality for climate change studies. Haylock and Nicholls (2000) updated this set to 1998 and examined the recent station history documentation to isolate stations of dubious quality. Ten stations were found that had either moved location in recent years or suffered from bad exposure and so these were rejected from the high quality daily data set.

The data contained tagged (and untagged – see below) days where the 24-hour total was an accumulation over several days. Previous studies had distributed these accumulations evenly over the preceding days. As data quality was the highest priority in this study, accumulations (and their preceding days) were set to missing. An examination of extremes is very dependent on the amount of missing data. Haylock and Nicholls (2000) therefore used a station selection process that was based on the probability of an extreme event being missing, resulting in 91 stations having sufficient data.

16 The study included methods suggested by Nicholls and Murray (1999): using raindays only to calculate percentiles and examining the fraction of total rainfall from extreme events.

The indices examined by Haylock and Nicholls were total rainfall, the number of raindays (>1mm), the intensity of extreme events, the frequency of extreme events and the proportion of the total rainfall from extreme events. Stations were grouped into four regions. Annual (May-April) results for the period 1910-1998 were presented. Haylock and Nicholls calculated and compared indices using three different methods:

i) examining the intensity and fraction of total rainfall from events above an annual percentile calculated from all days.

ii) as for i) but using just raindays to calculate the percentile threshold

iii) as for i) but using a long-term constant threshold

As some regions of Australia had experienced a significant trend in the number of raindays, the comparison of three methods was adopted to test for the sensitivity of the indices to this change.

The study confirmed the key findings of previous studies: an increase in total rainfall and raindays in the eastern part of the country and a decrease in the southwest. It also gave some examples of how the strong regionally dependent trend in raindays affects results. For example, Fig. 4 shows the extreme intensity index calculated using two different methods at a station that has experienced a significant reduction in the number of raindays. The figure compares the average intensity of the extreme events as measured by exceedences of an annual 99th percentile threshold calculated using all days (i.e. the highest 4 events) with the average intensity of the extreme events as measured by exceedences of an annual 95th percentile threshold calculated using just raindays (i.e. highest 5% of raindays). The trend is stronger when using a threshold based on raindays.

17 70 60 50 40 30

Intensity (mm) 20 10 0 1900 1920 1940 1960 1980 2000 Year

Figure 4: Average intensity of the four highest events (solid) and highest 5% of events > 1mm (dashed) for Goomalling with linear trends for 1910-1998

Generally, the sign of the intensity of extremes matched the total rainfall (and raindays). However, the effect of the changing number of raindays was evident when comparing methods, as in Fig. 4. The trend in the proportion of total rainfall from extreme events was generally opposite to the trend in total rainfall, but again this was sensitive to the methodology. This index highlighted the difference between the eastern and western regions. Only light rainfall events increased in the east leading to a reduction in the extreme proportion. In the west, extreme, as well as less intense events, decreased in frequency.

Correlations between the extreme indices and total rainfall show that for years with higher rainfall, there was rain on more days with higher average intensity in the highest events and more events above a long-term threshold. Significant correlations of opposite sign occurred between the total rainfall and the extreme proportion for two different methodologies, which highlighted the sensitivity to methodology of calculating the indices.

Success in achieving aims and progression of work Haylock and Nicholls (2000) were successful in that they confirmed previous studies using an updated and higher-quality data set. They were particularly successful in showing that, when examining extremes, the results can be very sensitive to the number of raindays, particularly when examining long-term trends when there are trends in the

18 number of raindays. Other studies have adopted a parametric approach to calculate extreme thresholds and percentiles (e.g. Groisman et al., 1999). Such studies would also be sensitive to changes in the number of raindays.

Since the completion of this study, the high-quality daily rainfall data set has seen continued development and analysis by the National Climate Centre at the Bureau of Meteorology. Although there are plans to routinely monitor and publish some of the extremes indices used in Haylock and Nicholls (2000), nothing is operational yet.

Although every attention to data quality was paid in this study, recent work by Viney and Bates (2004) has shown that there are still significant problems associated with untagged accumulations in the high-quality daily data set. They found that as many as 102 of 181 stations contain untagged accumulations. Estimates of extreme rainfall intensity (90th and 95th percentiles) for records with accumulations tended to be slightly too large for stations with high rainfall probability and too small for stations with low rainfall probability. The authors revisited past studies, including Haylock and Nicholls (2000), discussing how this problem affects the results. They concluded, “Although none of the reanalyses presented in this paper is likely to have substantially altered the broad conclusions of the respective studies, the detail at individual stations can be affected appreciably“. For climate change studies, they suggest midweek sampling (discarding Saturday-Monday), as these records are less likely to contain tagged or untagged accumulations. Although this could remove from the record real extreme events, this is outweighed by removing the uncertainty associated with possible false extremes from accumulations. This could only be done for climate indices not requiring a continuous record, such as those analysed in Haylock and Nicholls (2000).

How have extreme rainfall and temperature changed over Southeast Asia?

Motivation and state of knowledge By the late 1990s, there had still been only a few studies examining observed changes in extremes for single countries, including the USA (Karl and Knight, 1998) and Australia (Suppiah and Hennessy, 1998; Hennessy et al., 1999). There was even less published work looking at changes over larger regions. Notable exceptions were Plummer et al. (1999) who examined changes in extreme indices over Australia and New Zealand and Groisman et al. (1999), who examined extremes using gamma distribution statistical modelling for eight countries: Canada, the United States, Mexico, the former Soviet Union, China, Australia, Norway, and Poland. The study by Groisman et al. (1999) was the first attempt to extend an analysis of extremes to a large part of the globe. Work on

19 extremes was proceeding in Europe, with a meeting in Bracknell of the WMO working group devoted to climate indices (Folland et al., 1999a) and the establishment of the European Climate Assessment (Klein Tank and Wijngaard, 2000 – see http://eca.knmi.nl).

However there was a need to initiate extremes research in regions that had not received the attention of such studies. In particular, developing countries were lacking analyses due to insufficient resources to undertake them, limited access to data, fewer digitised records and reduced data quality to which extremes analyses are very sensitive. Southeast Asia and the Pacific was identified as a key region (Manton and Nicholls, 1999), in particular due to its vulnerability with regard to high population density, heightened rainfall variability due to ENSO, low-lying islands, coral reefs and exposure to tropical cyclones. Therefore, in 1998 the Asia-Pacific Network (APN) for Global Change Research funded a workshop on climate indices, hosted in Melbourne by the Australian Bureau of Meteorology and attended by 14 countries.

During this workshop, participants were given the opportunity to present results from analyses that they had undertaken. Most countries were aware of long-term trends in mean temperature and rainfall in their regions, as well as the relationship of these variables to ENSO, but very little work had been done on extremes. A main outcome was to undertake a regional analysis of extremes at a follow-up workshop.

In December 1999, the 2nd APN Workshop on Climate Extremes was hosted in Melbourne and attended by participants from 15 countries. The week long workshop enabled the participants to analyse daily rainfall and temperature data from their country, checking data for quality and calculating and analysing indices of extremes. As well as aiming to produce a consistent analysis across the region, a key aim of the workshop was to provide the tools and knowledge to enable the participants to continue similar analyses after the workshop. I acted in a supervisory role in this workshop as a member of the organising committee. As well as collaborating in the selection of the extreme indices, my responsibilities included the authoring and supervising of the running of the indices software as well as creating regional time series and maps of trends.

Key findings Chapter 6, Manton et al. (2001) (published in International Journal of Climatology) presents the results from the second workshop. Data quality checking and testing for

20 inhomogeneities occupied a large part of the workshop, during which many problems were identified. Stations were rejected due to the effects of urbanisation and site changes. Ninety-one stations, including rainfall from 13 countries and temperature from 12 countries, were identified as having data of sufficiently high quality to use in the analysis.

As well as examining total rainfall and the number of days with rainfall above 2mm, three rainfall extreme indices were examined: extreme frequency; extreme intensity; and extreme proportion, taken from Haylock and Nicholls (2000). The wide variety of climates studied meant only percentile-based thresholds could be used. The 1st and 99th percentiles were used, calculated using all days rather than just days with rain. Four temperature indices were examined, all relating to the frequency of days above/below the long-term 99th/1st percentile. These were referred to as hot days, warm nights, cool days, and cold nights. The availability of digitised data was a limiting factor on the period that could be examined. Therefore, only results for 1961-1998 were presented.

Maps of linear trends of the rainfall indices showed that annual totals generally decreased, mainly due to the dominance of El Niño events since the mid 1970s. This decrease was partly manifested by a significant reduction in the number of raindays throughout most of Southeast Asia and the western South Pacific.

Trends in extreme rainfall indices were less coherent. The extreme frequency generally declined, although some stations showed a significant increase in French Polynesia and Vietnam. In addition, there was a significant increase in extreme intensity at a couple of stations in French Polynesia and Japan. The proportion of total rainfall from extremes (extreme proportion) generally increased, with many stations showing a significant increase. This is probably due to the methodology of examining the contribution from a fixed number of extreme events when there has been a significant reduction in the number raindays, as discussed in Chapter 5.

The extreme temperature indices showed very coherent increases, associated with general warming across the region. Maps of the trends showed a high proportion of stations with significant trends associated with this warming. The coherence in trends enabled the calculation of country-average trends for the temperature extremes. There was generally a larger decrease in cold nights than cool days, except for the Philippines and Fiji. Fiji was the only country with an increase in either cold nights or cool days. There were stronger increases in the number of warm nights than hot days (with some exceptions). Neighbouring countries showed considerable consistency in sign and

21 magnitude of trends. Therefore, we created regional average indices of temperature extremes. These showed generally linear trends, apart from 1998 which was globally the warmest year on record and contained a far greater number of hot days and warm nights than any other year (Fig. 5). All regional temperature indices contained a statistically significant trend. The number of cold extremes decreased by half, while warm extremes doubled. The participants, most of whom had never published in an international journal, each wrote a paragraph discussing the observed trends for their country.

Warm Nights 18 15 12 9 6 Frequency 3 0 1960 1965 1970 1975 1980 1985 1990 1995 2000 Year

Figure 5: Time-series of the 1961-1998 all-station average of the frequency of warm nights. The thin line is a trend-line computed by least squares linear regression.

Although data quality may still pose problems with the results, the consistency across the region lends credibility. Of particular concern was some of the very large numbers of hot days or warm nights observed in 1998. However, this was seen across most of the region and was reflected in the regionally averaged indices.

Care is needed when drawing conclusions about long-term changes from this study, as the time period analysed was relatively short. For example, a comparison between these results and those of Haylock and Nicholls (2000) shows that Australian trends in total rainfall, dry days and extreme events at some stations for the period 1910-1998 tend to be of the opposite sign to 1961-1998.

Success in achieving aims and progression of work The 2nd APN Workshop and subsequent analysis was very successful on two accounts. Firstly, it provided a consistent regional analysis of extremes using the highest quality data available in a region with very little coverage in the literature. Manton et al. (2001)

22 plugged a large geographical hole and provided valuable input into the IPCC 3rd assessment report (Folland et al., 2001). Secondly, the workshop provided valuable capacity building and network building for climate scientists in developing countries.

The value of capacity building and the importance of developing a regional network were highlighted in three follow-up workshops, which took place in Melbourne from 2001-2004 and in all of which I have been heavily involved. While it would have been easy to accept the results from Manton et al. (2001) as an end to the exercise, the follow-up workshops enabled participants to present work that followed on from the 1999 workshop. In the 3rd workshop, participants showed the results of applying the Manton et al. (2001) methodologies to a larger station set, thereby verifying the earlier results. Participants were also equipped with other tools for analysing extremes developed in other projects. The 4th workshop examined the importance of metadata and provided a framework for digitisation of station metadata. Results are discussed in Page et al. (2004). The 5th workshop in March 2004 examined the relationship between the means and extremes as well as using online tools to begin examining the relationship between trends and interannual variability of extremes with large-scale forcings. Three publications on this topic are currently in preparation, all of which I am involved in as co-author (Della-Marta et al., 2004; Griffiths et al., 2004; Nicholls et al., 2004).

Following the publication of Manton et al. (2001), another near-global study of extremes was carried out by Frich et al. (2002), of which I produced rainfall results for Australia and rainfall and temperature results for Southeast Asia. The geographical coverage was greatly improved over Groisman et al. (1999), including large parts of North America, Europe, Asia, Australasia and the Pacific. Frich et al. (2002) provided a valuable contribution to the IPCC 3rd assessment report (Folland et al., 2001) with regards to the documentation of historical trends in extreme rainfall and temperature for a large part of the globe.

How has extreme rainfall changed over South America and what has caused this?

Motivation and state of knowledge Following the publication of Frich et al. (2002), significant gaps in the geographical coverage of documented trends in extremes were still apparent, especially Central and South America, Africa and central and western Asia. Therefore, the joint WMO Commission for Climatology (CCl)/CLIVAR Expert Team on Climate Change

23 Detection, Monitoring and Indices (ETCCDMI) - of which I am an external advisor - planned two workshops in Jamaica (Peterson et al., 2002) and Morocco (Easterling et al., 2003), modelled on the APN workshops. The Jamaican workshop invited scientists from the Caribbean region and the Morocco workshop covered northern Africa. The success of these two workshops was apparent in the publication of results, in particular Peterson et al. (2002). The ETCCDMI met again in Norwich, UK in November 2003 to plan further workshops. In May 2004, a workshop was held in South Africa and in August 2004 in Brazil. I was selected on the basis of my experience with the APN workshops to attend the Brazilian workshop as a member of the four-person organising committee.

Research into climate change in South America has seen many recent publications since studies in the late 1980s identified the important influence of ENSO on climate variability in the region (Aceituno, 1988; Rogers, 1988; Ropelewski and Halpert, 1987; Ropelewski and Halpert, 1989). Since then there has been attention focussed on trends in total rainfall and links with SST in the Amazon (Marengo, 2004), central South America (Liebmann et al., 2004b) and southern South America (Minetti et al., 2003; Rusticucci and Penalba, 2000). There has also been studies of particular dynamical features such as the South Atlantic Convergence Zone (Carvalho et al., 2004; Liebmann et al., 1999).

While there had been many studies of changes in total rainfall, there were markedly fewer looking at changes in extreme rainfall. Also these were more limited in geographical extent. The region of southeastern South America had received the most attention with several studies finding relationships between extreme rainfall in the region and features such as the South American low-level jet (Liebmann et al., 2004a), the South Atlantic Convergence Zone (Carvalho et al., 2002), the Madden Julian Oscillation (Carvalho et al., 2004) and ENSO (Grimm and Pscheidt, 2001). However, a regional study looking at changes in extreme rainfall over the entire region was still lacking.

Key findings Chapter 7, Haylock et al. (2004) (submitted to Journal of Climate) is a publication of the rainfall results from the ETCCDMI workshop in Brazil. Temperature results are presented in Vincent et al. (2004). With 75 references, Haylock et al. (2004) gave a comprehensive survey of past publications of rainfall variability in the region. The large number of diverse studies highlighted the importance of a cross-border regional analysis

24 to apply a consistent methodology across the region as well as to document historical changes in extremes.

Twenty-eight scientists from eight South American countries brought daily rainfall and temperature data to this week long workshop to be quality controlled and analysed for changes in 27 extreme indices. Over 70 stations were successfully analysed, most providing results from 1960 to 2000. Participants brought daily data to the workshop for calculating the indices but released only the derived annual indices. Therefore, I carried out all post-processing of the rainfall indices after the workshop without access to the daily data. Fifty-four stations were selected from the initial set of stations based on data quality and availability. They provided good coverage over a large part of the continent, except for Amazonia.

Haylock et al. (2004) presented maps of trends in 12 rainfall indices for the period 1960-2000. The maps showed that the trends in the extremes were generally the same as for total rainfall, with a change to wetter conditions in Ecuador and northern Peru and the region of southern Brazil, Paraguay, Uruguay and northern and central Argentina. A decrease was observed in southern Peru and southern Chile. Changes were generally spatially coherent with many stations showing statistically significant trends in some indices.

The regional coherence of the trends in most of the indices suggested that changes were driven by large-scale forcing, according to the hypothesis of Haylock and McBride (2001). Therefore Haylock et al. (2004) examined links between the rainfall indices and SSTs by performing a canonical correlation analysis (CCA) of each of the indices with SST. This was the same methodology as that adopted by Haylock and Goodess (2004) (Chapter 8).

The CCA revealed two coupled patterns that accounted for a large portion of the continental-scale interannual variability of extreme rainfall. Since both patterns were associated with canonical coefficients with significant trends and loading patterns that matched the observed trends in the indices, it was concluded that both coupled patterns were important for explaining the trends in the rainfall indices. The first pattern was related to ENSO, with a change to a generally more negative SOI during the period causing part of the observed trend in the indices. However, this pattern could not explain the significant decrease in many of the rainfall indices in Chile. This was found to be associated with a second pattern with a complex SST signal. However, compositing the mean sea level pressure (MSLP) according to the sign of the canonical

25 coefficient for the SST pattern revealed that this CCA pattern was related to a weakening of the continental trough (Fig. 7), thus confining the rain bearing fronts in the Southern Ocean to pass south of the continent. This regional pattern is consistent with a hemispheric-scale pattern of increases in pressure in middle latitudes and decreases at high latitudes associated with a trend in the Southern Annular Mode (Thompson et al., 2000). There is a striking similarity between the changes in MSLP found here and those found over Australia related to the strong decrease in winter rainfall in southwest Western Australia (Chapter 2). Both these coupled patterns were seen in total annual rainfall as well as many of the extreme rainfall indices. c

Figure 7: Annual average MSLP expressed as a difference between years when the 2nd SST-total precipitation canonical coefficient is positive and when it is negative. Shading shows region where difference is significant at p<0.05.

Success in achieving aims and progression of work The aims of the Brazilian workshop were threefold: to publish observed changes in extremes using a consistent methodology across the region; to establish a network of scientists with the hope of building on this in the future; and to provide capacity- building to countries with less resources for climate change research.

Haylock et al. (2004) has been submitted to Journal of Climate in time to be included in the IPCC 4th Assessment Report. The study addressed the first workshop aim, but went

26 further than just documenting trends in extreme rainfall. By finding two large-scale SST patterns associated with the observed rainfall trends, it tied in with the large body of work that has examined whether changes in these patterns are part of anthropogenic climate change and therefore whether they will continue to exhibit such trends in the future. With regards ENSO, a change to a generally more positive SOI in recent years has reduced the trend towards more frequent El Niños during the latter decades of the twentieth century, suggesting that this will reduce the trend in total and extreme rainfall observed in Haylock et al. (2004). However, regarding the changes in the Southern Annular Mode, several studies have associated this with decreases in stratospheric ozone (Gillett and Thompson, 2003; Sexton, 2001; Thompson and Solomon, 2002) or as a response to increasing greenhouse gases (Cai et al., 2003; Fyfe, 2003; Kushner et al., 2001).

The success of the Brazilian workshop with regards the other two aims will only be seen with time. Experiences with the APN network (Chapter 6), suggest that follow-up workshops are really needed to strengthen a scientific network. Hopefully funding will be available for such meetings. The APN network also showed the benefits of capacity building. Improvements to climate research methodologies in the participating countries were seen almost immediately after the first workshop, through improved knowledge and a greater access to analysis tools. However, funding within the countries, where climate change research usually plays a secondary role to forecasting, also plays a crucial role in determining how much time can be devoted to such studies.

Since the Brazilian workshop, two other workshops have been held: in Guatemala (for countries in Central America which I attended as an advisor); and Turkey (for countries in western Asia). A final workshop will be held in early 2005 in India for countries in central and southern Asia. It is hoped that published results from all these workshops will provide important input into the 4th IPCC assessment report.

How does extreme winter rainfall vary over Europe and what causes this?

Motivation and state of knowledge Climate modellers will readily admit that one of the greatest areas of model uncertainty is the parameterisation of the hydrological cycle. In a comparison of the performance of GCMs over Australia (which I co-authored), Whetton et al. (1994) found rainfall to be the least well simulated variable analysed and contained the greatest uncertainty in climate change prediction. The problem is accentuated for extreme rainfall. However, it

27 is usually through GCMs that we can obtain possible scenarios of future climate. Therefore, it is recognised that statistical or dynamical downscaling is needed when deriving local-scale rainfall data from a GCM.

There are many statistical downscaling methods, ranging from simple linear rescaling of the GCM rainfall, to non-linear neural network models working with dozens of input variables. Determining the best methodology for rainfall is a huge task. The European Commission-funded project Statistical and Regional Dynamical Downscaling of Extremes for European Regions (STARDEX) aims to develop improved methodologies for downscaling extreme rainfall and temperature from climate models.

STARDEX has approached the measurement of extreme rainfall using climate indices. Software was developed (which I co-authored and maintain) to calculate 33 indices of extreme rainfall. The software is distributed from the STARDEX web site (http://www.cru.uea.ac.uk/cru/projects/stardex/) and has widespread use in other parts of the globe, including Australia, Southeast Asia, Canada and Peru. Indices are calculated seasonally and include the indices examined by Haylock and Nicholls (2000) and Manton et al. (2001) as a subset.

Members of STARDEX are approaching the task of downscaling extreme rainfall from two angles: downscaling the daily rainfall using daily circulation and calculating the seasonal indices; and downscaling the seasonal indices directly using seasonal data.

There have been few studies of European indices of extreme rainfall and temperature. The most comprehensive is Klein Tank and Konnen (2003), who found a general increase in annual extreme rainfall for 1946-1999. No studies have looked at the variability of European extremes or their possible causes.

In contrast, there has been much work on variability of mean European rainfall and links with circulation. In particularly the NAO has received much attention with many studies finding it to be one of the major influences on European climate (e.g. Qian et al., 2000; Rogers, 1997; Trigo et al., 2002).

Key findings Chapter 8, Haylock and Goodess (2004) (published in International Journal of Climatology) was a study aimed at determining if there is sufficient regional variability in indices of extreme winter rainfall to a) determine the possible cause of observed trends; and b) find large-scale patterns associated with interannual variability of the indices.

28 Daily rainfall at 491 European stations was available. To maintain an even spatial distribution, this was thinned to 347 stations, covering a large area from the UK to Eastern Europe and the Mediterranean to Scandinavia.

Two indices were examined in detail: R90N, the number of days with rain above the 1961-1990 90th percentile calculated from wet days; and CDD, the maximum number of consecutive dry days. Linear trends for the period 1958-2000 for these indices showed a change to wetter conditions in the north (increase in R90N, decrease in CDD) and the opposite in the south. The high spatial coherence implied that the trends were associated with a large-scale change in the circulation. Although, Klein Tank and Konnen (2003) found a general increase in annual extreme rainfall, we showed that the trends in winter rainfall had a strong north-south divide.

Haylock and Goodess (2004) next carried out a Principal Component (PC) analysis of the indices. They rotated six PCs (determined objectively using a Monte Carlo method), which accounted for 52% of the total variance of CDD and 39% for R90N. These moderately high measures of variance indicated the importance of regional variability for this large and climatically heterogeneous area.

Only one PC had a significant trend, PC2 of R90N. Three PCs of CDD and three PCs of R90N had significant correlations with the NAO, suggesting that, though the NAO is an important predictor, analysis of the rainfall alone did not isolate the NAO as a separate mode. However, R90N PC2 had a much higher correlation (0.65) with the NAO than the other two PCs (0.38 and 0.46), which together with its significant trend implied that changes in the NAO had caused the observed regional trend in R90N. This is also supported by the fact that the NAO had a significant trend over this period and the factor loadings of R90N PC2 strongly resembled the linear trends. Fig. 7 shows the similarity in trend and variability of the DJF NAO index compared with PC2 of R90N.

29 NAO -R90NPC2 4 3 2 1 0 1955 1965 1975 1985 1995 -1 -2 -3

Figure 7: 2nd principal component of the number of very wet days (R90N) with the DJF NAO index. The sign of R90N PC2 has been reversed.

Potential large-scale predictors were sought by correlating the indices with surface and upper air variables. A methodology was developed to isolate large-scale relationships, revealing MSLP to be the best predictor.

Haylock and Goodess (2004) then carried out a canonical correlation analysis (CCA) between the indices and MSLP, to isolate statistically coupled modes. The pair of canonical coefficients with the highest canonical correlation explained the highest proportion of variance in MSLP (but not the index) and had the highest correlation with the NAO. Also, the first R90N and first two CDD coupled patterns reflected the NAO.

Two other coupled patterns were identified: a MSLP centre positioned over the North Sea which caused variability of CDD and R90N in the UK and central Europe; and a dipole with poles over the eastern Mediterranean and central North Atlantic leading to a southwest-northeast gradient in the variability of the indices.

Finally, the analyses for these two indices were confirmed by using two others: the 90th percentile rainfall intensity and the maximum 10-day rainfall.

Success in achieving aims and progression of work The important result in Haylock and Goodess (2004) was that there was enough regional-scale variability in some of the extreme indices in winter to perform a statistically meaningful regional study of interannual variability. Also, by isolating the

30 pattern of variability of R90N that had caused the trend as a single mode, we were able to determine the probable cause for the regional trend. We therefore showed that the NAO was important for regional variability in extreme rainfall, and had probably caused a large part of the observed trend.

This work has now been incorporated into a statistical downscaling model for extreme indices, based on CCA of the indices with MSLP and other predictors. Initial comparisons with other downscaling models over the UK show it to perform almost as well in DJF as much more sophisticated non-linear neural network downscaling models that work with daily rainfall. However, the performance of the CCA seasonal model is not as good as the non-linear models in JJA. The model will be applied to GCM data from climate change experiments, to determine possible station-scale changes under various emission scenarios.

Conclusions

I conclude by returning to the key question:

The seven papers presented here form a diverse collection of rainfall studies from four different continents, however, there are general conclusions that one can draw from them. Their geographic diversity is also their strength when it comes to trying to form general conclusions about climate change. Whereas most large-scale historical temperature studies show a general increase in mean and extreme temperatures, the seven rainfall studies presented here show that such general conclusions can not be drawn about rainfall. Although this goes against expectations that a warmer atmosphere means an enhanced hydrological cycle, the lack of coherent change shows that other factors are more important with regards rainfall variability.

The seven papers have contributed to a large extent in answering the first part of the key question. Together they give a detailed analysis of trends in mean and extreme rainfall in Australia, southeast Asia and the Pacific, South America and Europe. In all these regions there was no consistent large-scale increase in mean or extreme rainfall. However large-scale patterns were still apparent and we were able to use these to seek large-scale changes in the atmospheric and oceanic forcing.

31 All the regions studied, except Europe, are situated in the Pacific basin. Therefore, not surprisingly, ENSO dominated all large-scale variability in these regions. However this is a key result, that extreme rainfall exhibits large-scale spatial coherence influenced by large-scale climate signals. In the case of Europe we found that the NAO explained a large portion of the regional-scale variability in extreme winter rainfall.

The final part of the key question is also the most important. As stated in the introduction, a primary motive for studying the climate is to be able to predict what is likely in the future. By finding that the variability of mean and extreme rainfall is dominated more by large-scale signals such as ENSO than by changes to a warmer atmosphere, we have shifted the question to: what is the likely change to such climate signals? This is probably best answered by the climate models and I put more faith in these models being able to answer this question than being able to model changes in rainfall. To answer this, though, would be a thesis in itself, but currently expected changes in ENSO, the NAO, the Southern Annular Mode and other such large-scale signals identified in the papers discussed here are still uncertain in the models.

References

Aceituno P. 1988. On the Functioning of the Southern Oscillation in the South- American Sector .1. Surface Climate. Monthly Weather Review, 116, 505-524.

Allan RJ and Haylock MR. 1993. Circulation Features Associated with the Winter Rainfall Decrease in Southwestern Australia. Journal of Climate, 6, 1356-1367.

Allan RJ, Mitchell CD and Pittock AB. 1992. The greenhouse effect - Regional implications for Western Australia, annual report 1990-91., CSIRO Division of Atmospheric Research, 72 pp.

Berlage HP. 1927. East-Monsoon forecasting in Java. Koninklijk Magnetisch en Meteorologisch Observatorium te Batavia, Verhandelingen, 42.

Berlage HP. 1934. Further research into the possibility of long range forecasting in Netherlands–India. Koninklijk Magnetisch en Meteorologisch Observatorium te Batavia, Verhandelingen, 31.

Braak C. 1919. Atmospheric variations of short and long duration in the Malay Archipelago. Koninklijk Magnetisch en Meteorologisch Observatorium te Batavia, Verhandelingen, 57.

32 Cai WJ, Whetton PH and Karoly DJ. 2003. The response of the Antarctic Oscillation to increasing and stabilized atmospheric CO2. Journal of Climate, 16, 1525-1538.

Carvalho LMV, Jones C and Liebmann B. 2002. Extreme precipitation events in southeastern South America and large-scale convective patterns in the South Atlantic convergence zone. Journal of Climate, 15, 2377-2394.

Carvalho LMV, Jones C and Liebmann B. 2004. The South Atlantic convergence zone: Intensity, form, persistence, and relationships with intraseasonal to interannual activity and extreme rainfall. Journal of Climate, 17, 88-108.

Della-Marta P, Zhai P, Haylock MR, Nicholls N, Manton MJ, Baek HJ, Chambers LE, Gosai A, Choi Y, Collins D, Griffiths GM, Iga N, Lata R, Laurent V, Maitrepierre L, Nakamigawa H, Ouprasitwong N, Solofa D, Tahani L, Thuy DT, Tibig L, Trewin B and Vediapan K. 2004. The relationship between ENSO and precipitation extremes in the Asia-Pacific region. In preparation.

Drosdowsky W and Chambers LE. 2001. Near-global sea surface temperature anomalies as predictors of Australian seasonal rainfall. Journal of Climate, 14, 1677- 1687.

Easterling DR, Alexander L, Mokssit A and Detemmerman V. 2003. CCl/CLIVAR workshop to develop priority climate indices. Bulletin of the American Meteorological Society, 84, 1403-1407.

Folland CK, Horton EB and Scholefield P. 1999a. Report of WMO Working Group on Climate Change. Detection Task Group on Climate Change Indices, Bracknell, 1 September 1998, WMO TD 930.

Folland CK, Karl TR, Christy JR, Clarke RA, Gruza GV, Jouzel J, Mann ME, Oerlemans J, Salinger MJ and Wang SW. 2001. Observed Climate Variability and Change. Climate Change 2001: The Scientific Basis. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change, Houghton JT, Ding Y, Griggs DJ, Noguer M, van der Linden PJ, Dai X, Maskell K and Johnson CA, Eds., Cambridge University Press, 99-181.

Folland CK, Miller C, Bader D, Crowe M, Jones P, Plummer N, Richman M, Parker DE, Rogers J and Scholefield P. 1999b. Workshop on Indices and Indicators for Climate Extremes, Asheville, NC, USA, 3-6 June 1997 - Breakout Group C: Temperature indices for climate extremes. Climatic Change, 42, 31-43.

33 Frei C and Schar C. 2001. Detection probability of trends in rare events: Theory and application to heavy precipitation in the Alpine region. Journal of Climate, 14, 1568- 1584.

Frich P, Alexander LV, Della-Marta P, Gleason B, Haylock M, Klein Tank AMG and Peterson T. 2002. Observed coherent changes in climatic extremes during the second half of the twentieth century. Climate Research, 19, 193-212.

Fyfe JC. 2003. Extratropical southern hemisphere cyclones: Harbingers of climate change? Journal of Climate, 16, 2802-2805.

Gillett NP and Thompson DWJ. 2003. Simulation of recent Southern Hemisphere climate change. Science, 302, 273-275.

Gordon HB, Whetton PH, Pittock AB, Fowler AM and Haylock MR. 1992. Simulated Changes in Daily Rainfall Intensity Due to the Enhanced Greenhouse-Effect - Implications for Extreme Rainfall Events. Climate Dynamics, 8, 83-102.

Griffiths GM, Chambers LE, Haylock MR, Manton MJ, Nicholls N, Baek HJ, Gosai A, Choi Y, Collins D, Della-Marta P, Iga N, Lata R, Laurent V, Maitrepierre L, Nakamigawa H, Ouprasitwong N, Solofa D, Tahani L, Thuy DT, Tibig L, Trewin B, Vediapan K and Zhai PM. 2004. Change in mean temperature as a predictor of extreme temperature change in the Asia-Pacific region. International Journal of Climatology, Submitted.

Grimm AG and Pscheidt I. 2001. Atmospheric patterns associated with extreme rainfall events in the spring during El Niño, La Niña and neutral years in southern Brazil. Proceedings of the 9th Congress of the Latin-American and Iberian Federation of Meteorological Societies and 8th Argentinean Congress of Meteorology (on CD, in Portuguese), Buenos Aires, Centro Argentino de Meteorólogos e Federação Latino- Americana e Ibérica de Sociedades de Meteorologia.

Groisman PY, Karl TR, Easterling DR, Knight RW, Jamason PF, Hennessy KJ, Suppiah R, Page CM, Wibig J, Fortuniak K, Razuvaev VN, Douglas A, Forland E and Zhai PM. 1999. Changes in the probability of heavy precipitation: Important indicators of climatic change. Climatic Change, 42, 243-283.

Hastenrath S. 1987. Predictability of Java Monsoon Rainfall Anomalies - a Case-Study. Journal of Climate and Applied Meteorology, 26, 133-141.

34 Haylock M and Goodess C. 2004. Interannual variability of European extreme winter rainfall and links with mean large-scale circulation. International Journal of Climatology, 24, 759-776.

Haylock M and McBride J. 2001. Spatial coherence and predictability of Indonesian wet season rainfall. Journal of Climate, 14, 3882-3887.

Haylock M and Nicholls N. 2000. Trends in extreme rainfall indices for an updated high quality data set for Australia, 1910-1998. International Journal of Climatology, 20, 1533-1541.

Heddinghaus TR and Krueger AF. 1981. Annual and Interannual Variations in Outgoing Longwave Radiation over the Tropics. Monthly Weather Review, 109, 1208- 1218.

Hendon HH. 2003. Indonesian rainfall variability: Impacts of ENSO and local air-sea interaction. Journal of Climate, 16, 1775-1790.

Hennessy KJ, Suppiah R and Page CM. 1999. Australian rainfall changes, 1910-1995. Australian Meteorological Magazine, 48, 1-13.

IOCI. 2001. Indian Ocean Climate Iniative Second Research Report - Towards Understanding Climate Variability in south western Australia, Dept of Environment, Water and Catchment Protection, 193 pp.

Karl TR, Derr VE, Easterling DR, Folland CK, Hofmann DJ, Levitus S, Nicholls N, Parker DE and Withee GW. 1995. Critical issues for long-term climate monitoring. Climatic Change, 31, 185-221.

Karl TR and Knight RW. 1998. Secular trends of precipitation amount, frequency, and intensity in the United States. Bulletin of the American Meteorological Society, 79, 231- 241.

35 Kirono DGC, Tapper NJ and McBride JL. 1999. Documenting Indonesian rainfall in the 1997/1998 El Nino event. Physical Geography, 20, 422-435.

Klein Tank AMG and Wijngaard JB. 2000. European climate assessment. 3rd European Conference on Applied Climatology, Pisa, Italy.

Klein Tank AMGK and Konnen GP. 2003. Trends in indices of daily temperature and precipitation extremes in Europe, 1946-99. Journal of Climate, 16, 3665-3680.

Kushner PJ, Held IM and Delworth TL. 2001. Southern Hemisphere atmospheric circulation response to global warming. Journal of Climate, 14, 2238-2249.

Lavery B, Joung G and Nicholls N. 1997. An extended high-quality historical rainfall dataset for Australia. Australian Meteorological Magazine, 46, 27-38.

Lavery B, Nicholls N and Kariko A. 1992. A historical rainfall data set for Australia. Australian Meteorology Magazine, 40, 33-39.

Liebmann B, Kiladis GN, Marengo JA, Ambrizzi T and Glick JD. 1999. Submonthly convective variability over South America and the South Atlantic convergence zone. Journal of Climate, 12, 1877-1891.

Liebmann B, Kiladis GN, Vera CS, Saulo AC and Carvalho LMV. 2004a. Subseasonal variations of rainfall in South America in the vicinity of the low-level jet east of the Andes and comparison to those in the South Atlantic convergence zone. Journal of Climate, 17, 3829-3842.

Liebmann B, Vera CS, Carvalho LMV, Camilloni I, Hoerling MP, Barros VR, Báez J and Bidegain M. 2004b. An observed trend in central South American Precipitation. Journal of Climate, 17, 4357-4367.

Manton MJ, Della-Marta PM, Haylock MR, Hennessy KJ, Nicholls N, Chambers LE, Collins DA, Daw G, Finet A, Gunawan D, Inape K, Isobe H, Kestin TS, Lefale P, Leyu CH, Lwin T, Maitrepierre L, Ouprasitwong N, Page CM, Pahalad J, Plummer N, Salinger MJ, Suppiah R, Tran VL, Trewin B, Tibig I and Yee D. 2001. Trends in extreme daily rainfall and temperature in Southeast Asia and the South Pacific: 1961- 1998. International Journal of Climatology, 21, 269-284.

Manton MJ and Nicholls N. 1999. Monitoring trends in extreme climate events. APN Newsletter, 1-3.

Marengo JA. 2004. Interdecadal variability and trends of rainfall across the Amazon basin. Theoretical and Applied Climatology, 78, 79-96.

36 McBride JL, Haylock MR and Nichols N. 2003. Relationships between the maritime continent heat source and the El Nino-Southern Oscillation phenomenon. Journal of Climate, 16, 2905-2914.

Minetti JL, Vargas WM, Poblete AG, Acuna LR and Casagrande G. 2003. Non-linear trends and low frequency oscillations in annual precipitation over Argentina and Chile, 1931-1999. Atmosfera, 16, 119-135.

Munich Re. 2003. Topics. Annual Review: Natural Catastrophes 2002, Munich Re, 48 pp.

Nairn G. 2003. A Nation Charred: Report on the inquiry into the recent Australian bushfires, House of Representatives Select Committee into the recent Australian bushfires, 500 pp.

Nicholls N. 1981. Air-Sea Interaction and the Possibility of Long-Range Weather Prediction in the Indonesian Archipelago. Monthly Weather Review, 109, 2435-2443.

Nicholls N. 1995. Long-term climate monitoring and extreme events. Climatic Change, 31, 231-245.

Nicholls N, Baek HJ, Gosai A, Chambers LE, Choi Y, Collins D, Della-Marta P, Griffiths GM, Haylock MR, Iga N, Lata R, Laurent V, Maitrepierre L, Manton MJ, Nakamigawa H, Ouprasitwong N, Solofa D, Tahani L, Thuy DT, Tibig L, Trewin B, Vediapan K and Zhai PM. 2004. The El Niño – Southern Oscillation and daily temperature extremes in east Asia and the west Pacific. Geophysical Research Letters, Submitted.

Nicholls N, Gruza GV, Jonzel J, Karl TR, Ogallo LA and Parker DE. 1996. Observed climate variability and change. Climate Change 1995, Houghton JT, Meira Filho LG, Callander BA, Harris N, Kattenberg A and Maskell K, Eds., Cambridge University Press, 133-192.

Nicholls N and Murray W. 1999. Workshop on Indices and Indicators for Climate Extremes, Asheville, NC, USA, 3-6 June 1997 - Breakout Group B: Precipitation. Climatic Change, 42, 23-29.

Nicholls N, Trewin B and Haylock M. 1999. Climate extremes indicators for State of the Environment monitoring, Australia. State of the Environment Technical Paper Series (Environmental Indicators), Department of the Environment.

37 Page CM, Nicholls N, Plummer N, Trewin B, Manton M, Alexander L, Chambers LE, Choi Y, Collins DA, Gosai A, Della-Marta P, Haylock MR, Inape K, Laurent V, Maitrepierre L, Makmur EEP, Nakamigawa H, Ouprasitwong N, McGree S, Pahailad J, Salinger MJ, Tibig L, Tran TD, Vediapan K and Zhai P. 2004. Data rescue in the Southeast Asia and South Pacific region - Challenges and opportunities. Bulletin of the American Meteorological Society, 85, 1483-1489.

Peterson TC, Taylor MA, Demeritte R, Duncombe DL, Burton S, Thompson F, Porter A, Mercedes M, Villegas E, Fils RS, Tank AK, Martis A, Warner R, Joyette A, Mills W, Alexander L and Gleason B. 2002. Recent changes in climate extremes in the Caribbean region. Journal of Geophysical Research-Atmospheres, 107, art. no.-4601.

Pittock AB. 1975. Climatic Change and Patterns of Variation in Australian Rainfall. Search, 6, 498-504.

Pittock AB. 1983. Recent Climatic-Change in Australia - Implications for a Co2- Warmed Earth. Climatic Change, 5, 321-340.

Plummer N, Salinger MJ, Nicholls N, Suppiah R, Hennessy KJ, Leighton RM, Trewin B, Page CM and Lough JM. 1999. Changes in climate extremes over the Australian region and New Zealand during the twentieth century. Climatic Change, 42, 183-202.

Qian BD, Corte-Real J and Xu H. 2000. Is the North Atlantic Oscillation the most important atmospheric pattern for precipitation in Europe? Journal of Geophysical Research-Atmospheres, 105, 11901-11910.

Ramage CS. 1968. Role of a Tropical Maritime Continent in Atmospheric Circulation. Monthly Weather Review, 96, 365-&.

Rogers JC. 1988. Precipitation Variability over the Caribbean and Tropical Americas Associated with the Southern Oscillation. Journal of Climate, 1, 172-182.

Rogers JC. 1997. North Atlantic storm track variability and its association to the north Atlantic oscillation and climate variability of northern Europe. Journal of Climate, 10, 1635-1647.

Ropelewski CF and Halpert MS. 1987. Global and Regional Scale Precipitation Patterns Associated with the El-Nino Southern Oscillation. Monthly Weather Review, 115, 1606- 1626.

Ropelewski CF and Halpert MS. 1989. Precipitation Patterns Associated with the High Index Phase of the Southern Oscillation. Journal of Climate, 2, 268-284.

38 Rusticucci M and Penalba O. 2000. Interdecadal changes in the precipitation seasonal cycle over Southern South America and their relationship with surface temperature. Climate Research, 16, 1-15.

Sexton DMH. 2001. The effect of stratospheric ozone depletion on the phase of the Antarctic Oscillation. Geophysical Research Letters, 28, 3697-3700.

Suppiah R and Hennessy KJ. 1998. Trends in total rainfall, heavy rain events and number of dry days in Australia, 1910-1990. International Journal of Climatology, 18, 1141-1164.

Thompson DWJ and Solomon S. 2002. Interpretation of recent Southern Hemisphere climate change. Science, 296, 895-899.

Thompson DWJ, Wallace JM and Hegerl GC. 2000. Annular modes in the extratropical circulation. Part II: Trends. Journal of Climate, 13, 1018-1036.

Trigo RM, Osborn TJ and Corte-Real JM. 2002. The North Atlantic Oscillation influence on Europe: climate impacts and associated physical mechanisms. Climate Research, 20, 9-17.

Vincent LA, Peterson TC, Barros VR, Marino MB, Rusticucci M, Miranda GC, Ramirez E, Alves LM, Ambrizzi T, Barbosa de Brito JI, Berlato MA, Grimm AM, Jaildo dos Anjos R, Marengo JA, Meira PR, Molion L, Muncunil DF, Nechet D, Rebello E, Abreu de Sousa JR, Anunciação YMT, Quintana J, Santos JL, Ontaneda G, Baez J, Coronel G, Garcia VJ, Trebejo I, Bidegain M, Corradi V, Haylock MR and Karoly DJ. 2004. Observed trends in indices of daily temperature extremes in South America 1960-2000. Journal of Climate, In preparation.

Viney NR and Bates BC. 2004. It never rains on Sunday: the prevalence and implications of untagged multi-day rainfall accumulations in the Australian high quality data set. International Journal of Climatology, 24, 1171-1192.

Whetton PH, Rayner PJ, Pittock AB and Haylock MR. 1994. An Assessment of Possible Climate-Change in the Australian Region-Based on an Intercomparison of General-Circulation Modeling Results. Journal of Climate, 7, 441-463.

Wright PB. 1974a. Seasonal Rainfall in Southwestern Australia and General Circulation. Monthly Weather Review, 102, 219-232.

Wright PB. 1974b. Temporal Variations in Seasonal Rainfalls in Southwestern- Australia. Monthly Weather Review, 102, 233-243.

39 Chapter 2: Allan and Haylock, 1993

Allan RJ and Haylock MR. 1993. Circulation Features Associated with the Winter Rainfall Decrease in Southwestern Australia. Journal of Climate, 6, 1356-1367.

Chapter 3: Haylock and McBride, 2001

Haylock M and McBride J. 2001. Spatial coherence and predictability of Indonesian wet season rainfall. Journal of Climate, 14, 3882-3887. 3882 JOURNAL OF CLIMATE VOLUME 14

NOTES AND CORRESPONDENCE

Spatial Coherence and Predictability of Indonesian Wet Season Rainfall

MALCOLM HAYLOCK AND JOHN MCBRIDE Bureau of Meteorology Research Centre, Melbourne, Australia

31 August 2000 and 28 March 2001

ABSTRACT Rainfall from 63 stations across Indonesia is examined for the period 1950±98 to determine the spatial coherence of wet season anomalies. An example of almost unrelated anomalies at two neighboring stations is presented. Principal component analysis is used to quantify the spatial coherence across the entire region. The signi®cant components show high loadings over only a small region, suggesting that rainfall in only this small region varies coherently on an interannual timescale. Correlation with the Southern Oscillation index (SOI) shows that rainfall over only this same region is being largely governed by the El NinÄo±Southern Oscillation (ENSO) phenomenon. In contrast, a similar analysis for the transition season (Sep±Nov) rainfall shows coherence across almost the entire region and a similarly large area of high correlation with the SOI. Results for all seasons are summarized with the use of an all-Indonesian rainfall index constructed from an averaged percentile ranking of seasonal rainfall from each station across the region. At the times of the year when a large (small) percentage of the variance of rainfall is described by the lowest-order principal components, there is a large (small) correlation between the SOI and the all-Indonesian rainfall index. The implication is that wet season rainfall in Indonesia is inherently unpredictable.

1. Introduction the Maritime Continent region of Indonesia. As discussed by Nicholls (1981), Hastenrath (1987), and There have been several recent studies examining McBride (1999, Chapter 3A), seasonal forecasting in connections between observed rainfall and a number of Indonesia has a long history, extending back to Braak large-scale climate signals (e.g., Wang et al. 2000; Dros- (1919). The basis of forecasting in the region has been dowsky 1993; Montecinos et al. 2000; Makarau and the persistence of the Southern Oscillation, one simple Jury 1997; Nicholls 1981). The aim of these studies has generally been to develop, where possible, methods for index of which is the pressure at Darwin, Australia long-range forecasting. Studies have looked for predic- (Nicholls 1981; McBride and Nicholls 1983). Beginning tors based on correlation with raw station data (e.g., with the early colonial work of Braak (1919) and Ber- Montecinos et al. 2000; Nicholls 1981), area averages lage (1927, 1934) and continuing through to the major (e.g., Makarau and Jury 1997), or principal components works of the modern era (Nicholls 1981; Hastenrath of station rainfall (e.g., Drosdowsky 1993). 1987), exploration into seasonal forecasting was based These forecasting methods generally rely on search- on the simple concept of lag relationships between Dar- ing for only a few large-scale predictors to account for win pressure and seasonal rainfall. changes in rainfall across large areas. It is our hypothesis The meteorology of Indonesia has been described by that inherent in such schemes is the assumption that a Sukanto (1969), Hackert and Hastenrath (1986), and large proportion of the variation of rainfall is spatially McBride (1999, Chapter 3A). Straddling the equator, coherent. It is not possible for two (or more) stations much of the region has rain throughout the year. How- to have uncorrelated interannual rainfall for a particular ever, for most of the region the major wet season co- season, yet share the same predictors. incides with the Southern Hemisphere summer monsoon This paper examines the predictability of seasonal (McBride 1983; Murakami and Sumi 1982), with the rainfall during the wet season (or ``Musim Hujan'') over peak rainfall occurring during the southern summer months of December±February (DJF). Despite this, seasonal forecasting in the region has focused on the early Corresponding author address: Mr. Malcolm Haylock, Bureau of wet season, or ``transition season'' months of Septem- Meteorology Research Centre, GPO Box 1289K, Melbourne, Victoria 3001, Australia. ber±November (SON); see, for example, Nicholls E-mail: [email protected] (1981) and Hastenrath (1987). This focus is due simply

᭧ 2001 American Meteorological Society 15 SEPTEMBER 2001 NOTES AND CORRESPONDENCE 3883

FIG. 1. Location of 63 stations. Three stations mentioned in the text are shown. Shading indicates orography.

FIG. 2. DJF rainfall at Bandung and Jatiwangi. to the pragmatic reason that the lag correlations between indices of the Southern Oscillation and rainfall are large data were not included. All stations selected have less for SON rainfall and small for DJF rainfall (Hastenrath than 10% of months missing for their period of record 1987; Kirono et al. 1999). and no stations have more than 1 yr when the entire The current paper examines the reasons for this year is missing. change of predictability following the framework es- Next, stations were subjected to a statistical test of poused in the above opening paragraphs. Speci®cally, Buishand (1982) to check for arti®cial jumps, outliers, the relationship between observed rainfall and a large- and trends in the monthly series. This method examines scale climate signal (viz., the Southern Oscillation) de- the cumulative deviation of the series from the mean, pends on anomalies in the rainfall having a large-scale tested against critical values derived from randomly coherence. As will be shown, rainfall anomalies across generated samples. The test assumes a normally dis- Indonesia during the peak of the wet season (DJF) have tributed series, a condition that was examined using a low coherence, and therefore are hypothesized to be Kolmogorov±Smirnov test on the monthly and seasonal inherently unpredictable. series. Stations failing the Buishand test were not in- The following section describes the dataset used. In cluded in the ®nal set. More common statistical meth- section 3 coherence of rainfall anomalies in the two ods, such as those discussed in Peterson et al. (1998), seasons of SON and DJF is examined through both cor- were not applied due to the lack of a good station net- relation analysis and through the use of principal com- work with overlapping data in neighboring stations. ponent analysis. The consistency between anomaly co- Finally, the series of the monthly anomalies of each herence and association with the Southern Oscillation station was visually examined for obvious inhomoge- is demonstrated in section 4. Section 5 discusses the neities that were not indicated by the statistical tests. implications of the results and speculates on the rela- The ®nal set of 63 stations (Fig. 1) provides even tionship between this de®nition of predictability and spatial coverage over the Indonesian region. While some other frameworks that appear in the literature. The ®nal of the stations only have data since 1960, most have concluding section summarizes the results. almost complete records from 1950 to 1998.

2. Data 3. Coherence of rainfall anomalies Figure 1 shows the location of 63 rainfall stations for a. Local coherence which monthly rainfall were compiled by D. G. C. Ki- rono (2000, personal communication). This set is an As an example of low spatial coherence of DJF rain- extension of the 33 stations used by Kirono et al. (1999). fall, consider the two stations Bandung and Jatiwangi, Several sources were used: data up to 1988 from the the locations of which are shown in Fig. 1. A plot of Badan Meteorologi dan Geo®sika (BMG) publication the DJF rainfall for these two stations (Fig. 2) illustrates ``Rain Observations in Indonesia''; data to 1991 from the low correlation between the stations. Much of the the BMG publication ``Climate Data in Indonesia''; and correlation of 0.30 (p Ͻ 0.05) is due to the declining more recent data for BMG stations from the BMG da- trend in both the series. Removing this trend gives a tabase. correlation of 0.18 (p Ͼ 0.10). In selecting stations from an initial set of over 3000, The lack of coherent variation between the DJF rain- Kirono et al. (1999) gave priority to meteorological sta- fall at these two stations suggests that it is likely that tions that were generally of higher quality and with local effects are dominating the rainfall in the wet sea- trained observers. Stations were also selected with re- son. Any large-scale changes are perhaps masked by cent data to enable possible future updates. Since many stronger local effects. of the stations had data in a number of sources, any While the two stations are close (less than 100 km differences between the sources were reconciled and the apart), the surrounding areas are very different topo- longest record compiled. Stations with much missing graphically. The eastern station, Jatiwangi, is located on 3884 JOURNAL OF CLIMATE VOLUME 14

FIG. 4. Spatial pattern of loadings of the ®rst unrotated principal component of SON rainfall. Contour interval is 0.1, the zero contour is bold, negative contours are dashed, and areas above ϩ0.3 and below Ϫ0.3 are shaded.

ponents changes the patterns only very slightly. A varimax rotation of two components yielded the ®rst two eigenvectors appearing almost identical to the unrotated patterns, with variance explained 11.0% and 10.5%. Since we are primarily interested in regionwide varia- FIG. 3. Spatial pattern of loadings of the ®rst two unrotated principal components of DJF rainfall. Contour interval is 0.1, the zero contour tion, and the rotated results are similar to the unrotated is bold, negative contours are dashed, and areas above ϩ0.3 and results, we will consider only the unrotated results. below Ϫ0.3 are shaded. The fact that there is no high-variance principal component (rotated or unrotated) or any signi®cant component with high loadings over a large part of the region the northern coastal plain of Java at an elevation of 50 m. implies there is only limited coherence in the DJF rain- The western station, Bandung, is at an elevation of 791 fall. Therefore, no single predictor [e.g., Southern Os- m and has mountains to the north and south. Therefore, cillation index (SOI) or a single SST principal com- the varied topography in the area (and the island gen- ponent] seems likely to explain a high proportion of erally) could be contributing to the low coherence of rainfall variation over the entire region. rainfall variation between the two stations. The ®rst two components together explain only 21% Performing the same analysis for SON gives a much of the total variance, but both contain high loadings over higher correlation of 0.61 (p Ͻ 0.01) between the two small areas. Therefore, the variance explained for each stations with or without the trend removed. Therefore, pattern over the entire area might be small, but the var- the low correlation in DJF is not just a result of the iance explained for the region with high loadings is differing topography in the region, but is also due to probably high. The DJF rainfall at the station Ujung- the less coherent nature of the rainfall during these pandang (Fig. 1), which is located in the region with months. high loadings, has a correlation of 0.72 (p Ͻ 0.01) with the score series for the ®rst component. Therefore, the b. Regional coherence ®rst component accounts for almost 52% of the variance at this station. To examine how coherently the DJF rainfall of all the The fact that there is no high-variance component stations are varying, a principal component analysis was with high loadings over a large part of the region implies carried out. An analysis based on the correlation matrix there is no instantaneous large-scale coherence. This is was used to focus on coherent variation rather than the because all correlations are calculated with zero lag. covariance matrix, which is weighted toward stations There is a possibility that some areas may lag others, with higher variability. A scree plot of the eigenvalues but with annual data this would be unexpected. for this analysis (not shown) indicates that the ®rst two The low spatial coherence does not hold in other sea- components form a degenerate pair (North et al. 1982) sons. Figure 4 shows the ®rst unrotated principal com- and are the only signi®cant components. The proportion ponent for SON rainfall. This is the only signi®cant of total variance explained by the ®rst two eigenvectors component, determined by examining the scree plot of is 11.1% and 10.5%. eigenvalues. Note that the loadings are high over almost The spatial patterns of loadings of the ®rst two prin- the entire region, showing that much of the variation is cipal components are shown in Fig. 3. Both patterns coherent. This component accounts for 38% of the total have high loadings in a small region only; the ®rst pat- variance. All other components in this season account tern in a horseshoe-shaped pattern around the island of for less than 8% each. Sulawesi; and the second pattern over an area extending There is a possibility that a skewed rainfall distri- from southern Sumatra to western Borneo. A pattern bution could bias any correlation calculations toward containing high loadings in only a small area is termed wet years. Two alternative approaches were examined ``simple structure'' and means that rotation of the com- to account for this: ®rst, a cube-root transform was used 15 SEPTEMBER 2001 NOTES AND CORRESPONDENCE 3885

FIG. 5. Correlation between DJF rainfall and DJF SOI. Contour interval is 0.1, the zero contour is bold, negative contours are dashed, and areas above ϩ0.3 and below Ϫ0.3 are shaded. to reduce the skewness before calculating the correlation matrix in the principal component analysis (Stidd 1953); and second, a Spearman rank correlation coef®cient was used instead of the Pearson coef®cient. Neither approach had any major impact on the results. In both cases, the ®rst two principal components appeared very similar to the previous results.

4. Relationship with ENSO The score series for the components in DJF and SON were correlated with the SOI to determine if variations in seasonal rainfall in the areas with high factor loadings were related to the El NinÄo±Southern Oscillation (ENSO) phenomenon. In DJF, the ®rst two components (Fig. 3) have correlations of 0.76 (p Ͻ 0.01) and Ϫ0.28 (p Ͻ 0.10) with the SOI. The SON ®rst component has a correlation of 0.79 (p Ͻ 0.01) with the SOI. Therefore, FIG. 6. Scatterplot of all-Indonesian rainfall index and SOI for DJF in DJF and SON, the area of the ®rst component with and SON. high loadings shows rainfall that is to a large part being governed by ENSO. In SON this includes a large part of the region, while in DJF it includes a much smaller Figure 6 shows this series plotted against the SOI for area in the north of the region. each year. The much higher correlation between the rain- These results are re¯ected in the correlations between fall index and the SOI for SON is evident in this plot. the SOI and the station rainfall. Figure 5 shows the The correlation for SON is 0.81 (p Ͻ 0.01), while for correlation between the DJF rainfall and the DJF SOI. DJF it is 0.23 (p Ͻ 0.20). The area of strong positive correlation in the north cor- Figure 6 also illustrates the reduced spatial coherence responds to the area of high loading in the ®rst principal in DJF. The range of values along the x axis is an in- component. The area of moderately strong negative cor- dication of how coherent the rainfall is. For a low sta- relations to the west of this region corresponds to the tionwide average there needs to be more stations with area of high loading in the second principal component. low rainfall for a particular year. The increased range Since the second component is not as well correlated for SON compared with DJF shows that coherence is with the SOI, the correlations are not as strong in this higher for SON. region. In SON the correlations between the station rain- The correlation between the all-Indonesian rainfall fall and the SOI (not shown) are high across most of index and the SOI was investigated for the other two the region. seasons March±May (MAM) and June±August (JJA). We can further examine the relationship between rain- Figure 7 shows the correlation of the index for all four fall and ENSO by considering an all-Indonesian rainfall seasons. The correlation is lowest in DJF and MAM and index. For each station we converted the seasonal rain- higher for the other two seasons. fall total into a rank percentile (0%±100%) for that sea- Figure 7 also shows the percent of total variance that son. By averaging across all stations for each year, we is explained by the signi®cant principal components for constructed an annual time series of the index for each each season. In this ®gure, the variance has been pro- season. The index is in the range 0 to 100 and weights portioned into the number of signi®cant components for each station equally, independently of the stations' mean that season as determined from the scree plot of eigen- or variance of rainfall. values. The proportion of variance (and therefore the 3886 JOURNAL OF CLIMATE VOLUME 14

framework developed in the above literature. The simple association, however, is that the slowly evolving boundary dynamics is inherently large scale in nature. The ¯ow instabilities associated with the chaotic aspect of the ¯ow are usually assumed to stem from sub-synoptic- scale ¯uctuations (Charney and Shukla 1981), whereas the slower dynamics is assumed to evolve on scales at least as large as the deformation radius. Consistent with this, for interannual rainfall variations in a region to be predictable, they must be associated with the evolving boundary condition, which means they must be coherent in space. As shown in Fig. 7, the most potentially predictable rainfall in Indonesia is during the dry season (JJA) and FIG. 7. Correlation between all-Indonesian rainfall index and SOI, the least predictable is the wet season (DJF). The lack and percent of variance explained by signi®cant principal components of spatial coherence in the wet season may simply be for each season. a function of low-latitude convective dynamics, whereby the organization of the ¯ow lies mainly in the non- regional coherence) is highest in JJA and SON. This is balanced ¯ow regime at scales smaller than the defor- the time of year when the correlation between the all- mation radius, as described by Ooyama (1982). In the Indonesian rainfall index and the SOI is highest. There- speci®c case of the Indonesian wet season, the domi- fore, at the times of year when ENSO has a stronger nance of mesoscale and submesoscale effects will be in¯uence on the rainfall, the rainfall is more spatially ampli®ed by the geography, which includes a mixture coherent. of sea, islands, and high mountains. However, this raises the question, Is the lack of control by the Southern Oscillation caused by the fact that these mesoscale ef- 5. Discussion fects dominate or, alternatively, does the domination by To our knowledge the direct association proposed the mesoscale occur because there is little control by here between predictability and large-scale coherence the Southern Oscillation? There is no way of satisfac- has not been noted before. There is, however, an ex- torily answering this question given the evidence just tensive literature on seasonal predictability at low lat- presented. The fact that during DJF the amplitude of itudes. Most of the early development was based on the ®rst principal component is still very highly cor- analysis of time series data at a particular station, with related with the SOI implies that ENSO is still active the potentially predictable component of the signal be- in the north of the region where this component has ing quanti®ed in terms of the component of the variance high loadings. However, a much more vigorous study in the interannual range compared with the variance would be needed to determine if the dynamics of ENSO associated with weather systems on the timescale of are still present in the rest of the region but are being days (Leith 1975, appendix 2.2; Madden 1976; Madden dominated by mesoscale convective dynamics. and Shea 1978; Shukla and Gutzler 1983). An important step in the understanding of predictability was made by 6. Conclusions Charney and Shukla (1981). They demonstrated that at low latitudes a large part of the variability is associated Coherence of seasonal rainfall anomalies has been with the variations of the boundary conditions to the examined for 63 stations across Indonesia for the period atmosphere, speci®cally quantities such as sea surface 1950±98. During the peak of the wet season (DJF), the temperature, albedo, and soil moisture. According to the coherence is small with, for example, rainfall being al- Charney±Shukla framework, the variance associated most unrelated at two neighboring stations separated by with the slowly evolving boundary conditions repre- less than 100 km. sents the predictable component of the signal as distinct The coherence across the region was quanti®ed with from the chaotic component associated with internal principle component analysis. For the wet season DJF, ¯ow instabilities. This concept has led to the develop- the low-order components have high loadings over only ment of various frameworks to determine the component small areas of the country, which means that it is only of the signal governed by boundary dynamics as distinct these regions that have coherence in interannual rainfall from internal dynamics, both being measured through variation. Consistent with this, these same regions have analysis of multisample ensemble integrations of gen- rainfall variations correlating highly with the Southern eral circulation models (Shukla 1998; Rowell 1998; Oscillation index at that time of year. Frederiksen et al. 1999). In contrast, the dominant principal component for the We have not yet explored the relationship between transition season SON has high loadings over most of the predictability concept of the current paper and the the Maritime Continent; and again consistent with this, 15 SEPTEMBER 2001 NOTES AND CORRESPONDENCE 3887 rainfall over most of the region correlates highly with Leith, C. E., 1975: The design of a statistical±dynamical climate the SOI. model and statistical constraints on the predictability of climate. The Physical Basis of Climate and Climate Modelling, WMO The conclusion in this paper is that the lack of large- GARP Publication Series, No. 16, WMO, 137±141. scale coherence in the DJF rainfall anomalies means Madden, R. A., 1976: Estimates of the natural variability of time- rainfall at that time of year is inherently unpredictable. averaged sea-level pressure. Mon. Wea. Rev., 104, 942±952. To verify this we have used a simple all-Indonesian ÐÐ, and D. J. Shea, 1978: Estimates of natural variability of time- averaged temperatures over the United States. Mon. Wea. Rev., rainfall index constructed from an averaged percentile 106, 1695±1703. ranking of seasonal rainfall from each station across the Makarau, A., and M. R. Jury, 1997: Predictability of Zimbabwe sum- region. At the times of the year when a large (small) mer rainfall. Int. J. Climatol., 17, 1421±1432. percentage of the variance of rainfall is described by McBride, J. L., 1983: Satellite observations of the Southern Hemi- the lowest-order principal components, there is a large sphere monsoon during winter MONEX. Tellus, 35A, 189±197. ÐÐ, 1999: Indonesia, Papua New Guinea, and Tropical Australia: (small) correlation between the SOI and the all-Indo- The Southern Hemisphere summer monsoon. Meteorology of the nesian rainfall index. Southern Hemisphere, Meteor. Monogr., No. 49, Amer. Meteor. Soc., 89±99. Acknowledgments. This research is partially support- ÐÐ, and N. Nicholls, 1983: Seasonal relationships between Aus- ed by the Australian Centre for International Agricul- tralian rainfall and the Southern Oscillation. Mon. Wea. Rev., 111, 1998±2004. tural Research Project LWR2/96/215. Special thanks to Montecinos, A., A. Diaz, and P. Aceituno, 2000: Seasonal diagnostic Dewi Kirono from the Faculty of Geography, Gadjah and predictability of rainfall in subtropical South America based Mada University, Bulaksumur, Yogyakarta, Indonesia, on tropical Paci®c SST. J. Climate, 13, 746±758. for help in assembling the rainfall dataset and to Paulus Murakami, T., and A. Sumi, 1982: Southern Hemisphere monsoon Winarso and Dodo Gunawan of Badan Meteorologi dan circulation during the 1978±79 WMONEX. Part I: Monthly mean wind ®elds. J. Meteor. Soc. Japan, 60, 638±648. Geo®sika, Jakarta, Indonesia. New, M., M. Hulme, and P. Jones, 1999: Representing twentieth- century space±time climate variability. Part I: Development of a 1961±90 mean monthly terrestrial climatology. J. Climate, 12, REFERENCES 829±856. Berlage, H. P., 1927: East-Monsoon forecasting in Java. Koninklijk Nicholls, N., 1981: Air±sea interaction and the possibility of long- Magnetisch en Meteorologisch Observatorium te Batavia, Ver- range weather prediction in the Indonesian archipelago. Mon. handelingen No. 5, 42 pp. Wea. Rev., 109, 2435±2443. ÐÐ, 1934: Further research into the possibility of long range fore- North, G. R., T. L. Bell, R. F. Cahalan, and F. J. Moeng, 1982: Sam- casting in Netherlands±India. Koninklijk Magnetisch en Meteo- pling errors in the estimation of empirical orthogonal functions. rologisch Observatorium te Batavia, Verhandelingen No. 26, 31 Mon. Wea. Rev., 110, 699±706. pp. Ooyama, K. V., 1982: Conceptual evolution of the theory and mod- Braak, C., 1919: Atmospheric variations of short and long duration elling of the tropical cyclone. J. Meteor. Soc. Japan, 60, 369± in the Malay Archipelago. Koninklijk Magnetisch en Meteorol- 380. ogisch Observatorium te Batavia, Verhandelingen No. 5, 57 pp. Peterson, T. C., and Coauthors, 1998: Homogeneity adjustments of Buishand, T. A., 1982: Some methods for testing the homogeneity in situ atmospheric climate data: A review. Int. J. Climatol., 18, of rainfall records. J. Hydrol., 58, 11±27. 1439±1517. Charney, J., and J. Shukla, 1981: Predictability of monsoons. Mon- Rowell, D. P., 1998: Assessing potential seasonal predictability with soon Dynamics, Sir J. Lighthill and R. P.Pearce, Eds., Cambridge an ensemble of multidecadal GCM simulations. J. Climate, 11, University Press, 99±109. 109±120. Drosdowsky, W., 1993: Potential predictability of winter rainfall over Shukla, J., 1998: Predictability in the midst of chaos: A scienti®c southern and eastern Australia using Indian Ocean sea-surface basis for climate forecasting. Science, 282, 728±731. temperature anomalies. Aust. Meteor. Mag., 42, 1±6. ÐÐ, and D. S. Gutzler, 1983: Interannual variability and predict- Frederiksen, C. S., D. P. Rowell, R. C. Balgovind, and C. K. Folland, ability of 500 mb geopotential heights over the Northern Hemi- 1999: Multidecadal simulations of Australian rainfall variability: sphere. Mon. Wea. Rev., 111, 1273±1279. The role of SSTs. J. Climate, 12, 357±379. Stidd, C. K., 1953: Cube-root-normal precipitation distributions. Eos, Hackert, E. C., and S. Hastenrath, 1986: Mechanisms of Java rainfall Trans. Amer. Geophys. Union, 34, 31±35. anomalies. Mon. Wea. Rev., 114, 747±757. Sukanto, M., 1969: Climate of Indonesia. World Survey of Clima- Hastenrath, S., 1987: Predictability of Java monsoon rainfall anom- tology, H. Arakawa, Ed., Vol. 8, Climates of Northern and East- alies: A case study. J. Climate Appl. Meteor., 26, 133±141. ern Asia, Elsevier, 215±229. Kirono, D. G. C., N. J. Tapper, and J. L. McBride, 1999: Documenting Wang, B., R. Wu, and X. Fu, 2000: Paci®c±East Asian teleconnection: Indonesian rainfall in the 1997/1998 El NinÄo event. Phys. How does ENSO affect east Asian climate? J. Climate, 13, 1517± Geogr., 20, 422±435. 1536. Chapter 4: McBride et al., 2003

McBride JL, Haylock MR and Nichols N. 2003. Relationships between the maritime continent heat source and the El Nino-Southern Oscillation phenomenon. Journal of Climate, 16, 2905-2914. 1SEPTEMBER 2003 MCBRIDE ET AL. 2905

Relationships between the Maritime Continent Heat Source and the El NinÄo±Southern Oscillation Phenomenon

JOHN L. MCBRIDE,MALCOLM R. HAYLOCK, AND NEVILLE NICHOLLS Bureau of Meteorology Research Centre, Melbourne, Australia

(Manuscript received 26 November 2001, in ®nal form 16 January 2003)

ABSTRACT Various earlier studies have demonstrated that rainfall in the Maritime Continent±Indonesia region is strongly related to the El NinÄo±Southern Oscillation (ENSO) during the dry half of the year but has a very weak association with ENSO during the summer±wet season months. This relationship is investigated over a wider domain through the use of outgoing longwave radiation (OLR) data as a proxy for rainfall. Consistent with the hypothesis of Haylock and McBride, it is found that the large-scale structure of the low- order empirical orthogonal functions (EOFs) of OLR have a strong resemblance to the patterns of correlation between OLR and the Southern Oscillation index (SOI). This supports the hypothesis that the predictable component of rainfall is determined by the component that is spatially coherent, as quanti®ed through EOF analysis. As was found earlier with rainfall, the region of largest correlation between interannual OLR anomalies and the SOI lies in the winter hemisphere. The predictable component of OLR (or rainfall) remains in the region of the Maritime Continent throughout the year and thus does not accompany the minimum OLR (maximum rainfall) during its annual interhemispheric progression as the major monsoon heat source. The sign of the OLR±SOI relationship is such that the Maritime Continent has increased rainfall during a La NinÄa or cold event. Patterns of correlation between sea surface temperature and the SOI show the existence of a region to the east of the Maritime Continent whereby sea surface temperature anomalies are positive during these (La NinÄa) conditions. This is in the sense of a direct relationship, that is, positive sea surface temperature anomalies corresponding to increased rainfall. The annual cycle of the sea surface temperature structure of ENSO is represented by the ®rst EOF of the interannual sea surface temperature series for each separate calendar month. The region of the sea surface temperature anomaly giving the direct relationship with Maritime Continent rainfall is part of the ``boomerang- shaped'' pattern that lies between and has the opposite sign from the anomalies in the eastern-central Paci®c and the Indian Oceans. Besides being a fundamental component of the large-scale sea surface temperature structure of ENSO, the boomerang pattern goes through an annual cycle such that it has maximum amplitude in the winter hemisphere. This suggests that interannual variations of Maritime Continent rainfall are in direct response to upstream sea surface temperature anomalies in the ENSO boomerang pattern.

1. Introduction (1982). In recent years there have also been considerable advances in the understanding of the relationships be- The El NinÄo±Southern Oscillation (ENSO) phenom- tween ENSO variations and variations in convective ac- enon is a major controlling factor of interannual vari- tivity associated with the Northern Hemisphere branch ability of climate over the globe. A fundamental component of this coupled atmosphere±ocean phenomenon of the Asian monsoon (see, e.g., Kirtman and Shukla is the variability of equatorial convective rainfall in the 2000; Lawrence and Webster 2001; Lau and Wu 2001). west-central Paci®c, at approximately the longitude of The current paper addresses variability of convection the date line. On interannual timescales the magnitude and rainfall in a third tropical heat source: in the inter- of convective activity at this location is both a response vening longitudes of Australia and the Maritime Con- to variations in the underlying sea surface temperatures tinent. As discussed by Lau and Chan (1983b), Meehl (SSTs) as well as a driving force for the trade winds (1987), and McBride (1998), Maritime Continent con- across the eastern Paci®c, through the mechanisms elu- vection and northern summer monsoon convection can cidated by Gill (1980), Webster (1981), and Zebiak be both considered part of the planetary monsoon system, with the tropical convective maximum undergoing an annual migration from over northern India in July Corresponding author address: Dr. John L. McBride, Bureau of to Indonesia in November±December (Meehl 1987). Meteorology Research Centre, GPO Box 1289K, Melbourne 3001, Australia. The relationship between Maritime Continent con- E-mail: [email protected] vection and ENSO has been previously studied by Lau

᭧ 2003 American Meteorological Society 2906 JOURNAL OF CLIMATE VOLUME 16 and Chan (1983a,b). These authors documented the ex- studies of Simmons et al. (1983) and Ting and Sar- istence of a dipole in convective variation between the deshmukh (1993) demonstrate that atmospheric tele- Maritime Continent and the date line. Thus, during a connection patterns and extratropical responses to trop- warm event convection is enhanced over the date line ical heat anomalies are highly sensitive to the longitudes and depressed over the Maritime Continent, with the at which the heating anomalies occur. Thus, it is im- converse happening during a cold event. The association portant to understand the ENSO response of the major between ENSO and interannual variations in Maritime equatorial rainfall centers. Continent convection has the sense that low pressure is The main ®nding is that the phenomenon of the dry associated with high rainfall. Thus, Maritime Continent season being related to ENSO and the wet season being rainfall is greater than normal during a cold event when effectively independent of ENSO is a general result ex- the Jakarta, Indonesia, or Darwin, Australia, pressure tending beyond the island station network to cover the anomaly is negative and less than normal in the opposite entire region. Examination of the annual cycle of the (warm) phase when Jakarta has a positive pressure west Paci®c SST pattern associated with ENSO reveals anomaly. the existence of a related phenomenon in SST. Specif- It has long been known there is a strong seasonal ically, Maritime Continent rainfall seems to be con- variation in the association between Australian±Indo- trolled by ¯uctuations in a ``boomerang-shaped'' SST nesian rainfall and the Southern Oscillation (Nicholls anomaly pattern of opposite sign to the anomalies in the 1981; McBride and Nicholls 1983; Hastenrath 1987; equatorial eastern Paci®c. Considering the relationship Ropelewski and Halpert 1987; Kiladis and Diaz 1989; between coherence and predictability proposed by Hay- Kirono et al. 1999). Using monthly rainfall from 63 lock and McBride, the predictable component of mon- stations across Indonesia, Haylock and McBride (2001) soon convective activity is not in the major heat source demonstrated that rainfall in the dry and transition sea- of the planetary monsoon, but rather in the Maritime sons is highly correlated with ENSO; whereas in the Continent region and centered in the winter hemisphere. wet season of December±January there is little or no This has implications for understanding the interactions correlation. They used empirical orthogonal function between ENSO and the planetary monsoon. (EOF) analysis to demonstrate that interannual variations in wet season rainfall exhibit very little spatial coherence. Haylock and McBride hypothesised that pre- 2. Analysis dictability of rainfall in a region is dependent on co- a. Correspondence between coherence and herence of the rainfall anomalies. On this basis they predictability concluded that Maritime Continent wet season rainfall is inherently unpredictable. Haylock and McBride (2001), through their analysis The current note documents further exploration of the of Indonesian monthly station data, demonstrated a seasonal structure of ENSO±monsoon rainfall relation- close correspondence between regions showing large ships. Through the use of outgoing longwave radiation amplitude in the low-order EOFs and regions that are (OLR) as a proxy for convection, the relationships have simultaneously correlated with the SOI. To extend this been extended beyond the land points represented by beyond the station network, which is restricted to the the Haylock±McBride station network. The data used islands of Indonesia, we use OLR as a proxy for rainfall, in the paper are OLR data for the period 1974±2001 following the reasoning of Heddinghaus and Krueger (excluding 1978), from the National Oceanic and At- (1981). mospheric Administration (NOAA) polar-orbiting sat- Figure 1 shows the ®rst unrotated EOF of the cor- ellites as described by Gruber and Krueger (1984). Grid- relation matrix of OLR for the four calendar months of ded SST data for the period 1949±98 are from the Met January, April, July, and October over the region 50ЊS± Of®ce Global Sea Ice and Sea Surface Temperature da- 30ЊN and 70ЊE±180Њ. The variance explained by the ®rst taset (GISST; Parker et al. 1995). The index of ENSO EOF for the four months is 18.4%, 18.8%, 15.5%, and activity used is the Southern Oscillation index (SOI), 20.7%, respectively. The ®gures show the spatial pat- which is the normalized Tahiti-minus-Darwin pressure terns of the loadings for each EOF, with light and dark difference as described by Allan et al. (1991). In section shading representing loadings of opposite sign. Regions 2 the results of EOF analyses are shown for each cal- where there are large amplitudes in the EOF pattern endar month for both OLR and SST. These are related signify regions where the interannual anomalies for that to the spatial structure of the relationships between OLR month are highly correlated with the amplitude of that variability and ENSO. Section 3 summarizes and dis- EOF. Accordingly, the data points within these regions cusses these results. have anomaly time series that are coherent with each The focus of the study is on ENSO±OLR relationships other. speci®cally in the Maritime Continent. This is as a fol- For all 4 times of the year there is a dipole structure low-up to our earlier study. It is also because interannual in this ®rst EOF with a large structure of one sign (light variability of Maritime Continent convection could play shading) over the Maritime Continent and Indochina an important part in global dynamics. For example the longitudes and a structure of opposite sign (dark shad- 1SEPTEMBER 2003 MCBRIDE ET AL. 2907

FIG. 1. Spatial patterns of loadings for the ®rst unrotated principal components of the interannual time series of OLR for individual calendar months. The four panels are for the months of Jan, Apr, Jul, and Oct, as labeled. Loadings with magnitude greater than 0.4 are shaded, with shading intervals of 0.2. Dark shading and light shading represent loadings of opposite (arbitrary) sign. ing) over the equatorial Tropics in the eastern part of and extends westward almost to New Guinea. Once the domain. As discussed in the introduction, the two again the same features show up in the correlation pat- centers of the dipole are described here as being the tern. Maritime Continent convection and the date line con- According to the Haylock±McBride hypothesis, pre- vection, respectively. There are also a number of bands dictability for a large-scale domain requires spatial co- of structure of either sign extending from the equator herence over that domain, and that coherence can be toward the northeast and toward the southeast at various quanti®ed as having large amplitude in the low-order times of the year. EOFs. It would be expected that the predictable com- The four panels of Fig. 2 show the correlations be- ponent of the signal would be the same component that tween OLR and the SOI over the same period (1974± is correlated with indices of large-scale phenomena, 2001) for the same calendar months. The remarkable such as the SOI. The close similarity between the pat- characteristic of these four diagrams is the similarity of terns in each panel of Fig. 2 and the patterns in the structure to the corresponding panel in Fig. 1. For ex- corresponding panel of Fig. 1 provides support for the ample, in January the lowest-order EOF (Fig. 1) has the Haylock±McBride hypothesis. largest amplitude over northern Borneo and the southern There are some interesting differences between the Philippines with a tongue of amplitude of the same sign (OLR±SOI) correlation patterns in Fig. 2 and the cor- extending eastward. The same structure shows up in the responding (rainfall±SOI) correlations in Haylock and corresponding correlation map in Fig. 2. In July the McBride (2001) and in Kirono et al. (1999). Speci®- Maritime Continent branch of the dipole is a relatively cally, the correlation patterns for OLR are smoother and small region located over southern Borneo and Java, have larger-scale structures. This may re¯ect the fact while the date line branch spans a wide latitude range that OLR is on a regular grid of relatively coarse (2.5Њ) 2908 JOURNAL OF CLIMATE VOLUME 16

FIG. 2. Coef®cients of linear correlation between OLR and the SOI for the interannual time series of individual calendar months. Shading is for correlations of magnitude greater than 0.3, with shading intervals of 0.15. Light shading represents negative correlations (i.e., dry in an El NinÄo or warm event) while dark shading represents positive correlations. resolution. It also may be related to physical differences OLR with coherent anomalies remains over the Mari- whereby OLR is more spatially coherent than rainfall. time Continent; that is, it does not follow the annual Despite that, the key point remains that for either var- return excursion from the Maritime Continent to India± iable (OLR or rainfall) the lowest-order EOF bears a Indochina. As discussed above and proposed by Hay- strong structural similarity to the pattern of correlation lock and McBride (2001) the coherent part of the OLR between that variable and the SOI. anomaly is related to that component that is predictable. Using the magnitude of the linear correlation coef®cient with the SOI as a simple measure of predictability, the b. Predictability located in the winter hemisphere of seasonal cycle of the total signal and also of its pre- the Maritime Continent dictable component (maximum convection, predictable As has been described by Heddinghaus and Krueger convection) is shown in Fig. 3. For each month the (1981), Lau and Chan (1983b), and others, the major maximum convection is represented by the shaded area, region of convective activity associated with the plan- which denotes OLR values less than 220 W mϪ2 in the etary monsoon undergoes a seasonal variation such that long-term monthly mean. The predictable convection is in the northern summer it is located in the Northern represented by the hatched area that denotes locations Hemisphere with the maximum activity over the Bay of where that month's anomalies of OLR are negatively Bengal±India±Indochina region. In the southern sum- correlated with the SOI at magnitudes greater than 0.6. mer it is located in the Southern Hemisphere with max- A noteworthy aspect of the annual cycle of the region imum activity over southern Indonesia and Papua New with large magnitude (OLR±SOI) correlations is that it Guinea. is generally located in the winter hemisphere. Thus, dur- In contrast to this, Fig. 1 indicates that the region of ing January±April it is located north of the equator; and 1SEPTEMBER 2003 MCBRIDE ET AL. 2909

FIG. 3. The annual cycle of cold cloud over the Maritime Continent as represented by the presence of low values of OLR: the shaded area for each month being long-term mean OLR values less than 220 W m Ϫ2. Superimposed for each month is the area of OLR that is negatively correlated with the SOI, with the hatching representing correlations of magnitude greater than 0.6. in July±October it is located south of the equator. This ture is not related in any simple way to the seasonal is consistent with the ®ndings of Haylock and McBride evolution of the OLR itself. This immediately raises the (2001) that rainfall over most of their Indonesian station question as to how it is related to the seasonal evolution network is predictable in June±November, but only in of the ENSO phenomenon. the north of their region in December±February and Figure 4 shows maps of correlation between sea sur- nowhere during March±May. The Haylock±McBride face temperature and the SOI for the three southern station network lies mainly in the more populated parts summer months of January±March. Also shown (cross of Indonesia (Java, Nusa Tenggara), which are south of hatching) is the region of predictable OLR for each the equator, with a few stations in the Northern Hemi- month (as represented by negative OLR±SOI correla- sphere. tions of magnitude greater than 0.6). The light shading is for positive correlation; dark is for negative. There is dark shading through the islands of the Maritime Con- c. The boomerang pattern in sea surface temperature tinent and westward into the Indian Ocean, signifying anomalies that when the SOI is positive (i.e., during a La NinÄa Figures 2 and 3 reveal a strong seasonal cycle in the event), sea surface temperatures throughout the Mari- location and magnitude of correlations between OLR time Continent are cold. There is a region of (SST±SOI) and the SOI; and as shown in Fig. 3 this seasonal struc- correlation of the opposite sign (light shading) located 2910 JOURNAL OF CLIMATE VOLUME 16

FIG. 4. Coef®cients of linear correlation between SST and the SOI for the interannual time series of individual calendar months. Results are shown for the three southern summer months of Jan±Mar. Shading is for correlations of magnitude greater than 0.2, with shading intervals of 0.2. Dark shading is for negative correlations; light shading for positive. Also shown (hatching) is the region where OLR is negatively correlated with the SOI, with the hatching representing correlations of magnitude greater than 0.6.

FIG. 5. Same as Fig. 4, but for the months of Jul±Aug±Sep. north of the equator and extending northeast from the hatched region of high-magnitude (SOI±OLR) correlation. This light-shaded region is of the sign such that Figure 5 shows the equivalent ®gure for the time of during a La NinÄa, when rainfall is enhanced over the year when the ``predictable'' component of convection northern (winter hemisphere) portion of the Maritime lies over the Southern Hemisphere part of the Maritime Continent, the sea surface temperatures are warm. In Continent, that is, July±September. At this time of year these months (January±March) this region is collocated there is light shading (positive correlations) throughout with the winter northeasterly trade winds. Thus in a La the Maritime Continent signifying positive sea surface NinÄa/cold event the warm sea surface temperatures are temperatures during a La NinÄa. Here the positive cor- upstream of the region where the Maritime Continent relations span both sides of the equator; but as with the convection is enhanced. previous ®gure the correlation pattern extends eastward 1SEPTEMBER 2003 MCBRIDE ET AL. 2911

FIG. 6. Spatial patterns of loadings for the ®rst unrotated principal components (using the correlation matrix) of the interannual time series of SST for individual calendar months. Light and dark shadings represent loadings of opposite sign. Shading is for loading magnitudes greater than 0.2, with intervals of 0.2. and poleward (in this case southward) following the path tural maps (weights) for the resultant ®rst EOF are of the winter trade winds. shown in Fig. 6. This ®gure can be considered to be a representation of the seasonal cycle of the ENSO sea surface temperature pattern. The percentage of variance d. The seasonal cycle of the ENSO SST anomaly explained by the pattern varies from a minimum of pattern 15.1% in August series to a maximum of 20.4% for The seasonal variation of the correlations between the October. The interannual correlations between the am- SOI and sea surface temperature in Figs. 4 and 5 can plitude of this EOF and the SOI are, respectively, for be interpreted as a depiction of the seasonal structure, the 12 months: Ϫ0.71, Ϫ079, Ϫ071, Ϫ0.69, Ϫ0.44, for this sector, of the sea surface temperature pattern for Ϫ0.51, Ϫ0.41, Ϫ0.72, Ϫ0.73, Ϫ0.85, Ϫ0.65, Ϫ0.70. ENSO. A number of previous authors have represented The correlations with NinÄo-3.4 (5ЊN±5ЊS, 170Њ±120ЊW) the ENSO sea surface temperature pattern through the index are as follows: 0.86, 0.84, 0.77, 0.70, 0.76, 0.64, use of EOF analysis of sea surface temperatures from 0.54, 0.72, 0.83, 0.84, 0.83, 0.83. the Paci®c Ocean (Weare et al. 1976), combined Indian± The most familiar feature of the ®gure is the wedge- Paci®c Oceans (Drosdowsky and Chambers 2001), or shaped pattern in the eastern equatorial Paci®c, with a the globe (Hsiung and Newell 1983). In these studies broad pattern of the same sign over the equatorial Indian the seasonal cycle was removed through the analysis of Ocean. The eastern Paci®c anomaly has its maximum a time series of monthly anomalies, where in each case amplitude in December±January and minimum in the anomaly was formed by subtracting the mean for April±June, as is well known (Rasmusson and Carpenter that calendar month. For each of the above studies the 1982). The other important feature is the boomerang- ®rst EOF of sea surface temperature was associated with shaped anomaly in the western Paci®c that has the op- the ENSO phenomenon, and accounted for 23.1% of posite sign to the eastern Paci®c anomaly. It is present the variance for the Paci®c, 14.9% for the Paci®c±Indian all year and has two zones of maximum amplitude: in Oceans, and 7.7% for the globe. the North Paci®c during January±May; and in the south- Based on the above studies, we can identify the large- west Paci®c during August±November. The regions of scale sea surface temperature structure of ENSO with positive correlation between the SOI and sea surface the ®rst nonseasonal EOF of Paci®c±Indian Ocean sea temperature shown in Figs. 4 and 5 and discussed in surface temperatures. To document the seasonal struc- the previous section are part of this boomerang anomaly ture of this EOF, but with the mean seasonal cycle still pattern. As discussed above this anomaly pattern ex- removed, we have carried out separate EOF analyses of tends poleward and eastward from the Maritime Con- the correlation matrix of Indian±Paci®c Ocean sea sur- tinent convection and has maximum amplitude in the face temperatures for each calendar month. The struc- winter hemisphere. 2912 JOURNAL OF CLIMATE VOLUME 16

The evolution through the year of the boomerang pat- downstream hypothesis is correct, then the same direct tern has a number of interesting characteristics. During relationship applies in the central region of the pattern, April±June, the boomerang has a weak structure with that is, the Maritime Continent. the main amplitude in the northern branch but little con- In the western (i.e., Indian Ocean) pole this is not the nection to the equator. In fact, at this time of year the case. In the Indian monsoon a warm event corresponds eastern Paci®c anomaly extends into the Maritime Con- to decreased rain; that is, the (rainfall±SOI) relationship tinent and connects with the anomaly of the same sign has the same sign as for the Maritime Continent; yet as in the Indian Ocean. From July onward, the southern India lies in the western pole of the sea surface tem- branch of the boomerang builds in magnitude and perature pattern, the local (and immediate upstream) sea ``curls'' around northern Australia, and in September± surface temperature anomalies have the opposite sign. October actually extends through the Maritime Conti- Thus, the mechanisms governing ENSO±monsoon re- nent into the tropical western Indian Ocean. At the same lationships there are different and presumably involve time the anomaly of the opposite sign in the Indian remote forcing from the east Paci®c through atmospher- Ocean retreats westward giving a ``dipole structure'' to ic teleconnections, as proposed in the ``atmospheric the Indian Ocean ENSO signal, similar to that observed bridge'' hypothesis of Lau and Nath (1996) and Klein for these months by Saji et al. (1999) and Webster et et al. (1999). al. (1999). During December±January the Southern It is important to note that the direct downstream Hemisphere branch of the boomerang weakens and re- mechanism is purely a hypothesis at this stage. The more treats toward the southeast, while simultaneously, the traditional mechanism invoked for the response of the Northern Hemisphere arm increases in magnitude. Maritime Continent to a warm or cold event has been a direct Walker cell forcing associated with the dipole behavior of Maritime Continent and date line convec- e. The mechanisms by which the Southern Oscillation tion. Thus, in a warm event the date line convection is in¯uences Maritime Continent rainfall enhanced and the Maritime Continent is suppressed, the The motivation for this study was primarily to in- mechanism being through direct longitudinal (Walker) vestigate why in the Southern Hemisphere parts of the circulation (see, e.g., Lau and Chan 1983a,b). Maritime Continent the SOI±rainfall relationship is The fact that the boomerang pattern is part of ENSO strong during the winter dry season (June±November) and has the seasonal structure of a winter hemisphere but weak during the wet season. From the above anal- maximum is beyond doubt. Substantiation can be found ysis, the boomerang pattern is clearly a basic component in the warm event (El NinÄo) composites carried out by of the sea surface signal associated with ENSO; and it Rasmusson and Carpenter (1982). The region of neg- has the same seasonal behavior, namely, the anomalies ative sea surface temperature anomalies emanating from are large in the winter hemisphere and weak in the sum- the Maritime Continent and extending eastward and mer hemisphere. poleward in the winter hemisphere can be clearly seen Given the existence of equatorward and westward in their composites for the southern winter (upper panel atmospheric winter trade wind ¯ow, the boomerang sea of Rasmusson±Carpenter's Fig. 20) and for the northern surface temperature anomalies are located directly up- winter (upper panel of Rasmusson±Carpenter's Fig. 21). stream of the location of the Maritime Continent con- During July±October when the southern branch of the vection anomalies. Thus, it is possible that the inter- boomerang extends through the southern part of the annual variability of Maritime Continent convection is Maritime Continent, the OLR in this region is spatially simply a downstream response to the boomerang sea coherent and strongly correlated with ENSO. Converse- surface temperature pattern, which appears to be part ly when the northern branch extends into the region of the large-scale structure of ENSO. Thus, during an during January±April, the OLR in the northern part of E1 NinÄo, SSTs are lower than normal in the boomerang the Maritime Continent is spatially coherent and strong- region, the winter trades receive less water vapor ly correlated with ENSO. through evaporation from the underlying seas, and so the water vapor supply for rainfall over the winter hemisphere portion of the Maritime Continent is decreased. 3. Summary and discussion Conversely, during a La NinÄa, the boomerang is warm, The major results of this analysis can be summarized evaporation is increased in the winter trades, and rainfall as follows: is increased downstream over Indonesia. The overall ENSO pattern has an east±west plus-mi- 1) Haylock and McBride's hypothesis that predictabil- nus-plus structure, corresponding to the Indian Ocean ity is associated with spatial coherence has been ex- (plus), the Maritime Continent (minus), and the east- tended from station rainfall to large-scale OLR pat- central Paci®c (plus) regions, respectively. It is well terns. The strong resemblance between the structure documented (e.g., Rasmusson and Carpenter 1982) that of the ®rst EOF of OLR (Fig. 1) and the patterns of convection in the eastern pole of the pattern responds correlation between OLR and the SOI (Fig. 2) lend directly to sea surface temperature. If the previous strong support to this hypothesis. 1SEPTEMBER 2003 MCBRIDE ET AL. 2913

2) The predictable component of the OLR (as speci®ed SOI±rainfall relationship is weak. It also provides some by the component with a large correlation with the upper limit on the proportion of variance associated with SOI) does not follow the annual march of the major the predictable signal. monsoon heat source. Rather it stays in the Maritime The concept is still in a primitive form. In this paper Continent region, but in the winter hemisphere. This the coherence was related simply to the spatial structure extends to a much larger region the ®ndings of earlier of the ®rst unrotated EOF. As shown for rainfall data, authors (Nicholls 1981; Hastenrath 1987; Haylock however, in Haylock and McBride (2001), there can be and McBride 2001) that Indonesian rainfall is strong- localized but spatially coherent behavior in several of ly associated with ENSO in the winter±spring dry the higher-order EOFs; and clearly the size of the do- season, but not in the summer±autumn wet season. main will have some effect on the number of leading 3) Patterns of correlation coef®cients between sea sur- EOFs required to capture the dominant large-scale co- face temperature and the SOI show a region of cor- herence. The EOF calculations of Fig. 1 have been car- relation of the correct sign for a direct relationship ried out over a number of different domains of varying between sea surface temperature anomalies and rain- size and with various central locations, as a check on fall. This region is part of the boomerang pattern of the statements in section 2. In particular it was veri®ed the large-scale sea surface temperature structure of that the high loadings remained over the Maritime Con- ENSO; and it also has its maximum amplitude in the tinent during the northern summer. winter hemisphere. The boomerang sea surface tem- Finally, the results of this short study have implica- perature anomalies are ``upstream'' rather than being tions for understanding the mechanisms by which ENSO collocated with the Maritime Continent rainfall. A governs rainfall. In particular, it has been inferred that mechanism has been hypothesised whereby the in- Indonesian rainfall variations may be a direct down- terannual variability of Maritime Continent rainfall stream response to sea surface temperature anomalies is simply a downstream response to the boomerang in the close-by region of the ``boomerang pattern.'' This sea surface temperature pattern. The validity of this is rather than the variations being a remote response to hypothesis will be explored in future studies. equatorial central Paci®c sea surface temperature anomalies (via the Walker circulation). An important area of In our earlier study (Haylock and McBride 2001), further work is to understand how the boomerang sea based on the analysis of rainfall data it was concluded surface temperature anomaly pattern itself develops, that December±January Indonesian rainfall was inher- presumably in response to the evolution of ENSO-as- ently unpredictable. This conclusion was based on the sociated sea surface temperature anomalies in the equa- lack of large-scale coherence of interannual rainfall torial central and eastern Paci®c. anomalies in the region. The results of the current paper would imply that the lack of predictability is con®ned Acknowledgments. This research is partially support- to the summer hemisphere and that the parts of Indo- ed by the Australian Centre for International Agricul- nesia and/or the Maritime Continent north of the equator tural Research Project LWR2/96/215. Comments from are predictable at that time of year. Some care should Harry Hendon, Wasyl Drosdowsky, George Kiladis, and be taken, however, in relating OLR to rainfall. As a an anonymous referee are greatly appreciated. check on the strength of the relationship, correlations were calculated between monthly values of OLR and rainfall summed over all rainfall stations (from the Hay- REFERENCES lock±McBride set) in that OLR grid box. For the months Allan, R. J., N. Nicholls, P. D. Jones, and I. J. Butterworth, 1991: A of July±September when rainfall variations are spatially further extension of the Tahiti±Darwin SOI, early ENSO events coherent, the correlation between interannual rainfall and Darwin pressure. J. Climate, 4, 743±749. Drosdowsky, W., and L. E. Chambers, 2001: Near-global sea surface anomalies and interannual OLR anomalies exceeded 0.5 temperature anomalies as predictors of Australian seasonal rain- in magnitude for 72% of the grid boxes. For January± fall. J. Climate, 14, 1677±1687. March (when rainfall variations are not spatially co- Gill, A. E., 1980: Some simple solutions for heat induced tropical herent), the OLR±rainfall correlations exceeded 0.5 in circulation. Quart. J. Roy. Meteor. Soc., 106, 447±462. Gruber, A., and A. F.Krueger, 1984: The status of the NOAA outgoing magnitude for only 15% of the grid boxes. Thus, the longwave radiation data set. Bull. Amer. Meteor. Soc., 65, 958± magnitude of the relationship is quite weak at the time 962. of year when the convective activity is greatest. Hastenrath, S., 1987: Predictability of Java monsoon rainfall anom- The association between predictability and coherence alies: A case study. J. Climate Appl. Meteor., 26, 133±141. Haylock, M., and J. McBride, 2001: Spatial coherence and predict- also needs to be further explored. For predictability as- ability of Indonesian wet season rainfall. J. Climate, 14, 3882± sociated with a low-frequency phenomenon such as 3887. ENSO, the association is intuitively obvious. However, Heddinghaus, T. R., and A. F. Krueger, 1981: Annual and interannual it is an idea with important implications. For example, variations in outgoing longwave radiation over the tropics. Mon. Wea. Rev., 109, 1208±1218. in a region of the world where there is no large-scale Hsiung, J., and R. E. Newell, 1983: The principal nonseasonal modes coherence of anomalies, it discourages the search for of global sea surface temperature. J. Phys. Oceanogr., 13, 1957± further large-scale predictors in the instance where the 1967. 2914 JOURNAL OF CLIMATE VOLUME 16

Kiladis, G. N., and H. F. Diaz, 1989: Global climatic anomalies as- the tropical Paci®c and Indian Ocean regions. Mon. Wea. Rev., sociated with extremes in the Southern Oscillation. J. Climate, 115, 27±50. 2, 1069±1090. Nicholls, N., 1981: Air±sea interaction and the possibility of long- Kirono, D. G. C., N. J. Tapper, and J. L. McBride, 1999: Documenting range weather prediction in the Indonesian Archipelago. Mon. Indonesian rainfall in the 1997/1998 El NinÄo event. Phys. Wea. Rev., 109, 2435±2443. Geogr., 20, 422±435. Parker, D. E., C. K. Folland, A. C. Bevan, M. N. Ward, M. Jackson, Kirtman, B. P., and J. Shukla, 2000: In¯uence of the Indian summer and K. Maskell, 1995: Marine surface data for analysis of cli- monsoon on ENSO. Quart. J. Roy. Meteor. Soc., 126, 213±239. matic ¯uctuations on interannual-to-century time scales. Natural Klein, S. A., B. J. Soden, and N.-C. Lau, 1999: Remote sea surface Climate Variability on Decadal-to-Century Time Scales, Na- temperature variations during ENSO: Evidence for a tropical tional Research Council, 241±252. atmospheric bridge. J. Climate, 12, 917±932. Rasmusson, E. M., and T. H. Carpenter, 1982: Variations in tropical Lau, K.-M., and P. H. Chan, 1983a: Short-term climate variability sea surface temperature and surface wind ®elds associated with and atmospheric teleconnections from satellite-observed out- the Southern Oscillation/El NinÄo. Mon. Wea. Rev., 110, 354± going longwave radiation. Part I: Simultaneous relationships. J. 384. Atmos. Sci, 40, 2735±2750. Ropelewski, C. F., and M. S. Halpert, 1987: Global and regional scale ÐÐ, and ÐÐ, 1983b: Short-term climate variability and atmo- precipitation patterns associated with the El NinÄo/Southern Os- spheric teleconnections from satellite-observed outgoing long- cillation. Mon. Wea. Rev., 115, 1606±1626. wave radiation. Part II: Lagged correlations. J. Atmos. Sci., 40, Saji, N. H., B. N. Goswami, P. N. Vinayachandran, and T. Yamagata, 2751±2767. 1999: A dipole mode in the tropical Indian Ocean. Nature, 401, ÐÐ, and H. T. Wu, 2001: Principal modes of rainfall±SST variability 360±363. of the Asian summer monsoon: A reassessment of the monsoon± Simmons, A. J., J. M. Wallace, and G. W. Branstator, 1983: Barotropic ENSO relationship. J. Climate, 14, 2880±2895. wave propagation and instability, and atmospheric teleconnec- Lau, N.-C., and M. J. Nath, 1996: The role of the atmospheric bridge tion patterns. J. Atmos. Sci., 40, 1363±1392. in linking tropical Paci®c ENSO events to extratropical SST Ting, M., and P. D. Sardeshmukh, 1993: Factors determining the anomalies. J. Climate, 9, 2036±2057. extratropical response to equatorial diabatic heating anomalies. Lawrence, D. M., and P. J. Webster, 2001: Interannual variations of J. Atmos. Sci., 50, 907±918. the intraseasonal oscillation in the South Asian summer monsoon Weare, B. C., A. R. Navato, and R. E. Newell, 1976: Empirical region. J. Climate, 14, 2910±2922. orthogonal analysis of Paci®c sea surface temperatures. J. Phys. McBride, J. L., 1998: Indonesia, Papua New Guinea, and tropical Oceanogr., 6, 671±678. Australia. The Southern Hemisphere summer monsoon. Mete- Webster, P. J., 1981: Mechanisms determining the atmospheric re- orology of the Southern Hemisphere, Meteor. Monogr., No. 49, sponse to sea surface anomalies. J. Atmos. Sci., 38, 554±571. Amer. Meteor. Soc., 89±99. ÐÐ, A. M. Moore, J. P. Loschnigg, and R. R. Leben, 1999: Coupled ÐÐ, and N. Nicholls, 1983: Seasonal relationships between Aus- ocean±atmosphere dynamics in the Indian Ocean during 1997± tralian rainfall and the Southern Oscillation. Mon. Wea. Rev., 98. Nature, 401, 356±360. 111, 1998±2004. Zebiak, S. E., 1982: A simple atmospheric model of relevance to El Meehl, G. A., 1987: The annual cycle and interannual variability in NinÄo. J. Atmos. Sci., 39, 2017±2027. Chapter 5: Haylock and Nicholls, 2000

Haylock M and Nicholls N. 2000. Trends in extreme rainfall indices for an updated high quality data set for Australia, 1910-1998. International Journal of Climatology, 20, 1533-1541. INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 20: 1533–1541 (2000)

TRENDS IN EXTREME RAINFALL INDICES FOR AN UPDATED HIGH QUALITY DATA SET FOR AUSTRALIA, 1910–1998 MALCOLM HAYLOCK* and NEVILLE NICHOLLS Bureau of Meteorology, Melbourne, Australia

Recei6ed 1 September 1999 Re6ised 28 June 2000 Accepted 30 June 2000

ABSTRACT Daily rainfall was analysed at 91 high quality stations over eastern and southwestern Australia to determine if extreme rainfall had changed between 1910 and 1998. Three indices of extreme rainfall were examined: the number of events above an extreme threshold (extreme frequency); the average intensity of rainfall from extreme events (extreme intensity); and the proportion of total rainfall from extreme events (extreme percent). Several problems are discussed associated with designing such indices under a climate with significant trends in the number of raindays. Three different methods are used for calculating the extreme intensity and extreme percent indices to account for such trends in raindays. A separate analysis was carried out for four separate regions with significant results including a decrease in the extreme frequency and extreme intensity in southwest Western Australia and an increase in the extreme percent in eastern Australia. Trends in the extreme intensity and extreme percent are largely dependent on the method used to calculate the index. Total rainfall is strongly correlated with the extreme frequency and extreme intensity indices, suggesting that extreme events are more frequent and intense during years with high rainfall. Due to an increase in the number of raindays during such years, the proportional contribution from extreme events to the total rainfall depends on the method used to calculate this index. Copyright © 2000 Royal Meteorological Society.

KEY WORDS: extreme rainfall; Australia; trend analysis; rainfall intensity

1. INTRODUCTION

The Second Assessment of the Intergovernmental Panel on Climate Change (Nicholls et al., 1996) examined whether, globally, the climate was becoming more extreme or variable. At the time, data and analyses were considered inadequate to answer this question. Karl et al. (1995a) and Nicholls (1995) discuss some of the reasons for this. Improved studies of climate extreme trends, on high quality and consistent data, are needed if we are to be able to determine whether climate extremes are varying. Analyses of trends in rainfall extremes over the Australian region have mainly concentrated on changes in upper percentiles and changes in frequencies of extreme events using arbitrary thresholds. Hennessy et al. (1999) analysed daily rainfall from 1910 to 1995 on a regional and seasonal basis. They found some significant changes in percentiles and frequency of extreme events, but with the magnitude and the sign of the changes varying with the season and the region. Likewise, Plummer et al. (1999) found trends in percentiles that depended on the season and the region. Results that are more regionally consistent have been obtained by Suppiah and Hennessy (1998). They found increasing trends in 90th and 95th percentiles over most of Australia, for both summer and winter half-years, as well as increasing trends in the frequency of events above long-term mean percentiles. Outside Australia, studies of extreme rainfall have included other methods. In a study of 20th century trends in the USA climate, Karl et al. (1995b) found a steady increase in the percentage of annual precipitation derived from extreme events exceeding 50.8 mm, contributed mainly by changes in spring

* Correspondence to: Bureau of Meteorology, GPO Box 1289K, Melbourne, Victoria 3001, Australia. e-mail: [email protected]

Copyright © 2000 Royal Meteorological Society 1534 M. HAYLOCK AND N. NICHOLLS and summer rainfall. Similarly, Karl and Knight (1998) found increases in both the intensity and the frequency of extreme events over the USA using thresholds based on long-term mean percentiles. The present study was carried out in order to analyse several indices of extreme rainfall using reliable and consistent Australian data. This study differs from previous studies in the data used, the method of station selection and by considering several previously unpublished indices: the average intensity of rainfall from extreme events (rather than percentiles in isolation); and the proportion of annual total rainfall falling in the highest 5% of raindays and highest four events. A third index, the number of events above the long-term 95th percentile, is calculated from raindays only, rather than the less extreme 90th and 95th percentiles calculated from all days by Suppiah and Hennessy (1998).

2. DATA

An update to the Australian high-quality rainfall data provided the basis for this study. Lavery et al. (1992) compiled the original set of 24-h rainfall totals from a comprehensive search of Australian rainfall records. As well as an exhaustive investigation of station history documentation to remove stations with bad exposure, instrumentation or observer accuracy, a series of statistical tests was utilized to further check the station integrity. Specific tests were also performed to check the influence of the change from imperial to metric units in 1974 as well as to check for bad observer practice. In this study, recent additions to the station history documentation were examined. From the original set of 191 stations, ten stations were rejected after recent inspections revealed poor exposure or bad observer practice. Statistical tests were not used to reject stations during this update. The resulting updated data set of 181 stations provides reliable daily observations from May 1910 to April 1998, providing adequate coverage over much of Australia. A lack of stations between 120° and 130°E is an obvious limitation, as is the sparse coverage north of 25°S due to the rarity of stations with long reliable records in these areas. The daily rainfall records include days where data are either missing or are recorded as an accumulated value over several days. Previous studies have used different methods for dealing with accumulations and missing values. Suppiah and Hennessy (1996) and Hennessy et al. (1999) distributed accumulations evenly over the preceding missing days and rejected stations with a large number of accumulations or missing days. Karl et al. (1995b) and Karl and Knight (1998) filled missing values by generating artificial rainfall amounts based on analyses of the distributions. Suppiah and Hennessy (1996) found that, for stations with many accumulations, there is an impact on trends in percentiles when accumulations are either distributed or ignored. In the present study the aim was to use an accurate and consistent data set using a fixed number of stations with minimal artefacts. Therefore, any accumulations were defined as missing data and stations were selected from the full set of 181 only if they had less than 10 days missing per year for at least 80 years in the 88-year period. This resulted in 94 stations being selected. The value of 10 days was chosen using Equation (1), which gives the probability P that at least one value of the top four events will be missing for a year with n missing days.

365−n4 P=1− . (1) 365

Setting P to 0.1 gives n=9.5. Therefore a year’s record with 10 days missing is equivalent to there being more than a 10% chance of having at least one missing value in the top four events. For a selected station, this will occur in less than 10% of the years. Less-strict selection criteria would only marginally increase the spatial coverage (mainly in northwest Australia), while significantly reducing confidence in the results. Likewise, the method of distributing accumulations (e.g. Suppiah and Hennessy, 1998) was not used because it added uncertainty to the quality of the time series. Finally, three stations were rejected due to an insufficient number of raindays for this study (see later). The 91 stations selected with these criteria, shown in Figure 1, provide good coverage of eastern and southwestern Australia.

The stations were divided into the three broad climatological/geographical regions: north, south, and west (Figure 1). All the stations in the south received more than 50% of their mean annual rainfall in the months May to October (i.e. winter-dominated rainfall). The stations in the northern group (except one) received less than 50% in these months (i.e. summer-dominated rainfall). The stations in the west also received most of their annual rain from May to October. Four of the stations in the western group form the southwestern subgroup and are of particular interest because rainfall in this region has decreased markedly since about 1960 (Allan and Haylock, 1993). The grouping corresponds closely with state boundaries: all the stations in the west are located in Western Australia; the stations in the south (except one) are located in Victoria, South Australia and Tasmania; and all the stations in the north are located in New South Wales and Queensland. Time series of the number of non-missing days were constructed for each season and region. Despite the attempt to include only stations with little missing data, eight of the 12 time series show a statistically significant decline in the number of non-missing days. Therefore, based on the selection criteria stated earlier, one would expect a higher than 10% chance of having one of the highest four events missing in any year in the latter part of the record. Suppiah and Hennessy (1996) note that the decrease in observations was largely caused by the cessation of observations on weekends, related to changing work practices.

3. METHODS

Three indices of extreme rainfall were calculated for each year in the period: the number of events above the long-term 95th percentile, referred to as the extreme frequency; the average intensity of rain falling in the highest events, referred to as the extreme intensity; and the proportion of total rainfall falling in the highest events, referred to as the extreme percent. The extreme frequency index examines changes in the number of extreme events. In calculating this index, the authors elected to use the mean 95th percentile (which varies for each station), rather than following the method of Karl et al. (1995b) involving a fixed threshold for all stations. A fixed threshold is impractical for a country like Australia with a high spatial variation in rainfall intensity (the mean 95th percentile for the 91 stations varied from 14 to 48 mm/day). The index is calculated by counting the number of events in the year with intensities above this threshold. This approach is similar to that used by Karl and Knight (1998) who considered changes in frequencies of events above specified long-term percentiles.

Figure 1. Station locations for 91 high quality stations and their division into three regions and one subregion. Locations of the two stations Goomalling and Collarenebri are also shown

The extreme intensity describes changes in the upper percentiles and, unlike an analysis of a single percentile threshold (e.g. Hennessy et al., 1999), incorporates changes in all events above this percentile. This index was calculated using three different methods: averaging the highest four events for each year; averaging the highest 5% of daily rainfall totals above 1 mm; and averaging all events above the long-term 95th percentile. The relative merits of each approach are discussed below. The extreme percent reflects changes in the upper portion of the daily rainfall distribution. The percentage of the total rainfall from the higher events is an indictor of changes in the shape of the rainfall distribution. This index is calculated for each year by dividing the extreme intensity by the year’s total rainfall. Therefore, the extreme percent is calculated using the three methods used for calculating the extreme intensity, to examine if the methods lead to different trends. Raindays are defined as days with at least 1 mm of rain. Thresholds lower than 1 mm can introduce trends in the number of raindays, associated with changes from imperial to metric units in 1974 (Hennessy et al., 1999). In addition, the under-reporting of small rainfall amounts has been a problem at many rainfall stations. A study by Trewin (1999, personal communication) estimated that more than 40% of rainfall events less than 2 mm in district 17 (northeastern South Australia) were not reported. One of the principal criteria for selection of a station in the high-quality rainfall data set was that the frequency distribution of daily rainfalls at the station was reasonably smooth and peaked at low totals (Lavery et al., 1992). Therefore, this should not present a major difficulty for this study. Nevertheless, it is prudent to adopt 1 mm as the lower threshold for a rainday, in order to minimize any artificial suppression of rainday frequency by the undetected under-reporting of small rainfall amounts at any station. The problem of a changing number of raindays at many stations introduces difficulties into the formulation of rainfall indices. At a station with a trend in the number of raindays, the meaning of the extreme intensity and extreme percent indices will change depending on the method used to calculate the indices. Changes in the number of raindays over the past century are common. Figure 1 shows the location of two such stations: Goomalling and Collarenebri. The highly significant reduction in the number of raindays at Goomalling of over 23 days per 100 years (pB0.001), is shown in Figure 2. This trend is typical of many stations in southwestern Australia (Suppiah and Hennessy, 1998) and is more pronounced in the latter part of the record since about 1960. Similarly, the strong increase in the number of raindays at Collarenebri of over 8 days per 100 years (pB0.06), (not shown), is typical of many stations in eastern Australia. The extreme intensity can be sensitive to changes in the number of raindays, depending on the method used in its calculation. Figure 3 shows this index calculated at Goomalling using two methods: the solid line is the average of the four highest events; the dashed line is the average of the highest 5% of raindays.

Figure 2. Number of days with rainfall \1 mm for Goomalling with linear trend. Data are missing in 1910 and 1992

Figure 3. Average intensity of the four highest events (solid) and highest 5% of events \1mm (dashed) for Goomalling with linear trends

The difference in the two approaches is apparent in the linear trends of the two series. The index calculated using the highest 5% of raindays has a stronger positive trend. The reason for the discrepancy can be explained by considering a rainfall distribution whose shape remains constant over time but whose population (raindays) decreases. As the population decreases, the 4th highest event would be expected to decrease. This is shown in Figure 4 where the shape of the distribution has remained the same but the population has halved. The shaded area represents the top four events in both curves. The lower limit of the two shadings on the x-axis indicates the changing value of the 4th highest event in the two distributions. Therefore, at a station like Goomalling where the number of raindays has decreased, if the rainfall distribution has not changed shape then the average of the four highest events will have a more negative trend than the average of the highest 5% of raindays.

Figure 4. Relative frequency of daily rainfall at an idealized station showing effects of halving all frequencies. Shaded area represents four highest events

One important consequence of calculating the extreme intensity using only raindays is that a reduction (increase) in the number of days with low rainfall will lead to an increase (decrease) in this index. This is because the number of raindays will change which will affect the threshold used in the calculation of the index. This is as would be expected if interpreted in terms of the change in the rainfall distribution. With a reduction in the population but no change in the frequency of the higher events, the upper percentiles would be expected to rise. This is not the case if the extreme intensity is calculated using all days. One further problem introduced with the calculation of extreme percentiles using raindays is that a sufficient number of raindays must be present to accurately determine the value of the percentiles. A cut-off of 15 days was used as the minimum number of raindays for which the 95th percentile could be calculated. Three of the 94 stations contained less than 15 raindays for more than 10% of the years and were therefore rejected. For the remaining stations, if a year contained less than 15 raindays, then the value for the extreme index was not used in the calculation of the regional average of the extreme index for that year. This constraint on the minimum number of raindays also meant that only annual results could be considered. Although it would be desirable to include results for the summer and winter periods, many stations did not meet this constraint on the minimum number of raindays when seasons were considered (particularly during the dry season). The 95th percentile calculated from only raindays provides a reasonable cut-off for extreme rainfall. This percentile was chosen to enable a comparison with the indices calculated with the top four events for each year. The 95th percentile generally represented the top three to five events, which contributed around 20–25% of the total rainfall. A similar study could be carried out using the 90th percentile. This would be focusing on the top five to ten annual events, which contributed around 30–40% of the total rainfall. The 95th percentile represents rainfall that is considerably more extreme than the 90th percentile. Regional averages have been calculated as a linear average of the stations in that region and therefore are biased towards areas where the station density is highest. While this does not represent a true areal average, it removes the possibility that an isolated station will dominate the calculation, as can happen when using Thiessen polygons to weight stations in the areal-averaging process. A linear average also removes the need to define an area for each region, which is necessary for areal-averaging methods (e.g. Hennessy et al., 1999). All trends have been calculated using a least-squares linear regression with statistical significance determined using the Kendall-tau test (Press et al., 1986). Where a trend is indicated as ‘significant’, it has at least 95% significance using this test.

4. RESULTS AND DISCUSSION

Time series of the total rainfall, the number of raindays with at least 1 mm rainfall, and the extreme indices have been calculated for each of the four regions for the period May 1910 to April 1998. These were calculated for each year for the months May to April of the following year, rather than January to December so as not to split the season of high summer rainfall in northern Australia. Starting in May also ensures that the season of high winter rainfall in southern Australia is not split. A summary of the trends in total rainfall and number of raindays is shown in Table I. The trends in the extreme indices are summarized in Table II. Table I shows an increase in the total rainfall and number of raindays in the northern and southern regions and a decrease in the western and southwestern regions. Both the total rainfall and number of raindays have decreased significantly in the southwest while only the number of raindays has increased significantly in the north. These changes, as noted in previous studies (e.g. Hennessy et al., 1999), imply a significant change in the number of heavy events in the southwest while a significant change in only the number of lighter events in the north. This is confirmed in Table II, which shows the trends for the extreme indices. The number of extreme events (extreme frequency) in the southwest has strongly decreased. In the north the increase in the

Table I. Trends in total rainfall and the number of raindays for the four regions

Region Total rainfall Raindays (mm/100 years) (days/100 years)

North 87.33 9.18 South50.17 6.39 West −67.79 −6.55 Southwest −185.07 −12.73

Trends in bold are significant at the 95% level.

Table II. Trends in extreme intensity, extreme percent and extreme frequency calculated using three different methods for the four regions

Region Annual top four events Annual top 5% raindays \Long-term 95th percentile

ExtremeExtremeExtreme Extreme Extreme Extreme Extreme intensitypercent intensitypercent intensity percent frequency (mm/100 (%/100(mm/100(%/100 (mm/100 (%/100 (days/100 years) years) years) years) years) years) years)

North3.80 −2.60 2.28 1.35 2.71 1.48 0.58 South 0.65 −2.62 −0.30 −0.89 0.62 −1.41 0.20 West −1.17 1.94 −0.47 0.41 −1.17 0.55 −0.41 Southwest −4.66 1.05 −3.21 −0.54 −4.46 −4.03 −1.69

Trends in bold are significant at the 95% level. number of heavy events is not significant. Therefore, the significant trend in raindays in the northern region is largely reflecting changes in the number of lighter events. In most cases, the sign of the trend in the extreme intensity index matches the trend in raindays and total rainfall. The only significant change is in the southwest but strong changes have also occurred in the north. The effect of the changing number of raindays is evident in a comparison of the extreme intensity index calculated using the three different methods. In most cases, the trends are most similar between the index calculated from the four highest events and the events exceeding the long-term 95th percentile. The trend is generally strongest when the index is calculated using the four highest events. In the northern and southwestern regions where there has been significant changes in the number of raindays, the intensity is changing in the top four events as well as the events above an extreme threshold and in the highest 5% of raindays. The extreme percent shows changes of less than 5% per 100 years over most of the country. The fact that there have been significant decreases in this index calculated from the highest four events in the northern and southern regions is more an indication of the low interannual variability of this index in these regions. The sign of the trends in this index are generally opposite to that of the total rainfall. One interesting result in this index is in the southwest where a significant decrease in the extreme intensity divided by a significant decrease in the total rainfall results in weak and non-significant trends of mixed sign in the extreme percent, depending on the method used to calculate the index. Since the extreme percent is the extreme intensity/total rainfall, this implies that the proportional change in the intensity of the extreme events is similar to that of total rainfall. On the other hand, in the northern region where both total rainfall and extreme intensity are increasing, the significant decline in the extreme percent calculated from the highest four events suggests that the total rainfall is increasing at a proportionally greater rate. The problems associated with determining statistical significance of trends of series that are averages of other series have been addressed using an alternative test of significance. When averaging several series there is a possibility that series with a higher mean or variance will dominate the calculations. Therefore, each component series was normalized by replacing each value in the series with its rank, then applying

Table III. Correlations with total rainfall

Region Raindays Annual top four events Annual top 5% \Long-term 95th percentile raindays

ExtremeExtremeExtremeExtreme Extreme Extreme Extreme intensitypercent intensitypercent intensity percent frequency

North 0.93 0.88 −0.68 0.76 0.22 0.65 0.74 0.95 South 0.94 0.73 −0.65 0.53 0.01 0.45 0.47 0.88 West 0.92 0.69 −0.47 0.42 0.08 0.32 0.47 0.87 Southwest 0.89 0.70 −0.33 0.62 0.27 0.30 0.58 0.81

Numbers in bold indicate 95% significance. the Kendall-tau test to the average of the ranked series. All trends in Tables I and II which were found to be significant using the Kendall-tau test on the average of the raw series are also significant using this ‘rank’ test. Correlations with total rainfall for the number of raindays and the three extreme indices are shown in Table III. Highly significant correlations for raindays and the extreme frequency and extreme intensity suggest that years with high rainfall receive rain on more days, with a higher average intensity in the highest events and a larger number of events above an extreme threshold. Correlations between total rainfall and the extreme percent index are significant for two of the three calculation methods, with non-significant correlations between total rainfall and the extreme percent calculated using the highest 5% of raindays. One point of interest is the significant negative correlations between the total rainfall and the extreme percent calculated from the highest four events while the correlations with the extreme percent calculated from events exceeding the long-term 95th percentile are significantly positive. This suggests that years with higher rainfall receive a larger proportion of their rainfall from events above the long-term 95th percentile and less from the top four events. This is probably more from there being a larger number of events above the 95th percentile during wet years as shown by the high correlations between the total rainfall and the extreme frequency index. One question that arises is which is the best index of extreme rainfall intensity and how should it be calculated? The best guide of index design must be the final purpose of the index. If the aim is to use the index for climate change detection, then a complex index can be considered. On the other hand, an index for use by the public should be as clear and simple as possible. Explaining to a farmer that the proportion of annual rainfall from the highest 5% of events has increased but the actual number of events has decreased may be confusing. An index such as the amount of rain from the top four events is much clearer. However, if the desire is to find an index that reflects changes in the shape of a frequency distribution, then an index such as the average intensity of the highest 5% of events may be better. For analysis of shapes of distributions, parametric approaches might also be considered (e.g. Groisman et al., 1999).

5. CONCLUSION

As governments and other decision-makers begin to consider climate change when formulating policy, there is a strong need for clear and concise indices for monitoring of the environment. With analyses from global circulation models suggesting an increase in heavy rainfall and a decrease in light rainfall (e.g. Gordon et al., 1992), the challenge has been for scientists to determine if such a change exists in the historical record. Decoding such a signal from a record that not only contains high interannual variation but also includes possible trends in total rainfall and in the number of raindays poses many problems. The varying quality of daily rainfall data also poses problems for analyses of extreme rainfall. The three indices proposed in this study, which examine changes in both the intensity and frequency of extreme

Copyright © 2000 Royal Meteorological Society Int. J. Climatol. 20: 1533–1541 (2000) AUSTRALIAN EXTREME RAINFALL TRENDS 1541 events, as well as their contribution to the total rainfall, have been formulated to overcome these problems and calculated using a consistent set of high-quality stations with little missing data. These indices could be applied to other regions with such data.

ACKNOWLEDGEMENTS This study was partly funded by the State of the Environment Reporting Unit of Environment Australia. Comments from Tahl Kestin, Kevin Hennessy, Blair Trewin, Dean Collins and an anonymous reviewer were greatly appreciated.

REFERENCES Allan RJ, Haylock MR. 1993. Circulation features associated with the winter rainfall decrease in south-western Australia. Journal of Climate 7: 1356–1367. Gordon HB, Whetton PH, Pittock AB, Fowler AM, Haylock MR. 1992. Simulated changes in daily rainfall intensity due to the enhanced greenhouse effect: implications for extreme rainfall events. Climate Dynamics 8: 83–102. Groisman PY, Karl TR, Easterling DR, Knight RW, Jamason PF, Hennessy KJ, Suppiah S, Page C, Wibig J, Fortuniak K, Razuvaev VN, Douglas A, Forland E, Zhai P. 1999. Changes in the probability of heavy precipitaion: important indicators of climatic change. In Weather and Climate Extremes, Karl TR, Nicholls N, Ghazi A (eds). Kluwer Academic Publishers; 243–283. Hennessy KJ, Suppiah R, Page CM. 1999. Australian rainfall changes, 1910–1995. Australian Meteorology Magazine 48: 1–13. Karl TR, Knight RW. 1998. Secular trends of precipitation amount, frequency, and intensity in the United States. Bulletin of the American Meteorological Society 79: 231–241. Karl TR, Derr VE, Easterling DR, Folland CK, Hofmann DJ, Levitus S, Nicholls N, Parker DE, Withee GW. 1995a. Critical issues for long-term climate monitoring. Climatic Change 31: 185–221. Karl TR, Knight RW, Plummer N. 1995b. Trends in high-frequency climate variability in the twentieth century. Nature 377: 217–220. Lavery B, Kariko A, Nicholls N. 1992. A historical rainfall data set for Australia. Australian Meteorology Magazine 40: 33–39. Nicholls N. 1995. Long-term climate monitoring and extreme events. Climatic Change 31: 231–245. Nicholls N, Gruza GV, Jonzel J, Karl TR, Ogallo LA, Parker DE. 1996. Observed climate variability and change. In Climate Change 1995, Houghton JT, Meira Filho LG, Callander BA, Harris N, Kattenberg A, Maskell K (eds). Cambridge University Press: Cambridge; 133–192. Plummer N, Salinger MJ, Nicholls N, Suppiah R, Hennessy KJ, Leighton RM, Trewin B, Lough JM. 1999. Twentieth century trends in climate extremes over the Australian region and New Zealand. Climatic Change 42: 183–202. Press WH, Flannery BP, Teukolsky SA, Vetterling WT. 1986. Numerical Recipes: The Art of Scientific Computing. Cambridge University Press: Cambridge; 488–493. Suppiah R, Hennessy KJ. 1996. Trends in the intensity and frequency of heavy rainfall in tropical Australia and links with the Southern Oscillation. Australian Meteorology Magazine 45: 1–17. Suppiah R, Hennessy KJ. 1998. Trends in total rainfall, heavy-rain events and number of dry days in Australia, 1910–1990. International Journal of Climatology 10: 1141–1164.

TRENDS IN EXTREME DAILY RAINFALL AND TEMPERATURE IN SOUTHEAST ASIA AND THE SOUTH PACIFIC: 1961–1998 M.J. MANTONa,*, P.M. DELLA-MARTAb, M.R. HAYLOCKa, K.J. HENNESSYc, N. NICHOLLSa, L.E. CHAMBERSa, D.A. COLLINSb, G. DAWd, A. FINETe, D. GUNAWANf, K. INAPEg, H. ISOBEh, T.S. KESTINi, P. LEFALEj, C.H. LEYUk, T. LWINl, L. MAITREPIERREm, N. OUPRASITWONGn, C.M. PAGEc, J. PAHALADo, N. PLUMMERb, M.J. SALINGERd, R. SUPPIAHc, V.L. TRANp, B. TREWINb, I. TIBIGq and D. YEEr a Bureau of Meteorology Research Centre (BMRC), Australia b Bureau of Meteorology National Climate Centre (NCC), Australia c Commonwealth Scientific and Industrial Research Organisation (CSIRO) Atmospheric Research, Australia d National Institute of Water and Atmospheric Research (NIWA), New Zealand e Me´te´o-France, French Polynesia f Meteorological and Geophysical Agency, Indonesia g National Weather Ser6ice, Papua New Guinea h Japan Meteorological Agency, Climate Prediction Di6ision, Japan i Cooperati6e Research Centre for Southern Hemisphere Meteorology, Monash Uni6ersity, Australia j South Pacific Regional En6ironment Program, Samoa k Malaysian Meteorological Ser6ice, Malaysia l Department of Meteorology and Hydrology, Myanmar m Me´te´o-France, New Caledonia n Meteorological Department, Thailand o Fiji Meteorological Ser6ice, Fiji p Hydrometeorological Ser6ice, Vietnam q PAGASA, Philippines r Solomon Islands Meteorological Ser6ice, Solomon Islands

Recei6ed 28 February 2000 Re6ised 22 August 2000 Accepted 31 August 2000

ABSTRACT Trends in extreme daily temperature and rainfall have been analysed from 1961 to 1998 for Southeast Asia and the South Pacific. This 38-year period was chosen to optimize data availability across the region. Using high-quality data from 91 stations in 15 countries, significant increases were detected in the annual number of hot days and warm nights, with significant decreases in the annual number of cool days and cold nights. These trends in extreme temperatures showed considerable consistency across the region. Extreme rainfall trends were generally less spatially coherent than were those for extreme temperature. The number of rain days (with at least 2 mm of rain) has decreased significantly throughout Southeast Asia and the western and central South Pacific, but increased in the north of French Polynesia, in Fiji, and at some stations in Australia. The proportion of annual rainfall from extreme events has increased at a majority of stations. The frequency of extreme rainfall events has declined at most stations (but not significantly), although significant increases were detected in French Polynesia. Trends in the average intensity of the wettest rainfall events each year were generally weak and not significant. Copyright © 2001 Royal Meteorological Society.

KEY WORDS: Australasia; climate change; climate extremes; precipitation; Southeast Asia; South Pacific; temperature

1. INTRODUCTION

Extreme weather or climate events can have major impacts on society, the economy and the environment. Karl et al. (1999) assessed changes in climate extremes over many parts of the world during the past century. They reported a reduction in the number of extremely cold days, including fewer frosts and freezes, and an increase in the number of extremely hot days. As well, minima had increased more rapidly

* Correspondence to: Bureau of Meteorology, PO Box 1289K, Melbourne, Victoria 3001, Australia; e-mail: [email protected]

Copyright © 2001 Royal Meteorological Society M.J. MANTON ET AL. than maxima. Extreme precipitation had increased in the US, China, Australia, Canada, Norway, Mexico, Poland and the Former Soviet Union (Groisman et al., 1999). No clear tropics-wide trends have emerged in the number of tropical storms; Nicholls et al. (1998) found a slight increase in the number of intense tropical cyclones in the Australian region since 1969, while Landsea et al. (1996) reported a decline in the number of intense Atlantic hurricanes over a similar period. There is little evidence of a change in extra-tropical storms, but only a limited amount of data have been analysed. Fewer studies have examined trends in climate extremes, other than changes in mean values, largely a result of the extra demands this places on data quality and quantity. Even a relatively small amount of missing data, whilst not necessarily affecting the mean significantly, immediately raises the possibility that an extreme event has been missed. Also, when investigating trends in the extreme ends of a climatic distribution, the likelihood of complications resulting from erroneous data is increased because outliers can be incorrectly considered as true data extremes (or genuine extremes may be rejected as outliers). Applying consistent analyses across a wide region also requires that the criteria used to define extreme events are meaningful across the entire region. What is considered extreme in one part of the region might be considered quite routine in another. Further data issues in monitoring extreme events are discussed in Nicholls (1995). Little is known about trends in extreme temperature or rainfall in the Asia-Pacific region. To some extent, this is because the region covers a broad range of countries, some of which have poor data availability, quality and consistency. The region is particularly vulnerable to changes in climate extremes as a result of its generally high population density, exposure to tropical cyclones, strong links between rainfall and the El Nin˜o–Southern Oscillation, low-lying islands, coral reefs, and fire-prone environments. In addition, the Interdecadal Pacific Oscillation (Power et al., 1999) is an important source of climate variation in the region. In 1998, the Asia-Pacific Network (APN) for Global Change Research funded a workshop on climate extremes, hosted by the Australian Bureau of Meteorology Research Centre (BMRC). A major aim of the workshop was to encourage regional participation in international studies on monitoring and detecting changes in climate extremes (Manton and Nicholls, 1999). The meeting included representatives from 14 countries: Australia, Fiji, French Polynesia, Indonesia, Japan, Malaysia, Myanmar, New Caledonia, New Zealand, Papua New Guinea, the Philippines, Samoa, Thailand and Vietnam. A significant outcome of the workshop was the desire of all participants to contribute to an analysis of Asia-Pacific trends in extreme climate for the Third Assessment Report of the Intergovernmental Panel on Climate Change (IPCC). It was intended that this analysis should be consistent and relevant across the region, and simple to calculate. This led to a second workshop in December 1999, also funded by the APN and hosted by BMRC, attended by representatives from the same 14 countries plus the Solomon Islands. At this workshop, extreme rainfall and temperature trends were analysed by applying common quality control and analysis techniques across the region. While other variables could have been considered, rainfall and temperature were judged in the 1998 workshop as being the variables with the longest and most reliable records, and the most relevant for comparison with other international studies. This paper reports the results of the analyses conducted as a result of this second workshop. Quality control and analysis techniques are described in the next section, followed by trend results in the third section, and a discussion of the results in the fourth.

2. DATA ANALYSIS METHODS

2.1. Climate station selection Representatives from the national climate service of each country participating in the workshop provided daily rainfall and maximum and minimum temperature data for a small number of stations in their country. The stations satisfied the following criteria:

The records were as long as possible, and included the standard reference period of 1961–1990. Less than 20% of the daily values were missing in each year. The stations were of high-quality, preferably non-urban, and well maintained. The station, in most cases, had a documented history of changes such as those involving instrumentation, observation practices and the station’s immediate environment (metadata). In most cases, the station had been located at a single site during the period of record. Whilst participants were free to choose the stations that they felt best met these criteria, obvious candidates were those stations already selected for inclusion in the country’s Reference Climate Station (RCS) network or the Global Climate Observing System (GCOS) Surface Network (GSN) (Peterson et al., 1997). The primary purpose of such stations is to identify long-term climate trends and, consequently, these stations will have been chosen for having long, continuous and homogeneous records, for having minimal influence from urbanization, and for generally being high-quality stations.

2.2. Quality control Inhomogeneities or discontinuities in a climate record can be caused by any change to the station or its operation, including site location, exposure, instrumentation, or observational practice. These discontinuities will not only affect mean climatic values, but also the extremes of the climatic distribution, and may affect the extremes differently to the mean (Trewin and Trevitt, 1996). Numerous studies have used procedures such as visual examination of data, neighbouring station checks, and statistical tests to identify and adjust for inhomogeneities in seasonal or annual mean temperature and total rainfall (e.g. Lavery et al., 1992, 1997; Torok and Nicholls, 1996; Peterson et al., 1998). A few studies (e.g. Trewin, 2000) have made adjustments at the daily time-scale and allowed for different magnitudes of discontinuity at different parts of the temperature or rainfall distribution. Ideally, this would have been done for the climate records used in this analysis, but this was beyond the scope of the workshop. However, attempts were made to minimize the influence of inhomogeneities on the results of the analysis. The daily time-series from each station were first examined visually to identify any obvious outliers, trends and potential discontinuities. Annual mean series (annual total for rainfall) were then produced from the daily rainfall and maximum and minimum temperature time-series, and examined for discontinuities using a software package known as Multiple Analysis of Series for Homogenization (MASH) (Szentimrey, 1997). The MASH technique calculates the difference between series at a candidate station and at a number of climatically-similar reference stations to identify statistically significant shifts in the annual mean values of the candidate series. Potential break points were identified using MASH, and compared with available metadata. MASH quantifies the adjustment needed to create a homogeneous annual mean series. However, applying these adjustments to all daily data during a year is not an appropriate method of producing a homogeneous daily data series. The magnitude of the discontinuity will vary at different times of the year, and it is not known how an adjustment for the mean should be translated into an adjustment for the more extreme values in the distribution. If no major discontinuities were detected for a station, then it was accepted for analysis. If a discontinuity could be related to some change suggested by the station metadata, then the station was rejected. If a discontinuity could not be related to metadata, it was assumed that the climatic shift was real. The results presented here are derived from the most homogeneous data identified at the workshop. Finally, only stations where each year of record had at least 308 days of data available were included. This criterion equates to a probability of at least 50% that all of the four most extreme events for that year would be present in the data set. Based on the homogeneity testing and quality control, data for a set of 91 high-quality stations (Appendix A and Figure 1) were prepared for analysis. As most of the stations have data from 1961–1998, this period was used to investigate trends in extremes. A small proportion of stations did not have data for the complete period (see Appendix A). This should be borne in mind when the results of the analysis of trends are discussed. It should also be remembered that although we believe the data set

Figure 1. Locations of stations (*) used in this study used here is the best available, at the present time, other unidentified inhomogeneities probably exist in these data. Future work may refine the data set further.

2.3. Analysis methods Numerous extreme rainfall and temperature indices have been used in previous studies. Some indices involved arbitrary thresholds, such as the number of days each year with daily rainfall exceeding 25.4 mm or 50.8 mm (Groisman et al., 1999), or the number of winter days below 0°C (Jones et al., 1999). These are suitable for regions with little spatial variability in climate, but arbitrary thresholds are inappropriate for regions spanning a broad range of climates. For example, in a country like Australia that covers latitudes 12°–44°S, there is no single temperature or rainfall threshold that would be considered extreme in all regions. For this reason, some studies have used extreme indices based on statistical quantities such as the 10th or 90th percentile (e.g. Plummer et al., 1999). These upper and lower percentiles are extreme in all regions, but vary in absolute magnitude from site to site. For example, the 99th percentile of daily rainfall exceeds 100 mm in summer and autumn over north-eastern Australia, but falls below 30 mm in the south (Hennessy et al., 1999). As this study covers a broad region, extreme climate indices were based on the 1st and 99th percentiles. The percentiles were computed using all non-missing days. For rainfall, this included rain days and dry days. As there are 365 days in a year, the 1st percentile is the 4th lowest value, and the 99th percentile is the 4th highest. We have chosen the following eight indices of extremes: Frequency of daily rainfall exceeding the 1961–1990 mean 99th percentile (extreme frequency). Average intensity of events greater than or equal to the 99th percentile each year, i.e. in the four wettest events (extreme intensity). Percentage of annual total rainfall from events greater than or equal to the 99th percentile, i.e. received in the four wettest events (extreme proportion). Frequency of days with at least 2 mm of rain (rain days). Frequency of days with maximum temperature above the 1961–1990 mean 99th percentile (hot days). Frequency of days with minimum temperature above the 1961–1990 mean 99th percentile (warm nights). Frequency of days with maximum temperature below the 1961–1990 mean 1st percentile (cool days). Frequency of days with minimum temperature below the 1961–1990 mean 1st percentile (cold nights). Other indices, e.g. total rainfall, the 5th and the 95th percentile, were also calculated. Where appropriate, the results from these other indices are discussed.

The threshold of 2 mm rainfall in the definition of ‘rain days’ was used to avoid artificial trends. Such trends can arise from a tendency of some observers to fail to report small rainfall amounts (Lavery et al., 1992). Also, the threshold for reporting rainfall can alter with changes in observing equipment or observer instructions (Nicholls and Kariko, 1993). Groisman et al. (1999) also used a threshold to select ‘rain days’ to avoid such problems. An annual time-series of each index was computed for each station, without removing the seasonal cycle of temperature or rainfall. This means that changes in the hot day index will tend to be representative of changes in the hottest season, so changes in hot days during other seasons are effectively excluded. Hence, the indices convey information about events with the most extreme magnitude each year. From a practical viewpoint, the indices target events that, presumably, would have the greatest impact. If we had removed the seasonal cycle before computing indices, our time-series would have shown trends in extremes from seasons that are not the hottest, coldest or wettest for the year. These would be difficult to interpret in terms of practical impacts, although such an approach would have the advantage of increasing the sample size.

3. EXTREME RAINFALL AND TEMPERATURE TRENDS

3.1. Regional patterns of trends Before describing the direction and magnitude of trends at each station in each country, we present an overview of the geographical pattern of trends in each index. This provides a clear picture of the spatial variation in the direction of trends and their statistical significance. The Kendall-tau test was used to test significance of the trends. All trends noted as significant were significant at the 95% level. 3.1.1. Rainfall. Annual total rainfall in the region has generally decreased between 1961 and 1998 (not shown). This decrease is associated with the predominance of El Nin˜o events since the mid-1970s (Trenberth and Hoar, 1997). Total rainfall only increased significantly in southern New Zealand and some stations in French Polynesia. The number of rain days (with at least 2 mm of rain) has decreased significantly throughout most of Southeast Asia, and the western South Pacific (Figure 2). Trends in the extreme rainfall indices were less spatially coherent than were those for rain days and (as is shown later) extreme temperature. The frequency of extreme rainfall events has declined at the majority of stations, although significant increases in this index were detected in French Polynesia (Figure 3). There

Figure 2. Trend in the frequency of days with at least 2 mm of rain (rain days). The sign of the linear trend is indicated by +/− symbols at each site; bold indicates significant trends (95%). Data from 1961–1998

Figure 3. Trend in the frequency of daily rainfall exceeding the 1961–1990 mean 99th percentile (extreme frequency). The sign of the linear trend is indicated by +/− symbols at each site; bold indicates significant trends (95%). Data from 1961–1998 has been a weak decline (i.e. not statistically significant) in the average intensity of the highest four events each year over much of Southeast Asia (Figure 4). Weak increases in extreme intensity were found at some stations in Australia, Fiji, New Zealand and New Caledonia. Significant increases in extreme intensity were found at a couple of stations in French Polynesia and Japan. The proportion of annual rainfall from extreme events has generally increased, with a few stations exhibiting a significant increase (Figure 5).

3.1.2. Temperature. The number of hot days has increased at most stations (significantly at several stations), but there were significant decreases in northern Australia (Figure 6). Warm nights increased in frequency almost everywhere, and generally these increases were significant (Figure 7). Weak (not statistically significant) declines were found in southern Australia and New Zealand. Cool days (Figure 8) and cold nights (Figure 9) declined in frequency, with a few exceptions (almost entirely in Australia and New Zealand).

Figure 4. Trend in the average intensity of the events greater than or equal to the 99th percentile (extreme intensity). The sign of the linear trend is indicated by +/− symbols at each site; bold indicates significant trends (95%). Data from 1961–1998

Figure 5. Trend in the percentage of annual total rainfall from the events greater than or equal to the 99th percentile (extreme proportion). The sign of the linear trend is indicated by +/− symbols at each site; bold indicates significant trends (95%). Data from 1961–1998

Figure 6. Trend in the frequency of days with maximum temperature above the 1961–1990 mean 99th percentile (hot days). The sign of the linear trend is indicated by +/− symbols at each site; bold indicates significant trends (95%). Data from 1961–1998

As there is considerable spatial coherence in the sign of the temperature trends (with the notable exceptions being Australia and New Zealand), country-average trends were calculated, to allow comparison of the magnitudes of the trends. Figure 10 shows a larger decrease in the number of cold nights than cool days, except in the Philippines. Fiji is the only country with an increase in the number of either cool days or cold nights. Figure 11 shows that there has been a stronger increase in the number of warm nights than hot days except in the Philippines, Fiji, New Zealand and French Polynesia. Neighbouring countries exhibit considerable consistency in the magnitude, as well as the sign, of the trends. The spatial consistency of the trends exhibited in Figures 6–11 encouraged us to calculate simple regional averages of the trends in the numbers of hot/warm and cool/cold days and nights, by calculating arithmetic averages, for each year, of all stations. Area-weighting was considered inappropriate because much of the area is ocean and may not be well-represented by land-based stations. The time-series of these

Figure 7. Trend in the frequency of days with minimum temperature above the 1961–1990 mean 99th percentile (warm nights). The sign of the linear trend is indicated by +/− symbols at each site; bold indicates significant trends (95%). Data from 1961–1998

Figure 8. Trend in the frequency of days with maximum temperature below the 1961–1990 mean 1st percentile (cool days). The sign of the linear trend is indicated by +/− symbols at each site; bold indicates significant trends (95%). Data from 1961–1998 regional average indices are shown in Figure 12. This figure indicates that the decline in the numbers of cold days and cool nights, and the increase in the number of hot days and warm nights, has been reasonably linear (although with considerable year-to-year variability) through the 1961–1998 period. There was a dramatic increase in hot days and warm nights across the region in the last year examined (1998), coinciding with the globe’s warmest year in the period of instrumental records. The linear trends in all four time-series are statistically significant. The number of cold nights and cool days have been reduced by about half, while the numbers of hot days and warm nights have increased by a factor of about 2–3.

3.2. Trends for each country The maps and charts presented above summarize some interesting trends across the region. In this section, we briefly discuss the changes that have occurred in each country.

Figure 9. Trend in the frequency of days with minimum temperature below the 1961–1990 mean 1st percentile (cold nights). The sign of the linear trend is indicated by +/− symbols at each site; bold indicates significant trends (95%). Data from 1961–1998

Figure 10. Bar chart showing the trends in the number of cool days and cold nights, for each country. The countries are exhibited in a general west–east orientation, so the results can be compared with neighbouring countries

3.2.1. Australia. No spatially-consistent pattern of trends in the rainfall extremes indices emerged from the Australian stations. The only statistically-significant trends were a decrease in the proportion of rainfall from the wettest four events (extreme proportion) at Deniliquin in the southeast, and an increase at Mardie in the northwest. These results contrast somewhat with those reported by Hennessy et al. (1999) for a longer period starting in 1910. They found significant increases in the 99th percentile daily rainfall in parts of southeast Australia. This emphasizes that the trend results are influenced by the sampling period. The all-Australian average of trends in hot days was close to zero, owing to a mixture of positive and negative trends throughout the country. The largest positive trends for this index were at Bathurst and Port Macquarie in the southeast, with the trend at Port Macquarie significant at the 5% level. Significant negative trends were found at Kalumburu and Burketown, both in northern Australia. Trends in warm nights were generally positive throughout Australia and, consequently, the average for the whole country showed an increasing trend. Burketown, Barcaldine and Port Macquarie showed significant positive trends for this index.

Figure 11. As in Figure 9, but for hot days and warm nights

The frequency of cool days showed predominantly decreasing trends, but all trends were non-significant. Consequently the all-Australian average showed a decreasing trend for this index. The strongest and most consistent trends were in the frequency of cold nights with a decrease in the overall Australian average. Five stations (Wandering, Burketown, Barcaldine, Port Macquarie and Bathurst) recorded significant negative trends. 3.2.2. Fiji. Rotuma and Ono-I-Lau exhibited significant increases in the proportion of annual rainfall falling in the extreme events (‘extreme proportion’). Labasa Mill had a significant decrease in the number of rain days. There was a significant upward trend in warm nights and hot days for Laucala Bay, with a distinct rise in the trend from 1971. The frequency of hot days and warm nights rose sharply from the late 1980s. 3.2.3. French Polynesia. In Atuona, in the Marquesas Islands, there has been a significant increase in total rainfall, in the frequency of extreme rainfall, and in the number of rain days. Annual total rainfall increased by more than 50% after 1976. This climate shift, which at first sight appears to reflect an inhomogeneity, is confirmed at neighbouring stations. Elsewhere in French Polynesia, no extreme rainfall trends were significant. Significant increase in warm nights and decreases in cold nights were found at all stations. Significant increases in the number of hot days and decreases in the number of cool days were also detected at all stations except Rapa, where the trends were not significant. 3.2.4. Indonesia. There were no significant trends in any of the extreme rainfall indices in Indonesia, and homogenous temperature data were not available. 3.2.5. Japan. In general, rain days have decreased and the extreme rainfall indices have increased. Positive trends in the ‘extreme intensity’ and ‘extreme proportion’ indices were significant at one station in northern Japan. The frequency of cool days and cold nights has decreased. Negative trends in the frequency of cold nights were significant at most stations. The frequency of hot days and warm nights has increased. Significant positive trends have occurred in the frequency of warm nights at stations in southern Japan. 3.2.6. Malaysia. There has been a significant decrease in rain days at all stations, except Kuching. There were no other significant trends in extreme rainfall indices. There was a significant positive trend in the number of warm nights, with a large peak in 1998. Likewise, there exists a significant decrease in the frequency of cool days and cold nights.

Figure 12. Time-series of the regional averages of the frequency of hot and cold days and nights. The thin line is a trend-line computed by linear regression

3.2.7. Myanmar. There were no significant trends in extreme rainfall indices. At the single temperature station, warm nights have significantly increased in frequency. Cold nights have decreased in frequency. There are no significant trends in maximum temperature (hot days or cool days).

3.2.8. New Caledonia. For the eight stations analysed, there were no significant trends in extreme rainfall indices. Of the eight stations, seven exhibited (non-significant) increases in the ‘extreme proportion’ index and a (non-significant) decrease in the ‘extreme frequency’ index. There has been a (non-significant) decrease in the number of rain days on the west coast, and a (non-significant) increase on the east coast. Significant changes in the extreme temperature indices were found at all three stations. The clearest trend is for the number of warm nights with significant increases at all stations. Two stations had significant increases in the number of hot days, and significant decreases in the number of cool days. Only

Koumac, in the north of the country, had a significant decrease in the number of cold nights, although the other stations also exhibited decreases (but not significant).

3.2.9. New Zealand. Annual rainfall has decreased (non-significantly) in the two North Island stations (Ruakura and Gisborne), and increased in Invercargill (significantly) and Hokitika (non-significantly). A significant increase in the extreme frequency has occurred at Holitika in the west of the South Island. This rainfall pattern for the period 1961–1998 is consistent with an observed circulation change over New Zealand since 1976, to more westerly winds over southern New Zealand, and higher pressures over the North Island (Salinger and Mullan, 1999). Somewhat consistent with these circulation changes has been an increase in the North Island hot days at Gisborne, a location sheltered from the west. The frequency of hot days has significantly increased, and there has been a significant decrease in the frequency of cool days there. In contrast, there has been a significant decrease in hot days at Hokitika in the west. No significant signal is observed for cold nights there, however. Invercargill in the south shows no trends in extreme temperatures at all. However, Ruakura in the north shows a significant decrease in the frequency of cold nights.

3.2.10. Philippines. There has been a significant decrease in the number of rain days at three of the five stations: Baguio, in the north of the archipelago, Daet, which is situated along the east coast, and Dumaguete, in the middle islands. At Basco, there were significant increases in the frequency of warm nights and hot days (particularly high in 1998) and a decrease in the number of cool days and cold nights. At Baguio, the frequency of warm nights and hot days also exhibited significant increases (with large values in 1983 and 1998), but a significant decrease in the number of cool days. At Tuguegarao (located in a valley), the frequency of cold nights has shown a significant decrease.

3.2.11. Solomon Islands. The number of rain days has decreased at all stations, with Honiara, Kira Kira and Munda having significant decreases. At Honiara, the proportion of annual rainfall from extreme rainfall has increased significantly.

3.2.12. Thailand. At Nan, the number of rain days has decreased significantly, and the proportion of total rainfall from extreme rainfall has increased significantly. Prachuap Khiri Khan shows a significant decrease in rain days. There was a significant increase in extreme minimum temperature at both stations, partly owing to a peak in 1998. The number of warm nights increased, and the number of cold nights and cool days decreased. Maximum temperature increased significantly at Nan, while the increases at Prachuap Khiri Khan were not significant.

3.2.13. Vietnam. Phulien showed no significant trends in any of the rainfall indices, although there were decreases in all indices except the percent of total rainfall from extreme events (which showed no trend).

4. DISCUSSION

The patterns of trends presented in the previous section show, for some of the indices, considerable spatial consistency within countries and across regions. For instance, almost all stations exhibit increases in the frequency of hot extremes, and decreases in cold extremes, with many of these trends being statistically significant. The country-by-country analyses show considerable consistency, even in the magnitude of the extreme temperature trends (as well as the sign), with their neighbours. This consistency extended to the relative magnitudes between, for instance, the trends in the frequency of hot days and warm nights: in the west of the region, the frequency of warm nights was increasing faster, whereas the frequency of hot days was increasing faster in the east. Even the dramatic peak in the frequency of hot days and warm nights

Copyright © 2001 Royal Meteorological Society Int. J. Climatol. (in press) SE ASIAN EXTREME RAINFALL TRENDS in 1998 was reported at many stations, in various countries. The extreme rainfall indices showed less spatial consistency, except for the tropics-wide decrease in the number of rain days. Few of the rainfall indices, apart from the number of rain days, showed statistically significant trends. The general consistency, and the similarity of trends in neighbouring countries, lends credibility to the overall trends. This, in turn, suggests that the homogeneity testing and other approaches used in selecting and testing of station data prior to analysis has resulted in a data set useful for analysing trends and variations in extremes across the region. Further improvements of the homogeneity testing would, nevertheless, be worthwhile. For instance, a concerted effort to adjust the daily data for inhomogeneities, similar to the approach used by Trewin (2000) for Australia, would lead to an improved data set for analysis. However, such an approach demands considerable resources (computing and human). In the short term, the necessary resources seem unlikely to become available for such an analysis to be applied to all countries in the region. However, it would be feasible to apply our techniques of homogeneity testing and extremes analysis to larger numbers of stations, in each country. This would provide a more detailed analysis for each country, and allow increased confidence in the reality of the trends reported here. It is anticipated that individual countries, as the necessary resources become available, will undertake this task. The relatively short time frame of 38 years used in this study is likely to produce trends that are sensitive to the sampling period. For example, trends in total rainfall, dry days and extreme events in Australia from 1910–1995 tend to be in the opposite direction to those for 1961–1998. For some countries in the region, appropriate data do not exist from before the late 1950s. So extending the analysis (across the entire region) backwards is not possible. Nevertheless, it would be worthwhile, in the countries for which this is possible, to extend the analysis backwards, to provide a better understanding of the decadal fluctuations in extremes prior to the period considered here. This study has demonstrated that a concerted, regional, cooperative effort can produce a regional data-base of high quality that is useful for estimating recent trends in extremes. Extension of this approach to other parts of the world, where little is currently known regarding trends in extremes, would be worthwhile. A limitation, however, of such studies is the low spatial density of stations with homogenous data. This limitation is largely a consequence of departures from standards in instrumentation and/or observational practice that, while often unavoidable, significantly reduce the number of stations with homogeneous data. Even if an inhomogeneity in a station’s data series, the collection (and proper management) of associated metadata can provide sufficient information to obtain the real climatic signal. The United Nations Framework Convention on Climate Change urges countries to address the deficiencies in climate observing networks and, as Karl et al. (1995) stress, adhere to the requirements for long-term climate monitoring. No attempt has been made here to consider possible causes of the observed trends in extremes. We do note, however, that the observed increase in hot days and warm nights, and the decrease in cool days and cold nights is consistent with a number of simulations with global climate models, driven by increasing greenhouse gas levels (e.g. Hennessy et al., 1998).

ACKNOWLEDGEMENTS The Asia-Pacific Network (APN) for Global Change Research generously funded the two workshops that led to the production of this paper. The Bureau of Meteorology Research Centre (BMRC), Melbourne, Australia, hosted both workshops.

APPENDIX A

List of high-quality stations, their location and period of record. Absence of a period of record for a specific variable indicates that the station was not used to analyse that variable.

Country Station Latitude Longitude Rainfall Temperature

Australia Birdsville 25.90°S 139.33°E 1910–1998 1957–1998 Palmerville 16.98°S 144.07°E 1910–1998 1957–1998 Burketown 17.73°S 139.53°E 1910–1998 1957–1998 Barcaldine 23.55°S 145.28°E 1910–1998 1957–1998 Port Macquarie 31.43°S 152.92°E 1910–1998 1921–1998 Bathurst 33.43°S 149.57°E 1910–1998 1921–1998 Deniliquin 35.55°S 144.93°E 1910–1998 1957–1998 Orbost 37.68°S 148.45°E 1910–1998 1957–1998 Mardie 21.19°S 115.98°E 1910–1998 Gunbalunya 12.33°S 133.06°E 1910–1998 Menzies 29.69°S 121.03°E 1910–1998 Fingal Forestry 41.64°S 147.97°E 1910–1998 Boyup Brook 34.15°S 116.20°E 1910–1998 Hobart 42.88°S 147.32°E 1944–1998 Carnarvon 24.88°S 113.67°E 1945–1998 Wandering 32.67°S 116.67°E 1957–1998 Forrest 30.83°S 128.12°E 1947–1998 Giles 25.03°S 128.28°E 1957–1998 Darwin 12.42°S 130.88°E 1941–1998 Kalumburu 14.28°S 126.63°E 1957–1998 Fiji Rotuma 12.50°S 177.05°E 1961–1998 Suva 18.15°S 178.45°E 1942–1998 1942–1998 Rarawai Mill 17.55°S 177.73°E 1925–1998 Ono-I-Lau 20.67°S 178.72°E 1961–1998 Labasa Mill 16.45°S 179.35°E 1931–1998 French Polynesia Rapa 27.62°S 144.33cW 1953–1998 1953–1998 Faaa 17.55°S 149.60°W 1957–1998 1957–1998 Atuona 9.80°S 139.03°W 1960–1998 1961–1998 Indonesia Pangkal Pinang 2.17°S 106.13°E 1961–1998 Jakarta 6.17°S 106.82°E 1961–1998 Balikpapan 1.28°S 116.83°E 1960–1998 Menado 1.53°S 124.92°E 1961–1998 Ambon 3.70°S 128.08°E 1961–1998 Palu 0.70°S 119.73°E 1961–1998 Japan Hamada 34.90°N 132.07°E 1961–1998 1961–1998 Kushiro 43.33°N 145.58°E 1961–1998 1961–1998 Hachijo Island 33.10°N 139.78°E 1961–1998 1961–1998 Yamagata 38.03°N 140.04°E 1961–1998 1961–1998 Yonaguni Island 24.47°N 123.02°E 1961–1998 1961–1998 Malaysia Bayan Lepas 5.30°N 100.27°E 1951–1998 Kuching 1.48°N 110.33°E 1953–1998 Sitiawan 4.22°N 100.70°E 1968–1998 Miri 4.33°N 113.98°E 1953–1998 Kota Kinabalu 5.93°N 116.05°E 1953–1998 Sandakan 5.90°N 118.08°E 1954–1998 1954–1998 Mersing2.45°N 103.83°E 1968–1998

Myanmar Dawei 14.07°N 98.18°E 1960–1990 Taunggyi 20.78°N 97.05°E 1967–1990 Sittwe 20.15°N 92.90°E 1961–1990 Myitkina 25.36°N 97.35°E 1960–1990 Kaba Aye 16.87°N 96.18°E 1961–1990 1961–1990 Mandalay 21.98°N 96.10°E 1960–1990 New Caledonia Koumac 20.56°S 164.29°E 1951–1998 1954–1998 Noume´a 22.24°S 166.45°E 1951–1998 1951–1998 Ouanaham 20.78°S 167.26°E 1961–1998 1960–1998 Touho 20.78°S 165.23°E 1952–1998 Pone´rihouen 21.08°S 165.40°E 1952–1998 Yate´ 22.15°S 166.91°E 1951–1998 Toutouta 22.01°S 166.22°E 1951–1998 Paı¨ta 22.14°S 166.37°E 1951–1998 New Zealand Ruakura 37.78°S 175.31°E 1907–1996 1940–1998 Gisborne 38.68°S 178.01°E 1938–1998 1940–1998 Lincoln 43.63°S 172.47°E 1900–1998 1900–1998 Hokitika 42.72°S 170.99°E 1900–1998 1900–1998 Invercargill 46.42°S 168.33°E 1939–1998 1938–1998 Philippines Basco 20.45°N 121.97°E 1961–1998 1961–1998 Tuguegarao 17.62°N 121.73°E 1961–1998 1961–1998 Baguio 16.42°N 120.60°E 1961–1998 1961–1998 Daet 14.12°N 120.98°E 1961–1998 1961–1998 Dumaguete 9.37°N 123.28°E 1961–1998 1961–1998 Solomon Islands Taro Island 6.70°S 156.40°E 1975–1998 1975–1998 Munda 8.33°S 157.26°E 1962–1998 1962–1998 Honiara 9.41°S 159.97°E 1951–1998 1951–1998 Henderson 9.42°S 160.05°E 1974–1998 1974–1998 Auki 8.14°S 160.73°E 1962–1998 1962–1998 Kira Kira 10.42°S 161.92°E 1965–1998 1965–1998 Lata 10.70°S 165.80°E 1970–1998 1970–1998 Thailand Nan 18.77°N 100.77°E 1951–1998 1951–1998 Udon Thani 17.38°N 102.80°E 1951–1998 1951–1998 Suphan Buri 14.47°N 100.13°E 1951–1998 1951–1998 Chanthaburi 12.60°N 102.12°E 1951–1998 1951–1998 Prachuap Khiri Khan 11.83°N 99.83°E 1951–1998 1951–1998 Vietnam Phu Lien 20.80°N 106.63°E 1957–1998 1957–1998 Playcu 15.20°N 106.50°E 1959–1998 Van Ly 20.12°N 106.30°E 1957–1998 1959–1998

REFERENCES Groisman P, Karl T, Easterling D, Knight R, Jamason P, Hennessy K, Suppiah R, Page C, Wibig J, Fortuniak K, Razuvaev V, Douglas A, Førland E, Zhai P. 1999. Changes in the probability of extreme precipitation: important indicators of climate change. Climatic Change 42: 243–283. Hennessy KJ, Suppiah R, Page CM. 1999. Australian rainfall changes 1910 to 1995. Australian Meteorological Magazine 48: 1–13. Hennessy KJ, Whetton PH, McGregor JL, Katzfey JJ, Page CM, Nguyen K. 1998. Fine resolution climate change scenarios for New South Wales. CSIRO Atmospheric Research consultancy report for NSW EPA. Jones PD, Horton EB, Folland CK, Hulme M, Parker DE, Basnett TA. 1999. The use of indices to identify changes in climatic extremes. Climatic Change 42: 131–149. Karl TR, Derr VE, Easterling DR, Folland CK, Hofmann DJ, Levitus S, Nicholls N, Parker DE, Withee GW. 1995. Critical issues for long-term climate monitoring. Climatic Change 31: 185–221.

Karl TR, Nicholls N, Ghazi A. 1999. CLIVAR/GCOS/WMO workshop on indices and indicators for climate extremes. Climatic Change 42: 3–7. Landsea CW, Nicholls N, Gray WM, Avila LA. 1996. Downward trends in the frequency of intense Atlantic hurricanes during the part five decades. Geophysical Research Letters 23: 1697–1700. Lavery B, Kariko A, Nicholls N. 1992. A historical rainfall data set for Australia. Australian Meteorological Magazine 40: 33–39. Lavery B, Joung G, Nicholls N. 1997. An extended high-quality historical rainfall dataset for Australia. Australian Meteorological Magazine 46: 27–38. Manton MJ, Nicholls N. 1999. Monitoring trends in extreme climate events. Asia Pacific Network for Global Change Research, Tokyo. APN Newsletter 5(1): 1–3. Nicholls N. 1995. Long-term climate monitoring and extreme events. Climatic Change 31: 231–245. Nicholls N, Kariko A. 1993. East Australian rainfall events: interannual variations, trends, and relationships with the Southern Oscillation. Journal of Climate 6: 1141–1152. Nicholls N, Landsea C, Gill J. 1998. Recent trends in Australian region tropical cyclone activity. Meteorology and Atmospheric Physics 65: 197–205. Peterson TC, Harald D, Jones P. 1997. Initial selection of a GCOS surface network. Bulletin of the American Meteorological Society 78: 2145–2152. Peterson TC, Easterling DR, Karl TR, Groisman P, Nicholls N, Plummer N, Torok S, Auer I, Boehm R, Gullett D, Vincent L, Heino R, Tuomenvirta H, Mestre O, Szentimrey T, Salinger J, Førland E, Hanssen-Bauer I, Alexandersson H, Jones P, Parker D. 1998. Homogeneity adjustments of in situ atmospheric climate data: a review. International Journal of Climatology 18: 1493–1517. Plummer N, Salinger MJ, Nicholls N, Suppiah R, Hennessy KJ, Page CM, Leighton RM, Trewin B, Lough JM. 1999. Twentieth century trends in climate extremes over the Australian region and New Zealand. Climatic Change 42(1): 183–202. Power S, Casey T, Folland C, Colman A, Mehta V. 1999. Inter-decadal modulation of the impact of ENSO on Australia. Climate Dynamics 15: 319–324. Salinger MJ, Mullan AB. 1999. New Zealand climate: temperature and precipitation variations and their links with atmospheric circulation 1930–1994. International Journal of Climatology 19: 1049–1071. Szentimrey T. 1997. Statistical procedure for joint homogenization of climatic time series. In Proceedings of the Seminar for Homogenization of Surface Climatological Data. Hungarian Meteorological Service: Budapest; 47–62. Torok SJ, Nicholls N. 1996. A historical annual temperature dataset for Australia. Australian Meteorological Magazine 45: 251–260. Trenberth KE, Hoar TJ. 1997. El Nin˜o and climate change. Geophysical Research Letters 24: 3057–3060. Trewin BC, Trevitt ACF. 1996. The development of composite temperature records. International Journal of Climatology 16: 1227–1242. Trewin B. 2000. The frequency of extreme temperature events in Australia, PhD thesis, University of Melbourne (in preparation).

Haylock MR, Peterson T, Abreu de Sousa JR, Alves LM, Ambrizzi T, Anunciação YMT, Baez J, Barbosa de Brito JI, Barros VR, Berlato MA, Bidegain M, Coronel G, Corradi V, Garcia VJ, Grimm AM, Jaildo dos Anjos R, Karoly D, Marengo JA, Marino MB, Meira PR, Miranda GC, Molion L, Moncunill DF, Nechet D, Ontaneda G, Quintana J, Ramirez E, Rebello E, Rusticucci M, Santos JL, Trebejo I and Vincent L. 2004. Trends in total and extreme South American rainfall 1960-2000 and links with sea surface temperature. Journal of Climate, Submitted. Trends in total and extreme South American rainfall 1960-2000 and links with sea surface temperature Submitted to Journal of Climate. 6th December, 2004 Haylock M. R.c, Peterson T.q, Abreu de Sousa J. R. m, Alves L. M.b, Ambrizzi T.v, Anunciação Y. M. T. m, Baez J.g, Barbosa de Brito J. I.x, Barros V. R.d, Berlato M. A.z, Bidegain M.k, Coronel G.t, Corradi V.f, Garcia V. J.u, Grimm A. M.I, Jaildo dos Anjos R.m, Karoly D.o, Marengo J. A.b, Marino M. B.a, Meira P. R.m, Miranda G. C.s, Molion L.w, Moncunill D. F.j, Nechet D.y, Ontaneda G.n, Quintana J.e, Ramirez E.l, Rebello E.m, Rusticucci M.d, Santos J. L.h, Trebejo I.r, Vincent L.p Affiliations a Banco Nacional De Datos, Servicio Meteorológico Nacional, Argentina b Centro de Previsão de Tempo e Estudos Climáticos, Instituto Nacional de Pesquisas Espaciais, Brazil c Climatic Research Unit, University of East Anglia, UK d Departamento de Ciencias de la Atmósfera y los Océanos, FCEN, Universidad de Buenos Aires, Argentina e Dirección Meteorológica de Chile, Chile f Dirección Nacional de Meteorología, Uruguay g Dirección Nacional de Meteorología e Hidrología, DINAC, Paraguay h Faculty of Marine Sciences, Escuela Superior Politécnica del Litoral (Espol), Ecuador I Federal University of Parana, Brazil j Fundação Cearense de Meteorologia e Recursos Hídricos - FUNCEME, Brazil k Faculty of Science, Universidad de la Republica, Uruguay l Instituto de Hidráulica e Hidrológica UMSA, Bolivia m Instituto Nacional de Meteorologia INMET, Brazil n Instituto Nacional de Meteorología e Hidrología, Ecuador o University of Oklahoma, USA p Meteorological Service of Canada, Canada q National Climatic Data Center, USA r Servicio Nacional de Meteorología e Hidrología, Peru s Servicio Nacional de Meteorología e Hidrología, Bolivia t Universidad Nacional de Asunción u Universidad Nacional Agraria La Molina, Peru v Universidade de Sao Paulo, Brazil w Universidade Federal de Alagoas, Brazil x Universidade Federal de Campina Grande, Brazil y Universidade Federal do Para, Brazil z Universidade Federal do Rio Grande do Sul, Brazil Corresponding Author’s Address: Malcolm Haylock Climatic Research Unit University of East Anglia Norwich, NR4 7TJ United Kingdom email: [email protected] Phone: +44-1603-593857 Fax: +44-1603-507784 Abstract A weeklong workshop in Brazil in August 2004 provided the opportunity for twenty-eight scientists from southern South America to examine daily rainfall observations to determine changes in both total and extreme rainfall. Twelve annual indices of daily rainfall were calculated over the period 1960 to 2000, examining changes to both the entire distribution as well as the extremes. Maps of trends in the twelve rainfall indices showed large regions of coherent change, with many stations showing statistically significant changes in some of the indices. The pattern of trends for the extremes was generally the same as that for total annual rainfall, with a change to wetter conditions in Ecuador and northern Peru and the region of southern Brazil, Paraguay, Uruguay and northern and central Argentina. A decrease was observed in southern Peru and southern Chile, with the latter showing significant decreases in many indices. A canonical correlation analysis between each of the indices and sea surface temperatures (SST) revealed two large-scale patterns that have contributed to the observed trends in the rainfall indices. A coupled pattern with ENSO-like SST loadings and rainfall loadings closely resembling the pattern of the observed trend reveals that the change to a generally more negative SOI has had an important effect on regional rainfall trends. A significant decrease in many of the rainfall indices at several stations in southern Chile and Argentina can be explained by a canonical pattern reflecting a weakening of the continental trough leading to a southward shift in storm tracks. This latter signal is a change that has been seen at similar latitudes in other parts of the Southern Hemisphere.

1 1. Introduction Research into changes in rainfall extremes has seen an increase in published results in recent years since the IPCC 2nd Assessment Report (Nicholls et al. 1996) identified a scarcity of such studies. Studies from single countries have included a large part of the globe, including the USA (Karl and Knight 1998), Australia (Haylock and Nicholls 2000), the UK (Osborn et al. 2000) and single European countries e.g. Switzerland (Frei and Schar 2001), Italy (Brunetti et al. 2002), Norway (Benestad and Melsom 2002) and Belgium (Vaes et al. 2002). Cross-border regional studies are however less abundant, and the diverse nature of methods employed for the country specific studies makes comparison between countries difficult. The study by Plummer et al. (1999), who examined changes in climate extremes over Australia and New Zealand, was one of the earliest multi-country analyses. However it was Groisman et al. (1999) who attempted the first study of extreme rainfall for a globally diverse selection of countries. They examined extremes using gamma distribution statistical modelling for eight countries: Canada, the United States, Mexico, the former Soviet Union, China, Australia, Norway, and Poland. More recently work has been published by Klein Tank and Konnen (2003) and Haylock and Goodess (2004) examining trends and variability in extreme indices for almost all of Europe. During the late 1990s, several international workshops developed indices for climate extremes (Folland et al. 1999; Nicholls and Murray 1999). The aim was to create a set of indices that could be calculated for a variety of climates to enable inter-comparison between regions. Methodologies for calculating the indices were reported, highlighting that these indices should be designed to maximise their independence (low inter-index correlation). As well as a desire for more regional analyses, there was a need to include results from regions lacking in published studies. In particular developing countries were lacking analyses due to insufficient resources to undertake such analyses, limited access to data, fewer digitised records and reduced data quality to which extremes analyses are very sensitive. Southeast Asia and the Pacific was identified as a key region (Manton and Nicholls 1999), in particular due to its vulnerability with regard to high population density, heightened rainfall variability due to ENSO, low-lying islands, coral reefs and exposure to tropical cyclones. Therefore, in 1998 the Asia-Pacific Network (APN) for Global Change Research funded a workshop on climate indices and the results were published in Manton et al. (2001). In order to tie together work on extremes for the IPCC 3rd Assessment Report, a near-global analysis was undertaken by Frich et al. (2002), who analysed linear trends in ten climate indices for a large part of North America, Europe, Asia and the Pacific. The Expert Team on Climate Change Detection, Monitoring and Indices (ETCCDMI – http://www.clivar.org/organization/etccd/) was established as a joint CCl/CLIVAR venture to provide advice on climate monitoring indices. Following the publication of Frich et al. (2002), the ETCCDMI identified key regions in which to promote climate change studies: the Caribbean, Central and South America, Africa and West and Central Asia. They decided the best way to establish research in these regions was through regional workshops similar in format to the APN workshop (Manton et al. 2001). In 2001 two such workshops were held: in Morocco to cover northern African countries (Easterling et al. 2003); and in Jamaica to cover the Caribbean (Peterson et al. 2002). The ETCCDMI met again in 2003 to plan further workshops in time to produce results for the IPCC 4th Assessment Report. Four workshops were planned for 2004, to take place in South Africa for southern African countries, in Brazil for southern South American countries, in Turkey for western Asian countries, in Guatemala for Central and northern South American countries and in India for central and south Asian countries. This paper deals with rainfall results from the Brazilian workshop. Temperature results are presented in the companion paper (Vincent et al. 2004).

2 The five and a half day Brazilian workshop was held in August 2004 in Maceio and was attended by 28 scientists from eight South American countries: Argentina, Bolivia, Brazil, Chile, Ecuador, Paraguay, Peru and Uruguay. A new approach for this workshop was to invite a representative for each country from both a university and the national weather service. This was in order to strengthen ties between the two groups as well as to promote research in both the educational and government streams. The key aims of the workshop were: to provide a consistent methodology for analysing extremes across the region; to establish a network of scientists with the hope of building on this in the future; and to provide capacity-building to countries with less resources for climate change research.

Until recently, there had been little published work on rainfall extremes in South America. Due to the implementation of various projects on the Paraná-La Plata River Basin in southeastern South America, several recent papers have studied rainfall extremes in this basin and linked them to the regional circulation. Liebmann et al. (2004a) identified that extreme daily precipitation events in southeastern Brazil have a strong tendency to cluster in response to remote or local influences and that their occurrence is related to the frequency and intensity of the South American low-level jet (SALLJ) east of the Andes (Marengo et al. 2004). Intense/weak SALLJ episodes during the austral summer are related to enhanced/decreased moisture transport from the Amazon Basin and a larger/lower probability of extreme rainfall events in southeastern South America downstream of the jet. Previously, Carvalho et al. (2002) showed that most extreme events in the state of Sao Paulo in southeastern Brazil occur when the South Atlantic Convergence Zone (SACZ) is strong. Later, Carvalho et al. (2004) examined extreme precipitation events in relation to the Madden Julian Oscillation (MJO) and found that when convective activity linked to the MJO is weak over Indonesia, extreme rainfall events increase along the SACZ (and weaken in southeastern Brazil). They also concluded that extreme rainfall events in southeastern Brazil exhibit interannual variation related to large-scale forcing, with more days with extreme events during El Niño conditions than during La Niña. This is consistent with Grimm and Pscheidt (2001) who concluded that for southern Brazil there was a large increment in the frequency of extreme events during El Nino years in the 1963-92 period.

Previous studies have identified regional changes in total rainfall and links with sea surface temperatures (SST) and circulation. The late 1980s saw the earliest work directed at examining links between ENSO and South American rainfall. Studies by Aceituno (1988), Rogers (1988) and Ropelewski and Halpert (1987, 1989) identified a decrease in rainfall in northeast Brazil and an increase in southeast Brazil, northern Argentina, Paraguay and Uruguay associated with El Niño events. The links between ENSO and rainfall in the Amazon basin were examined by Marengo (1992), Marengo and Hastenrath (1993) and Marengo et al. (2001) who proposed a mechanism for the observed rainfall decrease during El Niño events. More recently the South Atlantic Convergence Zone (SACZ) has received attention, with Liebmann et al. (1999) and Carvalho et al. (2002, 2004) examining ENSO-induced changes in its position and intensity and the effect of such changes on convective rainfall. These studies, along with Doyle and Barros (2002), highlight the importance of SSTs in the western South Atlantic in influencing the SACZ. The South American summer monsoon has received particular attention, as this season is generally the wettest period over most of the northern and central part of the continent. Vera et al. (2004), Nogues-Paegle et al. (2002), Grimm (2003, 2004) and Grimm et al. (2000, 1998) present a comprehensive account of findings for the South American monsoon and southern Brazil.

While there has been much attention given to variability and links with large-scale forcing, several recent studies have focussed on the trends in total rainfall and rainfall extremes. In the Amazon basin, observational studies based on time series of rainfall and river streamflow data (Chen et al. 2001; Chu et al. 1994; Depaiva and Clarke 1995; Gentry and Lopez-Parodi 1980; Marengo 2004; Marengo et al. 1998; Matsuyama et al. 2002) have produced conflicting results, showing positive or negative trends of rainfall mainly due the use of different time periods. Marengo (2004) identifies a 3 weak positive rainfall trend in the whole Amazonia, with negative/positive rainfall trends in northern/southern section of the basin. He concludes that these trends are less important than the decadal scale rainfall variability in both sides of the basin. Previously, Hastenrath and Greischar (1993) and Marengo et al. (1998) have found positive rainfall trends in northeast Brazil, while positive rainfall trends have been also detected in southern Brazil and northern Argentina (Barros et al. 2000). Long-term trends in rainfall at stations in southern South America were examined by Minetti (1998) and Minetti et al. (2003). They found that for the period 1931-1999 there had been a steady decrease in annual rainfall for a large area west of the Andes and an increase to the east in central Argentina. They identified a third region in northern Argentina that had seen a steady increase up until the 1980s when strong El Niño events caused a general decline. Similar trends were observed at some stations in southern South America by Rusticucci and Penalba (2000). They noted that the Chilean station Valdivia (used in this study) had seen a large decrease in total annual precipitation over the period 1901-90, mainly due to a decrease in winter precipitation. They suggested this could be due to a change in the behaviour of mid-latitude frontal systems as had been found in Australia (Allan and Haylock 1993).

A recent study by Liebmann et al. (2004b) identified seasonal linear trends of precipitation from central South American during 1976-1999, and showed the largest positive trend occurred south of 20oS during January-March and centred over southern Brazil, while from 1948-1975 the trend is also positive, but with less than half the slope. The trend is due to an increase in the percent of rainy days, and an increase in the rainy day average. The precipitation trend is related to a positive sea- surface temperature trend in the nearby Atlantic Ocean, but apparently not causally. The trend in the Atlantic seems to result from a decrease in mechanical stirring and coastal upwelling associated with a decrease in the strength of the western edge of the circulation associated with the South Atlantic High.

However, all these studies do not show any indication on trends in extreme rainfall events. Therefore, this paper attempts to provide updated information on rainfall trends in South America, both for mean and extreme events. Section 2 discusses the data availability and quality checking. Section 3 examines the linear trends in the rainfall indices. Section 4 looks at links with SSTs and section 5 concludes the study. 2. Data availability and quality The focus on climate indices derived from raw daily data overcame a major problem inherent in regional studies. The participating countries’ fears of loss of control of daily data were allayed by the workshop format whereby participants brought data to be analysed at the workshop but released only derived climate indices. While the climate indices are very valuable to scientists for climate monitoring, they are of little commercial value. Participants were requested to bring at least five stations with daily rainfall and minimum and maximum temperature, as well as any available metadata for those stations. A long record was desirable from only high quality climate stations covering as much of the country as possible. Indices of extremes are sensitive to changes in station location, exposure, equipment and observer practice. Therefore, a large part of the workshop was devoted to testing stations for invalid measurements and inhomogeneities. Firstly negative rainfall amounts were set to missing and then a graph of the daily rainfall record was examined by eye. Outliers were considered more closely to see if they were part of a multi-day wet spell and not the result of several days accumulation of missing observations, as is present in other regions, for example Australia (Viney and Bates 2004). If a nearby station was available then unusually high records could be compared across stations. Statistical tests were not used to identify outliers due to the large variety of distributions from which daily rainfall is derived in such a climatically diverse region. 4 An inhomogeneity test (Easterling and Peterson 1995) was applied to the annual total rainfall, although the lack of surrounding stations from which to construct a reference series made this difficult to interpret. Also the large interannual variability of annual rainfall, particularly in the ENSO-dominated dry regions of western South America, lead to the detection of significant jumps in the record which coincided with wet El Niño years. It was therefore decided to examine the annual climate indices by eye to identify unusual behaviour possibly caused by changes to the station or observing practices. Stations with probable non-climate related changes were rejected. The inhomogeneity test was much more powerful in detecting jumps in the temperature record and will be discussed further in the companion paper (Vincent et al. 2004). Twelve rainfall indices were calculated and are listed in Table 1. Most of the indices relate to extreme rainfall, although two are more indicative of changes to the entire rainfall distribution: the total annual wet-day rainfall PRCPTOT and the average wet-day rainfall intensity SDII. A wet-day refers to days with at least 1mm precipitation. This relatively high threshold was used as previous studies have found that lower thresholds can be sensitive to problems such as underreporting of small rainfall amounts and changes in the units of measurement (e.g. Hennessy et al. 1999). In the workshops held prior to 2004, the data quality checking and calculation of indices was carried out using a Microsoft Excel spreadsheet “ClimDex”. This platform was chosen due to its familiarity and ease of use. However a change to how some of the indices were calculated, in particular the use of bootstrapping for calculating the exceedence of base period normals (Zhang et al. 2004), meant that Excel was no longer the best program. Therefore a version of ClimDex was created using the powerful and statistically robust open source “R” package (http://www.r- project.org). The new program “RClimDex” was provided with a graphical interface to run under a variety of operating systems. RClimDex is available for download from the ETCCDMI indices website (http://cccma.seos.uvic.ca/ETCCDMI/). After rejecting suspect stations, climate indices were calculated for an initial set of 68 stations. The climate indices are sensitive to missing data. Therefore the annual indices were not calculated for years where there were more than 15 days missing. The period of record for each station differed. Since it was desired to present results for a consistent period across the network, a methodology was adopted to find the best period and station set. The aim was to maximise the period length as well as the number of stations. Therefore for each period of at least 30 years from 1950-2003, the number of stations with at least 80% non-missing annual indices was calculated. Clearly the shorter the period chosen the more stations would pass the test. Since the aim was to optimise the period length as well as the number of stations, for each period we calculated the function: f(start year, end year)= (number of stations with > 80% data) * period length This had its maximum value for the period 1960-2000 (41 years) with 54 stations. In comparison, if we look for the maximum number of stations that passed in any 35-year period we only gain an additional 2 stations. A map of the final set of 54 stations is shown in Figure 1 and the stations are listed in Table 2. The start and end years listed in Table 2 are those for which the data were available. In all analyses we have confined the period to 1960-2000. Although there are large areas with no stations in parts of Brazil, coverage is still good over most of the region, including the southern tip of continent. 3. Linear trends Although RClimDex produces plots of the indices with the linear least squares trend, it was decided that the generally non-Gaussian distributed indices required a more statistically robust methodology for trend estimation. Also many of the indices contained much larger values for El Niño years which would have undue influence on the non-resistant least squares method. Kendall’s Tau was therefore adopted as a measure of the trend (Kendall 1938). This non-parametric statistic measures the relative ordering of all possible pairs of data points, where the year was used as the independent

5 variable and the extreme index as the dependent variable. Kendall’s Tau was used in preference to the Spearman correlation as the latter encounters problems with tied values, a situation that occurs often in the frequency related indices. Kendall’s Tau, expressed as a value between -1 and 1, also has the added benefit that it enables comparison of indices between stations with vastly different rainfall amounts. In contrast, a map of least-squares linear trends in rainfall would be dominated by the wetter stations. Figure 2 shows the trends for the twelve rainfall indices. In this figure we have elected, for ease of interpretation, to show just the sign of the trend as well as flag stations with trends significant at p<0.05. Figure 2 shows there to be large regions of spatial coherence in the sign of the trends, as well as general agreement between the indices. Note that in all indices except CDD, an increase corresponds to wetter conditions. The pattern of trends in the PRCPTOT index shows broad features that are present to varying degrees in the other indices: I) generally wetter in the central eastern region (southern Brazil, Paraguay, Uruguay and northeast Argentina); II) drier in the southwest (southern Chile and southwest Argentina); III) drier in southern Peru; IV) and wetter in northwest Peru and Ecuador. The southernmost station in Tierra del Fuego, Punta Arenas, has become significantly wetter. The stations in northern Brazil show no consistent sign. In all indices, the stations with a significant trend are located mainly in regions I and II. Punta Arenas also shows a significant trend in many indices. The positive trends in southern Brazil, Uruguay and Northeast Argentina agree with previous findings by Barros et al. (2000) and in central South America by Liebmann et al. (2004b) and Minetti et al. (2003). There are several notable differences in the patterns of change between the indices and the general observations mentioned above. For Paraguay, although the sign of the trends of most of the indices agrees with those of the surrounding regions, the trends in CDD and CWD are of the opposite sign. Therefore, although the rainfall has increased in intensity, the maximum wet/dry spell length has decreased /increased, although not significantly. This is also the case for the single Bolivian station, which shows an increase in all the indices (some significantly), except CWD. At most other stations the sign of CWD agrees with PRCPTOT and is opposite to CDD. The patterns of trends for the frequency of extremes indices, R10mm and R20mm, are very similar to PRCPTOT. Several stations show a significant trend in the extremes but not in the total e.g. in Peru and Ecuador for R10mm, but the reverse is also the case. The two percentile exceedence indices, R95p and R99p, show similar patterns of trends to the total rainfall, although there are less significant changes in the area of increase in the central eastern region (region I). In this region more stations show a significant increase in the rainfall exceeding the 99th percentile than the 95th, including a significant decrease in R99p at a station in western Uruguay which showed a non-significant increase in total rainfall. The two indices of maximum event intensity, RX1day and RX5day, show a less defined regional signal due to lower spatial coherence. No stations in region I show a significant trend and the trends here are of mixed sign. Several stations in the west show a significant change, although in several cases this is the opposite sign to PRCPTOT e.g. RX1day in central Peru and northern Chile. Also the stations in Ecuador show more of a non-significant decrease than the non-significant increase observed in PRCPTOT. The sign of the trends in the average intensity index, SDII, agrees mostly with PRCPTOT, although the significance is different for some stations. The two indices of the proportion of total precipitation from extremes, R95pTOT and R99pTOT, both show a similar pattern of trends to PRCPTOT. There are more stations showing a significant change for the 99th percentile than for the 95th, particularly in region I. One of the most consistent signals in Figure 2 is the decrease in total and extreme rainfall in the southwest, as evident in the westernmost Argentinean station and its two neighbouring Chilean stations. In all indices except CDD, these stations have showed a decrease, which in most cases is

6 significant. These findings are consistent with previously discussed studies noting a rainfall decrease in this region (Minetti 1998; Minetti et al. 2003; Rusticucci and Penalba 2000). 4. Links with SST The high spatial coherence of the sign of the trends of the rainfall indices suggests that there could be changes to the large-scale forcing of the rainfall. This section examines possible changes in forcing. The choice of predictors to examine is governed to a large extent by the time period over which the indices are calculated as well as the size of the region under examination. Although the rainfall indices are calculated annually, most of the indices are dominated by the rainfall during the wettest time of the year. However the diverse nature of the station network, which includes stations from the tropics to the mid-high latitudes, means that the wet season occurs at different times of the year between stations. Therefore our potential predictor should be fairly constant throughout the year. The large area under consideration also dictates that our forcing variable should also be large-scale. The most obvious candidate is the large-scale, slowly evolving SST anomalies. We quantified the relationship between the rainfall indices and SSTs using a canonical correlation analysis (CCA). The canonical patterns and coefficients were calculated using a singular value decomposition (SVD) of the cross-covariance matrix of the principal components (PC) of the two fields being examined. This is numerically more stable than the more common method of working with the joint variance–covariance matrix (Press et al. 1986) and also incorporates the pre-filtering of the data by using just the significant PCs. Bretherton et al. (1992) discuss the benefits of this methodology further and compare it with other methods of finding coupled modes. This CCA methodology requires that we first find the number of significant PCs. This was objectively decided by a Monte Carlo process, whereby 1000 PC analyses were carried out using data randomly resampled in time from the stations (Preisendorfer et al. 1981). In each of the 1000 analyses, 54 station annual series of length 41 years were generated with similar statistical properties to the original data but with random interstation correlations. Then, for each randomization, the eigenvalues were calculated. Each of the eigenvalues of the real observations was then compared against the distribution of the 1000 randomly generated values to determine whether they were greater than the rank 50 eigenvalue (equivalent to p < 0.05). This methodology is the same as that employed by Haylock and Goodess (2004). A separate CCA was carried out for each of the twelve rainfall indices with SST. Like least squares linear trends, CCA is sensitive to outliers. In order to reduce the effect of outliers in the CCA from the non-Gaussian distributed indices, we first converted the indices to their relative ranks. This method can be thought of as a CCA that optimises the Spearman correlation of the coupled modes, except that the SST data were not also ranked as these were more normally distributed. The SST data used were taken from Global Sea Ice Coverage and Sea Surface Temperature (GISST) data set version 2.3b (Parker et al. 1995) over the region 80 oW to 0oW, 60oS to 20oN. In order not to unduly weight SST grid points at higher latitudes, the data were first interpolated onto an equal-area grid over the same region of dimensions 280km x 280km at the equator. The rainfall indices were used at the 54 station locations. Although this will give more weight towards regions with higher station density, it was decided that this was better than either interpolating the stations to a regular grid or thinning the station network. Table 3 shows statistics of the CCA for each index. The table includes the number of coupled patterns, the canonical correlations, the proportion of variance explained by each canonical coefficient for both the SST and rainfall indices and the correlation of the canonical coefficients with the SOI. Note that in the case of the indices, the proportion of variance refers to the variance of the rank of the indices and not the raw indices. From here on we will just refer to this as the variance of the indices.

7 Seven PCs of SST were used in the CCA (selected by the above Monte Carlo methodology), accounting for 81.4% of the total SST variance. The objectively determined number of PCs of the indices varied from 2 to 4 PCs. Since this is smaller than the seven PCs of SST used in the CCA, it is the number of PCs of the indices that determines the number of coupled CCA patterns. Table 3 shows that the canonical coefficients accounted for between 16.3% and 44.1% of the total variance in the indices. While this at first seems low, it is an important conclusion that continental scale variability is still an important factor in interannual variability of extreme rainfall. Since the CCA truncates the number of components of SST from the initial 7, the proportion of variance explained by the SST canonical coefficients varied for each index, between 27.6% and 54.9%. Generally one would expect that variables with a higher spatial coherence to yield a larger number of significant PCs explaining a higher proportion of the variance. For example, although the SST observations were over a much larger area than the indices, the SST yielded a higher number of significant PCs accounting for a larger proportion of total variance than any of the rainfall indices. The index with the equal highest number of significant components, as well as the highest proportion of variance in both the SST and indices is the total annual precipitation PRCPTOT. Two of the extreme rainfall indices, R20mm and RX5day share the same number of significant components and almost as high a total variance as PRCPTOT. R95pTOT, R99p and RX1day are the three indices with the lowest variance and number of significant components. The maximum canonical correlation varies between the indices from 0.876 for PRCPTOT to 0.542 for RX1day. An examination of the canonical correlations in Table 3 shows no clear relationship between the correlation and the proportion of variance explained by the associated coefficients. For the two percentile exceedence indices R95p and R99p, the highest canonical correlation is related to the canonical coefficient that explains the highest variance in both SST and the rainfall index. For CDD, RX1day and RX5day, the highest correlation is related to the highest proportion of variance in SST only. For PRCPTOT, R20mm, R95pTOT and SDII, the highest correlation relates to the coefficient that explains the highest variance in the rainfall index only. For CWD, R99pTOT and R10mm the patterns with the highest correlation explain the highest variance in neither the SST nor the index. The three coupled patterns that account for the highest proportion of variance in the indices are the first patterns of PRCPTOT, R20mm and R95p. Figure 3 shows the factor loadings of the canonical coefficients for the SST and indices for these three cases. The loadings represent the correlations between the canonical coefficients and the SST or rank of the indices. The SST patterns in the three panels are very similar and resemble the classic ENSO response, for example the SST composited by the SOI (not shown but see Rasmusson and Carpenter (1982)). The familiar high loadings in the tropical eastern Pacific and loadings of opposite sign in the southeast Pacific are indicative of ENSO, however the moderate loadings to the east of the continent as well as those of opposite sign in the South Atlantic are less similar to ENSO. Table 3 shows that the canonical coefficients for these three SST patterns all have a correlation with the annual SOI above 0.7. The patterns of loadings for the three rainfall indices in Figure 3 resemble the trends of the indices (section 3), although the sign is reversed. The loadings are coherent over the central eastern region of southern Brazil, Paraguay, Uruguay and northern Argentina. The sign of the loadings in this region is the same as that over Ecuador and northern Peru and of opposite sign to northeast Brazil, southern Peru and southern Chile. The similarity in the patterns of the trends (Figure 2) and the ENSO-dominated CCA loadings (Figure 3) leads to the proposal that it is changes in the behaviour of ENSO that have contributed to the observed trends in these three rainfall indices. This hypothesis is examined further in Figure 4, which shows the annual SOI with the first canonical coefficients for PRCPTOT, R20mm and R95p. The figure shows the matching trend and interannual variability in the SOI and the three indices, supporting the conclusion that changes in ENSO have contributed to the observed trend in the indices. The trend in the first canonical coefficients for R20mm and R95p is significant at p<0.05, however not for PRCPTOT. Since the trend in the rainfall coefficients is negative, the contribution to the trend of the indices from this pattern is opposite to the sign of the 8 loadings in Figure 3. Therefore, from Figure 3, the trend from this pattern agrees with those outlined in section 3. The mechanisms for the observed ENSO-related pattern of rainfall changes have been discussed in several of the papers mentioned in the introduction. Marengo (1992) and Marengo and Hastenrath (1993) examined the reason for the variability of rainfall in the Amazon with ENSO. They proposed that during La Niña events in late summer, an anomalously southward-displaced ITCZ caused wetter conditions in the Amazon, with ascending motion over the northern Amazon and associated subsidence and drier conditions west of the Andes in Ecuador and Peru. SST anomalies in central equatorial Pacific explain less than 40% of rainfall variance in central Amazonia (Marengo 1992), and the fraction of variance explained by SST anomalies in the central Pacific is slightly higher during the period 1976-1998 compared to 1950-1975 as shown by Marengo (2004). This is due the presence of more frequent and intense El Niño events in 1976-98 (with an active role of the tropical Pacific), while during 1950-75 the tropical Pacific was less active. The mechanism and effect of a northward-displaced SACZ during La Niña conditions was examined by Carvalho et al. (2002; 2004), which was found to cause the observed decrease in rainfall over the region of southern Brazil and wetter conditions in the north (see also Liebmann et al. 2004b). Both these patterns are reproduced in our CCA results in Figure 3. An important difference between the factor loadings of the indices in Figure 3 and the trends in Figure 2 is that the stations in southern Chile and southwest Argentina that have seen a significant decrease in many of the indices are not represented by coherent high loadings in Figure 3. Therefore the patterns in Figure 3 can not explain the changes in this region. Figure 5 shows the loading patterns for the 2nd canonical coefficients for PRCPTOT. This pattern of loadings for the rainfall index shows large negative loadings for the stations in the southwest and generally positive elsewhere. Several other stations further north have negative loadings but these are of much smaller magnitude. Therefore this pattern represents behaviour in the south that is out of phase with the other stations. Interestingly the loading over the southernmost station, Punta Arenas, is small indicating that this mode is not important for variability at this station. Figure 6 shows the corresponding canonical coefficients with a strong positive trend in both series, leading to drier conditions in southern Chile and wetter in the rest of the region. The trends in both these series are significant (p<0.0001). Table 3 shows that the correlations between these coefficients and the SOI are low (0.144 for the SST and 0.046 for the rainfall). The SST pattern in Figure 5 is quite complex, showing areas of positive loadings centred west of Patagonia and the tropical South Atlantic along with regions of negative loadings west of the continent, in the tropical North Atlantic and in the higher latitudes of the South Atlantic. Since the canonical coefficients have a strong positive trend, the areas of positive (negative) SST loading are also areas of warming (cooling) with regard the contribution from this pattern. To determine the effect of this pattern of SSTs on the circulation we have composited the average annual mean sea level pressure (MSLP) for years when the SST coefficient in Figure 6 is positive and when it is negative. MSLP was taken from the NCEP reanalyses (Kalnay et al. 1996). These composites are shown in Figures 7a and 7b. The figures show the general pattern of low pressure centred over the Amazon basin, the ridge of the subtropical high at 30oS with a trough over the continent and a steady zonal gradient to lower pressures further south. Note that the stations in southern Chile and Argentina that have seen a strong decrease in rainfall are situated south of the subtropical high at 40oS-45oS. At first the MSLP patterns in Figures 7a and 7b appear very similar, however when one considers Figure 7c, which is the difference of the composite of years with positive SST canonical coefficient minus the negative years (Figure 7b - 7a), significant differences become apparent. In this figure the shaded area shows regions where the difference of the means is significant at p<0.05 using a t-test. The most important features with regards rainfall over the region are the significant change to higher pressures in the south of the continent and the change to lower pressures in the northeast. In the south the increase in pressure corresponds to a weakening of the trough in Figure 7b compared with Figure 7a. 9 This significant increase in MSLP over the south of the continent suggests that the northward cyclone propagation through South America may be affected, thus confining the rain-bearing systems to higher latitudes. This is supported in Figure 7c by the significant change to generally lower pressure at the highest latitudes of the region. Recently, Pezza and Ambrizzi (2003) indicated that the total number of Southern Hemisphere cyclones during the austral winter season had declined from 1973 to 1996. On the other hand, they also showed that the number of intense cyclones had increased. Our results suggest that an increase in baroclinic activity at higher latitudes could lead to the confinement of the transient systems to the south, thereby affecting the meriodional displacement of polar air masses to South America. The possibility that the climatological decrease in the number of cyclones over the Southern Hemisphere is linked to climate change has been suggested by several previous studies (e.g. Key and Chan 1999; Simmonds and Keay 2000; Simmonds et al. 1998). Interestingly a similar rainfall decrease has also been found in Australia in latitudes 30-40oS (Allan and Haylock 1993; Smith et al. 2000), being in agreement with the present analysis. This regional pattern is consistent with a hemispheric-scale pattern of increases in pressure in middle latitudes and decreases at high latitudes associated with a trend in the Southern Annular Mode (SAM, Thompson et al. 2000). Modelling and other studies have linked the trend in the SAM and southward shift in the storm track to a climate response to decreases in stratospheric ozone (Gillett and Thompson 2003; Sexton 2001; Thompson and Solomon 2002) or as a response to increasing greenhouse gases (Cai et al. 2003; Fyfe 2003; Kushner et al. 2001). In Figure 7c there is little change in MSLP over the region of the Patagonian station Punta Arenas, which has seen a significant rainfall increase over the period (Figure 2). Since the loadings for this station are small for this canonical pattern (Figure 5), we can not attribute this rainfall increase to these changes in MSLP. Also this station is located almost directly on the zero contour in Figure 7c, indicating that this canonical pattern has not lead to any change in pressure at this station. The significant decrease in pressure in northeast Brazil related to this pattern is probably related to the warming in the central Atlantic (Figure 5). The rainfall stations in this region all have a positive loading, corresponding to wetter conditions due to the significant trend in the canonical coefficients. Although we have only shown this pattern for PRCPTOT, this pattern is not limited to total rainfall but also to the extremes. A similar CCA pattern, along with canonical coefficients with a strong trend, occurs in some of the other indices, e.g. the 2nd canonical patterns of R10mm, R20mm and SDII. The two large-scale changes so far discussed, the changes in ENSO and the weakening of the continental trough, account for most of the regional coupled patterns identified by the CCA for all the indices. Other patterns listed in Table 3 that are not related to one of these two changes either account for a small proportion of total variance of the indices, or have a small canonical correlation reflecting a less significant statistical relationship between the SST and rainfall indices. It must be mentioned however that a low proportion of total variance might still indicate a pattern that is important for a particular area, but when considered over the entire region under study is less important. When performing an analysis like CCA that optimises linear correlation, there is a chance that any trends in the two datasets will unduly influence the results, leading to coupled patterns that have similar trends but low correlation in interannual variability. Figures 4 and 6 show that it is not the case, with the series having strong similarity in the interannual variability as well as the trends. Still, we checked the results by performing a CCA using detrended data with very similar results. 5. Conclusion This study presented results from a weeklong workshop attended by 28 scientists from southern South America. This is a region that was identified by the CCl/CLIVAR Expert Team on Climate

10 Change Monitoring Detection and Indices as lacking in coherent regional studies of long-term trends in climate extremes. The regional analysis, covering a large part of continental South America, was made possible through the workshop format. It showed that valuable results of changes in climate extremes could be obtained using a consistent methodology, but without individual countries having to surrender potentially commercially valuable daily data. This paper presented results on trends and variability of twelve annual rainfall indices, covering changes in both the entire distribution as well as its wet and dry extremes. Fifty-four stations were deemed to be of sufficiently high quality and to have sufficient observations to be used to assess changes for the period 1960-2000. Maps of trends in the twelve rainfall indices showed large regions of coherent change, with many stations showing statistically significant changes in some of the indices. The pattern of trends for the extremes was generally the same as that for total annual rainfall, with a change to wetter conditions in Ecuador and northern Peru and the region of southern Brazil, Paraguay, Uruguay and northern and central Argentina. A decrease was observed in southern Peru and southern Chile, with the latter showing significant decreases in many indices. To investigate the possible causes of these spatially coherent trends, we performed a CCA of each of the indices with annually averaged SST observations. This revealed two large-scale coupled patterns that we proposed have contributed to the rainfall changes. The first was a change during the period to more El Niño dominated conditions with a generally lower SOI. Through various previously documented mechanisms, such as increased subsidence and a northward shifted ITCZ over northeast Brazil and the Amazon basin and a southeast shift in the SACZ, this was shown to have partly caused the observed trend in the rainfall indices. The changes in ENSO could not account for the significant decrease over southern Chile. A separate CCA mode, uncorrelated with ENSO suggested that this was caused by a general weakening of the continental trough at higher latitudes, causing a southward shift in the storm tracks during the period. Both these large-scale SST signals were represented by canonical coefficients with statistically significant trends, which have contributed to the observed significant changes in the rainfall indices. While the workshop provided the mechanism to produce a regional study, it also provided the means to build a strong scientific network in the region. These published results, while presenting a valid and statistically sound picture, still only represent a small selection of stations from a vast region. We hope that this will only improve with time. Acknowledgements Support for the workshop was provided by the U.S. Department of State and the Inter American Institute for Global Change Research References Aceituno, P., 1988: On the Functioning of the Southern Oscillation in the South-American Sector .1. Surface Climate. Monthly Weather Review, 116, 505-524. Allan, R. J. and M. R. Haylock, 1993: Circulation Features Associated with the Winter Rainfall Decrease in Southwestern Australia. Journal of Climate, 6, 1356-1367. Barros, V., M. E. Casteñeda, and M. Doyle, 2000: Recent precipitation trends in southern South America east of the Andes: An indication of climatic variability. Southern Hemisphere paleo and neo-climates, P. P. Smolka and W. Volkheimer, Eds., Springer-Verlag. Benestad, R. E. and A. Melsom, 2002: Is there a link between the unusually wet autumns in southeastern Norway and sea-surface temperature anomalies? Climate Research, 23, 67-79. Bretherton, C. S., C. Smith, and J. M. Wallace, 1992: An Intercomparison of Methods for Finding Coupled Patterns in Climate Data. Journal of Climate, 5, 541-560.

11 Brunetti, M., M. Maugeri, T. Nanni, and A. Navarra, 2002: Droughts and extreme events in regional daily Italian precipitation series. International Journal of Climatology, 22, 543-558. Cai, W. J., P. H. Whetton, and D. J. Karoly, 2003: The response of the Antarctic Oscillation to increasing and stabilized atmospheric CO2. Journal of Climate, 16, 1525-1538. Carvalho, L. M. V., C. Jones, and B. Liebmann, 2002: Extreme precipitation events in southeastern South America and large-scale convective patterns in the South Atlantic convergence zone. Journal of Climate, 15, 2377-2394. Carvalho, L. M. V., C. Jones, and B. Liebmann, 2004: The South Atlantic convergence zone: Intensity, form, persistence, and relationships with intraseasonal to interannual activity and extreme rainfall. Journal of Climate, 17, 88-108. Chen, T. C., J. H. Yoon, K. J. St Croix, and E. S. Takle, 2001: Suppressing impacts of the Amazonian deforestation by the global circulation change. Bulletin of the American Meteorological Society, 82, 2209-2216. Chu, P. S., Z. P. Yu, and S. Hastenrath, 1994: Detecting Climate-Change Concurrent with Deforestation in the Amazon Basin - Which Way Has It Gone. Bulletin of the American Meteorological Society, 75, 579-583. Depaiva, E. M. C. D. and R. T. Clarke, 1995: Time Trends in Rainfall Records in Amazonia. Bulletin of the American Meteorological Society, 76, 2203-2209. Doyle, M. E. and V. R. Barros, 2002: Midsummer low-level circulation and precipitation in subtropical South America and related sea surface temperature anomalies in the South Atlantic. Journal of Climate, 15, 3394-3410. Easterling, D. R. and T. C. Peterson, 1995: A New Method for Detecting Undocumented Discontinuities in Climatological Time-Series. International Journal of Climatology, 15, 369-377. Easterling, D. R., L. Alexander, A. Mokssit, and V. Detemmerman, 2003: CCl/CLIVAR workshop to develop priority climate indices. Bulletin of the American Meteorological Society, 84, 1403- 1407. Folland, C. K., C. Miller, D. Bader, M. Crowe, P. Jones, N. Plummer, M. Richman, D. E. Parker, J. Rogers, and P. Scholefield, 1999: Workshop on Indices and Indicators for Climate Extremes, Asheville, NC, USA, 3-6 June 1997 - Breakout Group C: Temperature indices for climate extremes. Climatic Change, 42, 31-43. Frei, C. and C. Schar, 2001: Detection probability of trends in rare events: Theory and application to heavy precipitation in the Alpine region. Journal of Climate, 14, 1568-1584. Frich, P., L. V. Alexander, P. Della-Marta, B. Gleason, M. Haylock, A. M. G. K. Tank, and T. Peterson, 2002: Observed coherent changes in climatic extremes during the second half of the twentieth century. Climate Research, 19, 193-212. Fyfe, J. C., 2003: Extratropical southern hemisphere cyclones: Harbingers of climate change? Journal of Climate, 16, 2802-2805. Gentry, A. H. and J. Lopez-Parodi, 1980: Deforestation and Increased Flooding of the Upper Amazon. Science, 210, 1354-1356. Gillett, N. P. and D. W. J. Thompson, 2003: Simulation of recent Southern Hemisphere climate change. Science, 302, 273-275. Grimm, A. G. and I. Pscheidt, 2001: Atmospheric patterns associated with extreme rainfall events in the spring during El Niño, La Niña and neutral years in southern Brazil. Proceedings of the 9th Congress of the Latin-American and Iberian Federation of Meteorological Societies and 8th

12 Argentinean Congress of Meteorology (on CD, in Portuguese), Buenos Aires, Centro Argentino de Meteorólogos e Federação Latino-Americana e Ibérica de Sociedades de Meteorologia. Grimm, A. M., 2003: The El Nino impact on the summer monsoon in Brazil: Regional processes versus remote influences. Journal of Climate, 16, 263-280. Grimm, A. M., S. E. T. Ferraz, and J. Gomes, 1998: Precipitation anomalies in southern Brazil associated with El Nino and La Nina events. Journal of Climate, 11, 2863-2880. Grimm, A. M., V. R. Barros, and M. E. Doyle, 2000: Climate variability in southern South America associated with El Nino and La Nina events. Journal of Climate, 13, 35-58. Groisman, P. Y., T. R. Karl, D. R. Easterling, R. W. Knight, P. F. Jamason, K. J. Hennessy, R. Suppiah, C. M. Page, J. Wibig, K. Fortuniak, V. N. Razuvaev, A. Douglas, E. Forland, and P. M. Zhai, 1999: Changes in the probability of heavy precipitation: Important indicators of climatic change. Climatic Change, 42, 243-283. Hastenrath, S. and L. Greischar, 1993: Further Work on the Prediction of Northeast Brazil Rainfall Anomalies. Journal of Climate, 6, 743-758. Haylock, M. and N. Nicholls, 2000: Trends in extreme rainfall indices for an updated high quality data set for Australia, 1910-1998. International Journal of Climatology, 20, 1533-1541. Haylock, M. and C. Goodess, 2004: Interannual variability of European extreme winter rainfall and links with mean large-scale circulation. International Journal of Climatology, 24, 759-776. Hennessy, K. J., R. Suppiah, and C. M. Page, 1999: Australian rainfall changes, 1910-1995. Australian Meteorological Magazine, 48, 1-13. Kalnay, E., M. Kanamitsu, R. Kistler, W. Collins, D. Deaven, L. Gandin, M. Iredell, S. Saha, G. White, J. Woollen, Y. Zhu, M. Chelliah, W. Ebisuzaki, W. Higgins, J. Janowiak, K. C. Mo, C. Ropelewski, J. Wang, A. Leetmaa, R. Reynolds, R. Jenne, and D. Joseph, 1996: The NCEP/NCAR 40-year reanalysis project. Bulletin of the American Meteorological Society, 77, 437-471. Karl, T. R. and R. W. Knight, 1998: Secular trends of precipitation amount, frequency, and intensity in the United States. Bulletin of the American Meteorological Society, 79, 231-241. Kendall, M. G., 1938: A new measure of rank correlation. Biometrika, 30, 81-93. Key, J. R. and A. C. K. Chan, 1999: Multidecadal global and regional trends in 1000 mb and 500 mb cyclone frequencies. Geophysical Research Letters, 26, 2053-2056. Klein Tank, A. M. G. K. and G. P. Konnen, 2003: Trends in indices of daily temperature and precipitation extremes in Europe, 1946-99. Journal of Climate, 16, 3665-3680. Kushner, P. J., I. M. Held, and T. L. Delworth, 2001: Southern Hemisphere atmospheric circulation response to global warming. Journal of Climate, 14, 2238-2249. Liebmann, B., G. N. Kiladis, J. A. Marengo, T. Ambrizzi, and J. D. Glick, 1999: Submonthly convective variability over South America and the South Atlantic convergence zone. Journal of Climate, 12, 1877-1891. Liebmann, B., G. N. Kiladis, C. S. Vera, A. C. Saulo, and L. M. V. Carvalho, 2004a: Subseasonal variations of rainfall in South America in the vicinity of the low-level jet east of the Andes and comparison to those in the South Atlantic convergence zone. Journal of Climate, 17, 3829-3842. Liebmann, B., C. S. Vera, L. M. V. Carvalho, I. Camilloni, M. P. Hoerling, V. R. Barros, J. Báez, and M. Bidegain, 2004b: An observed trend in central South American Precipitation. Journal of Climate, 17, 4357-4367. Manton, M. J. and N. Nicholls, 1999: Monitoring trends in extreme climate events. APN Newsletter, 1-3.

13 Manton, M. J., P. M. Della-Marta, M. R. Haylock, K. J. Hennessy, N. Nicholls, L. E. Chambers, D. A. Collins, G. Daw, A. Finet, D. Gunawan, K. Inape, H. Isobe, T. S. Kestin, P. Lefale, C. H. Leyu, T. Lwin, L. Maitrepierre, N. Ouprasitwong, C. M. Page, J. Pahalad, N. Plummer, M. J. Salinger, R. Suppiah, V. L. Tran, B. Trewin, I. Tibig, and D. Yee, 2001: Trends in extreme daily rainfall and temperature in Southeast Asia and the South Pacific: 1961-1998. International Journal of Climatology, 21, 269-284. Marengo, J. A., 1992: Interannual variability of surface climate in the Amazon basin. International Journal of Climatology, 12, 853-863. Marengo, J. A., 2004: Interdecadal variability and trends of rainfall across the Amazon basin. Theoretical and Applied Climatology, 78, 79-96. Marengo, J. A. and S. Hastenrath, 1993: Case-Studies of Extreme Climatic Events in the Amazon Basin. Journal of Climate, 6, 617-627. Marengo, J. A., J. Tomasella, and C. R. Uvo, 1998: Trends in streamflow and rainfall in tropical South America: Amazonia, eastern Brazil, and northwestern Peru. Journal of Geophysical Research-Atmospheres, 103, 1775-1783. Marengo, J. A., W. R. Soares, C. Saulo, and M. Nicolini, 2004: Climatology of the low-level jet east of the Andes as derived from the NCEP-NCAR reanalyses: Characteristics and temporal variability. Journal of Climate, 17, 2261-2280. Marengo, J. A., B. Liebmann, V. E. Kousky, N. P. Filizola, and I. C. Wainer, 2001: Onset and end of the rainy season in the Brazilian Amazon Basin. Journal of Climate, 14, 833-852. Matsuyama, H., J. A. Marengo, G. O. Obregon, and C. A. Nobre, 2002: Spatial and temporal variabilities of rainfall in tropical South America as derived from Climate Prediction Center merged analysis of precipitation. International Journal of Climatology, 22, 175-195. Minetti, J. L., 1998: Trends and jumps in the annual precipitation in South America, south of the 15 degrees S. Atmosfera, 11, 205-221. Minetti, J. L., W. M. Vargas, A. G. Poblete, L. R. Acuna, and G. Casagrande, 2003: Non-linear trends and low frequency oscillations in annual precipitation over Argentina and Chile, 1931-1999. Atmosfera, 16, 119-135. Nicholls, N. and W. Murray, 1999: Workshop on Indices and Indicators for Climate Extremes, Asheville, NC, USA, 3-6 June 1997 - Breakout Group B: Precipitation. Climatic Change, 42, 23- 29. Nicholls, N., G. V. Gruza, J. Jonzel, T. R. Karl, L. A. Ogallo, and D. E. Parker, 1996: Observed climate variability and change. Climate Change 1995, J. T. Houghton, L. G. Meira Filho, B. A. Callander, N. Harris, A. Kattenberg, and K. Maskell, Eds., Cambridge University Press, 133-192. Nogues-Paegle, J., C. Mechoso, R. Fu, H. Berbery, C. Winston, T. Chao, K. Cook, A. Diaz, D. Enfield, R. Ferreira, A. Grimm, V. Kousky, B. Liebmann, J. Marengo, K. Mo, D. Neelin, J. Paegle, A. Robertson, A. Seth, C. Vera, and J. Zhou, 2002: Progress in Pan American CLIVAR Research: Understanding the South American Monsoon. Meteorologica, 27, 1-30. Osborn, T. J., M. Hulme, P. D. Jones, and T. A. Basnett, 2000: Observed trends in the daily intensity of United Kingdom precipitation. International Journal of Climatology, 20, 347-364. Parker, D. E., E. Jackson, and E. B. Horton, 1995: The 1961-1990 GISST2.2 Sea Surface Temperature and Sea-Ice Climatology. Climate Research Technical Note No. 63, Hadley Centre for Climate Prediction and Research. Peterson, T. C., M. A. Taylor, R. Demeritte, D. L. Duncombe, S. Burton, F. Thompson, A. Porter, M. Mercedes, E. Villegas, R. S. Fils, A. K. Tank, A. Martis, R. Warner, A. Joyette, W. Mills, L.

14 Alexander, and B. Gleason, 2002: Recent changes in climate extremes in the Caribbean region. Journal of Geophysical Research-Atmospheres, 107, art. no.-4601. Pezza, A. B. and T. Ambrizzi, 2003: Variability of Southern Hemisphere cyclone and anticyclone behavior: Further analysis. Journal of Climate, 16, 1075-1083. Plummer, N., M. J. Salinger, N. Nicholls, R. Suppiah, K. J. Hennessy, R. M. Leighton, B. Trewin, C. M. Page, and J. M. Lough, 1999: Changes in climate extremes over the Australian region and New Zealand during the twentieth century. Climatic Change, 42, 183-202. Preisendorfer, R. W., F. W. Zwiers, and T. P. Barnett, 1981: Foundations of Principal Component Selection Rules. SIO Reference Series 81-4, Scripps Institution of Oceanography, 192 pp. Press, W. H., B. P. Flannery, S. A. Teukolsky, and W. T. Vetterling, 1986: Numerical Recipes : The Art of Scientific Computing. Cambridge University Press, 818 pp. Rasmusson, E. M. and T. H. Carpenter, 1982: Variations in Tropical Sea-Surface Temperature and Surface Wind Fields Associated with the Southern Oscillation El-Nino. Monthly Weather Review, 110, 354-384. Rogers, J. C., 1988: Precipitation Variability over the Caribbean and Tropical Americas Associated with the Southern Oscillation. Journal of Climate, 1, 172-182. Ropelewski, C. F. and M. S. Halpert, 1987: Global and Regional Scale Precipitation Patterns Associated with the El-Nino Southern Oscillation. Monthly Weather Review, 115, 1606-1626. Ropelewski, C. F. and M. S. Halpert, 1989: Precipitation Patterns Associated with the High Index Phase of the Southern Oscillation. Journal of Climate, 2, 268-284. Rusticucci, M. and O. Penalba, 2000: Interdecadal changes in the precipitation seasonal cycle over Southern South America and their relationship with surface temperature. Climate Research, 16, 1- 15. Sexton, D. M. H., 2001: The effect of stratospheric ozone depletion on the phase of the Antarctic Oscillation. Geophysical Research Letters, 28, 3697-3700. Simmonds, I. and K. Keay, 2000: Variability of Southern Hemisphere extratropical cyclone behavior, 1958-97. Journal of Climate, 13, 550-561. Simmonds, I., D. A. Jones, and D. J. Walland, 1998: Multi-decadal climate variability in the Antarctic region and global change. Annals of Glaciology, Vol 27, 1998, 27, 617-622. Smith, I. N., P. McIntosh, T. J. Ansell, C. J. C. Reason, and K. McInnes, 2000: Southwest Western Australian winter rainfall and its association with Indian Ocean climate variability. International Journal of Climatology, 20, 1913-1930. Thompson, D. W. J. and S. Solomon, 2002: Interpretation of recent Southern Hemisphere climate change. Science, 296, 895-899. Thompson, D. W. J., J. M. Wallace, and G. C. Hegerl, 2000: Annular modes in the extratropical circulation. Part II: Trends. Journal of Climate, 13, 1018-1036. Vaes, G., P. Willems, and J. Berlamont, 2002: 100 years of Belgian rainfall: are there trends? Water Science and Technology, 45, 55-61. Vera, C., W. Higgins, J. Gutzler, J. A. Marengo, R. Garreaud, J. Amador, D. Gochis, J. Nogues- Paegle, C. Zhang, T. Ambrizzi, C. Mechoso, and D. Lettenmaier, 2004: A unified vision of the American monsoon systems. Journal of Climate, Accepted. Vincent, L. A., T. C. Peterson, V. R. Barros, M. B. Marino, M. Rusticucci, G. C. Miranda, E. Ramirez, L. M. Alves, T. Ambrizzi, J. I. Barbosa de Brito, M. A. Berlato, A. M. Grimm, R. Jaildo dos Anjos, J. A. Marengo, P. R. Meira, L. Molion, D. F. Muncunil, D. Nechet, E. Rebello, J. R.

15 Abreu de Sousa, Y. M. T. Anunciação, J. Quintana, J. L. Santos, G. Ontaneda, J. Baez, G. Coronel, V. J. Garcia, I. Trebejo, M. Bidegain, V. Corradi, M. R. Haylock, and D. J. Karoly, 2004: Observed trends in indices of daily temperature extremes in South America 1960-2000. Journal of Climate, In preparation. Viney, N. R. and B. C. Bates, 2004: It never rains on Sunday: the prevalence and implications of untagged multi-day rainfall accumulations in the Australian high quality data set. International Journal of Climatology, 24, 1171-1192. Zhang, X., G. Hegerl, F. W. Zwiers, and J. Kenyon, 2004: Avoiding inhomogeneity in percentile- based indices of temperature extremes. Journal of Climate, Submitted.

16 Tables ID Indicator name Definitions Units PRCPTOT Wet-day precipitation Annual total precipitation from wet days mm SDII Simple daily intensity index Average precipitation from of wet days mm/day CDD Consecutive dry days Maximum number of consecutive dry days days CWD Consecutive wet days Maximum number of consecutive wet days days R10mm Heavy precipitation days Annual count of days when RR>=10mm days R20mm Very heavy precipitation days Annual count of days when RR>=20mm days R95p Very wet day precipitation Annual total precipitation when RR>95th percentile of 1961-90 mm R99p Extremely wet day precipitation Annual total precipitation when RR>99th percentile of 1961-90 mm RX1day Max 1-day precipitation Annual maximum 1-day precipitation mm RX5day Max 5-day precipitation Annual maximum consecutive 5-day precipitation mm Percentage of annual total precipitation from days with R95pTOT Very wet day proportion RR>=95th percentile of 1961-90. % Percentage of annual total precipitation from days with R99pTOT Extremely wet day proportion RR>=99th percentile of 1961-90. % Table 1: Rainfall indices with their definition and units. RR is the daily rainfall rate. A wet day has RR>=1mm. A dry day has RR<1mm. All indices are calculated annually from January-December.

17 Country Station Start End Longitude Latitude Elev. (m) Argentina Ceres 1960 2002 -61.95 -29.88 88 Esquel 1959 2002 -71.15 -42.93 797 Jujuy 1960 2002 -65.08 -24.38 905 Laboulage 1960 2002 -63.37 -34.13 137 Mar Del Plata 1959 2002 -57.58 -37.93 21 Mendoza 1960 2002 -68.78 -32.83 704 Monte Caseros 1959 2002 -57.65 -30.27 54 Neuquen 1960 2002 -68.13 -38.95 271 Posadas 1959 2002 -55.97 -27.37 125 Salta 1959 2002 -65.48 -24.85 1221 Santiago 1959 2002 -64.30 -27.77 199 Trelew 1959 2002 -65.27 -43.20 43 Bolivia Patacamaya 1946 1999 -67.92 -17.20 3789 Brazil Aguafunda 1933 2003 -46.37 -23.39 779 Aquiraz 1912 2001 -38.38 -3.90 14 Bage 1960 2000 -54.11 -31.33 214 Belem 1951 2000 -48.48 -1.46 10 Cambara 1959 2003 -50.03 -23.00 450 Campinas 1890 2004 -47.05 -22.54 694 Crato 1912 2001 -39.38 -7.22 412 Curitiba 1961 2003 -49.28 -25.43 910 Iguatu 1912 2001 -39.30 -6.37 216 Julio Castilhos 1957 1999 -53.68 -29.23 514 Manaus 1954 1999 -59.96 -3.13 67 Passo Fundo 1950 2000 -52.41 -28.26 687 Pelotas 1950 2000 -52.35 -31.75 7 Ponta Grossa 1954 2001 -50.03 -25.23 880 Säo Borja 1957 2000 -56.00 -28.66 99 Veranopolis 1957 1998 -51.55 -28.94 705 Chile Concepcion 1950 2003 -73.02 -36.46 11 Coyahique 1961 2003 -72.12 -45.58 310 La Serena 1950 2003 -71.10 -29.90 146 Pta Arenas 1953 2000 -70.50 -53.00 16 Puerto Montt 1950 2003 -73.10 -41.43 58 Santiago 1950 2003 -70.76 -33.45 520 Valdivia 1951 2003 -73.08 -39.63 19 Ecuador Izobamba 1966 2002 -78.55 -0.35 3058 Loja 1964 2000 -79.20 -4.04 2160 Pichilingue 1965 2002 -79.48 -1.11 120 Paraguay Concepcion 1959 1999 -57.43 -23.43 70 Encarnacion 1951 1999 -55.50 -27.17 85 Mariscal 1950 1999 -60.59 -22.02 165 Pto Casado 1947 1999 -57.52 -22.17 80 Villarrica 1956 1999 -56.48 -25.81 110 Peru Canete 1939 2001 -76.32 -13.07 158 Huayao 1952 2002 -75.30 -12.03 3308 Imata 1937 2002 -71.08 -15.83 4519 Iquitos 1957 1994 -73.30 -3.70 125 Piura 1956 1994 -80.62 -5.08 49 Weberbauer 1966 2003 -78.50 -7.17 2536 Uruguay Carrasco 1960 2000 -56.00 -34.83 33 Mercedes 1951 2000 -58.07 -33.25 17 Paysandu 1950 2000 -58.03 -32.50 61 Rocha 1950 2000 -54.30 -34.48 18 Tabel 2: Location, elevation and period of data availability for 54 stations

18 index component r SST % index % r(SST,SOI) r(index,SOI) PRCPTOT 1 0.876 12.9% 16.4% 0.791 0.672 PRCPTOT 2 0.762 25.7% 10.1% 0.046 0.144 PRCPTOT 3 0.605 6.9% 7.2% -0.308 -0.130 PRCPTOT 4 0.402 9.4% 10.3% -0.041 0.131 Total 54.9% 44.1% SDII 1 0.777 19.3% 12.5% 0.705 0.511 SDII 2 0.657 24.1% 7.7% 0.100 0.141 Total 43.4% 20.2% CDD 1 0.742 29.5% 7.1% -0.348 -0.211 CDD 2 0.697 4.4% 10.1% -0.399 -0.411 CDD 3 0.419 9.2% 9.1% 0.236 0.220 CDD 4 0.287 5.3% 6.7% -0.245 -0.210 Total 48.5% 33.1% CWD 1 0.708 7.7% 7.6% -0.575 -0.480 CWD 2 0.591 24.1% 8.1% 0.304 -0.002 CWD 3 0.333 12.5% 7.9% 0.110 0.167 Total 44.3% 23.6% R10mm 1 0.865 9.8% 11.8% -0.757 -0.665 R10mm 2 0.749 23.2% 13.2% -0.114 -0.027 R10mm 3 0.476 9.6% 10.7% -0.169 -0.199 Total 42.6% 35.7% R20mm 1 0.865 19.9% 14.2% 0.715 0.584 R20mm 2 0.825 22.2% 7.1% 0.176 0.227 R20mm 3 0.553 4.7% 7.8% 0.123 0.123 R20mm 4 0.264 7.5% 8.9% -0.050 -0.094 Total 54.2% 38.0% R95p 1 0.862 16.2% 13.2% 0.729 0.576 R95p 2 0.385 12.0% 7.2% -0.003 -0.162 Total 28.2% 20.3% R99p 1 0.835 20.4% 10.6% 0.749 0.544 R99p 2 0.474 7.2% 6.6% 0.081 0.107 Total 27.6% 17.2% RX1day 1 0.542 17.0% 8.6% -0.299 -0.221 RX1day 2 0.330 11.0% 10.4% -0.248 -0.017 Total 28.0% 19.0% RX5day 1 0.546 18.2% 9.9% -0.276 -0.205 RX5day 2 0.509 12.4% 7.5% -0.022 -0.029 RX5day 3 0.320 8.2% 8.3% -0.020 0.066 RX5day 4 0.157 8.1% 10.9% 0.064 0.045 Total 47.0% 36.5% R95pTOT 1 0.676 9.0% 8.2% 3.7% 0.551 R95pTOT 2 0.263 25.4% 8.1% 0.6% 0.174 Total 34.4% 16.3% R99pTOT 1 0.697 9.8% 9.1% 4.4% -0.506 R99pTOT 2 0.674 15.3% 7.3% 3.3% 0.359 R99pTOT 3 0.512 10.3% 10.9% 2.9% -0.238 Total 35.5% 27.3% Table 3: canonical correlation, % variance and correlation between canonical coefficients and the SOI for each coupled canonical pattern

19 Figure Captions Figure 1: location of 54 rainfall stations Figure 2: Sign of the linear trend in rainfall indices as measured by Kendall’s tau. An increase is showed by a ‘+’, a decrease by a ‘O’. Bold values indicate significant at p<0.05. Figure 3: 1st CCA patterns for SST with a) PRCPTOT, b) R20mm and c) R95p. For the indices, a ‘O’ indicates a negative loading and a ‘+’ is positive. The size of the symbol is proportional to the magnitude, with the maximum symbol size given in the scale at the top right of each frame. Figure 4: 1st canonical coefficient for PRCPTOT, R20mm and R95p with the annual SOI Figure 5: As for Figure 3 but for the 2nd CCA pattern for SST-PRCPTOT Figure 6: 2nd CCA coefficients for SST-PRCPTOT Figure 7: Annual average MSLP composited for years when a) 2nd SST-PRCPTOT canonical coefficient < 0 and b) 2nd SST-PRCPTOT canonical coefficient > 0. c) difference b) - a). Shading shows region where difference is significant at p<0.05.

20 Figures

Ecuador

Peru Brazil

Bolivia

Paraguay

Chile

Uruguay

Argentina

Figure 1: location of 54 rainfall stations

21 Figure 2: Sign of the linear trend in rainfall indices as measured by Kendall’s tau. An increase is showed by a ‘+’, a decrease by a ‘O’. Bold values indicate significant at p<0.05.

22 a

23 c

Figure 3: 1st CCA patterns for SST with a) PRCPTOT, b) R20mm and c) R95p. For the indices, a ‘O’ indicates a negative loading and a ‘+’ is positive. The size of the symbol is proportional to the magnitude, with the maximum symbol size given in the scale at the top right of each frame.

3 15

2 10

1 5

0 0 SOI -1

Precip coefficient -5 -2

-10 -3

-4 -15

1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000

PRCPTOT R20mm R95p SOI Linear (PRCPTOT) Linear (R20mm) Linear (R95p) Linear (SOI)

Figure 4: 1st canonical coefficient for PRCPTOT, R20mm and R95p with the annual SOI

24 Figure 5: As for Figure 3 but for the 2nd CCA pattern for SST-PRCPTOT

2.5

1.5

0.5

0 1960 1965 1970 1975 1980 1985 1990 1995 2000 Canonical Coefficient

-0.5

-1

-1.5

-2

PRCPTOT SST

Figure 6: 2nd CCA coefficients for SST-PRCPTOT

25 a

26 c

Figure 7: Annual average MSLP composited for years when a) 2nd SST-PRCPTOT canonical coefficient < 0 and b) 2nd SST-PRCPTOT canonical coefficient > 0. c) difference b) - a). Shading shows region where difference is significant at p<0.05.

27 Chapter 8: Haylock and Goodess, 2004

Haylock M and Goodess C. 2004. Interannual variability of European extreme winter rainfall and links with mean large-scale circulation. International Journal of Climatology, 24, 759-776. INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 24: 759–776 (2004) Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/joc.1033

INTERANNUAL VARIABILITY OF EUROPEAN EXTREME WINTER RAINFALL AND LINKS WITH MEAN LARGE-SCALE CIRCULATION

M. R. HAYLOCK* and C. M. GOODESS Climatic Research Unit, University of East Anglia, Norwich, UK Received 22 September 2003 Revised 26 January 2004 Accepted 28 January 2004

ABSTRACT December–February (DJF) extreme rainfall was analysed at 347 European stations for the period 1958–2000. Two indices of extreme rainfall were examined: the maximum number of consecutive dry days (CDD); and the number of days above the 1961–90 90th percentile of wet-day amounts (R90N). A principal component analysis of CDD found six components that accounted for 52.4% of the total variance. Six components of DJF R90N were also retained that accounted for 39.1% of the total variance. The second component of R90N has a very significant trend and the factor loadings closely resemble the observed linear trend in this index, suggesting that the analysis has isolated the mode of variability causing the trend as a separate component. The principal components of the indices were correlated with surface and upper-air observations over the North Atlantic. The best correlations were generally found to be with sea-level pressure (SLP) observations. A separate canonical correlation analysis of each of the two indices with SLP revealed several coupled modes of variability. The North Atlantic oscillation (NAO) was isolated as the first canonical pattern for R90N. For CDD the first two canonical coefficients of CDD were significantly correlated with the NAO index. Generally, the canonical coefficients with the highest correlations with the NAO had the most significant trends, suggesting that the observed trend in the NAO has strongly contributed to the observed trends in the indices. Two other important canonical patterns were isolated: a pattern of anomalous mean SLP (MSLP) centred over the North Sea, which seems to be related to local sea-surface temperature over this region; and a dipole-like pattern of MSLP with poles over the eastern Mediterranean and the central North Atlantic. Repeating the canonical correlation analysis with two other indices of extreme rainfall, the 90th percentile of wet day amounts and the maximum 10 day rainfall total, gives very similar coupled patterns. Copyright  2004 Royal Meteorological Society.

KEY WORDS: Europe; extreme rainfall; trend analysis; principal component analysis; canonical correlation analysis; North Atlantic oscillation

1. INTRODUCTION

Studies of changes in climate extremes have become more prevalent since the Second Assessment of the Intergovernmental Panel on Climate Change (Nicholls et al., 1996) determined that analyses of extremes were inadequate. Since then, studies of regional and global-scale changes in extremes have attempted to answer the question of whether or not the climate is becoming more extreme and variable. Analyses of changes in extreme rainfall have used both parametric (e.g. Groisman et al., 1999) and non- parametric methods. The parametric methods usually involve ﬁtting a suitable distribution to the data then analysing changes in the distribution’s parameters. Non-parametric methods have utilized a large suite of indices of extreme rainfall, which calculate such quantities as seasonal and annual percentiles, numbers of very wet days and proportion of total rain from very wet days. Most studies of changes in extreme rainfall have focused on linear trends in the indices. The aim of these studies has been to determine whether there has been a statistically signiﬁcant shift in such indices of

* Correspondence to: M. R. Haylock, Climatic Research Unit, University of East Anglia, Norwich NR4 7TJ, UK; e-mail: [email protected]

Copyright  2004 Royal Meteorological Society 760 M. R. HAYLOCK AND C. M. GOODESS extremes. Regional studies of such trends have included work in North America (e.g. Karl and Knight, 1998), Europe (e.g. Klein Tank and Konnen, 2003), the UK (e.g. Osborn et al., 2000), Australia (e.g. Haylock and Nicholls, 2000), Central America (e.g. Peterson et al., 2002) and Southeast Asia and the Pacific (e.g. Manton et al., 2001). For the IPCC Third Assessment Report (Folland et al., 2001), a near-global analysis of trends in indices was conducted by Frich et al. (2002). However, little work has been done on explaining trends in the indices or looking at the interannual variability. Two notable exceptions are Peterson et al. (2002), who correlated indices of Caribbean hot days and rainfall intensity with near-global sea-surface temperature (SST), and Gershunov (1998), who examined El Nino–southern˜ oscillation-based predictability of extreme rainfall over North America. Studies of European non-extreme regional rainfall have looked at variability and climate forcing. The North Atlantic oscillation (NAO) has received particular attention, with many studies finding it to be one of the major influences on European climate (e.g. Rogers, 1997; Qian et al., 2000; Trigo et al., 2002). Other studies have looked at relationships between European precipitation and other variables, such as 500 hPa circulation (Wibig, 1999; Quadrelli et al., 2001) and SST (Ye, 2001; Lloyd-Hughes and Saunders, 2002). This study extends the search for regional climate forcing over Europe, focusing on extreme winter rainfall. Although we could have studied other seasons as well, it was decided that a more thorough analysis of just one season would be of more value. Also, the lower spatial coherence of the extreme indices in other seasons, particularly summer, may make such a study in these seasons impractical. Note that, in our search for external forcing of rainfall extremes, we were not looking for particular circulation types that cause individual extreme rainfall events, but rather changes in the mean circulation that lead to changes in extreme rainfall. Data and methodology are discussed in Section 2. Section 3 looks at linear trends in two indices of rainfall extremes and Section 4 extracts the major modes of variability in the indices. The relationship between the indices and mean atmospheric circulation is examined in the following sections, with Section 5 looking at the relationship between the principal components (PCs) of the indices and other surface and upper-air variables. Section 6 uses canonical correlation analysis (CCA) to quantify the relationship between the indices and sea-level pressure (SLP) and compares the results with two other indices of extreme rainfall.

2. DATA AND METHODOLOGY

The European Commission-funded project Statistical and Regional Dynamical Downscaling of Extremes for European Regions (STARDEX) was created to improve methodologies for downscaling extreme rainfall and temperature from climate models. As part of STARDEX, 491 daily rainfall stations were compiled by Fundacion´ para la Investigacion´ del Clima (FIC), Spain, which form the basis for this present study. Approximately half the stations were obtained from the European Climate Assessment (ECA) data set (Klein Tank et al., 2002) with additions from national meteorological institutes for 14 countries and three other organizations (see Acknowledgements). The long-term homogeneity of the ECA data set was examined by Wijngaard et al. (2003). They found that, for the period 1946–99, inhomogeneities in 13% of the precipitation stations in the set made those stations unsuitable for trend analysis and variability analysis of weather extremes. Since many of the inhomogeneities found by Wijngaard et al. (2003) were in the early part of the record and we were considering a later period, we decided to do our own quality control. Data were not adjusted for inhomogeneities, but stations were quality checked by FIC using the following tests:

• Stations were checked for negative rainfall values. No such values were found in the data. • An analysis of spatial coherence was undertaken, looking at mean annual precipitation, the proportion of days with precipitation and the proportion of days with trace precipitation. Stations with unusually low or high values were analysed further, looking for a meteorological explanation for their singularity. If such an explanation was not found then the station was rejected. • An analysis of the spatial coherence of the daily discretized values (rainfall <0.1 mm or rainfall ≥0.1 mm) was carried out. The temporal correlation of nearby station discretized series was calculated and stations

Copyright  2004 Royal Meteorological Society Int. J. Climatol. 24: 759–776 (2004) VARIABILITY OF EUROPEAN EXTREME WINTER RAINFALL 761

with low correlation values were analysed further, looking for a meteorological cause for their particular behaviour. If such an explanation (e.g. complex orography) was not found then the station was rejected. • Stations with too many missing observations were rejected. At ﬁrst, stations were rejected with at least 10% missing values, but this was relaxed in countries with less available data. All stations contain at least 83% non-missing data, with countries other than Italy, Portugal, Greece and some eastern European countries containing at least 90%.

For our analysis we required a uniform spatial distribution of stations so as not to bias results towards regions with higher station density. However, the spatial distribution of the 491 stations is irregular, with a higher station density over central Europe and southern Scandinavia. Rather than gridding the station data we decided to thin the station network, as the ultimate aim of the exercise was to analyse regional variation in the extreme indices at the station scale for downscaling. Stations were thinned by dividing the region into 1° × 1° boxes and selecting the station in each box with the most complete record. In order that we did not end up with two very close stations in neighbouring boxes, the exercise was repeated three times with the boxes offset by 0.5° longitude, then 0.5° latitude and finally 0.5° longitude and latitude. The final data set (Figure 1) of 347 stations provides good, even coverage over most of Europe for the period January 1958 to December 2000. The largest spatial gaps in the coverage are in the eastern Adriatic region. Two indices of climate extremes are examined in detail in this study: R90N, the number of days with rain above the 1961–90 90th percentile calculated from wet days; and CDD, the maximum number of consecutive dry days. The results from the study of these two indices are compared with two other indices: RQ90, the 90th percentile of wet-day amounts; and R10D, the maximum 10 day rainfall total. A wet day is defined as having rainfall of at least 1 mm. This relatively high threshold was used as previous studies have found that lower thresholds can be sensitive to problems such as underreporting of small rainfall amounts and changes in the units of measurement (e.g. Hennessy et al., 1999). A dry day is defined as having less than 1 mm rainfall.

Figure 1. Location of the 347 daily rainfall stations

The CDD index is calculated by determining the maximum number of consecutive days with rainfall less than 1 mm. If the longest dry period begins before 1 December or finishes after 28 February then it will only be counted within these dates. The R90N index is calculated by first determining the 90th percentile threshold of all events greater than 1 mm for December–February (DJF) over the period 1961–90, then for each winter counting the number of events above this threshold. The period 1961–90 is used, firstly as it is the standard period for calculating normals as recommended by the WMO, and secondly as it is preferable to using the entire period, as it will not change depending on the length of record. Although the index could change for some years depending on the normals period used, it will be very highly correlated to using our chosen period. The RQ90 index is the 90th percentile of all days with rainfall greater than 1 mm. The R10D index is the maximum rainfall calculated with a running 10 day window. All indices give one value for each DJF of each year, thus providing an annual series with which to work. The indices are sensitive to the number of missing days. Therefore, a year is set to missing if there are more than 20% days missing for that season. Also, R90N values are normalized for the number of missing days, e.g. if there are 95% non-missing observations for a particular season then the R90N index for that season is adjusted by a factor of 100/95. This adjustment is not applied to CDD, as this index requires non-missing observations to be consecutive. A missing observation in the middle of the longest dry spell will truncate the counting of the length of the spell to the missing day. This was considered preferable to the alternative of assuming there was no rain for the missing day, as it favours conservative estimates of the longest dry spell. For R10D, the rainfall is totalled over 10 consecutive days regardless of how many days are missing in that period. The four indices are part of a suite of 33 indices that are calculated by the STARDEX Diagnostics Extremes Indices Software (available at http://www.cru.uea.ac.uk/cru/projects/stardex/). The CDD and R90N indices were selected from the set for detailed analysis for a number of reasons. Firstly, it was decided that a concerted effort to analyse just two indices would be of more value than trying to cover a larger number, particularly since the indices themselves are to a large degree inter-correlated. Therefore, we selected one index representing dry conditions (CDD) and one for wet conditions (R90N). Both indices are counts of days and are fairly easy to calculate and interpret. There are also statistical reasons for choosing these two indices. Both indices could be considered indicators of moderate climate extremes: events that have a return period of 1 year or less. However, with only 43 years of data, it is not possible to look at more extreme events and their relation to climate variability with any statistical confidence. Frei and Schar (2001) show the difficulty in detecting trends in very rare events. They determined that the probability of detecting a factor of 1.5 change in seasonal counts of events with an average return period of 30 days is 0.6 for a 100 year record. For 100-day events this probability drops to 0.2. Still, the focus on moderate climate extremes does not detract from their importance. R90N examines the top 10% of rain events in a season. The average of this index over all stations and years is less than three events per season; with thresholds ranging from 4.1 to 48.8 mm and an average of 13.0 mm. Clearly, these are still important events from an impact point of view. CDD is looking at the length of the longest dry event each year, also a sufficiently extreme event to have important consequences. In Section 5, we relate the PCs of CDD and R90N to surface and upper-air observations of other climate variables. Upper-air observations used in the analysis were geopotential height, relative humidity and temperature at 500, 700 and 850 hPa as well as SLP taken from the National Centers for Environmental Prediction (NCEP) reanalyses (Kalnay et al., 1996). Some work has been done on verifying the NCEP reanalyses for these variables over Europe and the North Atlantic. Notably, Reid at al. (2001) show that mean SLP (MSLP) is generally well simulated, although there are periods and regions where the reanalyses diverge from observations. There have been several comparisons of NCEP upper-air temperatures, with satellite observations (e.g. Basist and Chelliah, 1997; Shah and Rind, 1998) showing that the NCEP analyses are in good agreement with observations. However, Santer et al. (1999) compared NCEP upper-air temperatures globally with radiosonde and satellite data and found only the post-1979 data to be reliably simulated. Despite these problems with data quality, the NCEP data set is still the best available upper-air data set for our analysis, although data quality will be an important consideration when looking for potential predictors for the extremes indices.

Copyright  2004 Royal Meteorological Society Int. J. Climatol. 24: 759–776 (2004) VARIABILITY OF EUROPEAN EXTREME WINTER RAINFALL 763

SST observations are also used in this study and were taken from the Global Sea Ice Coverage and Sea Surface Temperature (GISST) data set version 2.3b (Parker et al., 1995). In our analysis of the inﬂuence of the NAO on the indices, we have chosen to use Gibraltar–Iceland pressure difference (Hurrell, 1995; Jones et al., 1997). Although there are several ways of calculating an index of the NAO, Osborn et al. (1999) show that they are all very similar and they discuss the advantages of using our chosen index. The Gibraltar–Iceland difference has a stronger winter signal than using Azores as the southern station; and, unlike a PC analysis (PCA)-based approach, it is independent of the region used. Throughout our analysis we have used a DJF average of this index.

3. TRENDS IN DJF R90N AND CDD FOR 1958–2000

Figure 2 shows the linear trends in CDD and R90N for the 347 stations for the period 1958–2000. There have been coherent changes in both indices over the region, with a change to wetter conditions in the north (increased R90N and decreased CDD) and the opposite in the south. This is reflected by a similar pattern in the change in total rainfall (not shown). Superimposed on the large-scale north–south divide in trends are more local patterns of change, made apparent by the good spatial coverage of the network. A good example is the Iberian Peninsula, where the northwest has seen a small decrease in CDD (Figure 2(a)) while the rest of the peninsula has seen large increases in this index. In contrast, the R90N index (Figure 2(b)) shows a decrease over most of the peninsula (including the northwest), but a slight increase in the southeast. Table I gives a summary of the distribution of the trends. All trends have been calculated using a three-group resistant line method (Hoaglin et al., 1983) with statistical significance determined using the Kendall tau test (Press et al., 1986). Where a trend is indicated as ‘significant’, it has at least 95% significance using this test. The three-group resistant line method is more resistant to outliers than least-squares linear regression, a property derived from the fact that one of the most resistant measures of a sample is the median. This method divides the series (by time) into thirds and determines the trend of the line through the median of the first and last thirds. 50.1% of the stations show an increase in CDD and 47.6% for R90N. Whereas only 8.6% of stations show a significant trend in the CDD index, 19.0% exhibit a significant trend in R90N. Still, the

(a) (b)

Figure 2. (a) Linear trend in DJF CDD for 1958–2000. A ‘+’ signiﬁes an increase and a ‘ ’ shows a decrease. The size of the symbol is linearly proportional to the magnitude of the trend. Units are days/year and the maximum° trend magnitude is shown in the top right. (b) As for (a) but for R90N

Table I. Statistics of trends in DJF CDD and R90N for 1958–2000. Trends are in days/year

CDD R90N

Average 0.023 0.012 Average magnitude 0.104 0.039 Maximum magnitude 0.718 0.182 Proportion increasing 0.501 0.476 Proportion increasing significantly 0.040 0.141 Proportion decreasing 0.383 0.305 Proportion decreasing significantly 0.046 0.049 high spatial coherence suggests that many non-significant trends are regionally coherent, although the lack of significance suggests they are not sufficiently large compared with interannual variability. Nicholls (2001) discusses further the danger in adopting a dichotomous ‘significant/not-significant’ attitude to significance testing and using a common, but arbitrary (e.g. 5%), level. The almost equal number of stations showing increases and decreases in the two indices means the trends averaged over all stations are quite low (0.023 days/year for CDD and 0.012 days/year for R90N); however, ignoring the sign, the average magnitudes are about three to four times these values. The maximum station trends are much higher: 0.718 days/year for CDD and 0.182 days/year for R90N. It is difficult to compare these observed trends with previous studies as we have focused on DJF rainfall only, whereas other studies of extremes have looked at annual rainfall. Klein Tank and Konnen (2003) looked at trends in annual indices of the number of very wet days, the largest n-day rainfall totals and the fraction of total rainfall from extreme events. They found that there has been a general increase in annual extreme precipitation across Europe over the period 1946–99. Our results additionally show that there has been a strong north–south divide in the sign of the trend in extreme winter rainfall.

4. PRINCIPAL COMPONENTS OF R90N AND CDD

In order to determine the major modes of interannual variability in the two indices, the time series of the indices at the 347 stations were analysed using PCA. A separate analysis was carried out for the two indices using a singular value decomposition (SVD) of the matrix of standardized anomalies. The SVD approach to PCA produces similar results to ﬁnding the eigenvectors of the correlation matrix, but it is numerically more stable (Press et al., 1986). The PCs were normalized to unit variance by dividing by the square root of the corresponding eigenvalue. The factor loadings were calculated by scaling the eigenvectors by the square root of the eigenvalue. Therefore, the factor loadings represent the correlation between the PCs and the raw data. Factor loadings were rotated using varimax rotation in order to simplify the structure and relax the orthogonality constraint imposed on unrotated components (Richman, 1986). The number of components retained for rotation was selected by a Monte Carlo process, whereby 1000 PCAs were carried out using data randomly resampled in time from the stations (Preisendorfer et al., 1981). In each of the 1000 analyses, 347 station annual series of length 43 years were generated with similar statistical properties to the original data but with random interstation correlations. Then, for each randomization, the eigenvalues were calculated. Each of the eigenvalues of the real observations was then compared against the distribution of the 1000 randomly generated values to determine whether they were greater than the rank 50 eigenvalue (equivalent to p<0.05). This process resulted in six rotated components for CDD and six for R90N. Similar results are obtained by the more subjective method of determining a break in the scree plot of eigenvalues (Wilks, 1995). Table II summarizes the proportion of variance explained and the trends in the rotated PCs. Trends were calculated using the three-group resistant line method (Hoaglin et al., 1983), as described in Section 3. The higher proportion of variance explained for CDD (52.4%) compared with R90N (39.1%) implies that this index varies with greater spatial coherence. However, the total variance for both indices is still low, due to

Copyright  2004 Royal Meteorological Society Int. J. Climatol. 24: 759–776 (2004) VARIABILITY OF EUROPEAN EXTREME WINTER RAINFALL 765

Table II. Variance, trends and correlation with NAO for rotated PCs of R90N. Bold values are signiﬁcant at p<0.05

Variance (%) Trend p (trend) r (NAO)

CDD PC1 13.5 0.029 0.352 0.406 CDD PC2 10.2 −0.008 0.470 −0.185 CDD PC3 7.9 0.009 0.408 0.467 CDD PC4 7.2 −0.014 0.096 −0.061 CDD PC5 6.8 −0.016 0.164 −0.450 CDD PC6 6.8 −0.005 0.942 −0.080 Total 52.4 R90N PC1 10.3 0.001 0.205 0.378 R90N PC2 7.5 0.051 0.002 0.654 R90N PC3 7.1 0.004 0.746 −0.463 R90N PC4 6.1 −0.017 0.699 0.024 R90N PC5 4.1 0.015 0.096 −0.053 R90N PC6 4.0 0.003 0.875 −0.144 Total 39.1 the noisier nature of extremes compared with mean values and the large region under study. A separate PCA was carried out for total DJF rainfall (not shown), resulting in six significant components explaining 61.3% of the total variance. Although this is greater than CDD and R90N, it still leaves over a third of the variance unaccounted for. This suggests that DJF rainfall across the region is generally heterogeneous. Nevertheless, there is still sufficient variance explained by the PCs of CDD and R90N to show that regional modes of variability are important. For CDD and R90N, only one of the components has a significant trend (R90N PC2), suggesting that through this analysis we might only be able to explain the trend in R90N and not CDD. This is supported by the fact that, as stated earlier, a much higher proportion of stations have significant trends in R90N (19.0%) than CDD (8.6%) The factor loadings of the first two components for both the indices are shown in Figure 3. The first rotated component of CDD explains 13.5% of the variance and has no significant trend (Table II). The corresponding factor loadings (Figure 3(a)) have positive loadings over the central latitudes and southeast region and negative loadings over Scandinavia, the western UK and the southern Iberian Peninsula. It is interesting to note that this dominant mode of variability shows out-of-phase behaviour between the central region and the far north and southwest. This will be examined further in Section 6. The second PC of CDD explains 10.2% of the variance and has no significant trend (Table II). The highest loadings (Figure 3(b)) are all positive and concentrated in the central region, covering the UK, central and southern Scandinavia and central Europe, with smaller negative loadings over the southern region. The higher components of CDD (not shown) are more regionally focused: PC3 has its highest loadings over the western Mediterranean and southern Scandinavia, PC4 relates to eastern Europe, PC5 relates to the countries bordering the North Sea, and PC6 to Scandinavia. The first rotated component of R90N explains 10.3% of the variance and, like PC2 of CDD, has loadings (Figure 3(c)) that reflect an out-of-phase behaviour between the central and southern latitudes of the region. Similar to PC2 of CDD, the northern Norwegian coast and northern Finland also have loadings of opposite sign to the central part of the region. The second component of R90N explains 7.5% of the variance and has a statistically significant positive trend (Table II). The factor loadings (Figure 3(d)) resemble the linear trends in R90N (Figure 2(b)), with generally positive loadings in the north and negative in the south. The higher components of R90N (not shown) are more regionally focused: the third component, explaining 7.1% of the variance, has loadings very similar to the third component of CDD, except that the higher loadings cover most of the western part of the region rather than just the western Mediterranean and parts of Scandinavia.

(a) (b)

Figure 3. (a) Factor loadings of first rotated PC of CDD. A ‘+’ signifies a positive loading and a ‘ ’ indicates a negative loading. The size of the symbol is linearly proportional to the magnitude of the loading. The maximum loading° magnitude is shown in the top right. (b) As for (a) but for the second rotated PC of CDD. (c) As for (a) but for the first rotated PC of R90N. (d) As for (a) but for the second rotated PC of R90N

The fourth component, explaining 6.1% of the variance, is more local, with the highest loadings over the Alps and eastern Europe. The fifth and sixth components each explain less than 5% of the variance and give loadings with several small regions of coherent variability. In order to see how the PCs of one index relate to the other, a table of correlations between the PCs of both climate indices was constructed (Table III). Although there are several statistically significant correlations, the only correlation higher than 0.5 (25% explained variance) is between CDD PC3 and R90N PC3. The similarity in the pattern of factor loadings for these two components was noted earlier. Also worth mentioning is the correlation of 0.47 between CDD PC1 and R90N PC2, although the factor loadings show quite different patterns. The correlation of 0.39 between CDD PC2 and R90N PC1 reflects the similarity in the factor loadings discussed earlier.

Copyright  2004 Royal Meteorological Society Int. J. Climatol. 24: 759–776 (2004) VARIABILITY OF EUROPEAN EXTREME WINTER RAINFALL 767

Table III. Correlation between the rotated PCs of CDD and R90N. Bold values are signiﬁcant at p<0.05

CDD 1 CDD 2 CDD 3 CDD 4 CDD 5 CDD 6

R90N 1 −0.25 −0.39 0.11 −0.25 −0.23 −0.14 R90N 2 0.47 0.09 0.23 0.05 −0.29 −0.19 R90N 3 −0.20 0.22 −0.63 0.06 −0.14 0.09 R90N 4 −0.23 −0.12 −0.08 −0.28 −0.34 −0.03 R90N 5 −0.15 0.16 0.30 −0.43 0.21 0.14 R90N 6 0.17 −0.25 −0.14 −0.27 0.30 −0.09

5. RELATIONSHIP BETWEEN THE PRINCIPAL COMPONENTS OF CDD AND R90N AND OTHER VARIABLES

In order to determine possible large-scale forcing of the observed variability in the two indices, the PCs of the indices were first correlated with NCEP surface and upper-air observations and GISST observations. Upper-air observations used in the analysis were geopotential height, relative humidity and temperature at 500, 700 and 850 hPa. Surface observations included SLP and SST. These variables were chosen as being the most likely large-scale predictors of extreme rainfall. As we were looking for changes in extremes caused by changes in the mean circulation (see Section 1), all correlations were performed on DJF averages of the above variables, which were first interpolated to an equal-area grid (approximately 280 km × 280 km at the equator) over the region 20–80 °N, 60 °W–60°E. An equal-area grid was used so as not to bias regional averages to the higher latitudes. Since all correlations were calculated between concurrent predictor–predictand observations, we use the term ‘predictor’ not to imply that we would expect any skill with a lag correlation, but rather to imply that these variables are statistically linked to the extreme wet or dry conditions. The aim of this preliminary screening exercise was to assess which variables were generally the most important in forcing the indices, rather than identifying individual predictors for each of the PCs of the indices. Once the important predictors had been established, the relationship with the indices would be quantified using CCA (see Section 6). In this preliminary analysis we are only interested in patterns of high correlation that cover a region larger than a few grid cells and provide a physically explainable mechanism to affect extreme dry or wet conditions. Table IV shows a summary of the correlations between each PC of CDD and the 11 predictors. Since we are not interested in the sign of the correlation, we have considered the absolute values of the correlation. Table IV gives the absolute correlation averaged across the entire predictor grid, as well as the proportion of grid points with significant correlations. Where average absolute correlations are higher, the proportion of grid points with significant correlations is higher. The correlations varied widely, with generally the highest values for PC3 (0.19 to 0.37) and the lowest for PC6 (0.09 to 0.20). Correlations only showed slight differences for the same variable at different atmospheric levels, compared with the larger differences for different variables at the same level. For some PCs the upper levels had higher correlations (e.g. PC5), but for most components the lower levels were higher (e.g. PC1). In general, the correlations were highest for pressure parameters, followed by air temperature, humidity and SST. Table IV also shows the area-average absolute correlation and significant proportion averaged across all PCs. Averaging across all PCs showed MSLP to be the best predictor, with an average correlation magnitude of 0.22 and 28% of the grid points significant. SST was the poorest, with an average of 0.15 and only 10% of grid points significant. Although these average correlations are low, they reflect an average across the entire North Atlantic region and contain large-scale centres of low and high correlations that are much higher. An example is the map of correlations between MSLP and CDD PC3 (not shown): this has an average magnitude of 0.37 over the entire grid, but it contains a large centre of high positive correlation up to 0.74 located southwest of the UK and an equally strong centre of high negative correlation located north of Scandinavia. This pattern of correlation resembles the correlation between MSLP and the NAO, reflecting the significant correlation between CDD PC3 and the NAO (Table II).

Copyright  2004 Royal Meteorological Society Int. J. Climatol. 24: 759–776 (2004) 768 M. R. HAYLOCK AND C. M. GOODESS are (%) Significance r (%) Significance r (%) Significance r (%) 0.120.120.15 0.260.11 0.270.11 0.25 0.390.10 0.21 0.37 0.21 0.31 0.10 0.19 0.24 0.11 0.26 0.100.02 0.04 0.20 0.11 0.03 0.09 0.01 0.14 0.22 0.10 0.030.10 0.21 0.030.12 0.21 0.05 0.26 0.020.17 0.20 0.15 0.24 0.14 0.20 0.15 0.24 0.200.19 0.14 0.19 0.23 0.140.20 0.14 0.23 0.13 0.16 0.13 0.13 0.20 0.14 0.14 0.12 0.14 0.15 0.11 0.150.02 0.11 0.09 0.13 0.11 0.10 0.13 0.12 0.14 0.24 0.13 0.03 0.23 0.04 0.23 0.06 0.30 0.05 0.22 0.29 0.22 0.30 0.09 0.22 0.27 0.26 0.03 0.27 0.14 0.09 Significance so shows proportion of grid points with significant correlations. Results solute correlation and significant proportion averaged across all PCs 779 0.13 0.11 0.10 0.17 0.19 0.18 0.17 0.20 0.16 0.12 0.12 0.13 0.045 0.037 0.03 0.186 0.08 0.17 0.11 0.17 0.16 0.13 0.16 0.14 0.17 0.14 0.15 0.14 0.14 0.10 0.13 0.16 0.16 0.06 0.10 0.09 0.20 0.19 0.18 0.22 0.18 0.17 17 0.13 0.23 0.29 0.10 0.0515 0.22 0.11 0.28 0.15 0.13 0.14 0.09 0.24 0.31 r (%) Significance r (%) Significance r (%) PC1 PC2 PC3 PC4 PC5 PC6 Average Significance averaged across entire predictor grid. The last column shows the ab r Table IV. Absolute correlations between PCs of CDD, R90N and predictors. Al CDD MSLP 0.35 0.58 0.12 0.05 0.37 0.61 0. 850Z700Z 0.32500Z 0.30850T 0.29 0.49700T 0.30 0.44500T 0.29 0.42850RH 0.12 0.27 0.46 0.21700RH 0.12 0.41 0.19500RH 0.12 0.05 0.36 0.18SST 0.11 0.26 0.05 0.11R90N 0.20 0.04 0.32 0.15 0.10MSLP 0.19 0.04 0.14 0.30850Z 0.19 0.04 0.13 0.33 0.49700Z 0.11 0.02 0.15 0.31 0.20 0.05 0.43500Z 0.20 0.33 0.22 0.05 0.52850T 0.18 0.12 0.33 0.23 0.07 0.50 0.23 0.24700T 0.18 0.22 0.50 0.52 0.26 0.22500T 0.18 0.21 0.05 0.30 0.53 0.22850RH 0.17 0.48 0.31 0.23 0.25 0.19700RH 0.86 0.15 0.42 0.27 0.25 0.19 0.17500RH 0.33 0.16 0.27 0.83 0.1 0.27 0.19SST 0.33 0.18 0.33 0.70 0.1 0.31 0.20 0.14 0.51 0.1 0.29 0.14 0.30 0.21 0.51 0.32 0.48 0.31 0.47 0.10 0.27 0.34 0.42 0.07 0.45 0.24 0.33 0.55 0. 0.44 0.34 0.38 0.55 0.15 0.18 0.33 0.32 0.55 0.21 0.17 0.52 0.20 0.19 0.16 0.54 0.19 0.18 0.27 0.19 0.18 0.19 0.19 0.16 0.1 0.1 0.21 0.1 0.10

Copyright  2004 Royal Meteorological Society Int. J. Climatol. 24: 759–776 (2004) VARIABILITY OF EUROPEAN EXTREME WINTER RAINFALL 769

The same process was carried out for the PCs of R90N, and results are also summarized in Table IV. Correlations were generally slightly higher for R90N than for CDD, as reflected in the higher average correlations across all six PCs for each predictor. Values were generally highest for PC2 (0.18 to 0.50) and lowest for PC6 (0.09 to 0.16). Averaging across all PCs showed MSLP to have the highest correlations, followed by geopotential height, temperature, humidity and then SST. Like CDD, some PCs were better correlated with variables at a higher atmospheric level (e.g. PC1), but most were more highly correlated with lower level variables (e.g. PC2). It is interesting to note in the above analysis that even the predictor with the lowest average correlation, SST, contains large regions where the absolute correlation is greater than 0.3, which is statistically significant (p < 0.05). This occurs for five of the PCs of CDD and five of R90N. Therefore, the aim here is not just to isolate predictors with statistically significant correlations with the PCs of the extremes indices, but also to find the best predictor. MSLP has large areas of absolute correlation greater than 0.3 (up to 0.83) for all PCs of CDD and R90N. Table II shows the degree to which the CDD PCs are correlated with the NAO. Significant correlations between the PC and NAO occur for three of the PCs, suggesting that, though the NAO is probably an important predictor, the PC analysis of the predictand alone has not isolated the NAO as a separate mode of variability. Table II shows the correlation between the PCs of R90N and the NAO. Three of the components have significant correlations, suggesting that the NAO has not been isolated as an independent component by this analysis, although the correlation of 0.65 for PC2 is much higher than the others.

6. CANONICAL CORRELATION ANALYSIS OF CDD AND R90N WITH SEA-LEVEL PRESSURE

Section 5 shows that MSLP is statistically the best predictor for CDD and R90N when considering large-scale predictors related to regional variability in the indices. Studies of validation of the NCEP reanalyses also show MSLP to be one of the most reliable variables (see Section 2). In order to quantify the relationship between the indices and MSLP, a CCA was carried out between the two fields. MSLP was used over the same region as for Section 5. The canonical patterns and series were calculated using an SVD of the cross-covariance matrix of the PCs of the two fields. This is numerically more stable than the more common method of working with the joint variance–covariance matrix (Press et al., 1986) and also incorporates the pre-filtering of the data by using just the significant PCs (Barnett and Preisendorfer, 1987). First, a PCA was carried out for MSLP using the same SVD methodology as for CDD and R90N (Section 4). The number of components retained was determined by the Monte Carlo simulation discussed previously (Section 4), resulting in six PCs of MSLP that account for 87.2% of the total variance. A separate CCA was carried out using six PCs of MSLP with six of CDD and six PCs of MSLP with six of R90N. Table V summarizes the properties of the canonical correlation coefficients for CDD with MSLP and R90N with MSLP, showing the canonical correlations, the trends with p-value (using Kendall tau test) of the canonical coefficients, the proportion of variance explained by the coefficients and the correlation between the coefficients and the DJF NAO index. Again, trends were calculated using the three-group resistant line method (Hoaglin et al., 1983). For both indices, the pair of coefficients with the highest canonical correlation explains the highest proportion of variance in MSLP (but not the index) and has the highest correlation with the NAO. For R90N (Table V), only the first coefficient has a significant correlation with the NAO. In the case of CDD (Table V), the correlations with the NAO are not as high as for R90N, but the first, second and, to a lesser extent, fifth MSLP coefficients all have significant correlations with the NAO. Since only one of the MSLP coefficients in the MSLP–R90N analysis is significantly correlated with the NAO, the analysis has isolated the NAO signal for R90N but not for CDD. The first canonical patterns for MSLP and R90N are shown in Figure 4. The values plotted are equivalent to the correlations between the first canonical coefficients and the raw data. The MSLP pattern (Figure 4(a)) strongly resembles the well-documented NAO pattern (e.g. Osborn et al., 1999), calculated as either the

Table V. Properties of the canonical coefficients for the CDD–MSLP and R90N–MSLP CCA showing the canonical correlation, the trend with significance of the canonical coefficients, the variance explained by the canonical coefficients and the correlation between the canonical coefficients and the NAO. Bold values are significant at p<0.05

CCA R (CCA) Trend p (trend) Variance (%) r (NAO)

CDD–MSLP MSLP 1 0.028 0.124 26.7 0.582 0.946 CDD 1 0.013 0.310 13.4 0.529 MSLP 2 −0.029 0.100 16.2 −0.561 0.862 CDD 2 −0.011 0.458 7.0 −0.483 MSLP 3 0.005 0.892 8.8 0.184 0.828 CDD 3 −0.021 0.746 15.6 0.181 MSLP 4 −0.024 0.330 11.9 −0.253 0.696 CDD 4 −0.018 0.300 5.8 −0.177 MSLP 5 −0.052 0.001 15.8 −0.357 0.455 CDD 5 −0.011 0.408 6.0 −0.228 MSLP 6 0.014 0.221 7.7 −0.182 0.063 CDD 6 −0.004 0.958 4.7 0.024 R90N–MSLP MSLP 1 0.053 0.001 42.4 0.905 0.971 R90N 1 0.028 0.002 10.6 0.865 MSLP 2 −0.008 0.579 9.4 0.188 0.929 R90N 2 −0.011 0.942 8.8 0.154 MSLP 3 0.006 0.551 9.7 0.166 0.851 R90N 3 0.006 0.908 6.8 0.171 MSLP 4 0.035 0.043 11.9 0.158 0.599 R90N 4 0.025 0.088 4.7 −0.051 MSLP 5 0.043 0.022 7.3 0.066 0.342 R90N 5 0.002 0.699 4.7 0.005 MSLP 6 −0.018 0.092 6.6 −0.024 0.003 R90N 6 0.023 0.077 3.9 0.061

correlation between the NAO index and MSLP or the first PC of DJF MSLP. This pattern explains a high 42.4% of the variance in MSLP over the region. The R90N pattern (Figure 4(b)) resembles the trends in R90N (Figure 2(b)) and the second PC of R90N (Figure 3(d)). Table V shows that the first canonical coefficients of MSLP and R90N both have a significant trend. Also, the NAO has a significant trend over this period (not shown), with a tendency to lower pressures in the North Atlantic and higher pressure in mid latitudes. One can conclude that the observed regional trend in R90N over this period is caused to a large extent by the trend in the NAO. For CDD both the first two MSLP coefficients have correlations with the NAO of greater than 0.5 and the second coefficient has the most significant trend of the two coefficients (p < 0.11). The moderate correlation with the NAO means that the patterns (not shown) are slight variations on the NAO pattern, with out-of-phase behaviour between the high and mid latitudes. In the case of the first pattern, the poles of the oscillation are located over the Iberian Peninsula and the far northeast part of the region. For the second pattern, the southern pole is located further east. The variance contributed by the first two patterns of MSLP is 42.9%, close to the 42.4% explained by the first pattern of MSLP for the CCA with R90N. Therefore, the NAO signal for the CCA analysis with CDD is divided between the first two components. Explaining the trend in CDD is a little more difficult than for R90N, as none of the coefficients for CDD has a statistically significant trend. The fact that the trends in CDD (Figure 2(a)) have approximately the same pattern (but opposite sign) as for R90N (Figure 2(b)) suggests that the NAO is perhaps the largest cause of the observed trends in CDD, but this is not supported by the CCA with MSLP.

Copyright  2004 Royal Meteorological Society Int. J. Climatol. 24: 759–776 (2004) VARIABILITY OF EUROPEAN EXTREME WINTER RAINFALL 771

(a) (b)

Figure 4. (a) First canonical pattern of MSLP for analysis with R90N. Contour interval is 0.2 with negative loadings indicated by dashed contours. (b) First canonical pattern of R90N for analysis with MSLP. A ‘+’ signiﬁes a positive loading and a ‘ ’ indicates a negative loading. The size of the symbol is linearly proportional to the magnitude of the loading. The maximum loading° magnitude is shown in the top right

Figure 5. Second canonical pattern of MSLP for analysis with R90N. Contour interval is 0.2 with negative loadings indicated by dashed contours

The canonical pattern of MSLP for the third coefficient of the CDD analysis (not shown) is very similar to the second pattern for the R90N analysis (Figure 5). The associated patterns for CDD and R90N (not shown) are very similar to PC2 of CDD and PC1 of R90N (Figure 3(b) and (c)), with the highest loadings of the same sign located over the central and northern regions and opposite sign in the south and far north. As with the PCs, these canonical coefficients have fairly low correlations with the NAO. In Section 5, when exploring relationships between the PCs of the indices and surface and upper-air observations over the North Atlantic, relatively high correlations (0.5 to 0.7) were seen between the first PC of CDD and SST over the North Sea. This is a likely cause for this pressure pattern centred over this region. Anomalously high pressure centred over the North Sea is reflected in a general increase in CDD and decrease in R90N over the central part of the region; but the opposite occurs over southeastern Spain, where anomalous easterly winds bring wetter conditions, and similarly in northern Norway with anomalous westerly winds. The canonical pattern of MSLP for the fourth coefficient of the CDD analysis (not shown) is very similar to the fourth pattern for the R90N analysis (Figure 6). The pattern reflects a dipole-like behaviour in MSLP, with

Figure 6. Third canonical pattern of MSLP for analysis with R90N. Contour interval is 0.2 with negative loadings indicated by dashed contours poles over the eastern Mediterranean and the central North Atlantic and explains 11.9% of the total variance in MSLP. The corresponding canonical patterns for the indices show a corresponding northwest–southeast divide in the signs of the correlations. The variance explained by this pattern is not large. The fourth canonical pattern of R90N explains only 4.7% of the variance and the fourth canonical pattern of CDD explains only 5.8%. The only dominant canonical pattern not mentioned is the third pattern of the R90N analysis. The canonical pattern for MSLP (not shown) shows a latitudinal band of positive loadings across the centre of the region, explaining 9.7% of the variance in MSLP. The corresponding canonical pattern for R90N shows an out-of- phase relationship between the central latitudes and the northern and southern regions. The remaining two CDD and two R90N canonical patterns have much lower canonical correlations (<0.5) and proportion of variance and so will not be discussed. A possible reason why the NAO-related (first) R90N canonical pattern in the R90N–MSLP CCA (Figure 4(b)) is so similar to the pattern of linear trends for this index (Figure 2(b)) might be due to the correlation between the trends in the NAO and the index rather than the interannual variability. To test this, the R90N–MSLP CCA was repeated with detrended data. The results (not shown) reveal an almost identical series of canonical patterns. There is, however, a large drop in the variance explained by the first MSLP canonical coefficient, from 42.4% to 32.5%, as this coefficient no longer contains a significant trend. The variance explained by the corresponding R90N coefficient drops from 10.6% to 9.9%. In other studies, the NAO has been found to be the most important influence on regional mean winter rainfall. Wibig (1999) correlated rainfall with PCs of 500 hPa geopotential height and found the NAO-related component to have the highest correlations. Interestingly other major modes of variability in 500 hPa geopotential height (a pattern centred on Scandinavia and a central European blocking pattern) do not appear in our analysis of extremes. Qian et al. (2000) found the NAO to be only the second most important PC of MSLP affecting annual rainfall, suggesting that it is most active during winter. The most important influence on annual rainfall came from their North Sea pattern of MSLP, similar to our pattern of MSLP centred over this region. The similarity between the patterns from the CCA of CDD and R90N with MSLP suggests that we might see the same patterns in an analysis with other indices of extreme rainfall. To test this, the CCA with MSLP was repeated with two other indices: RQ90, the 90th percentile of wet-day amounts; and R10D, the maximum 10 day rainfall total. A more detailed discussion about the calculation of these indices is given in Section 2. Seven principal components of RQ90 and R10D were retained in the analysis, as determined by the Monte Carlo resampling method (Section 4), retaining 35.1% of the variance for RQ90 and 38.1% for R10D. Table VI gives a summary of the results for the two indices. An examination of Table VI and the canonical patterns (not shown) shows that the results are very similar to the CCA for R90N and MSLP. The first coefficients have a high correlation with the NAO and a significant trend. The corresponding canonical

Copyright  2004 Royal Meteorological Society Int. J. Climatol. 24: 759–776 (2004) VARIABILITY OF EUROPEAN EXTREME WINTER RAINFALL 773

Table VI. Properties of the canonical coefficients for the RQ90–MSLP and R10D–MSLP CCA showing the canonical correlation, the trend with significance of the canonical coefficients, the variance explained by the canonical coefficients and the correlation between the canonical coefficients and the NAO. Bold values are significant at p<0.05

CCA R (CCA) Trend p (trend) Variance (%) r (NAO)

RQ90–MSLP MSLP 1 −0.052 0.004 41.6 −0.899 0.917 RQ90 1 −0.036 0.027 7.5 −0.805 MSLP 2 −0.002 0.730 8.8 0.044 0.843 RQ90 2 −0.002 0.496 6.4 0.009 MSLP 3 −0.069 0.003 14.4 −0.176 0.583 RQ90 3 −0.022 0.074 4.6 −0.072 MSLP 4 0.003 0.683 5.8 0.203 0.576 RQ90 4 0.007 0.746 4.8 −0.001 MSLP 5 0.013 0.310 9.2 0.162 0.395 RQ90 5 −0.007 0.892 4.0 0.081 MSLP 6 0.018 0.077 7.4 0.066 0.040 RQ90 6 0.018 0.170 4.2 0.031 R10D–MSLP MSLP 1 −0.038 0.002 43.2 −0.812 0.974 R10D 1 −0.042 0.001 8.7 −0.782 MSLP 2 0.020 0.096 12.4 0.430 0.843 R10D 2 0.014 0.191 9.8 0.335 MSLP 3 −0.011 0.975 6.8 −0.194 0.758 R10D 3 −0.003 0.653 5.5 −0.113 MSLP 4 0.001 0.908 7.6 −0.156 0.626 R10D 4 −0.033 0.363 4.4 0.016 MSLP 5 0.042 0.034 10.4 0.022 0.375 R10D 5 0.011 0.668 4.8 −0.087 MSLP 6 −0.013 0.198 6.8 0.073 0.159 R10D 6 −0.022 0.114 4.8 −0.027 patterns for the indices show opposite signs in the north and south, resembling the long-term linear trend in the indices. The second canonical coefficients do not have a significant trend and have lower correlations with the NAO, although for R10D (Table VI) the correlations are still significant. The corresponding canonical patterns show high correlations centred over the North Sea for MSLP and, for the indices, the highest correlations are in central Europe, with opposite sign over the far north and south of the region. The MSLP dipole pattern, with poles over the eastern Mediterranean and the central North Atlantic, appears as the third canonical pattern for RQ90 and the fourth pattern for R10D.

7. CONCLUSIONS

Although extreme rainfall is generally not as spatially coherent as mean rainfall, we have shown that a spatial analysis of the interannual variability of indices of extremes can reveal interesting behaviour in the indices. Two indices of extreme DJF rainfall were analysed in detail: the maximum number of consecutive dry days (CDD) and the number of days above the 1961–90 90th percentile of wet-day amounts (R90N). Through a PCA of the indices, we extracted the major modes of interannual variability in these indices. In dealing with such a large and heterogeneous region, which includes stations from the Mediterranean to above the Arctic Circle, there will be constraints on the proportion of total variance that we can explain with just a small number of PCs. With six components of CDD we accounted for 52.4% of the total variance and six components of R90N explained 39.1%. Although these numbers are not large, they do show that

Copyright  2004 Royal Meteorological Society Int. J. Climatol. 24: 759–776 (2004) 774 M. R. HAYLOCK AND C. M. GOODESS regional-scale modes of variability are still important and extreme winter rainfall is not just controlled by local processes. Comparing these values with the still low 61.3% of variance explained by its PCs for total rainfall shows that the heterogeneous nature of the region is an important limiter of any spatial analysis. However, there are many indices of extreme rainfall that do not exhibit such a high average inter-station correlation and for which such an analysis would not be possible. Therefore, the relatively low variance with which we are working is a product of both the size of the region and the statistical nature of the extreme indices. The PCs of the indices were correlated with surface and upper-air observations of other variables. In general, atmospheric pressure parameters showed the highest correlations, followed by temperature, humidity and SST. Correlations were generally higher for lower atmospheric levels, although this may be partly due to possibly better quality data at the lower levels. A CCA of the indices with MSLP has revealed that the NAO is an important influence on extreme rainfall. The similarity between canonical patterns of the indices and the linear trends in the indices suggests that it is mainly changes in the NAO that have caused the observed trends in these indices. The analysis repeated with two other indices revealed very similar results. An obvious question raised by this analysis is how similar would results be for seasons other than DJF. We expect that such an analysis would be more difficult in June–August, with extreme rainfall caused more by convective processes and so probably less variance would be explained by the significant PCs. Also, the very dry conditions during these months in the Mediterranean region would pose problems for calculating both these indices. Therefore, a smaller region would probably be necessary. Whether MSLP is still the best predictor is uncertain, especially with a greatly reduced NAO signal in the summer half of the year. Still, such an analysis would be of great interest.

ACKNOWLEDGEMENTS

This work was funded by the Commission of the European Union under the STARDEX (Statistical and Regional Dynamical Downscaling of Extremes for European Regions) contract (EVK2-CT-2001-00115). Comments from Phil Jones, Anders Moberg and two anonymous referees were greatly appreciated. The daily rainfall data set was compiled by FIC, Spain, from the European Climate Assessment Project (European Climate Assessment & Dataset (ECA&D; http://www.knmi.nl/samenw/eca/index.html) and the following European countries’ national meteorological institutes:

Austria: Zentralanstalt fur¨ Meteorologie und Geodynamik Czech Republic: Cesk hydrometeorologick ustav´ Denmark: Danmarks Meteorologiske Institut Finland: Ilmatieteen laitos France: Meteo France Germany: Deutscher Wetterdienst Hungary: Orszagos´ Meteorologiai´ Szolgalat´ Netherlands: Koninklijk Nederlands Meteorologisch Instituut Norway: Meteorologisk institutt Portugal: Instituto de Meteorologia Russia (Former Soviet Union): Russian Meteorological and Hydrological Institute Spain: Instituto Nacional de Meteorolog´ıa Sweden: Sveriges Meteorologiska och Hydrologiska Institut Switzerland: MeteoSchweiz

Additionally, the Department of Meteorology and Climatology Aristotle University of Thessaloniki (UTH; 16 Greek stations), Servizio Meteorologico Regional, ARPA-Emilia Romagna, Italy (SMR; 10 Italian stations) and the Climatic Research Unit, University of East Anglia (UEA; 21 UK series) provided data.

Copyright  2004 Royal Meteorological Society Int. J. Climatol. 24: 759–776 (2004) VARIABILITY OF EUROPEAN EXTREME WINTER RAINFALL 775

REFERENCES

Barnett TP, Preisendorfer R. 1987. Origins and levels of monthly and seasonal forecast skill for United-States surface air temperatures determined by canonical correlation-analysis. Monthly Weather Review 115: 1825–1850. Basist AN, Chelliah M. 1997. Comparison of tropospheric temperatures derived from the NCEP/NCAR reanalysis, NCEP operational analysis, and the microwave sounding unit. Bulletin of the American Meteorological Society 78: 1431–1447. Folland CK, Karl TR, Christy JR, Clarke RA, Gruza GV, Jouzel J, Mann ME, Oerlemans J, Salinger MJ, Wang SW. 2001. Observed climate variability and change. In Climate Change 2001: The Scientific Basis. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change, Houghton JT, Ding Y, Griggs DJ, Noguer M, van der Linden PJ, Dai X, Maskell K, Johnson CA (eds). Cambridge University Press: 99–181. Frei C, Schar C. 2001. Detection probability of trends in rare events: theory and application to heavy precipitation in the Alpine region. Journal of Climate 14: 1568–1584. Frich P, Alexander LV, Della-Marta P, Gleason B, Haylock M, Tank AMGK, Peterson T. 2002. Observed coherent changes in climatic extremes during the second half of the twentieth century. Climate Research 19: 193–212. Gershunov A. 1998. ENSO influence on intraseasonal extreme rainfall and temperature frequencies in the contiguous United States: implications for long-range predictability. Journal of Climate 11: 3192–3203. Groisman PY, Karl TR, Easterling DR, Knight RW, Jamason PF, Hennessy KJ, Suppiah R, Page CM, Wibig J, Fortuniak K, Razuvaev VN, Douglas A, Forland E, Zhai PM. 1999. Changes in the probability of heavy precipitation: important indicators of climatic change. Climatic Change 42: 243–283. Haylock M, Nicholls N. 2000. Trends in extreme rainfall indices for an updated high quality data set for Australia, 1910–1998. International Journal of Climatology 20: 1533–1541. Hennessy KJ, Suppiah R, Page CM. 1999. Australian rainfall changes, 1910–1995. Australian Meteorological Magazine 48: 1–13. Hoaglin DC, Mosteller F, Tukey JW. 1983. Understanding Robust and Exploratory Data Analysis. Wiley. Hurrell JW. 1995. Decadal trends in the North-Atlantic oscillation — regional temperatures and precipitation. Science 269: 676–679. Jones PD, Jonsson T, Wheeler D. 1997. Extension to the North Atlantic oscillation using early instrumental pressure observations from Gibraltar and south-west Iceland. International Journal of Climatology 17: 1433–1450. Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D, Gandin L, Iredell M, Saha S, White G, Woollen J, Zhu Y, Chelliah M, Ebisuzaki W, Higgins W, Janowiak J, Mo KC, Ropelewski C, Wang J, Leetmaa A, Reynolds R, Jenne R, Joseph D. 1996. The NCEP/NCAR 40-year reanalysis project. Bulletin of the American Meteorological Society 77: 437–471. Karl TR, Knight RW. 1998. Secular trends of precipitation amount, frequency, and intensity in the United States. Bulletin of the American Meteorological Society 79: 231–241. Klein Tank AMG, Konnen GP. 2003. Trends in indices of daily temperature and precipitation extremes in Europe, 1946–1999. Journal of Climate 16: 3665–3680. Klein Tank AMG, Wijngaard JB, Konnen GP, Bohm R, Demaree G, Gocheva A, Mileta M, Pashiardis S, Hejkrlik L, Kern-Hansen C, Heino R, Bessemoulin P, Muller-Westermeier G, Tzanakou M, Szalai S, Palsdottir T, Fitzgerald D, Rubin S, Capaldo M, Maugeri M, Leitass A, Bukantis A, Aberfeld R, Van Engelen AFV, Forland E, Mietus M, Coelho F, Mares C, Razuvaev V, Nieplova E, Cegnar T, Lopez JA, Dahlstrom B, Moberg A, Kirchhofer W, Ceylan A, Pachaliuk O, Alexander LV, Petrovic P. 2002. Daily dataset of 20th-century surface air temperature and precipitation series for the European Climate Assessment. International Journal of Climatology 22: 1441–1453. Lloyd-Hughes B, Saunders MA. 2002. Seasonal prediction of European spring precipitation from El Nino–southern˜ oscillation and local sea-surface temperatures. International Journal of Climatology 22: 1–14. Manton MJ, Della-Marta PM, Haylock MR, Hennessy KJ, Nicholls N, Chambers LE, Collins DA, Daw G, Finet A, Gunawan D, Inape K, Isobe H, Kestin TS, Lefale P, Leyu CH, Lwin T, Maitrepierre L, Ouprasitwong N, Page CM, Pahalad J, Plummer N, Salinger MJ, Suppiah R, Tran VL, Trewin B, Tibig I, Yee D. 2001. Trends in extreme daily rainfall and temperature in Southeast Asia and the South Pacific: 1961–1998. International Journal of Climatology 21: 269–284. Nicholls N. 2001. The insignificance of significance testing. Bulletin of the American Meteorological Society 82: 981–986. Nicholls N, Gruza GV, Jonzel J, Karl TR, Ogallo LA, Parker DE. 1996. Observed climate variability and change. In Climate Change 1995, Houghton JT, Meira Filho LG, Callander BA, Harris N, Kattenberg A, Maskell K (eds). Cambridge University Press: 133–192. Osborn TJ, Briffa KR, Tett SFB, Jones PD, Trigo RM. 1999. Evaluation of the North Atlantic oscillation as simulated by a coupled climate model. Climate Dynamics 15: 685–702. Osborn TJ, Hulme M, Jones PD, Basnett TA. 2000. Observed trends in the daily intensity of United Kingdom precipitation. International Journal of Climatology 20: 347–364. Parker MD, Jackson E, Horton EB. 1995. The 1961–1990 GISST2.2 Sea Surface Temperature and Sea-Ice Climatology. Climate Research Technical Note No. 63, Hadley Centre for Climate Prediction and Research. Peterson TC, Taylor MA, Demeritte R, Duncombe DL, Burton S, Thompson F, Porter A, Mercedes M, Villegas E, Fils RS, Tank AK, Martis A, Warner R, Joyette A, Mills W, Alexander L, Gleason B. 2002. Recent changes in climate extremes in the Caribbean region. Journal of Geophysical Research — Atmospheres 107: 4601. Preisendorfer RW, Zwiers FW, Barnett TP. 1981. Foundations of Principal Component Selection Rules. SIO Reference Series 81–4. Scripps Institution of Oceanography. Press WH, Flannery BP, Teukolsky SA, Vetterling WT. 1986. Numerical Recipes: The Art of Scientific Computing. Cambridge University Press. Qian BD, Corte-Real J, Xu H. 2000. Is the North Atlantic oscillation the most important atmospheric pattern for precipitation in Europe? Journal of Geophysical Research — Atmospheres 105: 11901–11910. Quadrelli R, Pavan V, Molteni F. 2001. Wintertime variability of Mediterranean precipitation and its links with large-scale circulation anomalies. Climate Dynamics 17: 457–466. Reid PA, Jones PD, Brown O, Goodess CM, Davies TD. 2001. Assessments of the reliability of NCEP circulation data and relationships with surface climate by direct comparisons with station based data. Climate Research 17: 247–261. Richman MB. 1986. Rotation of principal components. Journal of Climatology 6: 293–335.

Rogers JC. 1997. North Atlantic storm track variability and its association to the North Atlantic oscillation and climate variability of northern Europe. Journal of Climate 10: 1635–1647. Santer BD, Hnilo JJ, Wigley TML, Boyle JS, Doutriaux C, Fiorino M, Parker DE, Taylor KE. 1999. Uncertainties in observationally based estimates of temperature change in the free atmosphere. Journal of Geophysical Research — Atmospheres 104: 6305–6333. Shah KP, Rind D. 1998. Comparing upper tropospheric and lower stratospheric temperatures: microwave sounding unit, radiosonde, COSPAR International Reference Atmosphere, and National Centers for Environmental Prediction–National Center for Atmospheric Research reanalysis monthly mean climatologies. Journal of Geophysical Research — Atmospheres 103: 31569–31591. Trigo RM, Osborn TJ, Corte-Real JM. 2002. The North Atlantic oscillation inﬂuence on Europe: climate impacts and associated physical mechanisms. Climate Research 20: 9–17. Wibig J. 1999. Precipitation in Europe in relation to circulation patterns at the 500 hPa level. International Journal of Climatology 19: 253–269. Wijngaard JB, Klein Tank AMG, Konnen¨ GP. 2003. Homogeneity of 20th century European daily temperature and precipitation series. International Journal of Climatology 23: 679–692. Wilks DS. 1995. Statistical Methods in the Atmospheric Sciences. Academic Press. Ye HC. 2001. Characteristics of winter precipitation variation over northern central Eurasia and their connections to sea surface temperatures over the Atlantic and Paciﬁc Oceans. Journal of Climate 14: 3140–3155.

Papers submitted as part of thesis

Paper Design Conduct Analysis Preparation

Allan and Haylock, 1993 30 80 50 20

Haylock and Goodess, 2004 90 100 100 90

Haylock and McBride, 2001 50 100 50 80

Haylock and Nicholls, 2000 75 100 90 90

Haylock et al., 2004 50 50 90 90

McBride et al., 2003 40 90 40 30

Manton et al., 2001 20 10 20 10

Other papers referenced in Chapter 1

Paper Design Conduct Analysis Preparation

Frich et al., 2002 10 15 15 10

Gordon et al., 1992 10 30 30 10

Nicholls et al., 1999 15 40 40 20

Whetton et al., 1994 10 60 40 20

Personal contribution (%) by author for each publication separated into: design of the investigation; conduct of the research; analysis of the outcome; and preparation of the manuscript.

+1-828-271-4287 [email protected] December 13, 2004

To whom it may concern:

Malcolm Haylock is the lead author of a 30 author paper entitled Trends in Total and Extreme South American Rainfall 1960-2000 and Links to Sea Surface Temperature. As the second of the 30 authors, I’d like to explain Malcolm’s role in the paper. The paper arises out of a regional climate change workshop that Malcolm participated in which brought together at least one national hydrometeorological service employee and one university professor from each country in the southern 80% of South America. At the workshop, we analyzed daily data from the region using a standardized suite of indices coordinated by a WMO Commission for Climatology/CLIVAR Expert Team. When the meeting ended, the participants agreed that we wanted Malcolm to create the rainfall indices paper. To do that, he had to carefully reanalyze all the QC and homogeneity work, reproduce the suite of indices, make sense out of what the indices were showing and then write the paper.

In my opinion, which I know is shared by other people I’ve talked to about this, Malcolm did an excellent job quite quickly (we are under deadline to get the work published in time for the results to contribute to the IPCC Fourth Assessment Report). He primarily worked alone on the paper although several of us had minor editing comments on the first draft of the paper. I would say that Malcolm’s part of the design of the investigation exceeded 50% (namely which indices to present and how to present them), conduct of the research exceeded 50%, analysis of the outcome was approximately 90%, and preparation of the manuscript exceeded 90%.

Regards,

Thomas C. Peterson, Ph.D.

Appendix 2: Citations

Paper Citations

Allan and Haylock, 1993 40

Frich et al., 2002 20

Gordon et al., 1992 68

Haylock and Goodess, 2004 0

Haylock and McBride, 2001 8

Haylock and Nicholls, 2000 8

Haylock et al., 2004 N/A

McBride et al., 2003 2

Manton et al., 2001 19

Nicholls et al., 1999 N/A

Whetton et al., 1994 20

Number of times each paper has been cited in the literature (from Web of Knowledge http://www.wok.mimas.ac.uk) Appendix 3: Publications

Allan RJ and Haylock MR. 1993. Circulation Features Associated with the Winter Rainfall Decrease in Southwestern Australia. Journal of Climate, 6, 1356-1367.

Frich P, Alexander LV, Della-Marta P, Gleason B, Haylock M, Tank AMGK and Peterson T. 2002. Observed coherent changes in climatic extremes during the second half of the twentieth century. Climate Research, 19, 193-212.

Haylock M and Goodess C. 2004. Interannual variability of European extreme winter rainfall and links with mean large-scale circulation. International Journal of Climatology, 24, 759-776.

Haylock M and McBride J. 2001. Spatial coherence and predictability of Indonesian wet season rainfall. Journal of Climate, 14, 3882-3887.

Haylock M and Nicholls N. 2000. Trends in extreme rainfall indices for an updated high quality data set for Australia, 1910-1998. International Journal of Climatology, 20, 1533-1541.

McBride JL, Haylock MR and Nichols N. 2003. Relationships between the maritime continent heat source and the El Nino-Southern Oscillation phenomenon. Journal of Climate, 16, 2905-2914.

Page CM, Nicholls N, Plummer N, Trewin B, Manton M, Alexander L, Chambers LE, Choi Y, Collins DA, Gosai A, Della-Marta P, Haylock MR, Inape K, Laurent V, Maitrepierre L, Makmur EEP, Nakamigawa H, Ouprasitwong N, McGree S, Pahailad J, Salinger MJ, Tibig L, Tran TD, Vediapan K and Zhai P. 2004. Data rescue in the Southeast Asia and South Pacific region - Challenges and opportunities. Bulletin of the American Meteorological Society, 85, 1483-1489.

Whetton PH, Fowler AM, Haylock MR and Pittock AB. 1993. Implications of Climate- Change Due to the Enhanced Greenhouse-Effect on Floods and Droughts in Australia. Climatic Change, 25, 289-317.

Whetton PH, Haylock MR and Galloway R. 1996. Climate change and snow-cover duration in the Australian Alps. Climatic Change, 32, 447-479. Whetton PH, Rayner PJ, Pittock AB and Haylock MR. 1994. An Assessment of Possible Climate-Change in the Australian Region-Based on an Intercomparison of General-Circulation Modeling Results. Journal of Climate, 7, 441-463. Acknowledgements

My decade of scientific learning has been guided by some very talented individuals, from my first job working with Rob Allan, Penny Whetton, Kevin Hennessy and the CSIRO Climate Impacts Group. My move to the Bureau of Meteorology Research Centre was thanks to Neville Nicholls, whose inspirational attitude to learning has shown me that science can be as elegant as a Bach prelude, and John McBride for introducing me to a whole new way of doing research. Thanks to the entire BMRC Climate Forecasting Group, especially Wasyl Drosdowsky who patiently answered questions about multivariate statistical techniques. I am sincerely grateful to Phil Jones and Clare Goodess at the Climatic Research Unit for giving me the opportunity to extend my learning to the North Atlantic. CRU houses some very talented scientists. Finally the biggest thanks to Kate, who has followed my journey to all the above places but led me to many new ones.