Uncertainties in Runoff Projections in Southwest Western Australian and Central Chilean Catchments

Uncertainties in runoff projections in Southwest Western Australian and Central Chilean catchments

Pilar Andrea Barría Sandoval

School of Earth Sciences

The University of Melbourne

Victoria, Australia

August, 2017

This thesis is submitted in fulfilment of the requirements of the degree of Doctor of Philosophy

Abstract

Important runoff reductions have been reported in mid-latitude, Mediterranean-like climate catchments of the Southern Hemisphere (SH), in particular in the southwest of Western Australia (SWA) and in central Chile (CC). These changes have been driven by decreases in rainfall since the mid-1970s. Despite regional rainfall and runoff projections from Global Climate Models (GCMs) indicating that the observed trends are expected to continue during the 21st century, the projections are affected by large uncertainties that limit their utility to decision makers. The main source of uncertainty in runoff projections are the GCMs used to produce future climate projections. However, uncertainties arise from the observations of the climate variables, the statistical methodology to downscale the GCM simulations to the catchment scale and the hydrological model used to simulate runoff. In particular, the short length (<50 years) and poor spatially distributed observed climatological variables in mountainous catchments, characterized by steep topography, hampers a deep analysis of runoff trends and runoff variability, such as the case of CC mountainous catchments.

The impact of the GCM uncertainty on runoff projections has mainly been assessed through comparison of multi-model runs of future climate with little exploration of uncertainties inside the models due to different parameterisations. This thesis seeks to investigate the uncertainty response of projected runoff due to both: perturbed physics parameter variations within a GCM using a novel 2500-member ensemble from the HadCM3L model, the climaprediction.net data (CPDN), termed the within-GCM uncertainties, and from a multi-model ensemble of different GCMs collated by the CMIP5 project, termed the between-GCM uncertainties.

The impact of GCM uncertainties on runoff modelling for pluvial regimes in southwest Western Australian and Central Chilean catchments was assessed. Both regions share similar trends and climatic features, with major decreases in winter precipitation and runoff since the mid-70s that have been related to a displacement of the Southern Hemisphere storm track. Nonetheless one important difference between SWA and CC catchments, is the presence of nivo- pluvial regimes located at the foothills of the Andes in CC, whose hydrology is poorly understood mainly due to the lack of well distributed and long gauge records that represent its variability.

The results presented in this thesis show that the impact of within-GCM uncertainties on runoff projections in SWA catchments is very large; larger than previous estimates of within- GCM uncertainties impact on runoff. The perturbed physics approach indicates that current water management assessments underestimate uncertainties in runoff projections. Regarding the comparison of the impact of between-GCM and within-GCM uncertainties on runoff projections in SWA catchments quantified as the difference between the 5th and the 95th percentile of simulations, the impact of within-GCM uncertainties on runoff projections range between 39% and 65%. Whereas the impact of between-GCM uncertainties on runoff projections range between 44% and 83% for the Representative concentration pathway 4.5 (RCP4.5) scenario and about 38% and 72% under the RCP8.5 for the period between 2050-2080 compared to 1970-2000. Regarding CC catchments, between-GCM uncertainties of about 55% and 51% in runoff projections using the RCP4.5 and the RCP8.5 scenarios were found. The results here reported indicate that the impact of within-GCM and between-GCM uncertainties in SWA catchments runoff projections is very similar. The results also indicate that because some GCMs in the CMIP5 ensemble have multiple runs, using different initial conditions, CMIP5 gives some insight into within-GCM uncertainty as well. For these reasons and because CMIP5 provides runs that represent all regions of the world, it is recommended for use in hydrological assessments of climate change impact and the uncertainties around the projections.

Finally, and aiming to improve the understanding on runoff variability in mountainous catchments of the temperate region of CC, this thesis includes the first high elevation runoff reconstruction in Chile using 300 years of tree ring chronologies of Araucaria araucana and Astroceudrus chilensis. The upper part of Biobío river melting season runoff (October-March) and pluvial season runoff (April-September) were reconstructed and analysed to investigate the influence of large scale climatic drivers on runoff generation, current drought trends and to improve the understanding of long term hydroclimate variability in this region. Important differences in the runoff variability of the upper and the lower elevation catchments were identified, which are in part influenced by the large scale climatic features that drive runoff generation in both regions.

Declaration

This is to certify that:

I. The thesis comprises only my original work towards the PhD except where indicated in the Preface,

II. Due acknowledgement has been made in the text to all other material used,

III. The thesis is fewer than 100 000 words in length, exclusive of tables, maps, bibliographies and appendices.

Pilar Andrea Barría Sandoval

Preface

This thesis comprises material that has been published in a peer-reviewed journal:

- Chapter 4 is based on Barria et al. (2015) paper, of which the content is original and the co-authors contributed 20 percent of the work: Barria, P., Walsh, K.J., Peel, M.C. and Karoly, D., 2015. Uncertainties in runoff projections in southwestern Australian catchments using a global climate model with perturbed physics. Journal of Hydrology, 529, pp.184-199. DOI: 10.1016/j.jhydrol.2015.07.040.

- Chapter 5 in this thesis is based on a manuscript currently accepted to be published in Journal of Southern Hemisphere Earth Systems Science, Barria et al (2017), of which the content is original and the co-authors contributed 15 percent of the work: Barria, P., Peel, M.C., Walsh, K.J., and Garreaud, R., 2017. Analysis of within and between-GCM uncertainties of runoff projections in Mediterranean-like catchments. Journal of Southern Hemisphere Earth Systems Science.

- Chapter 6 in this thesis is based on a paper currently published in the International Journal of Climatology, Barria et al. (2018), of which the content is original and the co- authors contributed 10 percent of the work: Barria, P., Peel, M.C., Walsh, K.J. and Muñoz, A., 2018. The first 300‐year streamflow reconstruction of a high‐elevation river in Chile using tree rings. International Journal of Climatology, 38, pp.436-451. DOI: 10.1002/joc.5186.

The research was funded, in part, by the Australian Research Council (ARC) Centre of Excellence for Climate System Science (grant CE110001028). Additional funds were provided by the scholarship CONICYT Becas Chile, CONICYT PAI/INDUSTRIA 79090016. Murray Peel is the recipient of an Australian Research Council Future Fellowship (FT120100130).

To my best friend, my confidante,

the person who has given me incessant and unconditional love and support since I was 2 years old. To my wonderful sister, Noe. Acknowledgments

Firstly I would like to thank my supervisors Dr. Kevin Walsh and Dr. Murray Peel for their support, understanding and wise advice during all the years of my PhD, especially for their patience during all the difficult moments I had throughout my PhD years.

Especial thanks to Conicyt that provided the Becas Chile scholarship 79090016 which allowed me to conduct my studies during more than 4 years.

My sincere thanks to all those people and institutions that provide data to pursue the aims of my research. In particular, I would like to thanks Katherine Sadler for providing me the AWAP data, Francois Delage and Ben Henley for helping me with the CMIP5 data, David Karoly for helping me to get the climateprediction.net data from the Met Office in the UK, Ariel Muñoz for providing the tree ring chronologies data and Ricardo Gonzalez who helped me with the observed runoff data of high elevation catchments in Central Chile.

Thanks to the co-authors of the three papers I wrote during my PhD, all of them published in different journals, thanks David Karoly, Ariel Muñoz and René Garreaud for your time, revisions, comments and discussion.

I would also like to thank my beloved friends, who walked with me during all the PhD years, thanks Katherine Lizama, Estephany Marillo, Raul Lugo, Toni Cox, Christopher Chambers, Andreas Nedegard, Daniel Pazmiño, Joshua Soderholm, Annie, Jan Tympel, Andrea Dittus, Javiera Jofré, May-Lin Tay, Mauricio Quezada, Fernando Chong, Francisco Sabat and Sandra Perez. Everything would be much more difficult without your friendship.

Finalmente quiero agradecer a mi familia, muchas gracias a mis queridos padres, amados hermanos Noe, Nacho y Rodrigo y a Emilia. Gracias por apoyarme en todos los proyectos que he emprendido en la vida, por quererme tal cual soy y por ayudarme en todos los momentos, buenos y malos.

Contents

Chapter 1. Introduction ...... 1

1.1 Hydroclimatic trends in Mediterranean-like climate catchments ...... 1

1.2 Objectives ...... 4

1.3 Thesis structure ...... 5

Chapter 2. Literature Review ...... 7

2.1 Observed trends and projections of hydroclimatic variables in SWA catchments …………………………………………………………………………………8

2.2 Observed trends and projections of hydroclimatic variables in Central Chile 12

2.3 Uncertainties in runoff projections ...... 17

2.4 Summary ...... 20

Chapter 3. Hydrometeorological DATA and catchment characteristics ...... 23

3.1 Observed meteorological and runoff data ...... 23

3.1.1 Southwest of Western Australia ...... 23

3.1.2 Central Chile ...... 29

3.2 GCM data ...... 38

3.3 Tree ring data ...... 44

3.4 Climatic features reconstruction time series ...... 47

Chapter 4. Uncertainties in runoff projections in southwestern Australian Catchments using a global climate model with perturbed physics ...... 49

4.1 Introduction ...... 49

4.2 Methodology ...... 50

4.2.1 Evaluation of CPDN output...... 50

4.2.2 Bias Correction Methodology ...... 53

4.2.3 Precipitation Evaporation Runoff Model Description ...... 54

4.3 Results ...... 57

4.3.1 Evaluation of CPDN data ...... 57

4.3.2 Runoff Modelling ...... 62

4.3.3 Comparison of within GCM uncertainties from stochastic generation of data …………………………………………………………………………….70

4.3.4 Comparison of within-GCM uncertainties from GCM perturbed physics . 71

4.4 Summary ...... 73

Chapter 5. Exploring uncertainties on runoff projections in Mediterranean like catchments ……………………………………………………………………………76

5.1 Introduction ...... 76

5.2 Methodology ...... 78

5.3 Results ...... 79

5.3.1 CMIP5 and CPDN evaluation in SWA and CC catchments ...... 79

5.3.2 PERM model calibration and evaluation in SWA and CC catchments ...... 82

5.3.3 Comparison of within and between-GCM uncertainties in runoff projections in SWA catchments ...... 84

5.3.4 Between-GCM uncertainty of runoff projections in Central Chilean Catchments …………………………………………………………………………….88

5.3.5 Analysis of interplay between ozone recovery and GHG in CC and SWA catchments ……………………………………………………………………………92

5.3.6 Discussion ...... 95

5.4 Summary ...... 96

Chapter 6. 300 years of Streamflow reconstruction of a high elevation catchment in central Chile using tree rings ...... 99

6.1 Introduction ...... 99

6.2 Methodology ...... 100

6.3 Results ...... 103

6.3.1 Reconstruction ...... 104

6.3.2 Links between runoff reconstructions and large scale climatic features .. 107

6.3.3 Spectral analysis of reconstructions ...... 111

6.3.4 Drought analysis of the reconstructions ...... 114

6.3.5 Comparison with the lower Biobío catchment reconstruction ...... 115

6.4 Summary ...... 117

Chapter 7. Conclusions and future work ...... 120

7.1 Within-GCM uncertainties in runoff projections in SWA catchments...... 121

7.2 Comparison of the between-GCM and within-GCM uncertainties on projected runoff in southwest Western Australia...... 123

7.3 Analyses of the between-GCM uncertainties on runoff projections in CC catchments. ………………………………………………………………………………124

7.4 Study of the runoff variability at lower frequencies in a high elevation catchment of CC using a multi-century reconstruction of runoff...... 125

7.5 Future Work ...... 127

Chapter 8. References ...... 130

III

List of Figures

FIGURE 1.1 GIORGI REGIONS, BASED ON GIORGI ET AL. (2001) AND EXTRACTED FROM ROWLANDS ET AL. (2012) ...... 3

FIGURE 1.2 THESIS STRUCTURE ...... 6

FIGURE 3.1 LOCATION OF SWA CATCHMENTS ...... 24

FIGURE 3.2 CLIMOGRAMS OF SWA CATCHMENTS ...... 26

FIGURE 3.3 SEASONAL VARIATION OF RUNOFF IN SWA CATCHMENTS ...... 26

FIGURE 3.4 ANNUAL GAUGED RUNOFF AND PRECIPITATION DATA. THE DOTTED LINE IS THE FITTED LINEAR CURVE TO THE

ANNUAL RUNOFF IN A) DONNELLY AT STRICKLAND, B) HELENA AT NGANGAGURINGURING AND C) DENMARK AT

KOMPUP ...... 28

FIGURE 3.5 LAND COVER IN SWA CATCHMENTS. EXTRACTED FROM FIG. 2, SILBERSTEIN ET AL. (2012) ...... 29

FIGURE 3.6 LOCATION OF CENTRAL CHILEAN CATCHMENTS ...... 32

FIGURE 3.7 CLIMOGRAMS FOR LOW ELEVATION CENTRAL CHILEAN CATCHMENTS ...... 33

FIGURE 3.8 SEASONAL VARIATION OF CENTRAL CHILEAN CATCHMENTS ...... 33

FIGURE 3.9 (A) SEASONAL VARIATION OF THE UPPER PART OF BIOBÍO RIVER RUNOFF (MM) OVER THE COMPLETE

HYDROLOGICAL YEAR (APRIL-MARCH). THE BARPLOT PRESENTS THE CLIMATOGRAM OF THE AREA WEIGHTED

PRECIPITATION (MM) GAUGED IN ABANICO AND POLCURA STATIONS (1966-2000) OVER THE HYDROLOGICAL YEAR.

2.(B) COMPARISON OF THE SEASONAL VARIATION OF OBSERVED RUNOFF (MM) IN THE UPPER AND THE LOWER PART OF

THE BIOBÍO RIVER (1960-2002) ...... 34

FIGURE 3.10 ANNUAL GAUGED RUNOFF AND PRECIPITATION DATA.THE DOTTED LINE IS THE FITTED LINEAR CURVE TO THE

ANNUAL RUNOFF AT A) CAUQUENES AT EL ARRAYÁN, B) CATO AT PUENTE CATO AND C) LUMACO AT LUMACO ...... 36

FIGURE 3.11 OBSERVED HIGH ELEVATION BIOBÍO RIVER SEASONAL RUNOFF AND TRENDS (1960-2015). THE FITTED LINEAR

TRENDS INDICATE REDUCTIONS OF 0.65 M3S-1 FOR ANNUAL RUNOFF, AN INCREASE OF ABOUT 0.07 M3S-1 FOR

PLUVIAL SEASON RUNOFF AND A REDUCTION OF ABOUT 1.4 M3S-1 FOR MELTING SEASON RUNOFF...... 37

FIGURE 3.12 LAND COVER IN THE CC REGION. DATA EXTRACTED FROM ZHAO ET AL. (2016) ...... 38

FIGURE 3.13 MAP SHOWING THE LOCATION OF THE UPPER PART OF THE BIOBÍO RIVER, THE RAINFALL AND TEMPERATURE

STATIONS (LIGHT BLUE CIRCLES), THE RUNOFF STATIONS (DARK BLUE CIRCLES), THE PRECIPITATION ISOHYETS (MM) AND

THE 12 TREE RING CHRONOLOGIES USED AS PREDICTORS ...... 45

FIGURE 4.1 BIAS CORRECTION METHODOLOGY FOR PRECIPITATION IN DONNELLY RIVER AT STRICKLAND DURING DECEMBER 54

FIGURE 4.2 PERM MODEL SCHEME ...... 56

FIGURE 4.3 COMPARISON OF RAW CPDN ANNUAL PRECIPITATION AND SCALED AWAP ANNUAL PRECIPITATION FOR SWA.

GREY LINES REPRESENT THE 2500 SIMULATIONS OF PRECIPITATION FROM CPDN. 95TH, 5TH PERCENTILES AND MEDIAN

OF THE SIMULATIONS ARE PRESENTED AS BLUE LINES AND AWAP ANNUAL PRECIPITATION IS PLOTTED WITH A RED LINE ...... 59

FIGURE 4.4 HISTOGRAM OF MEDIAN OF SIMULATED (CPDN) ANNUAL PRECIPITATION (MM) FOR THE PERIOD BETWEEN 1940

AND 2000. COMPARISON OF MEDIAN OF MODELLED BASED ON OBSERVED DATA (AWAP), AS INDICATED WITH THE

OPEN STAR, AND THE MEDIAN OF ALL THE SIMULATIONS DURING THE SAME PERIOD, INDICATED WITH AN ASTERISK. ... 59

FIGURE 4.5 COMPARISON OF RAW CPDN ANNUAL TEMPERATURE AND SCALED AWAP ANNUAL TEMPERATURE FOR SWA.

GREY LINES REPRESENT THE 2500 SIMULATIONS OF TEMPERATURE FROM CPDN. 95TH, 5TH PERCENTILES AND MEDIAN

OF THE SIMULATIONS ARE PRESENTED AS BLUE LINES AND AWAP ANNUAL TEMPERATURE IS PLOTTED WITH A RED LINE ...... 60

FIGURE 4.6 HISTOGRAM OF THE MEDIAN OF SIMULATED (CPDN) ANNUAL TEMPERATURE (°C) FOR THE PERIOD BETWEEN

1940 AND 2000. COMPARISON OF MEDIAN OF MODELLED BASED ON OBSERVED DATA (AWAP) AS INDICATED WITH

THE OPEN STAR AND THE MEDIAN OF ALL THE SIMULATIONS DURING THE SAME PERIOD, INDICATED BY AN ASTERISK ... 60

FIGURE 4.7 5TH AND 95TH PERCENTILES OF SEASONAL EMPIRICAL CUMULATIVE DISTRIBUTION FUNCTIONS FOR PRECIPITATION

SIMULATED USING CPDN. BLUE CURVES REPRESENT THE OBSERVED PERIOD AND RED CURVES THE PROJECTION FOR THE

PERIOD BETWEEN 2020-2080...... 63

FIGURE 4.8 5TH AND 95TH PERCENTILES OF SEASONAL EMPIRICAL CUMULATIVE DISTRIBUTION FUNCTIONS FOR TEMPERATURE

SIMULATED USING CPDN. BLUE CURVES REPRESENT THE OBSERVED PERIOD AND RED CURVES THE PROJECTION FOR THE

PERIOD BETWEEN 2020-2080 ...... 64

FIGURE 4.9 BIAS CORRECTED ANNUAL PRECIPITATION OVER DONNELLY RIVER AT STRICKLAND. GREY LINES REPRESENT THE

2500 SIMULATIONS OF BIAS CORRECTED PRECIPITATION FROM CPDN PROJECT. 95TH, 5TH PERCENTILES AND MEDIAN

OF THE SIMULATIONS ARE PRESENTED IN BLUE LINES AND AWAP ANNUAL PRECIPITATION IS PLOTTED IN RED LINE ..... 66

FIGURE 4.10 BIAS CORRECTED ANNUAL TEMPERATURE OVER DONNELLY RIVER AT STRICKLAND. GREY LINES REPRESENT THE

2500 SIMULATIONS OF BIAS CORRECTED TEMPERATURE FROM CPDN PROJECT. 95TH, 5TH PERCENTILES AND MEDIAN

OF THE SIMULATIONS ARE PRESENTED IN BLUE LINES AND AWAP ANNUAL TEMPERATURE IS PLOTTED IN RED LINE ...... 66

FIGURE 4.11 SIMULATED ANNUAL RUNOFF OVER DONNELLY RIVER AT STRICKLAND. GREY LINES REPRESENT THE 2500

SIMULATIONS OF RUNOFF USING PERM MODEL RUN WITH BIAS CORRECTED PRECIPITATION AND TEMPERATURE FROM

CPDN. 95TH, 5TH PERCENTILES AND MEDIAN OF THE SIMULATIONS ARE PRESENTED IN BLUE LINES AND OBSERVED

RUNOFF IS PLOTTED IN RED LINE ...... 68

FIGURE 4.12 HISTOGRAMS OF SEASONAL CHANGES IN PRECIPITATION AND RUNOFF OVER DONNELLY RIVER AT STRICKLAND 68

FIGURE 4.13 HISTOGRAMS OF ANNUAL CHANGES IN PRECIPITATION AND RUNOFF IN ALL OF THE CATCHMENTS ...... 69

FIGURE 4.14 BOXPLOT OF UNCERTAINTIES IN PRECIPITATION AND RUNOFF USING CPDN DATA AND STOCHASTIC GENERATION,

IN THE DONNELLY RIVER AT STRICKLAND FOR THE PERIOD 2035–2064 ...... 71

FIGURE 4.15 HISTOGRAMS OF ANNUAL CHANGES IN RUNOFF CONSIDERING ALL THE SIMULATIONS OF CPDN AND THE GROUPS

OF SIMULATIONS WITH DIFFERENT PERTURBATIONS OF THE PARAMETER RHCRIT. RED LINE REPRESENTS THE MEDIAN OF

THE WHOLE ENSEMBLE AND DOTTED RED LINE THE MEDIAN OF THE SIMULATIONS FOR A PARTICULAR PERTURBATION OF

RHCRIT...... 73

FIGURE 5.1 METHODOLOGY OF RUNOFF PROJECTIONS ...... 78

FIGURE 5.2 A) COMPARISON OF MEAN ANNUAL PRECIPITATION SIMULATED BY CMIP5 IN THE HISTORICAL PERIOD AND

OBSERVED MEAN ANNUAL PRECIPITATION. B) COMPARISON OF STANDARD DEVIATION OF ANNUAL PRECIPITATION

SIMULATED BY CMIP5 IN THE HISTORICAL PERIOD AND STANDARD DEVIATION OF OBSERVED ANNUAL PRECIPITATION. 80

FIGURE 5.3 A) COMPARISON OF MEAN ANNUAL TEMPERATURE SIMULATED BY CMIP5 IN THE HISTORICAL PERIOD AND

OBSERVED MEAN ANNUAL TEMPERATURE. B) COMPARISON OF STANDARD DEVIATION OF ANNUAL TEMPERATURE

SIMULATED BY CMIP5 IN THE HISTORICAL PERIOD AND STANDARD DEVIATION OF OBSERVED ANNUAL TEMPERATURE. 80

FIGURE 5.4 BIAS CORRECTED CMIP5 SIMULATED ANNUAL PRECIPITATION UNDER THE RCP8.5 SCENARIO OVER A) DONNELLY

AT STRICKLAND AND B) CAUQUENES AT EL ARRAYÁN ...... 82

FIGURE 5.5 HISTOGRAM OF PROJECTED DIFFERENCE IN MEAN ANNUAL PRECIPITATION AND RUNOFF FOR DONNELLY AT

STRICKLAND FOR THE PERIOD 2050-2080 COMPARED TO 1970-2000 USING CPDN AND CMIP5 GCMS ...... 85

FIGURE 5.6 HISTOGRAM OF PROJECTED DIFFERENCE IN MEAN ANNUAL PRECIPITATION AND RUNOFF FOR HELENA AT

NGANGAGURINGURING FOR THE PERIOD 2050-2080 COMPARED TO 1970-2000 USING CPDN AND CMIP5 GCMS ...... 86

FIGURE 5.7 HISTOGRAM OF PROJECTED DIFFERENCE IN MEAN ANNUAL PRECIPITATION AND RUNOFF FOR DENMARK AT

KOMPUP FOR THE PERIOD 2050-2080 COMPARED TO 1970-2000 USING CPDN AND CMIP5 GCMS ...... 87

FIGURE 5.8 EXCEEDANCE PROBABILITY OF PROJECTED MEAN ANNUAL RUNOFF AT SWA CATCHMENTS DURING 2050-2080 88

FIGURE 5.9 HISTOGRAM OF PROJECTED DIFFERENCE IN MEAN ANNUAL PRECIPITATION AND RUNOFF FOR CAUQUENES AT EL

ARRAYAN FOR THE PERIOD 2050-2080 COMPARED TO 1970-2000 USING CMIP5 GCMS ...... 89

FIGURE 5.10 HISTOGRAM OF PROJECTED DIFFERENCE IN MEAN ANNUAL PRECIPITATION AND RUNOFF FOR CATO IN PUENTE

CATO FOR THE PERIOD 2050-2080 COMPARED TO 1970-2000 USING CMIP5 GCMS ...... 90

FIGURE 5.11 HISTOGRAM OF PROJECTED DIFFERENCE IN MEAN ANNUAL PRECIPITATION AND RUNOFF FOR LUMACO IN

LUMACO FOR THE PERIOD 2050-2080 COMPARED TO 1970-2000 USING CMIP5 GCMS ...... 91

FIGURE 5.12 SIMULATION OF RUNOFF UNDER THE RCP4.5 SCENARIO USING CMIP5 MODELS CONSIDERING CHEM AND

NOCHEM MODELS IN A) CAUQUENES RIVER, B) CATO RIVER AND C) LUMACO RIVER ...... 93

FIGURE 5.13 SIMULATION OF RUNOFF UNDER THE RCP4.5 SCENARIO USING CMIP5 MODELS CONSIDERING CHEM AND

NOCHEM MODELS IN A) DONNELLY RIVER, B) HELENA RIVER AND C) DENMARK RIVER ...... 94

FIGURE 6.1 (A) COMPARISON OF MELTING SEASON RUNOFF RECONSTRUCTION MODEL AND OBSERVATIONS DURING THE

CALIBRATION PERIOD (1960-2002). THE RECONSTRUCTION HAS AN R2 OF 0.52 AND A RMSE OF 0.77 REGARDING

THE OBSERVED DATA. 4.(B) COMPARISON OF PLUVIAL SEASON RUNOFF RECONSTRUCTION MODEL AND OBSERVATIONS

DURING THE CALIBRATION PERIOD (1960-2002). THE RECONSTRUCTION HAS AN R2 OF 0.52 AND A RMSE OF 0.77

REGARDING THE OBSERVED DATA ...... 105

FIGURE 6.2 (A) MELTING SEASON RUNOFF RECONSTRUCTION OF THE UPPER PART OF BIOBÍO RIVER (M3SEC-1). THE LIGHT GREY

SHADING PRESENTS A +/- 2 STANDARD ERROR BAND. THE LIGHT BLUE LINE IS THE HISTORICAL MEDIAN OF THE

RECONSTRUCTED TIME SERIES AND THE DARK BLACK LINE CORRESPONDS TO THE 20-YEARS FITTED SPLINE. THE FITTED

LINEAR TRENDS INDICATE INCREASES OF 0.038 M3S-1 CONSIDERING THE PERIOD 1739-2002 AND REDUCTIONS OF

0.092 M3S-1 SINCE 1850. 5. (B) PLUVIAL SEASON RUNOFF RECONSTRUCTION OF THE UPPER PART OF BIOBÍO RIVER

(M3SEC-1). THE LIGHT GREY SHADING PRESENTS A +/- 2 STANDARD ERROR BAND. THE LIGHT BLUE LINE IS THE

HISTORICAL MEDIAN OF THE RECONSTRUCTED TIME SERIES AND THE DARK BLACK LINE CORRESPONDS TO THE 20-YEARS

FITTED SPLINE. THE FITTED LINEAR TRENDS INDICATE REDUCTIONS OF 0.026 M3S-1 CONSIDERING THE PERIOD 1739-

2002 AND INCREASES OF ABOUT 0.051 M3S-1 SINCE 1850. THE DARK LINE REPRESENTS THE 20 YEARS FITTED SPLINE

TO THE TIME SERIES...... 106

FIGURE 6.3 11 YEARS MOVING AVERAGE OF THE NORMALIZED RECONSTRUCTED UPPER BIOBIO MELTING SEASON RUNOFF AND

THE SAM AND THE PDO RECONSTRUCTIONS...... 108

FIGURE 6.4 20 YEARS MOVING AVERAGE OF THE NORMALIZED UPPER BIOBIO RECONSTRUCTED PLUVIAL SEASON RUNOFF AND

THE PDO RECONSTRUCTIONS ...... 109

FIGURE 6.5 EXTENDED MELTING SEASON RUNOFF TIME SERIES USING THE GAUGED RUNOFF UNTIL 2015 AND THE FITTED 11

YEARS MOVING AVERAGE OF THE TIME SERIES OF THE UPPER BIOBÍO RIVER ...... 111

FIGURE 6.6 EXTENDED PLUVIAL SEASON RUNOFF TIME SERIES USING THE GAUGED RUNOFF UNTIL 2015 AND THE FITTED 20

YEARS MOVING AVERAGE OF THE TIME SERIES OF THE UPPER BIOBÍO RIVER ...... 111

FIGURE 6.7(A) WAVELET POWER SPECTRUM OF THE NORMALIZED AND DETRENDED RECONSTRUCTED HIGH ELEVATION PART OF

BIOBÍO RIVER MELTING SEASON RUNOFF (1739-2002). 6.7 (B) WELCH SPECTRAL ANALYSIS OF THE NORMALIZED AND

DETRENDED RECONSTRUCTED HIGH ELEVATION PART OF BIOBÍO RIVER MELTING SEASON RUNOFF (1739-2002) ..... 113

FIGURE 6.8(A) WAVELET POWER SPECTRUM OF THE NORMALIZED AND DETRENDED RECONSTRUCTED HIGH ELEVATION PART OF

BIOBÍO RIVER PLUVIAL SEASON RUNOFF (1666-2002). 6.8.(B) WELCH SPECTRAL ANALYSIS OF THE NORMALIZED AND

DETRENDED RECONSTRUCTED HIGH ELEVATION PART OF BIOBÍO RIVER PLUVIAL SEASON RUNOFF (1666-2002) ...... 114

VII

List of Tables

TABLE 2.1 HISTORIC STREAMFLOW RECONSTRUCTIONS IN AUSTRALIA. THE LENGTH OF THE RECONSTRUCTIONS ARE PRESENTED

IN THE COLUMN ENTITLED LENGTH, THE INFORMATION ABOUT THE PROXY DATA USED TO DEVELOP THE

RECONSTRUCTION IS PRESENTED IN PROXY DATA COLUMN, AND THE EXPLAINED VARIANCE IN THE R2 COLUMN...... 12

TABLE 2.2 HISTORIC STREAMFLOW RECONSTRUCTIONS IN CHILE. THE LENGTH OF THE RECONSTRUCTIONS ARE PRESENTED IN

THE COLUMN ENTITLED LENGTH, THE MEAN ELEVATION OF THE CATCHMENT IN METERS ABOVE SEA LEVEL ARE

PRESENTED IN THE ELEVATION COLUMN AND THE EXPLAINED VARIANCE IN THE R2 COLUMN...... 17

TABLE 3.1 CHARACTERISTICS OF SWA RUNOFF STATIONS ...... 25

TABLE 3.2 CHARACTERISTICS OF CENTRAL CHILEAN CATCHMENTS ...... 31

TABLE 3.3 LIST OF PERTURBED ATMOSPHERIC PHYSICS PARAMETERS IN THE CPDN PROJECT. BOLD NUMBERS INDICATE

DEFAULT VALUES. EXTRACTED FROM SUPPLEMENTARY TABLE SI 1, ROWLANDS ET AL. (2012) ...... 40

TABLE 3.4 LIST OF PERTURBED SULPHUR CYCLE AND OCEAN PHYSICS PARAMETERS IN THE CPDN PROJECT. BOLD NUMBERS

INDICATE DEFAULT VALUES. EXTRACTED FROM SUPPLEMENTARY TABLE SI 2, ROWLANDS ET AL. (2012) ...... 41

TABLE 3.5 LIST OF CMIP5 GCMS CHARACTERISTICS ...... 42

TABLE 3.6 CHRONOLOGIES USED IN THE RECONSTRUCTIONS OF RUNOFF IN THE UPPER PART OF BIOBÍO RIVER ...... 46

TABLE 3.7 CROSS CORRELATION BETWEEN THE TREE RING CHRONOLOGIES ...... 47

TABLE 4.1 DIFFERENCES IN MEDIAN AND STANDARD DEVIATION OF AWAP TEMPERATURE AND CPDN TEMPERATURE

BETWEEN 1940 AND 2000 ...... 58

TABLE 4.2 DIFFERENCES IN MEDIAN AND STANDARD DEVIATION OF AWAP PRECIPITATION AND CPDN PRECIPITATION

BETWEEN 1940 AND 2000 ...... 58

TABLE 4.3 CORRELATION COEFFICIENTS BETWEEN SOUTHERN ANNULAR MODE, ENSO, PRECIPITATION AND TEMPERATURE

OBSERVED AND SIMULATED BY CPDN ...... 61

TABLE 4.4 PERM CALIBRATION RESULTS ...... 65

TABLE 4.5 PERM EVALUATION RESULTS ...... 65

TABLE 5.1 CALIBRATED PARAMETERS AND CALIBRATION STATISTICS FOR PERM IN THE CC AND SWA CATCHMENTS ...... 83

TABLE 5.2 EVALUATION OF PERM IN THE CC AND SWA CATCHMENTS ...... 83

TABLE 5.3 MEDIAN OF THE CHANGES IN PROJECTIONS OF RUNOFF AND PRECIPITATION FOR THE PERIOD BETWEEN 2050-2080

AND 1970-2000 IN SWA CATCHMENTS ...... 86

TABLE 5.4 MEDIAN OF THE CHANGES IN PROJECTIONS OF RUNOFF AND PRECIPITATION FOR THE PERIOD BETWEEN 2050-2080

AND 1970-2000 IN CC CATCHMENTS ...... 91

TABLE 6.1 CATCHMENTS AND STREAMFLOW CHARACTERISTICS ...... 100

TABLE 6.2 CORRELATIONS BETWEEN THE TREE RING CHRONOLOGIES AND SEASONAL RUNOFF OF THE UPPER PART OF BIOBÍO

RIVER ...... 101

TABLE 6.3 STATISTICS OF THE RECONSTRUCTIONS MODELS OF THE UPPER PART OF BIOBÍO RIVER ...... 104

TABLE 6.4 STATISTICALLY SIGNIFICANT CORRELATIONS BETWEEN OBSERVED SEASONAL RUNOFF IN THE UPPER AND THE LOWER

PART OF BIOBÍO RIVER AND CLIMATIC FORCINGS (1960-2000) ...... 107

VIII

TABLE 6.5 CORRELATIONS BETWEEN THE 11 YEAR (FOR THE SAM) OR 20 YEAR (FOR THE PDO) MOVING AVERAGE OF

SEASONAL RUNOFF RECONSTRUCTIONS AND CLIMATIC FORCING RECONSTRUCTIONS ...... 109

TABLE 6.6 Z TEST OF MEAN OF SEASONAL RECONSTRUCTED RUNOFF (M3S-1) FROM 1850 TILL 2002 ...... 110

TABLE 6.7 DROUGHTS OBSERVED IN BOTH MELTING SEASON AND PLUVIAL SEASON RUNOFF RECONSTRUCTIONS AND PHASE IN

THE CLIMATIC FORCINGS RECONSTRUCTIONS ...... 115

TABLE 6.8 COMPARISON OF HIGHEST AND LOWEST N YEAR MOVING AVERAGE OF THE RECONSTRUCTED LOWER AND UPPER

PART OF BIOBÍO RIVER (1739-2002) AND THE GAUGED RUNOFF (1960-2002). PERIOD 1-10 ARE THE N YEARS

MOVING AVERAGE, RANK 1-5 ARE THE EXTREME RECONSTRUCTED EVENTS AND OBSERVED ARE THE EXTREME GAUGED

EVENTS. BOLD VALUES INDICATE COINCIDENT PERIODS-YEARS AMONG THE DIFFERENT RECONSTRUCTIONS...... 116

List of Appendixes

APPENDIX 1 DONNELLY AT STRICKLAND OBSERVED TIME SERIES ...... 147

APPENDIX 2 HELENA AT NGANGAGURINGURING OBSERVED TIME SERIES ...... 148

APPENDIX 3 DENMARK AT KOMPUP OBSERVED TIME SERIES ...... 149

APPENDIX 4 CAUQUENES AT CAUQUENES OBSERVED TIME SERIES ...... 150

APPENDIX 5 CATO AT PUENTE CATO OBSERVED TIME SERIES ...... 151

APPENDIX 6 LUMACO AT LUMACO OBSERVED TIME SERIES ...... 152

APPENDIX 7 INFLOWS TO RALCO DAM OBSERVED TIME SERIES ...... 153

APPENDIX 8 INFLOWS TO LAJA DAM OBSERVED TIME SERIES ...... 154

APPENDIX 9 DONNELLY SEASONAL TRENDS ...... 155

APPENDIX 10 HELENA SEASONAL TRENDS ...... 156

APPENDIX 11 DENMARK SEASONAL TRENDS ...... 157

APPENDIX 12 CAUQUENES SEASONAL TRENDS ...... 158

APPENDIX 13 CATO SEASONAL TRENDS ...... 159

APPENDIX 14 LUMACO SEASONAL TRENDS ...... 160

APPENDIX 15 LAJA SEASONAL TRENDS ...... 161

APPENDIX 16 RALCO SEASONAL TRENDS ...... 162

APPENDIX 17 LARGE SCALE CLIMATIC FEATURES RECONSTRUCTIONS ...... 163

APPENDIX 18 HISTOGRAMS OF ANNUAL CHANGES IN RUNOFF CONSIDERING ALL THE SIMULATIONS OF CPDN AND THE GROUPS

OF SIMULATIONS WITH DIFFERENT PERTURBATIONS OF 4 ATMOSPHERIC PARAMETERS. RED LINE REPRESENTS THE

MEDIAN OF THE WHOLE ENSEMBLE AND DOTTED RED LINE THE MEDIAN OF THE SIMULATIONS FOR A PARTICULAR

PERTURBATION ...... 175

CHAPTER 1. INTRODUCTION

1.1 Hydroclimatic trends in Mediterranean-like climate catchments

Human history has been linked to the availability of fresh water which is vital for life and for the economic activities it supports. Continuous population growth, migration to cities and economic development increases the pressure on water availability and has been aggravated by the impacts of climate change, putting natural sources of fresh water in a vulnerable context. Water management arises from the need for an efficient, optimum and sustainable use of the resource that ensures water supplies for current needs and for the future human population.

Observed and projected hydrometeorological variables are fundamental inputs needed to address the water manager’s challenge. This challenge is particularly acute in mid-latitude and semi-arid regions of the Earth, where sustained increases in temperature, reductions in precipitation (Magrin et al., 2014; Reisinger et al., 2014) and reductions in streamflow (Henessy et al., 2007; Magrin et al., 2007) have been observed. Despite reductions in precipitation, increases in temperature and increases in potential evapotranspiration are widely recognized as causes related to drought conditions, the need to distinguish causation with respect to drought remains a challenge and a question to be addressed in future research (Kiem et al., 2016), challenge that is out of the scope of the present work. This work focuses on catchments located in two of these mid-latitude Mediterranean climate regions (Peel et al. (2007); temperature of the hottest month>10°C & 0°C

According to the World Bank, projections at a catchment scale are needed for policy makers to design adaptation measures (Alavian et al., 2009). Notably, a significant amount of research on regional hydrological projections has been developed during the last decades, but this work has inherent uncertainty associated with each stage of the modelling process. First, the internal variability and chaotic nature of the climate system represent a large source of uncertainty

in climate projections. Second, there are uncertainties in the future path of greenhouse gases (GHGs) emissions and their influence on the climate. The largest source of uncertainty arises from the global climate models (GCMs) used to simulate the climatic variables (Bates et al., 2008). For instance, there is evidence that GCMs do not satisfactorily represent the influence of large-scale physical processes that drive rainfall at the local scale in Australia (Randall et al., 2007; Kiem et al., 2008; Kiem and Verdon‐Kidd, 2011). Other sources of uncertainty in streamflow projections include the downscaling technique used to translate the coarse data from a GCM to the catchment scale, the hydrological model used to simulate runoff and the short length of observed variables, or even lack of gauged data in some regions like the mountainous areas of South America and in particular of Central Chile (Rubio and McPhee, 2010). Regarding observations, one important source of uncertainty is the poor understanding of the role of potential evapotranspiration (PET) in the reductions of runoff, regarding the causative order and interplay between the increases in temperature and PET, and the non-stationary relationship between rainfall and runoff in space and time (Kiem et al., 2016).

Besides the difficulties in producing robust climatic projections and accounting for the uncertainties associated to them, the sustainable freshwater resources management under the future hydrological uncertainty is a challenge itself (Poff et al., 2016). Model-based decision support systems and decision frameworks such as the eco-engineering decision scaling are tools explored by researchers to support water resources management under uncertain future climate (White et al., 2015; Poff et al., 2016). Decision-making tools are an active area of research out of the scope of the present study, however, some recommendations to deal with the uncertainties in runoff projections are provided in the Future Work Section (Chapter 7) of the thesis.

Climate change impact assessments have mainly focused on future runoff conditions rather than the uncertainty around those runoff projections. Peel et al. (2015) grouped uncertainty for GCMs into “between-GCM” uncertainties and “within-GCM” uncertainties. Between-GCM uncertainties refer to the differences in the climatic projections due to the structural differences in the GCMs for a given emissions scenario, which in this study has been addressed using the GCMs collated by the Coupled Model Intercomparison Project 5 (CMIP5; Taylor et al. (2012)). The true GCM uncertainty or within-GCM uncertainty corresponds to the different projections and trends that might be produced by one model, using different initial conditions and by adjusting the parameters that characterise the physics of the model in the range of plausible values for a given emissions scenarios. This is also known as ‘‘perturbed physics’’ analysis (Parker, 2013b).

The best source of data that has attempted to investigate the true GCM uncertainty has been the climateprediction.net project (CPDN; Rowlands et al. (2012)) and to date, this large ensemble of GCMs has not been used to quantify the impact of within-GCM uncertainty on runoff projections. CPDN data were released in “Giorgi regions” (Figure 1.1) based on the information 2

used by Giorgi et al. (2001). The SWA “Giorgi region” is adequate to represent SWA climate (Barria et al., 2015). However, the Southern South America region (SSA) of CPDN is very large, including central and southern Chile along with central and southern Argentina, regions that have influences of different climate features and significantly different signs in runoff changes. Thus for Central Chile, only the CMIP5 ensemble is feasible to be used in the study of impacts of GCM uncertainties on runoff projections. The difference in impact of between- and within-GCM uncertainties on runoff projections has not been addressed in existing hydrological assessments in either SWA or CC and as such is the focus of this thesis.

FIGURE 1.1 GIORGI REGIONS, BASED ON GIORGI ET AL. (2001) AND EXTRACTED FROM ROWLANDS ET AL. (2012) A further difficulty in assessing climate change impacts on hydrological variables arises from the lack of long (>50 years) and reliable (within feasible ranges) regional observed meteorological data, especially in rugged, steep topography mountainous areas where the installation and maintenance of hydrometeorological gauges is a challenge. In contrast to SWA, the climate, hydrology and topography of the temperate zone of CC is strongly influenced by the Andes mountain range and is characterized by a lack of observed hydrometeorological data where most of the Chilean rivers originate. Analysis of current changes and projections in streamflow require a better understanding of runoff generation, which involves study of the long term and spatially distributed records of runoff time series. Unlike SWA, this task is particularly challenging in CC high elevation catchments (>1000 masl) for the already mentioned scarcity of data. In this context, paleoclimatic methods are useful tools to understand the changes in climatic processes over time as tree ring chronologies have demonstrated to be a good proxy for

hydrometerological variables in central Chile (Muñoz et al., 2014), and they allow extension of runoff time series hundreds of years into the past. However, to date, no high elevation river time series has been reconstructed in Chile and the dendrochronological extension of streamflow in this region could be an important contribution to analyse multi-centennial runoff variability and lower frequency runoff processes. Reconstructed streamflow could give insights into current structural changes in the high elevation catchment hydrology and about current droughts, which are key issues to be addressed by water managers. Regarding Australia, results of exploratory reconstructions of runoff in temperate areas such as Western Tasmania (Allen et al., 2015) and Murray river (Gallant and Gergis, 2011) indicate that tree rings are able to reproduce around 30% of the natural variability, which is considerable lower than results obtained in CC temperate region (50%). Considering this evidence, reconstruction analysis is only developed in CC high elevation catchments in this thesis, with the aim of contributing to understanding the hydrology of this mountainous temperate region.

1.2 Objectives

This thesis intends to contribute to the understanding of runoff variability at different time scales in Mediterranean-like climate catchments of SWA and CC as both regions share similar climatic features and hydrological trends. This work addresses important gaps in the field of future water availability assessments through investigating four main research aims.

Aim 1: Investigate the impact of within-GCM uncertainties on runoff projections in southwest Western Australia.

• On a catchment scale, quantify how much runoff is expected to change by the end of the century, 2050-2080 compared to the historical period 1970-2000, by using a novel multi-thousand member ensemble of a GCM with perturbed physics (CPDN). • Implement a methodology to quantify the within-GCM uncertainties in runoff projections at an annual and seasonal time scale in SWA catchments. • Examine the differences between the within-GCM uncertainties using a perturbed physics ensemble of GCMs against current approximations that use statistical methodologies.

Aim 2: Assess the differences in impact of between-GCM and within-GCM uncertainties on projected runoff in southwest Western Australia.

• On a catchment scale, quantify how much runoff is expected to change by the end of the century, 2050-2080 compared to the historical period (1970-2000), using an ensemble of multiple GCMs (CMIP5). • Implement a methodology to quantify and compare the within and the between- GCM uncertainties in runoff projections at an annual and seasonal time scale in SWA catchments. • Investigate whether the number of different models and runs in the CMIP5 ensemble is a good approach for accounting for the within-GCM uncertainties.

Aim 3: Investigate the between-GCM uncertainties in runoff projections in Central Chilean catchments.

• On a catchment scale, quantify how much runoff is expected to change by the end of the century, 2050-2080 compared to the historical period (1970-2000), using an ensemble of multiple GCMs (CMIP5). • Quantify the uncertainties in runoff projections for the 21st century in Central Chilean catchments and compare them with the results obtained in SWA catchments.

Aim 4: Explore the runoff variability at lower frequencies in a high elevation catchment of CC (Biobío river) using a multi-century reconstruction of runoff.

• Are local tree ring chronologies informative proxies of climate variables for extending streamflow records in mountainous Chilean catchments? • Are current observed changes in the high elevation catchment part of a long term trend or part of the main cycles of natural variability of the catchment? How do they compare with the lower part of the catchment? • What are the large scale climatic features that drive the low frequency variability in the high elevation river?

1.3 Thesis structure

This thesis is structured in order to address the four research aims of the project as shown in Figure 1.2. First, in Chapter 2, a literature review is presented of the studies already undertaken in the field of hydrological variability and projections in Mediterranean-like catchments of the Southern Hemisphere, and the principal gaps in these studies are identified. Then, Chapter 3 includes the description of the data and the main characteristics of the regions of study.

Within-GCM uncertainties in runoff projections in SWA catchments are presented in Chapter 4 which addresses the first aim of the project. Quantification of within and between-GCM

uncertainties in SWA catchments and comparison of runoff projections among CC and SWA catchments are presented in Chapter 5, fulfilling aims 2 and 3 of the study. The fourth aim of the project is addressed in Chapter 6 where a 300 year reconstruction of pluvial season and melting season streamflow in a high elevation catchment in central Chile are presented. Finally, a discussion and conclusion section with the main findings of the project and the proposed future research are drawn and discussed in Chapter 7.

FIGURE 1.2 THESIS STRUCTURE

CHAPTER 2. LITERATURE REVIEW

The Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (Stocker et al., 2013) has stated that global warming is unequivocal, estimating that an increase in temperature is certain with a magnitude of about 0.85°C between 1888 and 2012. In contrast, observed changes in precipitation are not uniform with some regions presenting increases and others showing decreases in mean annual rainfall. Less confidence exists in the magnitude and direction of change in observed precipitation, mainly related to the lack of reliable and well distributed data, especially prior to 1951 (Stocker et al., 2013).

There is low confidence regarding the sign of any change in global streamflow. Dai et al (2009) in their analysis of 925 downstream gauged stations of rivers around the world found that decreases in streamflow have been observed in many low and mid-latitude catchments, where precipitation has decreased. Similar results were found by Milly et al. (2005) in their analysis of 165 world rivers. Runoff has larger natural variability relative to temperature and climate change can increase the hydrological variability with changes that depend on intrinsic characteristics of the catchment, like the hydrological regime of the water course. According to Alavian et al. (2009) snowmelt driven catchments have in general experienced earlier spring runoff due to the enhanced melt of glaciers and snowmelt, whereas in pluvial regime catchments the changes depend greatly on the sign of change in precipitation.

Despite the large variability regarding the magnitude and sign of change of observed runoff, there is evidence that the Southern Hemisphere mid-latitude regions, like southwest Western Australia (SWA) and central Chile (CC), have presented constant decreases in the average annual runoff during the last decades (Magrin et al., 2014; Reisinger et al., 2014). Barros et al. (2014) indicated that there is high confidence that SWA climate has experienced rising average temperature and reductions in precipitation during the last 50 years. Climatic projections inform us that temperature is expected to continue increasing during the 21st century. Barros et al. (2014) also noted that reductions in mean annual precipitation are projected with high confidence in SWA, affected by large uncertainties that challenge adaptation plans.

Negative trends in mean annual precipitation have been observed in Central Chile with a magnitude of -1 mm day1 (50 year)-1 during 1950-2008 (Barros et al., 2014). Conversely, positive trends have been observed in mean annual temperature from the mid-70s with increases that fluctuate regionally between 0.7° C and 1°C (40 year)-1. In contrast a cooling of the Chilean coast of about -1° (40 year)1 has been observed. There is high confidence in projections of decreasing runoff in the central Andes (Chile - Argentina) and medium confidence in increases in temperature (Barros et al., 2014).

SWA and CC are both areas experiencing prolonged drying (Boisier et al., 2016; IOCI, 2012). Increasing demand for water supply and decreasing runoff volumes due to a dry climate have placed these regions under water stress. Thereby, a better understanding of the natural variability of runoff at different time scales and analysis of projections of runoff at a catchment scale are critical for water resources allocation. This study focuses on the analysis of projections of runoff and the uncertainties around those projections in these two regions. Additionally, investigation of low frequency natural variability and long dry spells in a high elevation catchment in CC are performed. In the next two sections a review of the hydrological assessments already performed in SWA and CC are presented. The main literature undertaken in the field of GCM uncertainties on runoff projections are presented in section 2.3.

2.1 Observed trends and projections of hydroclimatic variables in SWA catchments

According to the Australian Bureau of Statistics (2012) over three quarters of the 2,451,400 inhabitants of Western Australia (September 2012) reside in the Greater Perth region, whose population is projected to increase to 3.1 million by 2050. This increased population will enhance water stress in the region where according to Climate Commission (2011) and Charles et al. (2007) average inflows to reservoirs have already declined since the mid-70s to around half historical levels (1911-1973). The drying trend in SWA has affected agriculture and urban drinking water supply, and according to the Climate Council of Australia (Climate Commission, 2011) the measures taken in this decade (2010-2020) are crucial as they will determine the future of the resource. Robust projections of runoff, defined as consistent in sign and including uncertainties quantification, are fundamental for planning these measures.

According to the Köppen-Geiger classification (Peel et al., 2007) SWA has a temperate Csb climate with a dry and warm summer. About 80% of total annual rainfall falls between May and October (Bates et al., 2010), which is the main driver of runoff generation in the region. Catchments in SWA present a pluvial regime of runoff which peak during the winter months (June-August) when precipitation is maximum. The mean annual rainfall in SWA ranges between around 500 mm year-1 in the north to around a maximum of 2300 mm year-1 in the Darling ranges (Silberstein et al., 2012). The hydrological sensitivity of runoff in SWA catchments to changes in precipitation is around 2.5 (Chiew, 2006; Jones et al., 2006), which means that a 1% change in precipitation is amplified into a 2.5% change in the runoff response. Understanding how changes in the large scale climatic features that have shaped current rainfall trends and how they are projected to evolve in the next decades is important in order to study future streamflow changes.

The El Niño Southern Oscillation (ENSO) has been identified as the main tropical driver of intra-seasonal and interannual precipitation in Australia with a major influence in the southeast and the north of the country (Allan, 1988; Nicholls et al., 1997; Risbey et al., 2009) but less influence on SWA. The Indian Ocean Dipole has been found to have an influence over precipitation during winter and early spring (June-October) in the south and west of Australia (Risbey et al., 2009). In addition, the Southern Annular Mode (SAM) has been distinguished as the most important extratropical feature that has been related to the interannual rainfall variability in the southern part of Australia during all the seasons, and particularly during winter (Hendon et al., 2007; Meneghini et al., 2007; Risbey et al., 2009).

An important shift in SWA climate has been detected and analysed by several researchers. Some authors identified the shift as beginning around the mid-1960s (Haylock, 1993; Ansell et al., 2000; Cai and Cowan, 2006), while others have indicated it started around the mid-1970s (Timbal, 2004; Frederiksen and Frederiksen, 2007; Hennessy et al., 2007; Charles et al., 2010b; Petrone et al., 2010; IOCI, 2012). This shift is related to changes in large scale climatic features that have produced a displacement of the Southern Hemisphere storm track (Frederiksen and Frederiksen, 2006; Frederiksen and Frederiksen, 2007; Frederiksen et al., 2011; Frederiksen and Frederiksen, 2011), which has led to reductions in autumn and winter precipitation in SWA of around 20% since the mid-1970s. A synchronicity in the historical extended dry spells across continental regions of the Southern Hemisphere including SEA, SWA and CC, has been reported by Verdon-Kidd and Kiem (2014) and Verdon-Kidd et al. (2014) which has been associated with changes in the ocean-atmospheric processes of the Pacific Ocean (ie. ENSO, Interdecadal Pacific Oscillation (IPO)), to generalised warming across the Indian Ocean and to a positive trend in the SAM.

The seasonality of SAM is highly correlated with winter precipitation over SWA (Cai and Cowan, 2006; Hendon et al., 2007) and current trends in rainfall have been associated with a positive trend in the Southern Annular Mode (Haylock, 1993; IOCI, 2002; Cai and Cowan, 2006; Delworth and Zeng, 2014) and with an intensification of the subtropical ridge (Timbal and Drosdowsky, 2013) that brings stable conditions and less storms to the region. The IOCI (2012) has also shown that a reduction in the strength of the Southern Hemisphere jetstream has been observed during the last decades, which means there is less potential energy in the atmosphere that affects storm formation.

A notable amount of research has been conducted that investigates the synoptic weather patterns related to precipitation in SWA. Raut et al. (2014) found that decreases in the frequency of strong fronts in June and weaker fronts in June-July, which are associated with a positive phase in SAM index enhanced when ENSO is in its neutral phase, are related to less rainfall in SWA. Similarly Pook et al. (2012) in their analysis of daily rainfall for April-October for the years 1965- 9

2009 concluded that reductions in the number of cutoff lows and fronts contribute greatly to precipitation reductions during the growing season (autumn and spring) and during winter respectively. Likewise, Hope et al. (2006) identified that the reduction in the frequency of deep low systems correlates very well with SWA rainfall reductions (r=0.7) in the period 1970-2000. However, rainfall reductions since 2000 are better correlated with increases in the daily occurrence of high pressure systems.

The decline in SWA runoff has been affected by the decreases in rainfall described above (Bates et al., 2010). Falling water tables have also contributed to declines in SWA runoff, which together with reduced soil moisture seems to have contributed to a change in the proportion of rainfall that becomes runoff (Petrone et al., 2010; Hughes et al., 2012). Hughes et al. (2012) found that for the Darling Range catchments, a rainfall threshold of between 1050-1400 mm per year is needed to increase groundwater storage, and sustained years of below- average rainfall has reduced the runoff ratio. Similarly, CSIRO. and BOM. (2009), in their analysis of historical and future streamflow in 13 surface water catchments, indicated that despite rainfall not decreasing during the period 1997-2007, runoff continued to decline, which suggest a change in the runoff ratio. This means that projections of runoff that use historical conditions in the catchments to calibrate hydrological models might overestimate future streamflow (Saft et al., 2016).

Regarding climatic projections, considerable research has been conducted evaluating the performance of different GCMs in Australia. Evans and Ji (2012) evaluated the performance of GCMs to simulate ENSO as the main driver of Australian climate interannual variability, precipitation, temperature, mean sea level pressure (MSLP) and daily rainfall among other variables over Australia. They found that HadCM3 is one of the models with better performance in simulating the climate of SWA. Alexander and Arblaster (2009) evaluated the CMIP3 (Coupled Model Intercomparison Project Phase 3; Meehl et al. (2007)) simulations of extreme temperature and precipitation over Australia, concluding that in general the ensemble of models can reproduce the observed trends in extreme temperature and precipitation but with large differences among the individual models.

Projections of the IOCI (2012) indicate that dry conditions are expected in southwest Western Australia for the next century compared to present day (1962-1999). Maximum and minimum temperatures are projected to increase during the next century. High pressure systems will become more prevalent, with a 70% increase in their incidence during 2081-2100 relative to 1961-2000, leading to reductions in precipitation. Alexander and Arblaster (2009) indicated that according to CMIP3 simulations, substantial increases in warm nights, heat wave duration, simple daily intensity in rainfall (ratio of annual total precipitation to number of days with precipitation higher than 1mm), consecutive dry days, very heavy precipitation contribution and decreases in frost days are projected for the 21st century. Moise and Hudson (2008) stated that according to 10

the CMIP3 simulations, reductions of about 25-30% in winter precipitation are projected for SWA under the A2 emission scenario (Nakicenovic et al., 2000).

Similar results have been reported from the CMIP5 ensemble. Projections are robust and consistent indicating persistent reductions in winter precipitation during the 21st century following the trends observed during the last 50 years in Southern Australia (Irving et al., 2012; Hope et al., 2015a; Hope et al., 2015b). Mean annual temperature over Australia has increased by 0.9°C in the period 1910-2011 and it is projected to increase by 0.6 to 1.7 °C for RCP2.6, 1.4 to 2.7 °C for RCP4.5 and 2.8 to 5.1 °C for RCP8.5 scenario (Van Vuuren et al., 2011) for the year 2090 (CSIRO and BOM., 2015).

In terms of changes in freshwater flows, Silberstein et al. (2012) reported that using CMIP3 GCMs, a median decline of 8% in rainfall by 2030 is projected in SWA, which would lead to a reduction of 25% in streamflow. Preston and Jones (2008) found that a median decrease of 20% in runoff is projected by 2030 in SWA catchments compared to the mean annual runoff of 1990 using 12 different GCMs under the A2 scenario. Finally, using CMIP5 projections, Hope et al. (2015a) noted with high confidence that increased potential evapotranspiration in combination with reduced rainfall will lead to reductions in soil moisture and runoff in SWA under scenarios RCP4.5 and RCP8.5. Hope et al. (2015a) emphasised that a more comprehensive analysis in runoff modelling is needed in order to assess projections of water resources in this region. Regarding analyses of long term hydrological data and multidecadal variability, some research has been performed using paleoclimate methods to reconstruct streamflow in Australia. They have been developed in few rivers, mainly located in the east coast of the country. From north to south and east to west, runoff time series for three rivers have been reconstructed: Burderkin (Isdale et al., 1998), the Murray river (McGowan et al., 2009; Gallant et al., 2011) and Western Tasmania (Allen et al., 2015). The main characteristics of these reconstructions are presented in Table 2.1. The results of these exploratory studies indicate that tree rings are able to reproduce around 30% of the natural variability, which is considerable lower than results obtained in the CC temperate region (50%). To date no SWA streamflow reconstruction has been developed, but tree ring chronologies of Callitris columellaris have been used to extend the autumn-winter rainfall time series for the Lake Tay region (Cullen et al., 2008) back to 1655. Analyses of the reconstructed rainfall time series reveal important multidecadal variability associated to El Niño-Southern Oscillation. Despite the little research performed in the field of streamflow reconstructions in SWA and the possibility of using tree ring chronologies to conduct the investigation, that analysis was not included in this thesis because the extension of runoff time series is focused in the study of long term variability in high elevation catchments with nivo- pluvial regime of runoff, characterised by a lack of observation records and thus understanding

of its temporal variability, rather than pluvial catchments such as the case of SWA. However, this is an interesting future work that would give important insights into multidecadal variability of runoff and drought analyses in SWA.

TABLE 2.1 HISTORIC STREAMFLOW RECONSTRUCTIONS IN AUSTRALIA. THE LENGTH OF THE RECONSTRUCTIONS ARE PRESENTED IN THE COLUMN ENTITLED LENGTH, THE INFORMATION ABOUT THE PROXY DATA USED TO DEVELOP THE RECONSTRUCTION IS PRESENTED IN PROXY DATA COLUMN, AND THE EXPLAINED VARIANCE IN THE R2 COLUMN.

Type of River Length Proxy data R2 Reference reconstruction

New Zealand Kauri, Indonesian Teak, Western Australia Callitris, Southeastern Australia Murray Annual (Aug- Huon pine, Southeastern Gallant et 1783-1988 0.22 river July) runoff Australia celery top pines, al. (2011) Eastern and Western Tasmania celery top pine tree rings. Fiji Tonga, Great Barrier and Bali corals

A. cupressoides, A. Summer Western selaginoides, P. Allen et al. (Dec- Jan) 1530-2007 0.35 Tasmania aspleniifolius, L. franklinii (2015) runoff tree ring chronologies

RMSE Murray Annual (Jan- 184,8 McGowan 1474-1994 river Dec) inflows Shen et al. (2006) PDO Gl et al. (2009) index (~5%) Burderkin Annual (Oct- Isdale et al. 1644-1980 0.5 river Sep) runoff Great barrier reef corals (1998)

Despite the extended and deep analysis of climatological features that shape natural variability and projections of runoff conducted in SWA, two gaps have been identified in this study which need to be filled in order to contribute with information for decision makers. First, the relationship between runoff projections and the impact of within-GCM uncertainties on these projections have not been addressed in SWA catchments. Second, a comparison of the between and the within-GCM uncertainties in runoff projections have not been quantified in SWA catchments. This is critical as reductions of runoff have been reported in the region and current assessments may be underestimating the water shortages for the next decades.

2.2 Observed trends and projections of hydroclimatic variables in Central Chile

Central Chile is the narrow band of land located between 33ºS and 39ºS on the west slopes of the Andes. According to the National Statistics Institute, 74% of the total Chilean population

resides in this region (INE, 2015). The largest agricultural productivity and around 70% of the total installed hydroelectric capacity of the country are produced in CC (Banco Mundial, 2011; Urrutia et al., 2011), which are the activities with largest consumptive and non-consumptive use of water in the country (MOP, 2012). According to MOP (2012), CC has experienced prolonged below-average runoff during the last decades, which is attributed to climate change, that together with water shortages due to disproportionate granted rights of water use has placed catchments in CC in a water scarcity condition, which is projected to continue into the future. Regional projections of runoff are crucial for planning and designing an adaptation strategy, however the high streamflow variability, the complex hydrological processes enhanced by the steep topography and the lack of well distributed stream gauging stations, especially in the upper part of the region, have challenged the analysis of statistically significant trends in runoff and the analysis of runoff projections in CC.

Central Chilean catchments encompass different flow regimes including pluvial, snowmelt and mixed rainfall-snowmelt catchments. Only pluvial and mixed rainfall-snowmelt catchments are analysed in this thesis. Rainfall variability dominates runoff generation in pluvial catchments, whereas both precipitation and temperature variability dominate runoff generation in the mixed rainfall-snowmelt catchments. Thus, changes in precipitation and temperature have the potential to affect the magnitude and the temporal variability of runoff in these rivers.

An analysis of 40 unimpaired CC rivers by Cortés et al. (2011) showed that considering the period between 1961 and 2006, the center of timing of the water year hydrograph (CT), defined as the center of mass of the annual flow, has statistically significant negative trends for all types of regimes. That means that a change in the distribution of the annual runoff with a movement toward winter months has been observed in CC since 1961. However, a break has been identified around the mid-70s characterised by no change in the CT in the period 1976-2006, which is likely associated with decreasing precipitation volumes since the mid-70s (Cortés et al., 2011). Furthermore, the authors indicated that rainfall dominated catchments present larger significant changes in runoff compared to snowmelt dominated catchments, suggesting a larger sensitivity to changes in rainfall than to temperature. Nonetheless, it is important to note that the Cortés et al. (2011) trend analysis is limited by the short and incomplete records of runoff in the high elevation part of CC where rivers are characterised by a mixed snowmelt-rainfall regime. Further analysis of observed changes and projections of rainfall are required in order to have a comprehensive understanding of runoff variability and to produce robust projections of runoff in this region.

According to the Köppen-Geiger climate classification, CC experiences a Csb temperate climate with a dry and warm summer (Peel et al., 2007). Precipitation in the region is mainly generated by warm and cold fronts associated with migratory surface cyclones (Garreaud et al., 13

2009) and ranges between around 500 to 5000 mm yr-1 (Rubio and McPhee, 2010). The Cordillera mountain range is an important feature as it provides conditions for orographic precipitation generating an increase in the mean annual precipitation from the Pacific coast to the foothills of the Andes with a factor of about 1.8 (Garreaud, 2007).

Two different patterns of precipitation have been detected in central Chile with a boundary at around 37°S. Mean annual precipitation occurs mostly during the austral winter months (June-August) in the region north of 37°, whereas south of 37° rainfall episodes are well distributed throughout the year with the largest volumes concentrated in winter. According to Quintana and Aceituno (2012), the interannual variability of the region located north of 37°S is dominated by ENSO activity, as represented in this work by the Southern Oscillation Index (SOI). The Pacific Decadal Oscillation (PDO; Trenberth and Hurrell (1994)) modulates the multidecadal variability of rainfall. Dry conditions are associated with a positive phase of the SOI (La Niña) and a negative phase of the PDO, while wet conditions are associated with a negative phase in the SOI (El Niño) and positive in the PDO. In the region south of 37°S, as in SWA, the SAM exerts a major influence in the interannual variability, with negative trends in precipitation being linked to a positive trend in the SAM.

Similar results have been observed in runoff variability. Rubio and McPhee (2010) studied the spatial variability of streamflow in catchments located in the region between 34°S and 45°S. The authors found that ENSO has a major influence on the runoff variability in the catchments located north of 37.5°S, whereas the SAM is linked with the runoff variability of the southern catchments. Rubio and McPhee (2010) also found significant negative trends in runoff in the region between 38°S and 40°S.

Increases in temperature and decreases in precipitation since mid-1970s have been observed in central Chile (Boisier and Aceituno, 2006; Carrasco et al., 2008; Garreaud, 2011), which might affect runoff generation. A positive trend in the equilibrium line altitude (ELA) has also been reported by Carrasco et al. (2008). The ELA corresponds to the altitude at which the accumulated mass in a glacier equals the melted mass, which is related to regional climatic characteristics. These changes have been associated with a shift in the PDO cycle identified in 1976/1977 (Carrasco et al., 2008). Furthermore, Masiokas et al. (2006) indicated that a strong positive correlation between the winter and annual precipitation in CC and the snowpack in the Central Andes has been observed in the period 1951 and 2005.

The summary of the project Climate variability in Chile: Assessment, Interpretation and Projections (Estudio de la variabilidad climática en Chile para el siglo XXI, 2016, September 18, retrieved from http://dgf.uchile.cl/PRECIS/articles-39442_pdf_Estudio_texto.pdf), noted that the lack of a long and well distributed observation dataset in Chile impeded the analysis of long term

trends. However, from the information available they discerned that there was a marked shift in the climate of Chile in the mid-1970s. Since then, there have been noticeable increases in temperature in north and central Chile, whereas a cooling has occurred on the coast with reductions in temperature by as much as -0.25°C/decade (Falvey and Garreaud, 2009b). A persistent drought, the so-called Megadrought, has been observed in central and southern Chile (2010-2015) with decreases in precipitation of about 30% with respect to the mean annual climatology of the period 1970-2000. Boisier et al. (2016) indicated that 50% of these changes in precipitation can be attributed to the PDO effect and they estimate that around a quarter is related to anthropogenic causes.

Regarding climatic projections, considerable research into the impact of climate change on central Chilean catchments has been conducted. However, these studies focus on the median of projections under a given scenario and they have not assessed uncertainties around them. Garreaud (2011) indicated that by using the PRECIS model (“Providing Regional Scenarios for Climate Studies”; Meteorological Office (2001)), increases in temperatures by 2100 of 4°C in summer and 5°C in winter under the A2 scenario are expected to occur in north and central Chile. Furthermore, Vicuña et al. (2010) reported that temperature is expected to increase by about 3- 4°C by 2100 along the subtropical Andes assuming the A2 and B2 scenarios (Nakicenovic et al., 2000) of the PRECIS model. Garreaud (2011) has indicated that reductions of 60-70% in mean annual rainfall are expected by the end of the century using the A2 scenario of PRECIS in the mountainous region of Central Chile. Projected changes in temperature and precipitation under both the A2 and the B2 scenarios would induce reductions in the mean annual runoff in the subtropical Andes and modify the hydrograph, with decreased runoff in spring/summer and increases in winter season (Vicuña et al., 2010).

Analyses have also been performed in CC using the CMIP3 (Meehl et al., 2007) and CMIP5 (Taylor et al., 2012) ensembles of GCM. Falvey and Garreaud (2009b) assessed CMIP3 temperature projections comparing them to observational records over Chile, indicating that even though the GCMs are skilful in representing current global temperatures, they are not able to reproduce the fine spatial differences in Chile. Demaria et al. (2013) assessed the impacts of climate change on runoff in one alpine catchment (Mataquito river) located in Central Chile using 12 ensemble members of CMIP3 and CMIP5 datasets under the Special report on emission scenarios (SRES) A1B, RCP 4.5 and SRES A2, RCP 8.5 scenarios respectively. They found that dry conditions are expected in the future with reductions in precipitation for the period 2070-2099 compared to 1960-1989 that fluctuate between 7% and 20%, and that differences in projections between the two datasets are negligible.

Regarding high elevation catchments, projections of surface runoff in the upper catchments of Central Chile suggest reductions over the next decades (Fuenzalida et al., 2007). 15

However, uncertainties arise from the short observational records and almost no coverage of meteorological data in this region. Understanding and attributing current changes in water availability requires analysis of long records in order to capture the natural temporal and spatial variability of runoff (Field, 2014), which extension of the instrumental record through paleoclimatic methods may assist. Tree rings are a commonly used proxy to reconstruct hydroclimatic variability (Urrutia et al., 2011) as tree growth depends on water availability and changes in temperature, which are the same variables that drive runoff generation in Central Chile. According to Boninsegna (2009), tree ring chronologies in South America constitute the highest elevation hydroclimatic data records worldwide.

Using paleoclimate methods, streamflow reconstructions in Chile have been developed in some rivers, mainly located in the central valley and along the coast. From north to south, runoff time series for four rivers have been reconstructed: Maule (Urrutia et al., 2011), the lower Biobío river (Muñoz et al., 2016), Puelo (Lara et al., 2008) and Baker (Lara et al., 2015). The main characteristics of these reconstructions are presented in Table 2.2. To date no Chilean streamflow series has been reconstructed for a catchment located over 1000 meters above sea level (masl), which is defined as a high elevation catchment.

The upper part of the Biobío has been chosen to reconstruct runoff in this project because water availability is critical for irrigation purposes and because this catchment is where 50% of Chilean hydroelectric energy is generated (Lara et al., 2003). The Biobio river is part of the ecoregion denominated by the Valdivian Rainforest, which is located between 35-48°S (Lara et al., 2005), primarily in Chile with parts extending into Argentina. One of the main differences between the upper and lower Biobío is the absence of a dry season in the upper part of the Biobío river. The upper Biobío receives mean annual precipitation of 2266 mm, with 25% occurring during the summer half of the year. Also, the upper and the lower Biobío river have different runoff seasonality with the upper having a double peak and the lower a single peak. The seasonality of runoff in the Biobío river is driven by precipitation and temperature: autumn-winter precipitation generates what it’s call here “pluvial runoff” (runoff between April and September). Precipitation falling above the 0oC isotherm in autumn-winter months contributes to the accumulation of snow. During spring and summer as the temperature increases the accumulated snow melts and in combination with spring and summer rainfall forms the “melting season runoff” or streamflow during spring and summer months (October to March).

In summary, the topography and weather characteristics of the high elevation parts of central Chile have made it difficult to install and maintain weather stations, which explains the lack of reliable and long term climatic records in this region. Long term regional information is fundamental to extending our understanding of runoff variability in the past and to analyse the observed changes in the hydrology and how they are related to climatic features. 16

TABLE 2.2 HISTORIC STREAMFLOW RECONSTRUCTIONS IN CHILE. THE LENGTH OF THE RECONSTRUCTIONS ARE PRESENTED IN THE COLUMN ENTITLED LENGTH, THE MEAN ELEVATION OF THE CATCHMENT IN METERS ABOVE SEA LEVEL ARE PRESENTED IN THE ELEVATION COLUMN AND THE EXPLAINED VARIANCE IN THE R2 COLUMN. Type of Elevation River Length R2 Reference reconstruction (masl) Summer (Dec- Puelo 1599-2002 314 0.42 Lara et al. (2007) May) runoff Annual (April- Maule 1590-2000 290 0.42 Urrutia et al. (2011) March) runoff Summer (Jan- Baker 1765-2004 261 0.53 Lara et al. (2015) April) runoff Lower Annual (April- Biobío 1599-2003 205 0.48 Muñoz et al. (2016) March) runoff catchment Upper Pluvial season Biobío (April-September) 1712-2002 1505 0.49 This work catchment runoff Upper Melting Biobío season(October- 1665-2003 1505 0.49 This work catchment March) runoff

To this point, two gaps have been identified in assessments of water resources in CC which challenge the delivery of accurate information needed to design adaptation measures under a water scarcity scenario and which this study seeks to address. First, water resources studies are affected by a lack of well distributed and long records of reliable observed data, especially in high elevation catchments. An extended regional dataset is needed in order to improve our understanding of current changes in the hydrology of the region.

Second, analyses of uncertainties on runoff projections in Central Chilean catchments have not been assessed. Only the median of models’ ensemble projections have been reported for specific catchments, which limits informed decision-making under water scarcity conditions.

2.3 Uncertainties in runoff projections

Quantification of regional runoff projections for the next decades (second half of the 21st century) are a fundamental input for designing adaptation policy frameworks. The IPCC 5th Assessment report (Stocker et al., 2013) has indicated that a better understanding of the climatic system and the ability to make projections under non-stationary conditions remain a key and active area of research. The characterization of precipitation, the most important driver of runoff, remains uncertain (Milly et al., 2005; Nohara et al., 2006; Kundzewicz et al., 2008). There is low confidence regarding the long term trends of rainfall mainly due to the inconsistencies in the observations, the lack of meteorological stations in complex topography places and due to uncertainties in the measuring methodologies (Stocker et al., 2013).

Modelling runoff under changing conditions involves uncertainties (Peel and Blöschl, 2011) including uncertainties in the Global Climate Models (GCMs) used in the simulation, the downscaling methodology that is used to translate the GCM outputs to the regional or catchment scale, and due to the hydrological model used to simulate runoff. In particular, limitations in the GCM representation of the influence of large -scale physical processes which drive rainfall at the local scale in Australia (Randall et al., 2007; Kiem et al., 2008; Kiem and Verdon‐Kidd, 2011) have been detected.

There are two primary downscaling approaches used to produce climate projections: the dynamical and the statistical methods. The dynamical method involves running a regional high- resolution model driven by boundary conditions from a GCM to produce climatic simulations at a catchment scale. On the other hand, the statistical method consists in the development of statistical relationships between the GCM output and the local climate variables. Because statistical downscaling is relatively easy to produce and less computationally expensive than dynamical downscaling, that is the methodology commonly used by researchers in their assessments of runoff projections (Mizukami et al., 2016). Another important source of uncertainty in the downscaling of the GCM data is the selection of the base period used to apply the methodology and the assumption that the bias correction are stationary over time (Kiem and Verdon‐Kidd, 2011; Peel et al., 2015).

Moreover, streamflow projections are also affected by the problem of identifying a representative hydrologic model structure and model parameters (Uhlenbrook et al., 1999; Vogel and Sankarasubramanian, 2003; Efstratiadis and Koutsoyiannis, 2010; Mendoza et al., 2015), which according to Mendoza et al. (2015) are substantial sources of uncertainty and as comparable to the GCM uncertainties in the estimated hydrologic projections. Also, uncertainties arise from the observed data against which the model is calibrated (Andréassian et al., 2004; McMillan et al., 2010), the poor understanding of controlling variables such as the role of potential evapotranspiration (PET) in the reductions in runoff (Kiem et al., 2016) and the calibration method and objective functions chosen to find the representative hydrologic model (Efstratiadis and Koutsoyiannis, 2010).

Nevertheless, considerable research has indicated that the largest source of uncertainty in the cascade process of simulating streamflow projections arise from the GCM used in the modelling (Chiew et al., 2008; Prudhomme and Davies, 2008; Ardoin-Bardin et al., 2009; Chiew et al., 2009; Chen et al., 2011; Xu et al., 2011; Teng et al., 2012b; Lafaysse et al., 2014) which is the source of uncertainties this thesis seeks to assess.

Global climate models solve the mathematical relations that represent the physics of the atmosphere and its interaction with Earth: the conservation of momentum, mass and energy

equations. However, the coarse scale of the grid they use (a couple of hundred square kilometres) that is restricted by the available computational tools and the time needed to solve the equations, the parameterisation used to simulate some of the processes that occur in smaller scales, and the feedback processes or the interactions between all the components of the system, all generate uncertainty (Räisänen, 2007; Prudhomme and Davies, 2008; Parker, 2013b).

Mote et al. (2011), Hawkins and Sutton (2009) and Hawkins and Sutton (2011) have indicated that the GCM uncertainties can be partitioned into three main groups. First, the unpredictable path of future emissions, which depends greatly on socioeconomic factors, remains a large source of uncertainties. To date, the future path of GHG has been assessed by defining different emission scenarios. Second, uncertainties arise from the representation of the climate sensitivity in the GCMs; defined as the response of the climatic system to the GHG emissions. This has been enhanced by an inadequate understanding of some relevant atmospheric processes like the aerosol-cloud interaction and the changes in the components of the water cycle (Stocker et al., 2013). Finally, the internal variability of the climate system, defined as how the natural climatic system can mask the effect of forced drivers of climate change challenges GCM formulation and is an important source of uncertainties.

Two approaches have been used to address GCM uncertainties in runoff projection studies: “between-GCMs” uncertainties and “within-GCMs” uncertainties (Parker, 2013a; Peel et al., 2015). The between-GCMs analysis of uncertainty considers the spread of a multi-model ensemble of different GCM projections, for the same emissions scenario, as a first order uncertainty; it quantifies the uncertainty due to different GCM formulations. Assessments of uncertainties in runoff projections mostly rely upon the ensembles of opportunity (Dobler et al., 2012; Steinschneider et al., 2012; Bosshard et al., 2013; Lafaysse et al., 2014) represented by the GCM runs collated in the Coupled Model Intercomparison Project Phase 3 (CMIP3; Meehl et al. (2007)) and Phase 5 (CMIP5; Taylor et al. (2012)).

On the other hand, ‘‘within-GCM’’ uncertainty here defined as the range of possible projections from a single model, for the same emissions scenario, that might be produced by a plausible range of values of the adjustable parameters within the model, when combined with uncertainties due to internal variability and initial conditions. This is a ‘‘perturbed physics’’ analysis which involves using the same model but changing in each simulation a selected set of the parameters that characterize the model physics (Parker, 2013a). So far, the only source of data that has attempted to investigate the true uncertainty in a GCM in this way has been the climateprediction.net project (Rowlands et al., 2012), which through parameter variation has quantified the physical aspects of uncertainty (Stainforth et al., 2005; Mote et al., 2011).

Peel et al. (2015) quantified an approximation to the within-GCM uncertainties in runoff projections in 17 worldwide catchments. The authors developed non-stationary stochastic replicates of monthly precipitation and temperature from several GCMs of the CMIP3 ensemble and ran a hydrological model with the 100 stochastically replicated monthly data. Peel et al. (2015) found that on average the within-GCM uncertainties in projections of mean annual precipitation for 2015-2044, computed as the standard deviation expressed as a percentage of the mean annual precipitation, are 4.2% which translates into within-GCM uncertainty of 10.1% in the mean annual runoff. Peel et al. (2015) emphasized that their analyses might be underestimating the true within-GCM uncertainties as the stochastic model does not sample uncertainties in the trends simulated by the GCMs.

The main limitation of ensembles of opportunity that explore between-GCM uncertainties is that they are not a random nor systematic statistical sample of models. Many of the GCM models collated in CMIP5 and CMIP3 are not independent (Jun et al., 2008; Masson and Knutti, 2011; Pennell and Reichler, 2011; Knutti et al., 2013). On the other hand, perturbed physics experiments are systematically constructed (Rougier, 2007; Sansó and Forest, 2009) but they do not take into account the range of structural uncertainties as they use just one GCM to be sampled in their parameters. Uncertainties in runoff projections are an active area of research and a more detailed analysis that involves structural uncertainties and parametric uncertainties is required to enhance informed decision making.

Two gaps have been identified in the field of projections of runoff which this study seeks to address. First, within-GCM uncertainties have not been quantified in hydrological assessments and there is limited information about within-GCM uncertainties in runoff projections (Barria et al., 2015; Peel et al., 2015). The results presented in Chapter 4 of this study and published by Barria et al. (2015), quantified the within- GCM uncertainties on runoff projections in SWA catchments using the novel perturbed physics approach (CPDN ensemble of GCMs). Second, comparisons of the differences among the between and the within-GCM uncertainties in runoff projections have not been evaluated in water resources assessments. As reported before, mostly climate variables from ensembles of opportunity (CMIP3 and CMIP5) have been used to quantify uncertainties which might result in an underestimation of the future water availability.

2.4 Summary

Because runoff projections are a fundamental tool for the design of strategies to deal with water shortages, understanding future behaviour of precipitation and temperature sensitive catchments has become a needed task in water stressed regions. According to the review presented in the previous sections, this is particularly important for rivers located in SWA and CC, where a

negative sign in the annual runoff has been observed during the last decades and where reductions are projected to continue during the next decades. To date, uncertainties around climatic projections have not been quantified in a comprehensive manner and the most important limitations of current assessments which this study addresses are summarised below:

• SWA catchments have a pluvial hydrological regime and catchments are sensitive to changes in precipitation and temperature. A reduction of about 50% in the mean annual runoff has been observed in SWA since mid-70s which is dramatic. Reductions in runoff in SWA catchments have been attributed to a displacement of the storm track which is associated with a positive trend in SAM. • A considerable amount of research allows us to conclude that streamflow projections in SWA indicate continuous reductions till the second half of the 21st century. However, the impact on runoff of true within-GCM uncertainties has not been quantified. This deficiency will be addressed in Chapter 4. • Comparison among the between and the within-GCM uncertainties have not been developed in SWA catchments. Hydrological assessments have focused on quantification of uncertainties in runoff projections using multi-model ensembles. This deficiency will be addressed in Chapter 5. • CC catchments have different regimes of runoff and both pluvial and mixed rainfall- snowmelt catchments are studied in this thesis. Rainfall is the largest driver of streamflow changes in the region. • CC has a temperate climate; ENSO and SAM are the main drivers of interannual variability such as in SWA, and similar trends in precipitation and runoff have been observed with reductions since the mid-70s. • The steep CC topography and the short available records in the upper lands of CC (compared to SWA catchments) challenge the study of trends in hydroclimatological variables and the analysis of low frequency processes and drought. Assessments have indicated that tree ring chronologies can be a good proxy to extend runoff records in high elevation catchments. A high elevation streamflow reconstruction for the BioBio River will be presented in Chapter 6. It’s worth mentioning that acknowledging the influence of the Andes mountain range in the hydroclimatic variability of CC, especially in mountainous catchments characterised by scarce observations, paleoclimatic reconstructions have been restricted to this region and avoiding SWA. • Quantification of the between-GCM uncertainties have not been performed in hydrological assessments in CC catchments. This deficiency will be addressed in Chapter 5.

• A thorough assessment of projections of runoff in SWA and CC needs to be performed, which includes analysis of projections of runoff for different time windows and the uncertainties around these projections in order to provide the information that is fundamental for the planning of the water resources in a climate change context. This deficiency will be addressed in Chapter 5.

CHAPTER 3. HYDROMETEOROLOGICAL DATA AND CATCHMENT CHARACTERISTICS

Observed and modelled hydrometeorological data have been used to address the research questions raised in the Introduction and the Literature Review sections. Three sources of data have been consulted in two regions of the Southern Hemisphere: SWA and CC. Gauged and modelled meteorological and runoff data on a monthly basis that represent the climate of the region of study were obtained from the Department of Water and the weather services of Australia and Chile. Monthly precipitation and temperature from ensembles of GCMs were used to analyse the uncertainties in historical and future projections of climatic variables: temperature, precipitation and runoff. Finally, annually resolved tree ring chronologies from long lived trees that grow in the high altitudes of the temperate Andes mountain range were used to reconstruct seasonal runoff in a high elevation catchment in Central Chile. Description of these data is presented in the following sections.

3.1 Observed meteorological and runoff data

3.1.1 Southwest of Western Australia

Monthly gridded modelled data based on spatial interpolation of observed temperature and precipitation data from the Australian Water Availability Project (AWAP; Jones et al. (2009)) were used to calibrate the hydrological model used to produce projections of runoff. AWAP data was produced by the Bureau of Meteorology (BOM) and the Commonwealth Scientific and Industrial Research Organization (CSIRO) of Australia using ~7500 gauges within the BOM network (Tozer et al., 2012). Anomalies of daily/monthly meteorological data were interpolated using the Barnes successive correction technique (Jones and Weymouth, 1997), whereas the monthly climatological averages were interpolated using three dimensional smoothing splines (Jones et al., 2009). It’s important to mention that despite the methodology used to generate the AWAP data has been found to be robust in the representation of the climatology of Australia in the long term, its accuracy is limited in regions with insufficient station networks, where the available data is insufficient to resolve detail. Examples of regions with limited AWAP representations are the coastal region of Western Australia (Jones et al., 2009) and the steep topography zones such as the Snowy Mountains in South East Australia. AWAP data has also been found to have limitations in the representation of extreme dry or wet conditions (Tozer et al., 2012; King et al., 2013), which is not the case of the SWA region studied in this thesis. ARCGIS 9.3 was used to delineate the catchment boundaries using the two-second Shuttle Radar Topography Mission (SRTM) Smoothed Digital Elevation Model (DEM-S) version 1.0

(Geoscience Australia, 2010) and area-weighted temperature and precipitation values were calculated for each catchment under study. More information is provided in Chapter 4.

Without loss of generality three unimpaired catchments that encompass zones with different amounts of annual precipitation and temperature, and different coefficients of variation of runoff were analysed in SWA, Donnelly, Helena and Denmark rivers (Peel et al., 2000; Zhang et al., 2013), which are presented in Figure 3.1 and their characteristics summarized in Table 3.1. An unimpaired river is defined here as meaning the streamflow is not subject to human regulation, thus making them good indicators of climate change as the magnitude and variability of streamflow should be only sensitive to the changes of the meteorological variables.

FIGURE 3.1 LOCATION OF SWA CATCHMENTS

TABLE 3.1 CHARACTERISTICS OF SWA RUNOFF STATIONS

Mean Mean Standard Mean Coeff. Period of annual Area annual deviation of annual Catchment of observations precipitation (km2) runoff annual runoff temperature variation (mm/year) (mm/year) (mm/year) (ºC)

Donnelly at 1961-1992 1004 780 162.72 69.8 0.43 15.2 Strickland Helena at 1973-2000 665 327 6.41 5.6 0.87 17.00 Ngangaguringuring Denmark at 1961-2000 835 502 58.93 34.3 0.58 15.3 Kompup

Regarding the hydrometorological characteristics of the catchments summarized in Table 3.1, the total annual precipitation ranges between a maximum of around 1000 mm in the wettest catchment (Donnelly) to a minimum of around 665 mm in the driest catchment (Helena) (time series presented in Appendices 1-8). On average 48% of the total annual precipitation falls during austral winter months (June-July-August) in the three catchments and only 7% of the total annual rainfall falls during summer months (December-January-February). The mean annual runoff and the coefficient of variation range between ~6 mm to ~163 mm, and between 0.87 to 0.43 at the driest and the wettest catchment (Helena and Donnelly respectively), indicating that the drier the catchment the larger its interannual variability. The mean monthly temperature in the region peaks around February and then gradually decreases to its minimum value in July. Helena catchment, located in the northeast part of SWA, has the lowest annual precipitation and the maximum temperature among the three rivers. The climograms of the catchments which illustrate the monthly precipitation and temperature in every catchment are presented in Figure 3.2.

FIGURE 3.2 CLIMOGRAMS OF SWA CATCHMENTS The seasonal variation of runoff in SWA catchments is presented in Figure 3.3. The three catchments have a single peak around austral winter months June-August, when precipitation is maximum. Runoff is a minimum during summer months when precipitation is a minimum and temperature is a maximum. There is absence of snowmelt runoff in the SWA rivers.

FIGURE 3.3 SEASONAL VARIATION OF RUNOFF IN SWA CATCHMENTS The observed annual runoff along with the annual precipitation for the three catchments under study is presented in Figure 3.4. These plots indicate reductions in annual rainfall and in annual runoff which according to the Mann-Kendall test (Wilks, 2011) are statistically significant only in Donnelly at Strickland River at the 95% level of significance. The dotted lines in Figure 26

3.4 indicate reductions of runoff of about 4.3 mm year-1, 0.1 mm year-1 and 0.57 mm year-1 during the observed period at Donnelly, Helena and Denmark catchment respectively. The seasonal analysis of gauged runoff and precipitation are presented in Appendices 9-16. Overall, small reductions have been observed in mean autumn rainfall and runoff and increases seen in mean spring rainfall and runoff, which are not statistically significant. Particularly concerning are the strong reductions in winter season runoff because this represents 61% of the total annual runoff and the largest contribution to reservoirs in SWA.

FIGURE 3.4 ANNUAL GAUGED RUNOFF AND PRECIPITATION DATA. THE DOTTED LINE IS THE FITTED LINEAR CURVE TO THE ANNUAL RUNOFF IN A) DONNELLY AT STRICKLAND, B) HELENA AT NGANGAGURINGURING AND C) DENMARK AT KOMPUP According to Silberstein et al. (2012), land cover in SWA (Figure 3.5) consists of 67% native forest vegetation which is mainly Eucalyptus marginata and marri E. calophylla. Only 1%

of the total SWA land area is covered by commercial plantations and around 2% corresponds to irrigated agriculture. However, a more extensive area dedicated to dryland agriculture is located in the Northern Perth basin and in the Swan Coastal plains. Silberstein et al. (2012) also reported that SWA has a complex geology formed by some of the oldest rocks in the world and that together with the topography, marked by the uplands of the Darling Plateau, has important effects on the hydrology of the region.

FIGURE 3.5 LAND COVER IN SWA CATCHMENTS. EXTRACTED FROM FIG. 2, SILBERSTEIN ET AL. (2012)

3.1.2 Central Chile

Five catchments were analysed in Central Chile, which span a 5°latitude range and different altitudes ranging from coastal to high elevation catchments (from ~60 masl. to ~1500 masl). The location of the catchments is presented in Figure 3.6.

Following the methodology adopted in SWA, and without loss of generality, the gauged runoff of three unimpaired low elevation catchments, Cauquenes at El Arrayán, Cato at Puente Cato and Lumaco at Lumaco (Rubio and McPhee, 2010) which span zones with different annual precipitation, annual temperature and coefficients of variation of runoff, were used to calibrate the hydrological model that was then run under different climatic scenarios until the end of the 21st century in order to analyse projections of runoff and the uncertainties around those projections. The high elevation catchments gauged data (inflows to Laja and inflows to Ralco)

corresponding to the upper part of the Biobío river were used to develop multi-centennial reconstructions of seasonal runoff using tree ring chronologies that provided information about long-term hydrological variability in the region. Monthly rain and temperature data for the five catchments were obtained from the records of two Chilean agencies: the Directorate of Water Resources (Dirección General de Aguas, DGA) and the weather service (Dirección Meteorológica de Chile, DMC).

ARCGIS 9.3 was used to delineate the catchments boundaries using the ASTER Global DEM (ASTER, 2009) and area-weighted temperature and precipitation values were calculated for each catchment using the meteorological stations indicated in Figure 3.6.

Low elevation catchment runoff data were obtained from the Directorate of Water Resources (Dirección General de Aguas, DGA) in Chile. The analysis of the upper part of Biobio river used 56 years (1960/1961-2015/2016) of monthly runoff data recorded at two stations: inflows to Laja Dam and inflows to Ralco dam. The data were provided by the private company ENDESA Chile, and correspond to undammed data, which means that the effect of the operation of the dams has been removed from the records (R. Gonzalez, personal communication, October 12, 2013). The time series of inflows to Laja and Ralco are highly correlated at the monthly and seasonal scales and they represent the same seasonal variation. Therefore, one time series representing the temporal variability of the entire region was constructed by taking the area weighted average of both the Laja and Ralco inflows. The characteristics of the 5 catchments and the combined series representative of the upper part of Biobio river are presented in Table 3.2.

The meteorological and hydrological time series for each station are presented in Appendices 1-

TABLE 3.2 CHARACTERISTICS OF CENTRAL CHILEAN CATCHMENTS Standard Mean Mean Mean annual deviation of Coeff. Period of Area annual annual Catchment precipitation annual of observations (km2) runoff temperature (mm/year) runoff variation (mm/year) (ºC) (mm/year) Cauquenes en el 1966-2000 715 619 463.6 260.4 0.56 8.7 Arrayán Cato en 1966-2005 1694 987 1297.2 489.7 0.38 8.7 Puente Cato Lumaco en 1966-2005 1053 869 620.6 198.4 0.32 10.1 Lumaco Inflows to 1960-2005 2286 967 1803.6 439.5 0.24 4.8 Laja dam Inflows to 1960-2011 2274 5115 1594.3 403.6 0.25 6.2 Ralco dam Combined series, upper 1960-2011 - - 1627.6 405.53 0.25 - part of the Biobío river

FIGURE 3.6 LOCATION OF CENTRAL CHILEAN CATCHMENTS The CC mean annual precipitation ranges between a maximum of around 2300 mm per year in the wettest catchment (Laja) to around a minimum of 700 mm in the driest catchment (Cauquenes at El Arrayán). In general, precipitation increases with altitude and with longitude in the region. Similarly to the SWA catchments, 51% of mean annual precipitation falls during austral winter months (June-July-August) in the low elevation Cauquenes and Cato catchments and only 22% of the annual rainfall falls during spring and spring-summer months (September- February). On the other hand, Lumaco at Lumaco and the high elevation catchments Laja and Ralco have a larger proportion of spring-summer rainfall (30%) but still winter precipitation is dominant (Appendices 9-16). The monthly variation of temperature has a maximum in January and a minimum in July; the driest catchment is Cauquenes which also presents the largest runoff coefficient of variation. Climograms of mean monthly precipitation and temperature for the low elevation catchments are presented in Figure 3.7.

Similarly to the SWA catchments, low elevation CC rivers present a pluvial regime of runoff with a peak during winter months, centred on July, following the maximum month of rainfall which is June (Figure 3.8). The high elevation CC catchments present a double peak

regime of runoff with a maximum during winter months centred around July and a second peak during the melting season when temperatures increase, which can be seen in Figure 3.9

FIGURE 3.7 CLIMOGRAMS FOR LOW ELEVATION CENTRAL CHILEAN CATCHMENTS

FIGURE 3.8 SEASONAL VARIATION OF CENTRAL CHILEAN CATCHMENTS

(a)

(b)

FIGURE 3.9 (a) Seasonal variation of the upper part of Biobío river runoff (mm) over the complete hydrological year (April-March). The barplot presents the climatogram of the area weighted precipitation (mm) gauged in Abanico and Polcura stations (1966 -2000) over the hydrological year. 2.(b) Comparison of the seasonal variation of observed runoff (mm) in the upper and the lower part of the Biobío River (1960-2002) The instrumental records presented in Figure 3.10 for the low elevation catchments and in Figure 3.11 for the high elevation part of Biobío river indicate continuous reductions in the mean annual rainfall and runoff. The dotted lines presented in Figure 3.10 are the fitted linear curves to the mean annual runoff and they indicate reductions of about 1.9 mm year-1, 0.9 mm year-1 and 2.2 mm year-1 for Cauquenes, Cato and Lumaco respectively. There are differences in the trends between the low elevation and the high elevation catchments. The instrumental records of the lowland catchments indicate reductions in the mean winter and summer rainfall and runoff which are not statistically significant (Appendices 9-16). Overall, slighter reductions have been observed in mean autumn rainfall and runoff and not statistically significant changes in mean spring rainfall and runoff. On the other hand, the fitted linear curves to the annual and seasonal

runoff in the high elevation catchments (Figure 3.11) indicate reductions in the mean annual and summer runoff of about 0.4 mm year-1 and 1.7 mm year-1 respectively but increases during the winter months of about 0.4 mm year-1. This has been associated with an increase in the snow line and a reduction in the pluvial area of the catchment (Fuenzalida et al., 2007) and it is one of the research questions that this study seeks to address through the extension of the seasonal time series.

FIGURE 3.10 ANNUAL GAUGED RUNOFF AND PRECIPITATION DATA.THE DOTTED LINE IS THE FITTED LINEAR CURVE TO THE ANNUAL RUNOFF AT A) CAUQUENES AT EL ARRAYÁN, B) CATO AT PUENTE CATO AND C) LUMACO AT LUMACO

FIGURE 3.11 OBSERVED HIGH ELEVATION BIOBÍO RIVER SEASONAL RUNOFF AND TRENDS (1960- 2015). THE FITTED LINEAR TRENDS INDICATE REDUCTIONS OF 0.65 M3S-1 FOR ANNUAL RUNOFF, AN INCREASE OF ABOUT 0.07 M3S-1 FOR PLUVIAL SEASON RUNOFF AND A REDUCTION OF ABOUT 1.4 M3S-1 FOR MELTING SEASON RUNOFF. Regarding land cover, Underwood et al. (2009) indicated that Central Chilean catchments have an urban coverage that fluctuates between 0.7% and 1.5%, and an area with intensive agricultural activity that fluctuates between 24 to 29%. In particular, the driest catchment, Cauquenes, has 59% of the total area covered by urban, farm and pasture lands which primarily consist of pine and Eucalyptus plantations (Cooper et al., 2013).

The Biobío high elevation catchment has two dominant tributaries, the Laja River which drains the higher part of the northern region of the catchment and the Vergara River which flows from the south central valley. There is little agriculture in the catchments, with land use mainly mountainous with limited vegetation. Two dams, the Laja lagoon and the Ralco lagoon, have been

built for hydroelectricity and downstream irrigation purposes in this catchment. The land cover of the whole CC region was obtained from Zhao et al. (2016) and is presented in Figure 3.12.

FIGURE 3.12 LAND COVER IN THE CC REGION. DATA EXTRACTED FROM ZHAO ET AL. (2016)

3.2 GCM data

Monthly temperature and precipitation data from the GCMs collated by the CPDN and CMIP5 ensembles were used in this thesis. The CPDN experiment was set before the CMIP5 data and the RCP scenarios were fully released to the scientific community, thus the SRES scenarios were considered in the project. One of the key differences between the SRES and the RCPs scenarios is the lack of mitigation policies implied in any of the former, while the RCPs seek to be compatible with the full range of stabilisation and mitigation scenarios currently available in the literature (Moss et al., 2010). Although the differences between the SRES and the RCPs full range of scenarios, the A1B scenario used in this investigation projects a radiative forcing of about 6 Wm-2 by 2100, which is within the range of 4.5 Wm-2 and 8.5 Wm-2 radiative forcing spanned by the RCP4.5 and RCP8.5 scenarios of the CMIP5 data used in the thesis. Considering this, the

investigation accepts the feasibility of using both ensembles (CPDN and CMIP5) for the sake of focusing on the uncertainties that arise from the GCMs structures and parameterisation rather than on the emission scenarios they consider.

CPDN uses the HadCM3L model, a reduced ocean resolution version of the HadCM3 model (Gordon et al., 2000), which through perturbed physics was run 2500 times generating climatic simulations for the period between 1920 until 2080 (Rowlands et al., 2012). The resolution of the model is 2.5º latitude by 3.75º longitude with 19 vertical levels in the atmosphere and 2.5º latitude by 3.75º longitude and 20 vertical in the ocean (Rowland et al., 2012). Each individual ensemble member was run under control forcing (representative of 1900 conditions) and transient forcing (time-varying concentrations of greenhouse gases) for the period between 1920 and 2080. CPDN was run under the A1B scenario (Nakicenovic et al., 2000), which specifies the emissions of greenhouse gases and aerosols for a future world of rapid economic growth. More information about CPDN is presented in Chapter 4.

The atmospheric parameters and the sulphur cycle and ocean parameters that were perturbed in the CPDN experiment and the plausible values that were considered in the project are presented in Table 3.3 and in Table 3.4, extracted from the project design paper (Rowlands et al., 2012).

TABLE 3.3 LIST OF PERTURBED ATMOSPHERIC PHYSICS PARAMETERS IN THE CPDN PROJECT. BOLD NUMBERS INDICATE DEFAULT VALUES. EXTRACTED FROM SUPPLEMENTARY TABLE SI 1, ROWLANDS ET AL. (2012)

TABLE 3.4 LIST OF PERTURBED SULPHUR CYCLE AND OCEAN PHYSICS PARAMETERS IN THE CPDN PROJECT. BOLD NUMBERS INDICATE DEFAULT VALUES. EXTRACTED FROM SUPPLEMENTARY TABLE SI 2, ROWLANDS ET AL. (2012)

Furthermore, a list of the 106 simulations from the 41 CMIP5 GCMs used in this thesis, along with the GCMs characteristics, is presented in Table 3.5. The CMIP5 ensemble mainly provides insight into between-GCM uncertainty. However, as some GCMs have several different runs for the same emissions scenario, the ensemble also provides some insight into within-GCM uncertainty, due to different parameterizations and initial conditions. CMIP5 models were run under control forcings (pre-industrial conditions) for the period between 1900 and 2005 and transient forcing (time-varying concentrations of greenhouse gases) for the period between 2006 and 2100. The anthropogenic forcing scenarios considered in this study correspond to the RCP4.5 (Van Vuuren et al., 2011), which is a moderate scenario that assumes a path of emissions that peaks by 2040 and then starts decreasing. The RCP8.5 scenario (Van Vuuren et al., 2011) was also evaluated, which assumes that the emissions continue to rise thorough to 2100, which gives a more conservative (i.e. risk averse) perspective regarding future runoff reductions. The CMIP5 data were resampled to a resolution of 1.5° latitude by 1.5° longitude and were provided by the Australian Bureau of Meteorology (F. Delage, personal communication, October, 9, 2015).

TABLE 3.5 LIST OF CMIP5 GCMS CHARACTERISTICS

Climate modelling centre and Stratospheric GCM Ensemble member Resolution Reference location Ozone Atmospheric: Centre for Australian Weather and Dix et al. ACCESS1-0 r1i1p1 1.875°x1.25°, Ocean: No Chem.

Climate Research, Australia (2013) 1°x1° Atmospheric: Centre for Australian Weather and Dix et al. ACCESS1-3 r1i1p1 1.875°x1.25°, Ocean: No Chem.

Climate Research, Australia (2013) 1°x1° College of Global Change and Earth Atmospheric:

BNU-ESM System Science, Beijing Normal r1i1p1 2.7906°x2.8125°, Ji et al. (2014) Chem. University, China Ocean: 1°x1° r1i1p1, r2i1p1, Atmospheric: National Centre for Atmospheric Meehl et al. CCSM4 r3i1p1, r4i1p1, 0.9424°x1.25°, Ocean: Chem. Research, USA (2012) r5i1p1, r6i1p1 1°x1° Atmospheric: CESM1- Community Earth System Model Gent et al. r1i1p1 0.9424°x1.25°, Ocean: Chem.

BGC Contributors (2011) 1°x1° Atmospheric: CESM1- Community Earth System Model r1i1p1, r2i1p1, Gent et al. 0.9424°x1.25°, Ocean: Chem. CAM5 Contributors r3i1p1 (2011) 1°x1°

CMCC- Centro Euro-Mediterraneo per I Atmospheric:3.7111°x Vichi et al. r1i1p1 No Chem.

CMS Cambiamenti Climatici, Italy 3.75° (2011)

Centro Euro-Mediterraneo per I Atmospheric: Vichi et al. CMCC-CM r1i1p1 No Chem. Cambiamenti Climatici, Italy 0.7484°x0.75° (2011)

Atmospheric: CNRM- Centre National de Recherches Voldoire et al. r1i1p1 1.4008°x1.40625°, Chem.

CM5 Meteorologiques, France (2013) Ocean: 1°x1° r1i1p1, r2i1p1, Commonwealth Scientific and r3i1p1, r4i1p1, Atmospheric: CSIRO- Industrial Research Organization in Rotstayn et al. r5i1p1, r6i1p1, 1.8653°x1.875°, Ocean: No Chem.

Mk3-6-0 collaboration with Queensland Climate (2012) r7i1p1, r8i1p1, 1.875°x1.875° Change Centre of Excellence, Australia r9i1p1, r10i1p1 Atmospheric: r1i1p1, r2i1p1, Canadian Centre for Climate Modelling 2.7906°x2.8125°, Arora et al. CanESM2 r3i1p1, r4i1p1, No Chem.

and Analysis, Canada Ocean: (2011) r5i1p1 0.9303°x1.1407°

r2i1p1, r8i1p1, Atmospheric: Hazeleger et EC-EARTH EC-EARTH consortium, Europe No Chem.

r9i1p1, r12i1p1 1.1215°x1.125° al. (2011)

LASG, Institute of Atmospheric Atmospheric:

FGOALS-g2 Physics, Chinese Academy of Sciences r1i1p1 2.7906°x2.8125°, Li et al. (2013) No Chem. and CESS, Tsinghua University, China Ocean: 1°x1° LASG, Institute of Atmospheric Atmospheric: Physics, Chinese Academy of Sciences, Bao et al. FGOALS-s2 r1i1p1 1.6590°x2.8125°, No Chem.

China, The First Institute of (2013) Ocean: 1°x1° Oceanography, SOA, China The First Institute of Oceanography, r1i1p1, r2i1p1, Qiao et al. FIO-ESM Atmospheric: 2°x2° No Chem.

SOA, China r3i1p1 (2013) NOAA Geophysical Fluid Dynamics Atmospheric: 2°x2.5°, Donner et al. GFDL-CM3 r1i1p1 Chem.

Laboratory, USA Ocean: 0.3344°x1° (2011) Atmospheric: GFDL- NOAA Geophysical Fluid Dynamics Dunne et al. r1i1p1 2.0225°x2°, Ocean: No Chem.

ESM2G Laboratory, USA (2012) 0.375°x1° Atmospheric: GFDL- NOAA Geophysical Fluid Dynamics Dunne et al. r1i1p1 2.0225°x2.5°, Ocean: No Chem. ESM2M Laboratory, USA (2012) 0.3344°x1°

Climate modelling centre and Stratospheric GCM Ensemble member Resolution Reference location Ozone

GISS-E2-H- NASA Goddard Institute for Space Atmospheric: 2°x2.5°, Schmidt et al. r1i1p1 No Chem.

CC Studies, USA Ocean: 1°x1° (2006)

r1i1p1, r2i1p1, GISS-E2-H- NASA Goddard Institute for Space Atmospheric: 2°x2.5°, Schmidt et al. r3i1p1, r4i1p1, No Chem. p1 Studies, USA Ocean: 1°x1° (2006) r5i1p1 r1i1p2, r1i1p3, GISS-E2-H- r2i1p2, r2i1p3, p2 and NASA Goddard Institute for Space Atmospheric: 2°x2.5°, Schmidt et al. r3i1p2, r3i1p3, Chem. GISS-E2-H- Studies, USA Ocean: 1°x1° (2006) r4i1p2, r4i1p3, p3 r5i1p2, r5i1p3 GISS-E2-R- NASA Goddard Institute for Space Atmospheric: 2°x2.5°, Schmidt et al. r1i1p1 No Chem. CC Studies, USA Ocean: 1°x1.25° (2006) r1i1p1, r2i1p1, GISS-E2-R- NASA Goddard Institute for Space Atmospheric: 2°x2.5°, Schmidt et al. r3i1p1, r4i1p1, No Chem. p1 Studies, USA Ocean: 1°x1.25° (2006) r5i1p1, r6i1p1 r1i1p2, r1i1p3, GISS-E2-R- r2i1p2, r2i1p3, p2 and NASA Goddard Institute for Space r3i1p2, r3i1p3, Atmospheric: 2°x2.5°, Schmidt et al. Chem. GISS-E2-R- Studies, USA r4i1p2, r4i1p3, Ocean: 1°x1.25° (2006) p3 r5i1p2, r5i1p3, r6i1p3 National Institute of Meteorological Atmospheric: HadGEM2- Martin et al. Research, Korea Meteorological r1i1p1 1.25°x1.875°, Ocean: No Chem.

AO (2011) Administration, Korea 1°x1° Atmospheric: HadGEM2- Martin et al. Met Office Hadley Centre, UK r1i1p1 1.25°x1.875°, Ocean: No Chem.

CC (2011) 1°x1° Atmospheric: HadGEM2- r1i1p1, r2i1p1, Collins et al. Met Office Hadley Centre, UK 1.25°x1.875°, Ocean: No Chem.

ES r3i1p1, r4i1p1 (2011) 1°x1° Atmospheric: IPSL- r1i1p1, r2i1p1, Dufresne et al. Institut Pierre Simon Laplace, France 1.8947°x3.75°, Ocean: Chem.

CM5A-LR r3i1p1, r4i1p1 (2013) 2°x2° Atmospheric: IPSL- Dufresne et al. Institut Pierre Simon Laplace, France r1i1p1 1.2676°x2.5°, Ocean: Chem. CM5A-MR (2013) 2°x2° Atmospheric: IPSL- Dufresne et al. Institut Pierre Simon Laplace, France r1i1p1 1.8947°x3.75°, Ocean: Chem. CM5B-LR (2013) 2°x2° Japan Agency for Marine-Earth Science and Technology, Atmosphere Atmospheric: MIROC- and Ocean Research Institute (The 2.7906°x2.8125°, Watanabe et r1i1p1 Chem.

ESM-CHEM University of Tokyo), and National Ocean: al. (2011) Institute for Environmental Studies, 0.5582°x1.40625° Japan Japan Agency for Marine-Earth Science and Technology, Atmosphere Atmospheric: MIROC- and Ocean Research Institute (The 2.7906°x2.8125°, Watanabe et r1i1p1 Nochem.

ESM University of Tokyo), and National Ocean: al. (2011) Institute for Environmental Studies, 0.5582°x1.40625° Japan Japan Agency for Marine-Earth Atmospheric:1.4008°x r1i1p1, r2i1p1, Watanabe et MIROC5 Science and Technology, Atmosphere 1.40625°, Ocean: Nochem.

r3i1p1 al. (2011) and Ocean 0.5°x1.40625° MPI-ESM- Max Planck Institute for Meteorology, r1i1p1, r2i1p1, Atmospheric:1.8653°x Giorgetta et al. No Chem.

LR Germany r3i1p1 1.875° (2013) MPI-ESM- Max Planck Institute for Meteorology, r1i1p1, r2i1p1, Atmospheric:1.8653°x Giorgetta et al. No Chem.

MR Germany r3i1p1 1.875° (2013)

Climate modelling centre and Stratospheric GCM Ensemble member Resolution Reference location Ozone

MRI- Meteorological Research Institute, Atmospheric:1.8653°x Yukimoto et r1i1p1 No Chem.

CGCM3 Japan 1.875° al. (2012)

NorESM1- Atmospheric:1.8947°x Iversen et al. Norwegian Climate Centre, Norway r1i1p1 Chem.

ME 2.5° (2013) NorESM1- Atmospheric:1.8947°x Iversen et al. Norwegian Climate Centre, Norway r1i1p1 Chem.

M 2.5° (2013)

bcc-csm1-1- Beijing Climate Center, China Atmospheric:2.7906°x

r1i1p1 Wu (2012) No Chem. m Meteorological Administration, China 2.8125°, Ocean: 1°x1°

Beijing Climate Center, China Atmospheric:2.7906°x

bcc-csm1-1 r1i1p1 Wu (2012) No Chem. Meteorological Administration, China 2.8125°, Ocean: 1°x1° Russian Institute for Numerical Atmospheric:1.5°x2°, Volodin et al. inmcm4 r1i1p1 No Chem.

Mathematics, Russia Ocean: 0.5°x1° (2010)

3.3 Tree ring data

Paleoclimatic proxy records are a useful tool to analyse the long term variability of climate variables. In this study, tree ring chronologies of A. araucana and A. chilensis (Villalba and Veblen, 1997; Le Quesne et al., 2000) are used. A. araucana is a long lived conifer that grows in the Patagonian Andes. Muñoz et al. (2014) studied the growth pattern of araucarias and indicated that there are significant positive correlations between the tree ring width and precipitation, and significant negative correlations between the tree ring width and temperature, linking the growth of araucarias with the availability of soil moisture. Astroceudrus Chilensis is a deciduous species that grows in a broad latitudinal range from 32°40'S to 43°30'S spanning the valleys and the mountainous area. A. Chilensis is a moisture-sensitive long lived tree (Austrocedrus chilensis, 2016, December 12, retrieved from http://www.conifers.org/cu/Austrocedrus.php).

A tree-ring chronology is a non-dimensional index that represents departures of growth for any one year compared to an average growth of a group of trees. The set of 12 chronologies used here were sampled in different sectors of the Andes near the upper Biobío catchment spanning the area between latitude 37°S and 39°28'S (Figure 3.13). The characteristics of the 12 chronologies are presented in Table 3.6 and the correlation between them (average of 0.22) in Table 3.7, with mean sensitivity that ranges between 0.18 and 0.29 when considering both species. The mean sensitivity is the average fractional relative change in the ring-width from one year to the next (Fritts, 2012). A sensitivity of around 0.2 is common for Araucaria Araucana and Astrocedrus Chilensis and indicates that the ring growth has low year-to-year variability.

FIGURE 3.13 MAP SHOWING THE LOCATION OF THE UPPER PART OF THE BIOBÍO RIVER, THE RAINFALL AND TEMPERATURE STATIONS (LIGHT BLUE CIRCLES), THE RUNOFF STATIONS (DARK BLUE CIRCLES), THE PRECIPITATION ISOHYETS (MM) AND THE 12 TREE RING CHRONOLOGIES USED AS PREDICTORS

TABLE 3.6 CHRONOLOGIES USED IN THE RECONSTRUCTIONS OF RUNOFF IN THE UPPER PART OF BIOBÍO RIVER

N° 1st order Mean Code Species Period Lat (oS) Long (oW) Author samples autocorrelation sensitivity

Holmes, A. 1664- LON 56 38°23′ 71°34 0.557 0.177 Dunwiddie, araucana 1975 Gutierrez A. 1585- RAL 90 38°03' 71°18' 0.582 0.232 Christie, D. chilensis 2005 A. 1444– CAV 62 37°52′ 71°01′ 0.533 0.194 Mundo, I. araucana 2003 Mundo, I. ,Lamarche, V.C.; Holmes, A. 1712- PIN 41 39°18' 71°17' 0.553 0.228 R.L.; Ambrose, araucana 2006 J.E.; Boninsegna, J.A. A. 1246- CHE 112 38° 06´ 70° 53' 0.498 0.203 Mundo, I. araucana 2007 Mundo, I., Lamarche, V.C.; Holmes, A. 1591- RAH 51 39°24' 70°48' 0.611 0.188 R.L.; Ambrose, araucana 2006 J.E.; Boninsegna, J.A. A. 1375- NAL 31 38°20′ 71°21′ 0.519 0.178 Muñoz A. araucana 2002 A. 1339- NIT 109 37°41' 71°17' 0.594 0.289 Christie, D. chilensis 2005 A. 1250- RIK 76 38°18' 71°42' 0.434 0.193 Muñoz A. araucana 2007 A. 1664- CAP 51 38°39' 71°42' 0.544 0.176 Muñoz A. araucana 2007 A. 1264- PAR 36 38°49' 71°04' 0.542 0.227 Mundo araucana 2006 A. 1608- COL 30 37°25' 71°36' 0.440 0.184 Lara, A. araucana 2002

TABLE 3.7 CROSS CORRELATION BETWEEN THE TREE RING CHRONOLOGIES Name LON RAL CAV PIN CHE RAH NAL NIT RIK CAP PAR COL chronology LON 1.00 0.12 0.36 0.28 0.20 0.27 0.32 0.25 0.27 0.87 0.19 0.26 RAL 0.12 1.00 0.25 0.36 0.17 0.35 0.06 0.73 0.11 0.01 -0.01 -0.07 CAV 0.36 0.25 1.00 0.30 0.52 0.41 0.36 0.11 0.28 0.28 0.29 0.25 PIN 0.28 0.36 0.30 1.00 0.00 0.39 0.27 0.29 0.24 0.13 0.01 0.14 CHE 0.20 0.17 0.52 0.00 1.00 0.30 0.18 0.20 0.13 0.13 0.44 0.13 RAH 0.27 0.35 0.41 0.39 0.30 1.00 0.20 0.16 0.26 0.14 0.11 0.12 NAL 0.32 0.06 0.36 0.27 0.18 0.20 1.00 0.07 0.00 0.22 0.20 0.74 NIT 0.25 0.73 0.11 0.29 0.20 0.16 0.07 1.00 0.06 0.11 0.09 -0.10 RIK 0.27 0.11 0.28 0.24 0.13 0.26 0.00 0.06 1.00 0.17 0.32 -0.04 CAP 0.87 0.01 0.28 0.13 0.13 0.14 0.22 0.11 0.17 1.00 0.14 0.14 PAR 0.19 -0.01 0.29 0.01 0.44 0.11 0.20 0.09 0.32 0.14 1.00 0.02 COL 0.26 -0.07 0.25 0.14 0.13 0.12 0.74 -0.10 -0.04 0.14 0.02 1.00

Tree ring chronologies of Araucaria Araucana and Astrocedrus Chilensis were used to reconstruct the melting season and the pluvial season runoff of the upper part of Biobío river. The tree ring chronologies were sampled in a non-destructive way; the ring core samples from trees were measured with a 0.001 mm precision and then cross-dated visually in order to assign a calendar year following the Schulman’s convention (Schulman, 1956) for the Southern Hemisphere. The quality of the chronologies and the elimination of the age and forest impacts were checked with the software COFECHA (Holmes, 1983) and ARSTAN (Cook, 1985). Following the methods of Cook and Kairiukstis (1990) a negative exponential curve was fitted to the tree rings and then was used to detrend them. This methodology has the aim of removing the biological trend in the tree ring time series through the difference detrending technique, which computes the difference of the time series and the trend at every point in time. Then these tree rings were grouped and averaged using a mean value function which is evaluated with the Expressed Population Signal (EPS statistic), which evaluates the common signal between rings in time (Wigley et al., 1984).

3.4 Climatic features reconstruction time series

Reconstructions of SAM and PDO were obtained from the Climatic Reconstructions Dataset of the National Oceanic and Atmospheric Administration (NOAA). The Villalba et al. (2012) SAM reconstruction, the Shen et al. (2006) PDO reconstruction and the D'arrigo et al. (2001) PDO reconstructions were used to assess the contribution of the long-term climatic features to the long-term streamflow variability.

The Villalba et al. (2012) SAM reconstruction used 108 different chronologies of 6 different species that were sampled in the mid-latitudes of the Southern Hemisphere from Tasmania, New Zealand and South America. None of these chronologies were used here in the melting season runoff reconstruction, which is the season in which SAM most influences runoff in this river. Discussion regarding the similarities and discrepancies among different SAM indexes used by the scientific community and its influence on Australian hydroclimatic variability is presented by Ho et al. (2012). According to their analysis, the more realistic indexes to account for the physical processes behind the SAM are those characterised by the mean sea level pressure (MSLP) measures and the station based indexes such as the NCEP and the Marshall indexes. Note that the Villalba et al. (2012) reconstructions used in this project are based on the recommended SAM indexes.

The Shen et al. (2006) PDO reconstruction used data from a drought/flood index obtained from Chinese historical documents (Zhang et al., 2003), which is a proxy for summer rainfall, to reconstruct the PDO index till 1470. Finally, the D'arrigo et al. (2001) PDO reconstruction used a set of tree ring chronologies sampled in the forests of western North America.

The proxies or chronologies used in the SAM and PDO reconstructions are independent from the chronologies used to reconstruct the upper Biobío runoff, and these reconstructions can be compared without risk of circularity.

All the extended time series used here are presented in Appendix 17 of the document.

CHAPTER 4. UNCERTAINTIES IN RUNOFF PROJECTIONS IN SOUTHWESTERN AUSTRALIAN CATCHMENTS USING A GLOBAL CLIMATE MODEL WITH PERTURBED PHYSICS

4.1 Introduction

This chapter addresses the first aim of this thesis presented in Chapter 2. It is a reproduction of the published paper Barria P, Walsh KJE, Peel MC, Karoly D (2015), Uncertainties in runoff projections in southwestern Australian catchments using a global climate model with perturbed physics, Journal of Hydrology, 529, 184-199. That has been reformatted to align with the thesis formatting. The section 1 “Introduction” and the section 2 “Data” have not been included as that material is available in Chapters 2 and 3 respectively, which have been removed to avoid repetition.

In SWA, water resources are one of the sectors more affected by the observed increase in temperature and decrease of precipitation caused by the anthropogenic climate change. Reduction by as much as 50% in the inflows to SWA dams since the mid-70s has been reported by the Climate Council of Australia (Climate Commission, 2011) which has negative implications for the water supply, agriculture and mining sectors, among others.

Considerable research has indicated that the changes in precipitation that SWA has faced since mid-70s are strongly related to three main features that bring stable atmospheric conditions and less storm formation to the area. First, the observed positive trend in the SAM (Haylock, 1993; IOCI, 2002; Cai and Cowan, 2006; Delworth and Zeng, 2014). Second, the intensification of the subtropical ridge (Timbal and Drosdowsky, 2013) and finally, the strength of the Southern Hemisphere jetstream, which means less potential energy is available in the atmosphere to lead to storm formation or intensification.

As reported in Chapter 2, particularly concerning are the SWA runoff projections which indicate continuous reductions till the end of the 21st century in the region. Future projections of water supply under climate change scenarios are a fundamental tool for efficient water resources planning. However, runoff projections are affected by uncertainties in the modelling process that limit their utility to decision makers. The main source of uncertainty in runoff projections are the Global Climate Models (GCMs) used to produce future climate projections; however, uncertainties are also associated with the observations of the climatic variables, the downscaling methodology used to fit the GCM output to the catchment scale and the hydrological model used to project runoff.

To date, the uncertainties on projected runoff have mainly been assessed through comparison of multi-model runs of future climate with little exploration of uncertainties inside the models due to different parameterisations. In this chapter the first aim of the thesis presented in Chapter 1 is addressed, which seeks to investigate the uncertainty response of projected runoff due to perturbed physics parameter variations within a GCM using a novel 2500 member ensemble from the HadCM3L model to quantify the within-GCM uncertainties on runoff projections in SWA.

Within the following sections the methodology used to pursue the objectives is first detailed. Then, in section 4.3 the main results which include the evaluation of the ability of the CPDN to simulate the SWA climate and the performance of the hydrological model Precipitation, Evaporation, Runoff model (PERM) to simulate the observed SWA hydrology are presented. The projections of runoff in the three SWA catchments for the period 2050-2080 compared to 1970 and 2000 and the within-GCM uncertainties due to perturbed physics are shown in section 4.3.3. In this section also, a comparison between the within-GCM uncertainties obtained using a perturbed physics ensemble of GCM and a statistical approach to account for within-GCM uncertainties is presented. Finally, an exploratory analysis of the differences in the mean and the spread of the runoff projections due to the different plausible values of 5 physics parameters is analysed in section 4.3.4. The summary and main conclusions of the chapter are shown in section 4.4.

4.2 Methodology

A three stages methodology often used in projections of water resources research has been followed, but here the novel approach of assessing the impact of perturbed physics on runoff projections was introduced, and compared against results obtained from previous methods developed in this field. First, the perturbed physics GCM data are evaluated over the region of study. Second, a bias correction to the GCM data is applied to scale the GCM data and match each observed catchment data, and finally a calibrated hydrological model is used to simulate runoff for the future using the GCM precipitation and temperature data.

4.2.1 Evaluation of CPDN output

The ability of the CPDN data to represent the climate of SWA was evaluated, first comparing the raw precipitation and temperature with observed data. Then, the relation between ENSO and SAM with precipitation and temperature using observed datasets and the CPDN data was assessed in order to evaluate whether the CPDN model output data represent the relation between the main drivers of climate in the region and the variables in study. Although SAM has been identified as one of the main large scale drivers influencing precipitation in SWA, ENSO 50

usually have some signatures in the region as well (Risbey et al., 2009). It’s worth noting that GCMs do not aim to reproduce the observed climate variables at any point of time of space, they rather seek to reproduce long term mean annual statistics that are widely representative of observed variables across continental or large range of locations. In this context, this investigation considers that despite ENSO does not have a large influence in SWA rainfall variability, an appropriate GCM should be able to represent the canonical driver of Australian rainfall variability properly. The CPDN data were released in ‘‘Giorgi regions’’, based on the regions analysed by Giorgi et al. (2001) that include both land and sea. Thus to compare land-based AWAP precipitation and air temperature at 1.5 m to CPDN data the application of correction factors is required. The procedure consisted of calculating a scale factor that represents the difference between the air temperature at a height of 1.5 m from HadCRUT3v (Jones, 1994; Brohan et al., 2006) over the same spatial extent as the CPDN Giorgi region (which include land and sea), and the HadCRUT3v temperature over just the land in this region. This factor was calculated and applied seasonally to the AWAP data. The scaled AWAP temperature data were then compared to the CPDN ensemble. Monthly gridded precipitation data from the Global Precipitation Climatology Project (Adler et al., 2003) were used to calculate similar scaling factors for precipitation to account for the difference in precipitation in the CPDN Giorgi region and precipitation over just land. The scaled AWAP precipitation was then compared to the CPDN data.

ENSO and SAM are the main drivers of precipitation and temperature variability in Australia (Hendon et al., 2007; Risbey et al., 2009; Arblaster et al., 2011). In SWA it is of particular interest for evaluating the relation between SAM and precipitation, because of the influence the SAM has had on reductions in rainfall (Raut et al., 2014). The Climateprediction.net dataset was evaluated to determine its skill in simulating large scale circulation patterns and their relationship with local variables over the area of study. Three correlation coefficients (Kendall, Spearman and Pearson) were used to analyse the relation between the variables in this region and circulation patterns over the observed period (1940–2000).

There are numerous indexes used in the literature to represent the SAM and its teleconnections, which might generate inconsistencies in the assessment of the impact of SAM on Australian hydroclimatic variability. Ho et al. (2012) performed a comparison of 7 different SAM indexes based on various definitions and sources of data, recommending the use of Marshall or Visbeck SAM indexes when assessing the impact of SAM on Australian variability. These indices are station-based indexes that use the Gong and Wang (1999) definition, recognized as the most suitable to represent the nature of the large scale feature. Adhering to Ho et al. (2012) results, the Southern Annular Mode is defined here as the difference in the normalized zonal average mean sea level pressure between 40°S and 65°S. Gong and Wang (1999) developed an

Index that measures SAM, the regional Antarctic Oscillation Index, detailed in the following equation:

̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ 푀푆퐿푃40°푆(푡) − 푀푆퐿푃40°푆(푡) 푀푆퐿푃65°푆(푡) − 푀푆퐿푃65°푆(푡) 퐴푂퐼푅푡 = − (1) 𝜎푀푆퐿푃 40°푆(푠푒푎푠표푛) 𝜎푀푆퐿푃 65°푆(푠푒푎푠표푛)

Where AOIRt is the Antarctic Oscillation Index for month t, MSLP40°S is the mean sea level pressure (MSLP) at 40°S averaged over all longitudes, MSLP65°S is the MSLP at 65°S ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ averaged over all longitudes, MSLP40°S(t) is the MSLP at 40°S averaged over all longitudes and ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ the whole period of time under analysis for month t, MSLP65°S(t) is the MSLP at 65°S averaged over all longitudes and the whole period of time under analysis for month t, σMSLP 40°S(season) is the standard deviation of MSLP at 40°S averaged over all longitudes for a particular season, and σMSLP 65°S(season) is the standard deviation of MSLP at 65°S averaged over all longitudes for a particular season.

An approximation to the AOIR index was used in this paper, in order to fit the Giorgi regions in which CPDN data were released. The approximation consists of replacing the

MSLP40°S(t) by the MSLP averaged over all SWA (-38°S to -28°S, 110°E- 125°E) and

MSLP65°S(t) by the MSLP over the Antarctic region (-90°S to -55°S, 0°e to 360° E).

Several indexes and methodologies have been used in assessments of the influence of ENSO in the Australian hydroclimatic variability, and the subjective choosing of any of them could lead to an underestimation of the real impact the driver has in the region where the assessment is being undertaken (Kiem and Franks, 2001). These indices are based on the difference between the sea level pressure at Papeete and Darwin (SOI; Troup (1965)), the sea surface temperature anomalies in different regions of the tropical Pacific (NIÑO3, NIÑO3.4, NIÑO4; Trenberth (1997)), and multiple climatic variables such as the sea level pressure, SST, surface zonal and meridional wind components, air temperature and cloudiness over the tropical Pacific, which define the Multivariate ENSO index (MEI; Wolter and Timlin (1993) and Wolter and Timlin (1998)). Without loss of generality, the Niño 3.4 index was used in this study, because it is a more direct measure of the Pacific Ocean temperature anomaly itself. The Niño 3.4 was calculated following the methodology described by Trenberth (1997). The first stage was computing the total area averaged SST from the Niño 3.4 region (5°N-5°S, 120°-170°W). Then the anomalies were calculated by subtracting the monthly mean SST for the period 1950 to 1979 from the data. Finally, the series were smoothed with a 3-months running mean. Anomalies higher/lower than 0.5°C/-0.5°C are defined as El Niño and La Niña respectively.

4.2.2 Bias Correction Methodology

GCM data have a different spatial scale (hundreds of kilometres) to the catchment scale and generally require downscaling and bias correction. In this thesis a Quantile-Quantile methodology (Themeßl et al., 2011), which is a type of statistical downscaling method, was used. The procedure is based on matching the empirical cumulative distribution functions of the GCM simulations to the empirical cumulative distribution function of the observations. This is a direct method because the predictor and the predictand are the same variable (temperature or precipitation) and has the advantage of being parameter-free because the empirical cumulative distribution is matched for each variable. Themeßl et al. (2011) described the methodology of correction using the following equations for the calibration period (in this case the period in which observed runoff data are available for each catchment), which in this project will be applied on a monthly basis and three times, one for each catchment:

푐표푟 푟푎푤 푌푡,푐푎푡푐ℎ푚푒푛푡 = 푋푡,푐푎푡푐ℎ푚푒푛푡 + 퐶퐹푡,푐푎푡푐ℎ푚푒푛푡 (2)

표푏푠,푐푎푙−1 퐶퐹푡,푐푎푡푐ℎ푚푒푛푡 = 푒푐푑푓푚표푦,푐푎푡푐ℎ푚푒푛푡(푃푡,푐푎푡푐ℎ푚푒푛푡) (3) 푚표푑,푐푎푙−1 − 푒푐푑푓푚표푦,푐푎푡푐ℎ푚푒푛푡(푃푡,푐푎푡푐ℎ푚푒푛푡)

푚표푑,푐푎푙 푟푎푤 푃푡,푐푎푡푐ℎ푚푒푛푡 = 푒푐푑푓푚표푦,푐푎푡푐ℎ푚푒푛푡(푋푡,푐푎푡푐ℎ푚푒푛푡) (4)

푐표푟 Where 푌푡,푐푎푡푐ℎ푚푒푛푡 is the bias corrected (cor) GCM variable over the catchment under 푟푎푤 study for month t, 푋푡,푐푎푡푐ℎ푚푒푛푡 is the raw GCM variable over the catchment under study for month t, 퐶퐹푡,푐푎푡푐ℎ푚푒푛푡 is the correction factor calculated during the calibration period for month t, 푚표푑,푐푎푙 푒푐푑푓푚표푦,푐푎푡푐ℎ푚푒푛푡 is the empirical cumulative distribution function for the GCM variable in a 표푏푠,푐푎푙 particular month of the year (moy) for the calibration period (cal), and 푒푐푑푓푚표푦,푐푎푡푐ℎ푚푒푛푡 is the empirical cumulative distribution function for the observed variable in a particular moy for the calibration period.

The difference between the inverse ecdf (ecdf-1) of the simulation and the observation over the calibration period for every probability represents the bias correction. An example is presented in Figure 4.1, in which each grey line represents the inverse ecdf for one of the 2500 CPDN simulations of precipitation for December (1940-2000) and the red line represents the inverse ecdf of gridded AWAP precipitation for December (1940-2000). The red arrow represents the correction for one simulation, which is the difference between the simulated and observed inverse ecdfs for that particular probability. For future projections, the same correction factors are applied to every simulation, matching the simulated values of the variable for the projection with the calibration period. For new extremes in the modelled period (values that have not been observed during the calibration period), higher or lower than observations, the correction factor 53

of the highest or the lowest observed value is applied respectively. Note that the bias correction methodology applied in the example presented in Figure 4.1 is the 1940-2000 period, this investigation avoids using the first 20 years of AWAP data (1920-1940) due to inconsistencies and noise observed in that period. However, the bias correction analysis performed in this thesis used the period in which observed runoff data is available in every catchment.

FIGURE 4.1 BIAS CORRECTION METHODOLOGY FOR PRECIPITATION IN DONNELLY RIVER AT STRICKLAND DURING DECEMBER

4.2.3 Precipitation Evaporation Runoff Model Description

The hydrological model used in this work is the Precipitation Evaporation Runoff model (PERM; Peel et al. (2015)). A schematic diagram of the model structure is presented in Figure 4.2. This is a lumped conceptual monthly model that uses as input monthly precipitation and temperature data observed or modelled over the catchment. Because the CPDN data has been released in a monthly time-step, a monthly hydrologic model was used in this assessment. It’s important to mention that by using a monthly rather than daily model, some of the parameters that account for the different water balance components (such as the soil moisture capacity) might be underrepresented. To minimize the limitations, the entire observed runoff record is used in the calibration of PERM, and the parameters are optimized to reduce and objective function that incorporates penalties which seek to improve the representation of the mean annual runoff and the observed annual coefficient of variation of the catchments. More information regarding the model parameter optimization process is provided below.

PERM has five model parameters that require calibration: the rate of snowmelt and potential evapotranspiration (ETrate, mm/°C/month); the proportion of snowmelt volume to runoff (Melt); the soil moisture storage capacity (Smax, mm); the baseflow linear recession parameter (K); and the interception storage capacity (Imax, mm). The model considers three main stores: the vegetation interception store (IC), the soil moisture store (SMS) and the snow store (ACCUM). When temperature is >0°C then all precipitation enters the interception store. Once the maximum capacity (Imax) is reached any remaining precipitation becomes throughfall (TFall) to the soil moisture store. Snow only accumulates when temperature is ≤ 0 and the snow begins to melt once temperature rises >0°C. Melting snow either becomes runoff (SnowF) or infiltrates (SMI) into the soil moisture store. Runoff is calculated through a volume balance and considers snow flow, partial area flow (PAreaF), soil moisture excess flow (SMF), once the soil moisture store reaches it maximum value (Smax), and base flow (BF). Evaporation from the interception store (AETINT) and soil moisture store (AETSOIL) are calculated as a linear function of temperature and water availability. In this analysis, snow does not occur in the three catchments being modelled, which effectively reduces PERM to a four parameter model as the Melt parameter becomes redundant due to ACCUM being zero. An automatic pattern search optimization method was used to calibrate the five model parameters (see Peel et al., 2015 for details). Ten different parameter sets were used as starting points to increase the likelihood of finding the global optimum of parameter values. The calibration sought to minimise the objective function defined as the sum of squared differences between the estimated and observed annual runoff. Although the model is run on a monthly time step, it will be used to simulate annual runoff for future climates. Therefore, penalties were applied to the objective function (OBJ) in order to ensure that the calibrated model reproduces summary statistics of observed annual runoff.

Precip AETINT

Imax TFall IC Interception store SnowF Snow accumulation Accum

SMI AETSOIL PAreaF

SMF Smax SMS BF Soil moisture store

Runoff

FIGURE 4.2 PERM MODEL SCHEME Peel et al. (2015) summaries these penalties as:

If the estimated and observed mean annual runoff differ by;

• More than 5% then OBJ = OBJ × 5 • More than 10% then OBJ = OBJ × 25 • More than 20% then OBJ = OBJ × 125

If the estimated and observed annual coefficient of variation differ by;

• More than 5% then OBJ = OBJ × 5 • More than 10% then OBJ = OBJ × 25 • More than 20% then OBJ = OBJ × 125

The evaluation method used for the model is a K-Fold cross-validation (Efron and Tibshirani, 1993), with K = 3. The entire time series is divided in three groups, two of them are used to calibrate and the third one to evaluate the model performance. This process is repeated 3 times until all thirds have been used to evaluate model performance. The metrics used to evaluate model performance are the Nash & Sutcliffe Efficiency value (Nash and Sutcliffe, 1970), which compares observed and modelled annual runoff, and the annual R2 (square of the correlation coefficient) between observed runoff and modelled runoff. Details of the calibration methodology can be found in Peel et al. (2015).

4.3 Results

Modelled and observed temperature, precipitation and runoff for the three catchments under study are presented in this section. First the ability of un-bias corrected CPDN data to simulate the climate of the study region during the observed period was evaluated. Then, how uncertainties in precipitation and temperature from CPDN translate into uncertainties in runoff projections for the next decades was investigated. Also, the perturbed physics results are compared against an approximation based on stochastic generation of data. Finally, the uncertainties in runoff projections due to specific physical GCM parameters is assessed.

4.3.1 Evaluation of CPDN data

Seasonal and annual raw (not bias corrected) precipitation and temperature simulated by the CPDN project were compared to scaled AWAP data in order to assess how these GCM runs simulate SWA climate. The annual analyses correspond to the aggregated variables between April and March (water year). The scale factor was introduced to make the observed data, which has been measured over land only, consistent with the raw CPDN data that includes both land and ocean. After incorporating the scale factor, a good agreement between observed and modelled data was observed for annual and seasonal climate variables. The annual values are presented in Figure 4.3 for precipitation and Figure 4.5 for temperature. In these figures, all the CPDN simulations of annual precipitation and temperature for the period between 1940 and 2080 were plotted in grey, including in blue the median of the simulations and the 5th and 95th percentiles. The red line shows the gridded AWAP annual precipitation and temperature between 1940 and 2000.

In Figs. 4.4 and 4.6, histograms of annual medians and standard deviations of precipitation and temperature respectively simulated by CPDN between 1940 and 2000 (the observed period) and the annual median and standard deviation obtained from AWAP data are shown. In terms of temperature the annual median of the CPDN simulations for the observed period (1940–2000) underestimates by 0.23°C the observed annual median of temperature during the same period. Substantial differences are observed in the annual standard deviation over the same period, with CPDN underestimating annual standard deviation of temperature by 0.14°C relative to the observed value (see the results presented in Table 4.1). In terms of precipitation, the results indicate that the annual median of CPDN simulations between 1940 and 2000 differs by 6.2% from the observed data, and the CPDN annual standard deviation underestimates the observed value by 45%, (Table 4.2).

TABLE 4.1 DIFFERENCES IN MEDIAN AND STANDARD DEVIATION OF AWAP TEMPERATURE AND CPDN TEMPERATURE BETWEEN 1940 AND 2000 Median Median Dif. Standard dev. Standard dev. Dif. AWAP data CPDN Median AWAP data CPDN data Standard (°C) (°C) (°C) (°C) (°C) dev. (°C) Raw temperature 16.09 15.86 -0.23 0.45 0.31 -0.14 over SWA Bias corrected temperature over 15.20 15.18 -0.02 0.42 0.43 0.01 Donnelly River at Strickland

TABLE 4.2 DIFFERENCES IN MEDIAN AND STANDARD DEVIATION OF AWAP PRECIPITATION AND CPDN PRECIPITATION BETWEEN 1940 AND 2000 Median Median Dif. Standard dev. Standard dev. Dif. AWAP data CPDN Median AWAP data CPDN data Standard (mm) (mm) (%) (mm) (mm) dev. (%) Raw precipitation 429.24 402.69 6.19 100.33 54.58 45.6 over SWA Bias corrected precipitation over 1003.90 995.93 0.79 150.46 156.02 3.7 Donnelly River at Strickland

In Figure 4.3 the median of the CPDN simulations of annual precipitation is similar to the median value of the scaled modelled based on observed (AWAP) annual precipitation, and the median of the simulations reproduce the reductions in observed precipitation since mid-1960s (Allan and Haylock, 1993; Ansell et al., 2000; Cai and Cowan, 2006). Results presented in Figure 4.5 indicate that the median of the CPDN simulations of annual temperature approximately fits the scaled observed annual temperature, and that the median of the simulations is consistent with the positive trend in scaled observed temperature. The results indicate that CPDN is better at reproducing annual temperature over SWA than annual precipitation, and that the medians of the climate variables are better simulated than the variability, measured by the standard deviation. Prior to the hydrologic modelling these differences were resolved using the bias correction methodology described in Section 4.2.

FIGURE 4.3 COMPARISON OF RAW CPDN ANNUAL PRECIPITATION AND SCALED AWAP ANNUAL PRECIPITATION FOR SWA. GREY LINES REPRESENT THE 2500 SIMULATIONS OF PRECIPITATION FROM CPDN. 95TH, 5TH PERCENTILES AND MEDIAN OF THE SIMULATIONS ARE PRESENTED AS BLUE LINES AND AWAP ANNUAL PRECIPITATION IS PLOTTED WITH A RED LINE

FIGURE 4.4 HISTOGRAM OF MEDIAN OF SIMULATED (CPDN) ANNUAL PRECIPITATION (MM) FOR THE PERIOD BETWEEN 1940 AND 2000. COMPARISON OF MEDIAN OF MODELLED BASED ON OBSERVED DATA (AWAP), AS INDICATED WITH THE OPEN STAR, AND THE MEDIAN OF ALL THE SIMULATIONS DURING THE SAME PERIOD, INDICATED WITH AN ASTERISK.

FIGURE 4.5 COMPARISON OF RAW CPDN ANNUAL TEMPERATURE AND SCALED AWAP ANNUAL TEMPERATURE FOR SWA. GREY LINES REPRESENT THE 2500 SIMULATIONS OF TEMPERATURE FROM CPDN. 95TH, 5TH PERCENTILES AND MEDIAN OF THE SIMULATIONS ARE PRESENTED AS BLUE LINES AND AWAP ANNUAL TEMPERATURE IS PLOTTED WITH A RED LINE

FIGURE 4.6 HISTOGRAM OF THE MEDIAN OF SIMULATED (CPDN) ANNUAL TEMPERATURE (°C) FOR THE PERIOD BETWEEN 1940 AND 2000. COMPARISON OF MEDIAN OF MODELLED BASED ON OBSERVED DATA (AWAP) AS INDICATED WITH THE OPEN STAR AND THE MEDIAN OF ALL THE SIMULATIONS DURING THE SAME PERIOD, INDICATED BY AN ASTERISK In order to evaluate whether the CPDN data represent the relation between the main drivers of climate in the region, the relation between ENSO and SAM with precipitation and temperature using gridded station-based data and CPDN data was assessed. ENSO is one of the main drivers of rainfall in Australia (Allan, 1988; Nicholls et al., 1997; Wang and Hendon, 2007; Risbey et al., 2009). Nicholls et al. (1997) indicated that the relation between ENSO and rainfall over Australia shows a considerable multi-decadal variability, driving wet and dry periods in its

different phases (El Niño and La Niña). However, the main area of influence of ENSO in terms of precipitation is north and eastern Australia (Nicholls and Lavery, 1992; Risbey et al., 2009), with no significant influence over SWA, since SAM is the main driver of precipitation in that zone (Raut et al., 2014). ENSO inter-decadal variability has been also found to be associated with variations in surface temperature in Australia (Power et al., 1999a; Power et al., 1999b). Power et al. (1999a) found a correlation of 0.1 between annual temperature and SOI (Southern Oscillation Index) when the Inter-decadal Pacific Oscillation (IPO) is over the threshold of 0.5. Numerous studies have indicated the IPO, described as exhibiting ‘‘ENSO-like’’ decadal behaviour (Power and Colman, 2006; Power et al., 2006), modulates the influence and intensity of ENSO events (Power et al., 1999a; Vance et al., 2015; Westra et al., 2015; Ashcroft et al., 2016).

Considering the importance of SAM in SWA rainfall variability, and of ENSO over most of Australian temperature variability, the correlation between annual and winter precipitation and SAM, and annual temperature and ENSO were explored. The median of the correlation coefficients between SAM and annual and winter precipitation obtained using three different methodologies (Spearman, Kendall and Pearson correlation) for all the CPDN simulations are presented in Table 4.3. These correlations are statistically significant at the 95% level. The mean correlation between annual SAM and annual precipitation, considering all the statistical significant simulations and three methodologies, is -0.3, while for winter SAM and winter precipitation it is -0.52.

TABLE 4.3 CORRELATION COEFFICIENTS BETWEEN SOUTHERN ANNULAR MODE, ENSO, PRECIPITATION AND TEMPERATURE OBSERVED AND SIMULATED BY CPDN Observed Observed Observed CPDN annual CPDN winter CPDN annual annual winter annual precipitation - precipitation - temperature – precipitation – precipitation temperature - SAM SAM ENSO SAM – SAM ENSO Spearman -0.33 -0.57 0.30 -0.30 -0.46 0.17 Kendall -0.23 -0.40 0.21 -0.19 -0.31 0.12 Pearson -0.34 -0.60 0.27 -0.27 -0.41 0.18 The correlations between the SAM index as defined by Nan and Li (2003) and annual and winter gridded based on observed precipitation (AWAP) considering Pearson, Kendall and Spearman correlation coefficients were also calculated. The correlation between observed winter precipitation and winter SAM is statistically significant at -0.39. These results are presented in Table 4.3. The influence of SAM on SWA precipitation is greatest in the winter season as this is the period in which precipitation is mostly concentrated in this region. The medians of statistical significant correlation coefficients, for the three methodologies, between annual ENSO and annual temperature in SWA are presented in Table 4.3.

The mean of the three medians is 0.26, which is larger than the correlation of 0.16 (not statistically significant) obtained from the observed NIÑO 3.4 Index and AWAP data. These results indicate that CPDN reproduces the observed negative correlation between annual precipitation, winter precipitation and SAM, which is the main driver of the changes in precipitation during the last decades. The conclusion supported in these analyses is that CPDN represents appropriately the climate of SWA, and is suitable for studying the uncertainties in future projections in this region.

4.3.2 Runoff Modelling

The PERM model was calibrated for each catchment using gridded based on observed area-weighted temperature and rainfall from the AWAP data as input to PERM for the period in which observed runoff data are available.

For the Donnelly River at Strickland, 32 years of observed runoff data in the period 1961– 1992 were used to calibrate the model. Twenty-eight years of observed runoff were used for modelling the Helena River at Ngangaguringuring, which corresponds to the period between 1973 and 2000 and finally 40 years of observed runoff were available in the Denmark River at Kompup, which corresponds to the period between 1961 and 2000. PERM is run on a monthly time step, however the model performance during calibration and evaluation on an annual basis was assessed. In the following analyses the majority of runoff results presented are for annual data with some seasonal data. Therefore, the evaluation of PERM’s performance at annual intervals is consistent with the later analysis. The calibrated parameters and the performance of the model for every catchment is given in Table 4.4, the coefficient of determination (R2) and the annual Nash & Sutcliffe coefficient of efficiency (N&SE) represent the evaluation of the model skill in simulating runoff in the catchments, where a value of 1 in each of them means a perfect match. The model was evaluated using a K-Fold cross-validation with K = 3 (Efron and Tibshirani, 1993), where the evaluation N&SE was calculated from the 2 thirds not used to calibrate the model. Results of the model performance during calibration and evaluation are presented in Tables 4.4 and 4.5 respectively. A good model performance during calibration and evaluation for the Donnelly River at Strickland and the Denmark River at Kompup was found, but poorer performance for the Helena River at Ngangaguringuring. Despite this poorer performance of PERM in representing the runoff generation in Helena catchment, this was included in the analysis because it is an important catchment, one of the tributaries that supply water to Perth and because it is a dry catchment, thus more susceptible to climate change impacts.

The calibrated model was then run 2500 times using the bias corrected precipitation and temperature data from CPDN to simulate runoff from 2000 to 2080 under the A1B scenario in each catchment in SWA. When performing the bias correction methodology, it was found that for

precipitation the ecdf of monthly future simulations of winter CPDN is drier or is shifted to less precipitation than the ecdf of precipitation during the observed period (Figure 4.7). The main difference lies in the extreme lower quantile, where the ecdf of precipitation of future simulations presents a shift to lower precipitation than the observed and in some cases to higher extremes as well. For temperature the ecdf of monthly temperature shows a shift to higher temperatures with higher hot extremes. The CPDN precipitation and temperature monthly ecdfs for the observed and future (2000–2080) periods are available in Figure 4.8.

FIGURE 4.7 5TH AND 95TH PERCENTILES OF SEASONAL EMPIRICAL CUMULATIVE DISTRIBUTION FUNCTIONS FOR PRECIPITATION SIMULATED USING CPDN. BLUE CURVES REPRESENT THE OBSERVED PERIOD AND RED CURVES THE PROJECTION FOR THE PERIOD BETWEEN 2020-2080.

FIGURE 4.8 5TH AND 95TH PERCENTILES OF SEASONAL EMPIRICAL CUMULATIVE DISTRIBUTION FUNCTIONS FOR TEMPERATURE SIMULATED USING CPDN. BLUE CURVES REPRESENT THE OBSERVED PERIOD AND RED CURVES THE PROJECTION FOR THE PERIOD BETWEEN 2020-2080

After introducing the bias correction it was found that the difference between the medians of the simulated temperatures (CPDN) and the gridded based on observed annual temperature (AWAP) for the period between 1961 and 1992 (which corresponds to the period in which observed runoff data are available in Donnelly catchment) is 0.02°C. The difference in the standard deviation of the CPDN simulations of annual temperature and gridded based on observed annual temperature (AWAP) is 0.01 °C, much lower than the 0.14 °C difference for the raw series. For precipitation the differences for the same period are 0.79% and 3.7% for median and standard deviation respectively, improving considerably the 6.2% and 45% obtained from raw data. See this information in Tables 4.1 and 4.2.

TABLE 4.4 PERM CALIBRATION RESULTS Catchment Parameters Evaluation annual modelled and observed runoff

Smax Etrate K Melt Imax R2 N&SE Helena River at 200.00 60.00 0.41 0.02 100.00 0.62 0.59 Ngangaguringuring Donnelly River at 767.81 7.56 0.00 0.27 23.88 0.94 0.94 Strickland Denmark River at 936.56 12.22 0.27 0.83 92.75 0.78 0.78 Kompup

TABLE 4.5 PERM EVALUATION RESULTS Obs. Mod. Dif. obs. Dif. obs. and Annual Catchment MAR MAR and Mod. Obs. Cv Mod. Cv Mod. Cv N&SE (mm) (mm) MAR (%) (%)

Helena River at 0.60 6.41 6.10 4.93 0.89 0.88 0.84 Ngangaguringuring Donnelly River at 0.92 162.72 162.98 0.16 0.43 0.41 5.00 Strickland Denmark River at 0.70 58.93 57.92 1.70 0.58 0.56 4.83 Kompup Bias corrected annual precipitation and temperature CPDN simulations were compared against gridded based on observed data (the AWAP weighted average) for each catchment, with the results for Donnelly catchment presented in Figure 4.9 (precipitation) and Figure 4.10 (temperature). The bias corrected methodology was applied to the complete period of record, but the correction was developed only over the period of observed runoff data available in each catchment, which in Figs. 4.9 and 4.10 corresponds to the period between 1961 and 1992. In these figures, grey lines correspond to the 2500 bias corrected simulations from CPDN, blue lines are the median, 5 and 95 percentiles of the data and the red line the AWAP modelled data. In both cases a good graphical agreement between the observed data (red line) and the median of the CPDN data (blue line) was obtained for the period in which the bias correction was developed (period in which runoff data were available).

FIGURE 4.9 BIAS CORRECTED ANNUAL PRECIPITATION OVER DONNELLY RIVER AT STRICKLAND. GREY LINES REPRESENT THE 2500 SIMULATIONS OF BIAS CORRECTED PRECIPITATION FROM CPDN PROJECT. 95TH, 5TH PERCENTILES AND MEDIAN OF THE SIMULATIONS ARE PRESENTED IN BLUE LINES AND AWAP ANNUAL PRECIPITATION IS PLOTTED IN RED LINE

FIGURE 4.10 BIAS CORRECTED ANNUAL TEMPERATURE OVER DONNELLY RIVER AT STRICKLAND. GREY LINES REPRESENT THE 2500 SIMULATIONS OF BIAS CORRECTED TEMPERATURE FROM CPDN PROJECT. 95TH, 5TH PERCENTILES AND MEDIAN OF THE SIMULATIONS ARE PRESENTED IN BLUE LINES AND AWAP ANNUAL TEMPERATURE IS PLOTTED IN RED LINE According to Figure 4.10, positive trends for temperature in Donnelly River at Strickland have been seen during the observed period (1940–2000) and faster increases are shown in the later period (2000–2080), which can be noticed from the blue lines that represent the median and the 5 and 95 percentiles. This result was also observed in Helena and Denmark catchments for annual and for seasonal results. Conversely, negative trends in annual precipitation are shown (Figure 4.9) during the observed period (1940–2000) and faster decreases are expected for the later period (2000–2080) for Donnelly catchment, which can be seen from the blue lines that

represent the median and percentiles of the simulations. These results were also observed in Helena and Denmark rivers (not shown). The decrease in precipitation is concentrated in the winter season, with changes in summer precipitation being almost negligible (not shown), which is in agreement with the literature (Nicholls and Lavery, 1992; Alexander et al., 2007; Charles et al., 2010a).

The annual runoff simulations using the CPDN climate data are shown in Figure 4.11. Similar results were obtained for the other two catchments so only the figures for Donnelly catchment are presented. From Figure 4.11, negative trends in Donnelly at Strickland runoff are observed in the projections for the period 2000–2080, concordant with the trends in precipitation, the main driver of runoff in the area. This result was also observed in the Helena River at Ngangaguringuring and Denmark River at Kompup. The projections of annual precipitation, annual runoff, seasonal precipitation and seasonal runoff were computed for the period 2050– 2080 and compared to 1970–2000 for all catchments, then the histograms including all the simulations were computed. The histograms for seasonal differences in the Donnelly catchment are presented in Figure 4.12. Since similar results were obtained for the other two catchments, just results for Donnelly River at Strickland are shown. Decreases in precipitation with a median of around 20% are projected for precipitation in winter, spring and autumn, which leads to decreases in runoff of around 50% for winter and autumn. The response of spring runoff to reduced precipitation is larger than the other seasons, which might be driven in part by reduced catchment wetness at the beginning of spring following reduced precipitation in autumn and winter and for the increases in temperature and potential evapotranspiration during these months. Summer precipitation shows no real change but has a large spread of results, which is explained due to the summer rainfall in these catchments being negligible. A large spread in the projections of runoff is also observed, with a median decrease of around 50%, which is caused by the combination of reduced precipitation in most seasons and higher temperatures in all seasons.

FIGURE 4.11 SIMULATED ANNUAL RUNOFF OVER DONNELLY RIVER AT STRICKLAND. GREY LINES REPRESENT THE 2500 SIMULATIONS OF RUNOFF USING PERM MODEL RUN WITH BIAS CORRECTED PRECIPITATION AND TEMPERATURE FROM CPDN. 95TH, 5TH PERCENTILES AND MEDIAN OF THE SIMULATIONS ARE PRESENTED IN BLUE LINES AND OBSERVED RUNOFF IS PLOTTED IN RED LINE

FIGURE 4.12 HISTOGRAMS OF SEASONAL CHANGES IN PRECIPITATION AND RUNOFF OVER DONNELLY RIVER AT STRICKLAND

Histograms of annual runoff and rainfall difference for the three catchments are presented in Figure 4.13. In all catchments a reduction in annual precipitation with a median of around 20% is expected, which leads a reduction in runoff of more than double (~50%). The Helena catchment, which is the driest catchment among the three, is the most sensitive to changes in precipitation, with a median reduction in runoff of around 65%, driven by a reduction in rainfall with a median of around 20%. The results indicate that the drier the catchment, the more sensitive the response in runoff to changes in precipitation and the larger the spread of the projections, which is consistent with the understanding of runoff sensitivity to changes in precipitation increasing as the humidity ratio decreases (Dooge, 1992; Dooge et al., 1999; Sankarasubramanian et al., 2001).

FIGURE 4.13 HISTOGRAMS OF ANNUAL CHANGES IN PRECIPITATION AND RUNOFF IN ALL OF THE CATCHMENTS

4.3.3 Comparison of within GCM uncertainties from stochastic generation of data

Here a comparison of results from a GCM with perturbed physics to that from the stochastic generation of GCM data is presented in order to identify any differences in the quantification of uncertainties in precipitation and runoff modelling for the Donnelly catchment. The GCM perturbed physics results from the previous section are shown in box-plot form (Figure 4.14) for annual precipitation and runoff over the Donnelly catchment. Stochastic generation results are drawn from Peel et al. (2015), where monthly precipitation and temperature data from 5 GCMs, selected for their good performance in simulating observed climate (McMahon et al., 2015), were stochastically replicated 100 times each. The stochastic methodology of Peel et al. (2015) is to de-trend the GCM data, replicate the de-trended data (both the signal and the noise around the trend) and add the trend to the stochastic data to form the stochastic replicate of the GCM run. Thus each stochastic replicate for a given GCM has the same trend but different stochastic data around the trend. Peel et al. (2015) indicated that they expected their stochastic method to under-estimate the true within-GCM uncertainty due to not replicating the trend. Results in Figure 4.14 indicate that the range of uncertainty using perturbed physics is larger than that using stochastic generation of data. The boxplots present the range of different simulations of rainfall and runoff, where the 25th and 75th percentiles are represented in the box, and the 97.5th and 2.5th percentiles in the lines. Considering the differences between the 75th and the 25th percentiles as a percentage of the median, a 12% range in precipitation for the 2035–2064 period is obtained using stochastic generation, compared to a 22% range using CPDN data for the same period. These results are amplified when analysing runoff: a range of 28% of the median runoff modelled for the period between 2035 and 2064 using stochastic generation is considerably smaller than the 57% range for the same variable and the same period using perturbed physics GCM data.

FIGURE 4.14 BOXPLOT OF UNCERTAINTIES IN PRECIPITATION AND RUNOFF USING CPDN DATA AND STOCHASTIC GENERATION, IN THE DONNELLY RIVER AT STRICKLAND FOR THE PERIOD 2035–2064

4.3.4 Comparison of within-GCM uncertainties from GCM perturbed physics

As the main interest is quantifying uncertainties in water projections due to within-GCM uncertainties the analyses have mostly focused on the whole range of plausible projections rather than the spread of results caused by a single parameter. However, it was also explored single parameter uncertainties relative to the entire CPDN ensemble for a few parameters that were identified as important for the simulation of precipitation.

Regarding uncertainties in input forcing in the GCM, Rowlands et al. (2012) presented an analysis of the main drivers of uncertainties in the CPDN simulations. They performed a linear variance decomposition of all the CPDN ensemble grouping the physical parameters into three groups. Solar, volcanic and sulphur cycle parameters were grouped as forcing parameters, with climate sensitivity used as a proxy for atmospheric physics and vertical diffusivity as representative of ocean physics. Their results showed that the atmospheric physics parameterisations accounted for most of the uncertainty among the ensemble (50% during the last 20 years), followed by uncertainties in forcing and in last position the ocean parameters. This was explained as due to the longer time-scale responses of oceans in the climate system.

As an exploratory analysis the histograms of the differences in runoff for the periods between 2050–2080 and 1970–2000 were calculated for the different plausible values of five parameters: rhcrit (critical relative humidity for cloud formation), cw_land (threshold cloud water content for rain over land), eacf (large-scale cloud coverage when the specific humidity in the grid box is equal to the saturation value), ct (accretion constant – time constant for conversion of cloud droplets to rain) and vf1 (cloud ice fall speed). All parameters were selected for their importance in the GCM parameterization of rainfall, which is the main driver of runoff in SWA. 71

In Figure 4.15, histograms of change in mean annual runoff obtained by selecting different values of the parameter rhcrit are presented, which show the spread of runoff change and the median change relative to the whole ensemble of CPDN. Figures for the other parameters are contained in the Appendix 18 of this thesis. The histograms in Figure 4.15 show the whole ensemble of simulations in blue (2500) and the simulations using each one of the different rhcrit values in grey. The median of the whole ensemble has also been plotted (red line) and the median of the group of simulations obtained using an rhcrit value (red dotted line). The three different values of rhcrit produce results with differences in median of 9%. The median of the reductions in runoff for the lowest rhcrit is -46%, and for the highest rhcrit is -55%. The spread of simulations moves to higher reductions in runoff when rhcrit is larger. This is due to the importance of rhcrit for cloud formation, as rhcrit increases cloud formation becomes harder, which results in less precipitation and less runoff. Among all the parameters analysed, rhcrit produced the largest impact on runoff reductions, showing how relevant parameterisations of cloud formation are for projections of climate and runoff.

FIGURE 4.15 HISTOGRAMS OF ANNUAL CHANGES IN RUNOFF CONSIDERING ALL THE SIMULATIONS OF CPDN AND THE GROUPS OF SIMULATIONS WITH DIFFERENT PERTURBATIONS OF THE PARAMETER RHCRIT. RED LINE REPRESENTS THE MEDIAN OF THE WHOLE ENSEMBLE AND DOTTED RED LINE THE MEDIAN OF THE SIMULATIONS FOR A PARTICULAR PERTURBATION OF RHCRIT.

4.4 Summary

In this chapter the analysis of how the uncertainties in the parameters specified within- GCM physics translates into uncertainties in runoff projections has been studied to address one of the gaps detected and presented in the Chapter 2 of this thesis. This gap is the lack of research that investigates the relationship between runoff projections and the impact of the within-GCM uncertainties on these projections in SWA catchments, which is useful to compare with studies that assess only multi-model uncertainties and is a critical issue under the current water shortages scenario the region is facing.

Summarizing, the novel results presented in Chapter 4 of this thesis show that:

- Uncertainties in projections of runoff, precipitation and temperature for the second half of this century using perturbed physics in a single GCM are very large. - Uncertainties in GCM parameterisations in precipitation and temperature translate into even larger uncertainties in runoff projections (approximately doubled), with a hydrological sensitivity of around 2.5, which is consistent with the values calculated by Chiew (2006) and Jones et al. (2006). It is important to emphasize that these uncertainties correspond just to uncertainties within a single GCM, without considering uncertainties in emissions scenario, downscaling/bias correction or

hydrological modelling. However, due to the large number of perturbations and initial conditions explored in the CPDN data, it is possible that these simulations match other GCMs, covering between-GCM uncertainties as well. - Plausible projections in annual precipitation indicate a reduction between the years 2050–2080 compared to 1970–2000 that range between 0% and 40% for the Donnelly and Denmark Rivers and between 0% and 50% for the drier Helena catchment. - The range of projected decrement in annual precipitation drives a runoff decrement between 10% and 80% over the same period, and this reduction is also larger in the Helena catchment (0–90%). A reduction of around 22% in annual and winter precipitation leads to a decrease of 50% in runoff which is a larger reduction than that found using stochastic generation of data, as an approach for assessing within-GCM uncertainties. - Winter reductions in precipitation and runoff for every catchment are very similar to the annual changes, mainly because around 80% of annual precipitation falls in winter months in these catchments. - Summer and spring runoffs are most sensitive to changes in precipitation due to increases in temperature during these months and the drier conditions in the catchment resulting from the reduced spring and summer rainfall, which increase sensitivity to changes in precipitation, as reported by Chiew (2006). - The driest catchments are more sensitive to changes in precipitation: for example, simulations in the Helena River show that the change in modelled runoff is more than double the change in precipitation. This is concordant with the results presented by Dooge et al. (1999) and Sankarasubramanian et al. (2001) which indicated that the drier the catchment the more sensitive the runoff to changes in precipitation. - The perturbed physics GCM results showed that uncertainties, quantified as the range between 25th and the 75th percentiles in the histogram of plausible projections using CPDN data, are approximately double those from stochastic generation of GCM data. This confirms that the stochastic approximation of within-GCM uncertainty of Peel et al. (2015) underestimates true within-GCM uncertainty as expected. In both cases the range of projections for runoff is larger, more than double the range in precipitation projections, indicating the sensitivity of runoff to precipitation. - The difference between the 10th and 90th percentiles of modelled annual runoff for the period between 2046–2065, compared to 1961–2000, using CPDN data averaged over the three catchments is around 78%, which is much larger than the 50% obtained by Teng et al. (2012a) using the GCM ensembles runs available in CMIP3 and CMIP5 for Australian catchments.

- Considering the analysis of uncertainties due to the variation of single parameters in the GCM physics parameterisations, it was found that of the five atmospheric parameters investigated the parameter representing critical relative humidity for cloud formation produced the largest changes in terms of precipitation and runoff. Rhcrit produced reductions in the median of runoff projections that fluctuated between -46% and -55%, indicating the relevance of this parameter to cloud formation processes in the GCM. - The results from the perturbed physics approach here presented indicate that current studies of future runoff under climate change, which solely consider between-GCM variability, tend to underestimate the uncertainty in runoff.

Finally, the methodology presented in Chapter 4, to explore the within-GCM uncertainties in runoff projections, can be extended to other catchments located in Mediterranean like climate regions where the CPDN data or a comparable ensemble with perturbed physics is available. This methodology is suitable to be used in assessments that explore the between-GCM uncertainties as well. In fact, the following chapter addresses the between-GCM uncertainties in SWA and in CC catchments as well.

CHAPTER 5. EXPLORING UNCERTAINTIES ON RUNOFF PROJECTIONS IN MEDITERRANEAN LIKE CATCHMENTS

5.1 Introduction

This chapter addresses the second and third aim of this thesis presented in Chapter 2. It is a reproduction of the manuscript accepted to be published in the Journal of Southern Hemisphere Earth Systems Science, titled Analysis of within and between-GCM uncertainties of runoff projections in Mediterranean-like catchments. The manuscript has been reformatted to align with the thesis formatting. The section 1 “Introduction” and the section 2 “Data” have not been included as that material is available in Chapters 2 and 3 respectively, which have been removed to avoid repetition.

In this chapter, the investigation of both the between-GCM uncertainties and the within- GCM uncertainties in runoff projections of Mediterranean-like climate catchments of the Southern Hemisphere, in particular catchments located in SWA and CC, are presented. This analysis builds upon the need for detailed analysis of streamflow projections and the inherent uncertainties in the modelling process in these regions.

As indicated in Chapter 1 of the thesis, the SWA and CC regions are both mid-latitude Mediterranean climate regions characterised by a warm and dry summer and where most of precipitation (~80% of the total annual volume) falls during autumn-winter months. According to the literature, the observed dry trends both regions have experienced since mid-70s are projected to continue for the next decades (Samaniego et al., 2009; Prosser, 2011; Teng et al., 2012a; Demaria et al., 2013; Prudhomme et al., 2014) which makes crucial the analysis of water projections and the uncertainties around them.

Regarding CC, increases in temperature and decreases in precipitation since mid-1970s have been observed (Carrasco et al., 2008; Falvey and Garreaud, 2009a; Boisier et al., 2016) along with a positive trend in the equilibrium line altitude (ELA) reported by Carrasco et al. (2008). However, large regional differences in temperature have been reported with a cooling by as much as -0.25°C/decade on the coastal part of CC (Falvey and Garreaud, 2009a). Fuenzalida et al. (2007) observed a marked shift in the Chilean climate in the mid-1970s, which has been associated with an intensified South Pacific subtropical anticyclone (SPSA) and a concurrent shift in the PDO cycle to its negative phase in 1976/1977 (Quintana and Aceituno, 2012). Furthermore, Boisier et al. (2016) reported a persistent ongoing rainfall deficit affecting CC since 2010, termed a ‘megadrought’, with decreases in precipitation of about 30% with respect to the mean annual

climatology of the period 1970-2000. Similarly to SWA, the SAM has been linked to the runoff reductions in CC, in particular in the catchments located in the region south of 37.5°S.

The results of this chapter seek to extend the Barria et al. (2015) assessment presented in Chapter 4, addressing both the between-GCM and within-GCM uncertainties of runoff projections for Mediterranean-like catchments of the Southern Hemisphere. In particular, between and within-GCM uncertainties in SWA and between-GCM uncertainties in CC are assessed, where similar climatic trends have been observed. Runoff projections were obtained by running a lumped hydrological model, the Precipitation Evaporation Runoff model (PERM; Peel et al. (2015)) forced by bias corrected CMIP5 (in SWA and CC) and CPDN runs (only in SWA). This comparison aims to provide quantitative information about the impact of within-GCM and between-GCM uncertainties on projected runoff to inform water management adaptation measures. Second, the comparison seeks to identify whether the number of multiple model runs in the CMIP5 ensemble could be an approximation that adequately represents the within-GCM uncertainty. If it does, then future hydrologic assessments of within-GCM uncertainty in extratropical regions could use CMIP5 model runs, which cover the world and are easily accessible, rather than CPDN. Further information about these ensembles of GCM runs is presented in the Data chapter. Since the CPDN data were saved and released over Giorgi regions (Giorgi et al., 2001), which are too coarse for representing the steep topography and resulting complex CC climate, only CMIP5 GCM runs were used in this area. CC is covered by the SSA Giorgi region which includes all southern Chile and part of Argentina. It does not reproduce the spatial variability of temperature and rainfall of the study area, and therefore is not suitable to produce runoff projections. Hence, the aim of the study is to expand the analysis performed in SWA into CC, quantifying how much runoff is expected to change by the second half of the century in the region, including the analysis of the between-GCM uncertainties around those projections, while considering the comparison among the impact of the within and the between- GCM uncertainties obtained in SWA.

Besides the quantification of uncertainties in runoff projections, it’s also relevant to understand which parameters or processes solved in the models have a major impact on the final simulations of runoff. Acknowledging the importance of the stratospheric ozone depletion and recovery in the Southern Hemisphere atmospheric circulation changes (Yang et al., 2007; Polvani et al., 2011a; Eyring et al., 2013), the effect of the differences in the structure GCMs use to account for that process is tested in the simulations of runoff totals. As aforementioned in the Data section, some of the GCMs collated in the CMIP5 ensemble have several runs, thus accounting for uncertainties due to different initial conditions rather than due to structural differences in the models. Then, another possibility is analysing separately individual ensemble members for models that have several runs which would give extra insights into different sources

of uncertainties. However, because the aim of the thesis is the quantification of the within and the between-GCM uncertainties in runoff projections, the assessment focuses in the whole range of simulations using all the GCMs available rather than groups of models that use different initial conditions or groups that fulfil certain specific criteria such as goodness of fit to observations among others.

In the following sections the methodology used to obtain the modelled runoff projections and uncertainties at a catchment scale are presented in section 5.2. Then, the results of the quantification of the between and within-GCM uncertainties on runoff projections in SWA, and the between-GCM uncertainties on CC runoff projections are presented in section 5.3 along with an exploratory analysis of the impact of the parametrisation of one of the process the GCMs solve on runoff simulations. Finally, the main findings and conclusions of the chapter are detailed in section 5.4.

5.2 Methodology

Runoff projections and quantification of between–GCM uncertainties for three SWA and three CC catchments for the second half of the century are made through hydrological modelling. The stages in the modelling methodology are presented in Figure 5.1 and they are described below.

FIGURE 5.1 METHODOLOGY OF RUNOFF PROJECTIONS A statistical downscaling methodology was used to fit the coarse outputs from the GCM (hundreds of km) to the catchment scale. A direct and parameter free Quantile-Quantile methodology was applied (Themeßl et al., 2011) which has been explained in detail in Chapter 4.

A lumped conceptual hydrological model was used to simulate runoff with the monthly bias corrected precipitation and temperature data from the GCM. The precipitation, evaporation, runoff model (PERM) which was developed and described by Peel et al. (2015). The description of the structure of the model and the calibration and validation methodology it uses are described in Chapter 4.

The within and between-GCM uncertainties of projected variables were calculated using the range between the 5th and the 95th percentiles of the change in precipitation or runoff for the

period 2050 to 2080 compared to 1970 to 2000. The projected rainfall and temperature were obtained from CPDN and the CMIP5 ensemble of GCM runs.

The two-sided Kolmogorov Smirnov test was used to compare the histograms of change in annual runoff using CPDN and CMIP5, to indicate whether they are from the same continuous distribution.

5.3 Results

Results of the quantification of GCM-uncertainty of projected runoff for SWA and CC are presented in the following five sub-sections. First, the ability of the raw CPDN and CMIP5 data to simulate the climate of the studied regions during the observed period is evaluated. Then, the calibration and evaluation results of modelled runoff for the six study catchments are shown in the second section. The third section presents the comparison of the effect of the within and between-GCM uncertainties of SWA runoff projections. Only between-GCM uncertainty of runoff projections are analysed for CC, presented in the fourth section. Finally, a sensitivity of runoff projections in CC to ozone characterisation within the GCM models is presented in section 5.3.5.

5.3.1 CMIP5 and CPDN evaluation in SWA and CC catchments

The performance of both ensembles CMIP5 and CPDN in simulating the annual (water year; from April to March), autumn (April-Jun), winter (July-August), spring (September- November) and summer (December-February) historical climatic variables over SWA, and the performance of CMIP5 in simulating the annual and seasonal historical variables over CC catchments are presented and compared in this section. Although CMIP5 and CPDN are structurally different ensembles, both aim to provide an approximation to quantify uncertainties in climatic projections. Previous research has indicated that CMIP5 correctly reproduces the annual cycle of precipitation, the winter precipitation and the annual and seasonal temperature over CC (Zazulie et al., 2017). Regarding SWA, previous investigation has indicated the CMIP5 models are acute in representing the historical phase and amplitude of seasonal temperature, the observed spatial precipitation pattern and the decline in precipitation in the region (Irving et al., 2012). Besides historical temperature and precipitation data provided by the CMIP5 ensemble has been previously assessed, seasonal precipitation and temperature CMIP5 historical data were compared to station-based gridded climatological data at the catchment scale in both regions, CC and SWA, to assess the ability of the ensemble of simulating these variables which are input to the hydrologic model. The bias in the mean state and in the variability (median and standard deviation) of both variables at the catchment scale were assessed. Boxplots of mean and standard deviation of annual precipitation simulated by the CMIP5 GCM ensemble during the observed 79

period are presented in Figures 5.2a and 5.2b respectively. The evaluation of the mean and standard deviation of annual temperature simulated by CMIP5 GCMs is presented in Figures 5.3a and 5.3b respectively.

a) b)

FIGURE 5.2 A) COMPARISON OF MEAN ANNUAL PRECIPITATION SIMULATED BY CMIP5 IN THE HISTORICAL PERIOD AND OBSERVED MEAN ANNUAL PRECIPITATION. B) COMPARISON OF STANDARD DEVIATION OF ANNUAL PRECIPITATION SIMULATED BY CMIP5 IN THE HISTORICAL PERIOD AND STANDARD DEVIATION OF OBSERVED ANNUAL PRECIPITATION.

a) b)

FIGURE 5.3 A) COMPARISON OF MEAN ANNUAL TEMPERATURE SIMULATED BY CMIP5 IN THE HISTORICAL PERIOD AND OBSERVED MEAN ANNUAL TEMPERATURE. B) COMPARISON OF STANDARD DEVIATION OF ANNUAL TEMPERATURE SIMULATED BY CMIP5 IN THE HISTORICAL PERIOD AND STANDARD DEVIATION OF OBSERVED ANNUAL TEMPERATURE. Overall, the CMIP5 GCMs underestimate mean annual precipitation by 31% on average in the SWA catchments. The GCM ensemble is better at representing the variability (annual standard deviation) than the mean of annual precipitation in the SWA catchments, with a 14% underestimation on average. Regarding mean annual temperature, the GCMs tend to overestimate mean annual temperature in SWA catchments. However, the ensemble reproduces the decrease in precipitation and the increase in temperature (not shown) during the observed period (1970- 2000). Barria et al. (2015) evaluated the CPDN dataset and found that the perturbed physics

ensemble was better at simulating observed temperature (a 1.4% difference in the average of the simulation of the median) than observed precipitation (a 6.2% of difference in the average of the simulation of the median) in the SWA catchments.

The CMIP5 ensemble tends to overestimate mean annual precipitation at the Cauquenes and Lumaco catchments during the observed period (an overestimation of 58% on average) and to underestimate mean annual precipitation for the Cato River (24%). Results presented in Figure 5.2b indicate that CMIP5 is more accurate at simulating the standard deviation of annual precipitation for the Cauquenes and Lumaco catchments (a 4.7% difference on average), than for the Cato catchment (44% difference). The CMIP5 ensemble is more accurate in simulating the standard deviation than the mean of annual temperature in the CC catchments. Despite the differences between CMIP5 and the observed mean and standard deviation of the annual climatic variables, the ensemble of models did reproduce reductions in precipitation and increases in temperature during the historical period (1970-2000). Larger spatial variability is observed in CC than in SWA which might be related to the larger bias in the GCMs when attempting to simulate steep topography such as in CC.

The monthly GCM precipitation and temperature data for the six catchments were translated to the catchment scale using the quantile-quantile bias correction methodology (Themeßl et al., 2012), prior to being input to the hydrological model. One example of bias corrected data is presented in Figure 5.4a and b, where annual precipitation at Donnelly and Cauquenes catchment respectively is shown. According to Figure 5.4 the mean and the percentiles of the simulations represented by the blue lines fit the observed annual rainfall over the catchment after using the bias correction methodology. Similar results were found in the other catchments for precipitation and temperature (not shown).

FIGURE 5.4 BIAS CORRECTED CMIP5 SIMULATED ANNUAL PRECIPITATION UNDER THE RCP8.5 SCENARIO OVER A) DONNELLY AT STRICKLAND AND B) CAUQUENES AT EL ARRAYÁN

5.3.2 PERM model calibration and evaluation in SWA and CC catchments

The PERM model was run on a monthly basis and calibrated against annual data for the three SWA catchments using observed area-weighted temperature and rainfall from the AWAP data (Jones et al., 2009). The calibrated parameters and calibration model performance statistics are presented in Table 5.1, while the evaluation of model performance is presented in Table 5.2. PERM simulates annual runoff well in the three catchments, with a mean coefficient of determination (R2) of 0.79 and annual Nash & Sutcliffe Efficiency (N&SE) of 0.78. The poorest

PERM performance was observed in the driest catchment, Helena River, which is still considered in our analysis because it is one of the most susceptible to climate change. The N&SE of the K- Fold cross-validation of the model is presented in Table 5.2, which averaged 0.74 over the three catchments, and indicates that the model is accurate at representing the monthly runoff and the hydrological processes of SWA catchments. The average difference between modelled and observed mean annual runoff and annual coefficient of variation across the three catchments was

2.3% and 3.6% respectively during the calibration period.

TABLE 5.1 CALIBRATED PARAMETERS AND CALIBRATION STATISTICS FOR PERM IN THE CC AND SWA CATCHMENTS

Calibration annual Parameters modelled and Catchment observed runoff Smax Etrate K Melt Imax R2 N&SE Donnelly River at Strickland 767.81 7.56 0.00 0.27 23.88 0.94 0.94 Helena River at 200.00 60.00 0.41 0.02 100.00 0.62 0.59 Ngangaguringuring Denmark River at Kompup 936.56 12.22 0.27 0.83 92.75 0.78 0.78 Cauquenes River at el 50.00 3.31 0.85 0.71 121.25 0.82 0.80 Arrayán Cato River at Puente Cato 154.38 3.69 0.13 0.62 250.00 0.89 0.87 Lumaco River at Lumaco 1109.69 5.00 0.20 0.83 4.00 0.60 0.60

TABLE 5.2 EVALUATION OF PERM IN THE CC AND SWA CATCHMENTS

PERM evaluation results Obs. Mod. Obs. Mod. MAR MAR Dif. Obs Dif. Obs. Annual Cv Cv. Catchment April- April- and Mod. And Mod. N&SE April- April- March March MAR (%) CV (%) March March (mm) (mm) Donnelly River at 0.70 58.93 57.92 1.70 0.58 0.56 4.83 Strickland Helena River at 0.60 6.41 6.10 4.93 0.89 0.88 0.84 Ngangaguringuring Denmark River at 0.92 162.72 162.98 0.16 0.43 0.41 5.00 Kompup Cauquenes River at el 0.79 439.93 432.00 1.80 0.60 0.48 20.00 Arrayán Cato River at Puente 0.86 1301.32 1296.60 0.36 0.38 0.31 18.42 Cato Lumaco River at 0.59 624.05 609.43 2.34 0.33 0.31 6.06 Lumaco PERM was also calibrated for the three CC catchments using monthly observed area- weighted temperature and rainfall data. PERM model was found to be accurate at representing 83

the annual hydrology of the CC catchments, with an average coefficient of determination (R2) of 0.77 and Nash & Sutcliffe Efficiency value (N&SE) of 0.76 across the three catchments during calibration (see Table 5.1). Regarding the evaluation of PERM, the average N&SE of the K-Fold cross-validation was 0.75 over the three catchments, which indicates the model performs well in these catchments (see Table 5.2). This is very similar to SWA were the average N&SE of the K- Fold cross-validation was 0.74 among the three catchments. The average difference between modelled and observed mean annual runoff and annual coefficient of variation across the three catchments was 1.5% and 14% respectively during the calibration period.

5.3.3 Comparison of within and between-GCM uncertainties in runoff projections in SWA catchments

Differences in projected mean annual precipitation and mean annual runoff between the period 2050–2080 and 1970–2000 for the three SWA catchments using first, the ensemble of CPDN GCM runs and then the ensemble of CMIP5 GCM runs were assessed. Histograms of the results for Donnelly catchment using the CPDN (from Barria et al. (2015)) and the two emissions scenarios from CMIP5 are presented in Figure 5.5. The 2500 runs using CPDN indicate reductions in mean annual precipitation for the period 2050-2080 compared to 1970-2000 with a median of around 21%, which leads to decreases in mean annual runoff of around 50% for the same period. This amplification in the response of runoff to changes in precipitation is known as runoff sensitivity which has been demonstrated to increase when the humidity ratio decreases, such as in the case of dry catchments or during dry months (Dooge, 1992; Dooge et al., 1999; Sankarasubramanian et al., 2001). The projections using the 106 CMIP5 GCM runs under the RCP4.5 scenario indicate a median reduction of around 12.5% in mean annual precipitation for the same period, which leads to a median decrease in mean annual runoff of about 34%. Simulations under the RCP8.5 scenario indicate a median reduction in mean annual precipitation of around 19%, which leads to a median reduction in mean annual runoff of around 46%, which is very similar to the results obtained for the A1B scenario from the CPDN dataset. Similar results to Donnelly are observed for change in mean annual precipitation and mean annual runoff for the period 2050-2080 relative to 1970-2000 for the Helena River and Denmark River using CPDN and CMIP5, which are presented in Table 5.3. The histograms of mean annual precipitation and runoff changes for these catchments are presented in Figures 5.6 and 5.7 respectively.

FIGURE 5.5 HISTOGRAM OF PROJECTED DIFFERENCE IN MEAN ANNUAL PRECIPITATION AND RUNOFF FOR DONNELLY AT STRICKLAND FOR THE PERIOD 2050-2080 COMPARED TO 1970-2000 USING CPDN AND CMIP5 GCMS The quantification of uncertainties in precipitation and runoff projections in the three SWA catchments using the CPDN and the CMIP5 ensembles is presented in Table 5.3. The results indicate that the within-GCM uncertainty for runoff projections at Donnelly catchment is about 39% (range using CPDN), whereas the mainly between-GCM uncertainty for runoff projections are about 41% (an average of the CMIP5 RCP4.5 and RCP8.5 scenarios). Within-GCM uncertainty for runoff projections at Helena and Denmark catchments of 65% and 56% were obtained, whereas between-GCM uncertainty of 78% and 64% under the CMIP5 RCP4.5 and RCP8.5 scenarios were obtained at Helena and Denmark catchment respectively. The results of the ratio between changes in runoff and changes in precipitation presented in Table 5.3 indicate that the drier the catchment the more sensitive it is to changes in precipitation. For instance, a given change of 1% in mean annual precipitation in Helena catchment leads to an average reduction of 2.8% in mean annual runoff.

TABLE 5.3 MEDIAN OF THE CHANGES IN PROJECTIONS OF RUNOFF AND PRECIPITATION FOR THE PERIOD BETWEEN 2050-2080 AND 1970-2000 IN SWA CATCHMENTS

Ratio Ratio Ratio Changes in Precipitation Changes in Runoff median median median Runoff Runoff Runoff Catchment Percentile /Prec. /Prec. /Prec. CPDN CMIP5 CMIP5 CPDN CMIP5 CMIP5 CPDN CMIP5 CMIP5 A1B RCP4.5 RCP8.5 A1B RCP4.5 RCP8.5 4.5 8.5 5th -32.96 -22.47 -31.24 -66.98 -53.48 -64.4 Median -20.78 -12.45 -19.26 -49.69 -33.59 -46.13 Donnelly at 2.39 2.70 2.40 Strickland 95th -9.15 -2.19 -8.19 -28.14 -9.05 -26.72 Range 5- 23.81 20.28 23.05 38.84 44.43 37.68 95th 5th -36.17 -23.13 -30.22 -84.23 -66.75 -77.66

Helena at Median -22.59 -12.46 -18.42 -60.45 -37.08 -52.69 Ngangaguringuri 95th -8.69 0.75 -2.93 -19.36 16.49 -5.81 2.68 2.98 2.86 ng Range 5- 95th 27.48 23.88 27.29 64.87 83.24 71.85 5th -29.45 -18.59 -26.94 -73.99 -56.18 -70.05 Median -18.08 -11.08 -16.42 -50.45 -30.49 -42.15 Denmark at 2.46 2.75 2.57 Kompup 95th -6.72 -0.97 -5.92 -18.15 9.00 -7.09 Range 5- 95th 22.73 17.62 21.02 55.84 65.18 62.96

FIGURE 5.6 HISTOGRAM OF PROJECTED DIFFERENCE IN MEAN ANNUAL PRECIPITATION AND RUNOFF FOR HELENA AT NGANGAGURINGURING FOR THE PERIOD 2050-2080 COMPARED TO 1970-2000 USING CPDN AND CMIP5 GCMS

FIGURE 5.7 HISTOGRAM OF PROJECTED DIFFERENCE IN MEAN ANNUAL PRECIPITATION AND RUNOFF FOR DENMARK AT KOMPUP FOR THE PERIOD 2050-2080 COMPARED TO 1970-2000 USING CPDN AND CMIP5 GCMS Results of the Kolmogorov-Smirnov test (not shown) applied to the histograms of changes in projected precipitation and runoff at the three SWA catchments indicate that projections of precipitation and runoff at Donnelly catchment under CPDN and CMIP5 RCP8.5 scenario are from the same distribution at the 5% significance level. Regarding Denmark catchment, the Kolmogorov-Smirnov test indicates that projections of precipitation using CPDN and CMIP5 RCP8.5 scenario are from the same distribution, however the hypothesis is rejected when comparing projections of runoff. Finally, the hypothesis that precipitation and runoff projections under the CPDN and CMIP5 ensemble of GCMs are from the same distribution is rejected at the 5% level of significance at Helena catchment. From these results, it can be concluded that the drier the catchment the more independent the simulations of precipitation and runoff obtained using CPDN and CMIP5 ensembles.

A final comparison is presented in Figure 5.8, where the mean exceedance probability curve obtained from the ensemble of runoff projections between 2050 and 2080 using CPDN and CMIP5 under the RCP4.5 and RCP8.5 scenarios is shown. The exceedance probability of annual observed runoff is also shown for comparison. The three ensembles of GCMs project reductions in runoff during 2050 and 2080 across all percentiles relative to the observed data. When considering extreme runoff, namely the 5% probability of exceedance runoff and the 95% probability of exceedance runoff, the differences between CPDN and the CMIP5 RCP8.5 scenario

ensemble are very small. On average, the difference in the projected 5% probability of exceedance runoff between CPDN and CMIP5 RCP8.5 ensemble is 7%, whereas the difference in the projections of the 95% probability of exceedance runoff is 1%, highlighting the similarities between the projections obtained from these two GCM ensembles.

FIGURE 5.8 EXCEEDANCE PROBABILITY OF PROJECTED MEAN ANNUAL RUNOFF AT SWA CATCHMENTS DURING 2050-2080

5.3.4 Between-GCM uncertainty of runoff projections in Central Chilean Catchments

As indicated in the introduction, CPDN is not suitable for projections in CC because the resolution of the output of the model (the Giorgi regions) is too coarse to simulate the complex Chilean topography that helps determine the climate of the region. Therefore, only the effect of between-GCM uncertainty on runoff projections was assessed in this region using the CMIP5 ensemble. Histograms of the percentage change in mean annual precipitation and runoff between 2050-2080 and 1970-2000 obtained from the CMIP5 ensemble under RCP4.5 and RCP8.5 scenarios are presented in Figure 5.9 for the Cauquenes River. The histograms of Cato and Lumaco River are presented in Figures 5.10 and 5.11 respectively. The 106 runs using CMIP5 RCP4.5 scenario over Cauquenes River indicate a median reduction of around 14% for mean annual precipitation, which leads to a median decrease in mean annual runoff of around 25%. On average, reductions in mean annual precipitation for the period between 2050 and 2080 relative to 1970 and 2000 for the three CC catchments are around 13% which leads to reductions in mean annual runoff of around 21%, with a ratio of change or sensitivity of runoff to changes in 88

precipitation of about 1.6 (Table 5.4). However, a considerable number of models projects increases in precipitation and runoff for the period 2050-2080 compared to 1970-2000, which is further analysed in the following sub-section.

FIGURE 5.9 HISTOGRAM OF PROJECTED DIFFERENCE IN MEAN ANNUAL PRECIPITATION AND RUNOFF FOR CAUQUENES AT EL ARRAYAN FOR THE PERIOD 2050-2080 COMPARED TO 1970-2000 USING CMIP5 GCMS

FIGURE 5.10 HISTOGRAM OF PROJECTED DIFFERENCE IN MEAN ANNUAL PRECIPITATION AND RUNOFF FOR CATO IN PUENTE CATO FOR THE PERIOD 2050-2080 COMPARED TO 1970-2000 USING CMIP5 GCMS

FIGURE 5.11 HISTOGRAM OF PROJECTED DIFFERENCE IN MEAN ANNUAL PRECIPITATION AND RUNOFF FOR LUMACO IN LUMACO FOR THE PERIOD 2050-2080 COMPARED TO 1970-2000 USING CMIP5 GCMS

TABLE 5.4 MEDIAN OF THE CHANGES IN PROJECTIONS OF RUNOFF AND PRECIPITATION FOR THE PERIOD BETWEEN 2050-2080 AND 1970-2000 IN CC CATCHMENTS

Ratio median Ratio median Changes in Changes in Runoff Runoff/Prec. Runoff/Prec. Precipitation CMIP5 4.5 CMIP5 8.5 Catchment Percentile CMIP5 CMIP5 CMIP5 CMIP5

RCP4.5 RCP8.5 RCP4.5 RCP8.5

5% -34.92 -45.48 -51.03 -67.40 Median -14.09 -26.19 -25.01 -42.04 Cauquenes 1.78 1.61 95% 15.59 -0.11 17.98 -2.60 Dif. 5-95% 50.51 45.37 69.01 64.80 5% -31.29 -39.06 -39.56 -51.68 Median -13.05 -21.65 -19.34 -31.80 Lumaco 1.48 1.47 95% 6.97 -7.46 3.26 -14.61 Dif. 5-95% 38.26 31.60 42.82 37.07 5% -30.24 -39.75 -43.51 -59.18 Median -12.22 -22.81 -19.71 -35.72 Cato 1.61 1.57 95% 10.45 0.23 10.91 -7.24 Dif. 5-95% 40.69 39.98 54.42 51.94

The runs using the CMIP5 RCP8.5 scenario over Cauquenes River indicate a median reduction in mean annual precipitation of around 26%, which leads to a median decrease in mean annual runoff of around 42%. The results presented in Table 5.4 indicate that on average, reductions in mean annual precipitation of about 24% in CC catchments are projected, which leads to reductions in mean annual runoff of about 37%, with a hydrological sensitivity of 1.55, a similar value to that found using the RCP4.5 scenario.

According to the results presented in Table 5.4, the between-GCM uncertainty in mean annual precipitation in Cauquenes River under the RCP4.5 and the RCP8.5 scenarios are about 51% and 45% respectively. Furthermore, the between-GCM uncertainty in runoff projections at Cauquenes River is about 69% and 65% under the RCP4.5 and RCP8.5 scenarios respectively. Between-GCM uncertainty in CC catchments is very large, with an average across the three catchments of about 55% under the RCP4.5 scenario and of about 51% under the RCP8.5 scenario.

5.3.5 Analysis of interplay between ozone recovery and GHG in CC and SWA catchments

Although runoff projections for CC and for SWA under the RCP4.5 scenario (CMIP5) suggest reductions in the period 2050-2080 compared to 1970-2000 with a median of 22% and 34% respectively, some GCM runs project increases of 20% in mean annual runoff for the same period in CC and SWA. To understand this result, an exploratory analysis was conducted for the catchments to investigate the importance to local climate of the interplay between the rate of ozone recovery and the increase in greenhouse gases (Polvani et al., 2011b). All CMIP5 GCMs include the effects of time-varying ozone with stratospheric ozone depletion in the past and stratospheric ozone recovery in the future. Following Eyring et al. (2013), the CMIP5 ensemble members were divided into CHEM and NOCHEM models. The CHEM models have either interactive ozone through a coupled chemical climate model or prescribed stratospheric ozone that varies according to the different RCP scenario used in the GCM (Table 3.5). In contrast, the NOCHEM models have one prescribed time-varying ozone path that is used for all RCP scenarios.

Histograms of percentage change in mean annual runoff for the period 2050-2080 relative to 1970-2000 for a) Cauquenes catchment; b) Cato catchment and c) Lumaco catchment, and for a) Donnelly; b) Helena and c) Denamrk rivers, for both CHEM and NOCHEM models in the CMIP5 ensemble for RCP4.5 scenario are presented in Figure 5.12 and Figure 5.13 respectively.

FIGURE 5.12 SIMULATION OF RUNOFF UNDER THE RCP4.5 SCENARIO USING CMIP5 MODELS CONSIDERING CHEM AND NOCHEM MODELS IN A) CAUQUENES RIVER, B) CATO RIVER AND C) LUMACO RIVER

FIGURE 5.13 SIMULATION OF RUNOFF UNDER THE RCP4.5 SCENARIO USING CMIP5 MODELS CONSIDERING CHEM AND NOCHEM MODELS IN A) DONNELLY RIVER, B) HELENA RIVER AND C) DENMARK RIVER According to the Kolmogorov-Smirnov test, there are significant differences between the two groups of models (Chem and NoChem GCMs) for all CC catchments at the 5% level of significance, and only for Denmark catchment in SWA. The CHEM models project a median change in mean annual runoff of -11.6% averaged over the three CC catchments and -30.4% in SWA catchments; while the NOCHEM models project a median change in mean annual runoff of -25.6% and -37% in CC and SWA catchments respectively. Thus, CMIP5 models with interactive or semi-offline ozone (stratospheric ozone levels respond to changes in GHG concentrations) project lesser reductions in mean annual precipitation and hence mean annual runoff for the second half of the century than GCMs with prescribed stratospheric ozone recovery in CC catchments. Fewer differences are observed in SWA catchments, which indicates the sign of reductions is much lesser sensitive to changes in ozone than in CC catchments.

5.3.6 Discussion

The results presented in this Chapter indicate that uncertainties in precipitation projections for the second half of the 21st century in SWA and CC catchments are very large (a spread of ~23% and ~40% under the RCP8.5 scenario), which translate into larger uncertainties in runoff projections for the same period (a spread of ~50% in the simulations of both regions).

Regarding the CMIP5 GCM ensemble, most of the runoff simulations under the RCP8.5 scenario suggest reductions during the 21st century with only a few exceptions. Whereas, for the RCP4.5 scenario results are more mixed, with reduced median projections of precipitation and runoff for the period 2050-2080 in the six study catchments, but some CC and SWA projections have an increase in runoff up to 20% relative to the 1970-2000 mean annual runoff.

Analysis of the GCMs used to project CC and SWA runoff under the RCP4.5 scenario suggests the divergence might be related to differences in the structure of the different GCMs, in particular how they prescribe the stratospheric ozone rate recovery. The divergence is observed in the RCP4.5 scenario, which simulates a radiative forcing of 4.5 W/m2 by the end of the 21st century, whereas it is not observed in the RCP8.5 scenario, which assumes a radiative forcing of 8.5 W/m2, which might be masking the effect of the ozone recovery in most of models.

According to the results presented in Section 5.3.5, the CC runoff simulations that use GCMs with prescribed ozone chemistry (independent of the GHG emission scenario) project major reductions in runoff for the second half of the century, whereas models with interactive ozone project lower reductions. This difference is associated to the well know interplay between the impact of GHG emissions and ozone recovery in the stratosphere on climatic simulations (Polvani et al., 2011b; Eyring et al., 2013). This result was also observed in Denmark catchment in SWA. The fact that GCMs with interactive ozone project lower runoff reductions compared to prescribed ozone chemistry GCMs simulations is consistent with Eyring et al. (2013), who indicated that in general the CMIP5 models fit climatic observations reasonably, but models with interactive ozone (CHEM) present larger deviations from the observations.

The precipitation reductions in the mid to high latitudes of the Southern Hemisphere, such as SWA and CC, are largely related to the observed positive trend in the SAM, characterised by a poleward displacement of the westerly jet and the consequent poleward movement of the storm track (Gillett et al., 2006; Hendon et al., 2007; Meneghini et al., 2007; Polvani et al., 2011b; Ho et al., 2012; Lim et al., 2016). Although both Antarctic ozone stratospheric depletion and increases in GHGs have contributed to these changes, the former has been largely reported as the main driver of the climatic changes in the region (Kang et al., 2011; Purich and Son, 2012; Gillett et al., 2013; Lee and Feldstein, 2013). However, stratospheric ozone is projected to recover by the mid-21st century, which is expected to partially offset the impacts of increases in GHGs on SH

mid-latitude climate features (Thompson et al., 2011; Eyring et al., 2013; Barnes et al., 2014; Wang et al., 2014).

Furthermore, Neely et al. (2014) found that a coarse temporal resolution of the ozone specified in CMIP5 GCMs, in particular those that don’t use an interactive atmospheric chemical coupled model, is related to bias in representing climatic trends in the SH. They highlight the importance of ozone characterisation in climatic projection assessments in the SH. It is important to note that the analysis presented in this article is only exploratory and further work needs to be done to understand the interplay between stratospheric ozone recovery and GHG emissions and their influence over CC and SWA precipitation and runoff. However, it highlights the importance of evaluating scenario adequacy and GCM ensembles when using projections for practical purposes.

5.4 Summary

The analyses presented in this chapter seek to address the two gaps identified in Chapter 2 regarding climatic projections in SWA and CC. First, the comparison of the within and between- GCM uncertainties on runoff projections in SWA has been performed. Second, the first quantification of the between-GCM uncertainties on runoff projections in CC catchments has been developed, a region where to date only the median of models’ ensemble projections for specific catchments have been performed. According to the results presented in the chapter, the main conclusions are:

- On average, reductions of about 47% and 37% in mean annual runoff for the period 2050-2080 relative to 1970-2000 are projected under the RCP8.5 scenario for SWA and CC catchments respectively.

- Uncertainties around the precipitation and runoff projections in SWA are very large. Within-GCM uncertainty for runoff projections in three SWA catchments, for the period 2050-2080, range between 39% and 65%. Furthermore, between-GCM uncertainty for runoff projections range between 44% and 83% for the RCP4.5 scenario and about 38% and 72% under the RCP8.5 scenario in SWA catchments for the same period.

- Within and between-GCM uncertainties for SWA catchment precipitation simulations are very similar, especially the results obtained from the CPDN (within- GCM uncertainty) and the CMIP5 (mainly between-GCM uncertainty) under the RCP8.5 scenario. Differences between the 95th and the 5th percentile of rainfall

simulations are 25% and 24% on average from CPDN and CMIP5 under the RCP8.5 scenario respectively.

- Regarding runoff projections in SWA, on average the within and between-GCM uncertainties are 53.2% and 57.5% respectively.

- The results indicated that the drier the catchment the larger the uncertainty in runoff projections and the larger the differences among between-GCM and within-GCM uncertainties. In SWA, Helena catchment is the driest and has the largest difference with 65% and 72% of within and between-GCM uncertainties respectively. This might be related to the larger hydrological sensitivity of dry catchments.

- At Donnelly catchment both, the CPDN and CMIP5 RCP8.5 scenario projections are from the same distributions according to the Kolmogorov-Smirnov test, which indicates that the within and between-GCM uncertainties of runoff projections are of similar magnitude for this catchment. Furthermore, comparison of the projected 5% and 95% probability of exceedance runoff for the period 2050-2080 in SWA from CPDN and CMIP5 GCM ensembles indicates projected extreme runoffs are very similar, with a difference of 1% for the A1B and RCP8.5 scenarios.

- Because some GCMs in CMIP5 have multiple runs using different initial conditions, CMIP5 gives some insight into within-GCM uncertainty as well. Given the easy access to CMIP5 runs that represent all regions of the world, it is recommended them as a good sample to be used in hydrological assessments.

- On average, between-GCM uncertainty in CC catchments of about 55% and 51% in runoff projections using the RCP4.5 and the RCP8.5 scenarios were found.

- Central Chilean catchments have a larger spread in runoff projections for the second half of the century than SWA. According to the results, precipitation and runoff is expected to continue decreasing during the whole 21st century in CC.

- Additional investigation of runoff projections using CMIP5 in CC and SWA catchments, under the RCP4.5 scenario, revealed differences in sign of projected changes for the second half of the 21st century in CC catchments might be related to differences in the rate of stratospheric ozone recovery in the models.

- According to the results, the dichotomy in SWA runoff projections can’t be attributed to the stratospheric ozone definition. This highlights the importance of an accurate

simulation of this feature in GCMs as the interplay between recovery of the ozone layer and increases in GHG influences climatic variability in the Southern Hemisphere.

CHAPTER 6. 300 YEARS OF STREAMFLOW RECONSTRUCTION OF A HIGH ELEVATION CATCHMENT IN CENTRAL CHILE USING TREE RINGS

6.1 Introduction

This chapter is a reproduction of the published paper Barria, P., Peel, M.C., Walsh, K.J. and Muñoz, A., 2018. The first 300‐year streamflow reconstruction of a high‐elevation river in Chile using tree rings. International Journal of Climatology, 38, pp.436-451, that has been reformatted to align with the thesis formatting. The section 1 “Introduction” and the section 2 “Data” have not been included as that material is available in Chapters 2 and 3 respectively, which have been removed to avoid repetition.

The results presented in this chapter arise from the research questions presented in Chapter 2, related to the lack of extended, well distributed and reliable streamflow records in the high elevation catchments in Central Chile, which have made difficult the analysis of current trends and the multidecadal variability of the catchments located in this area. As distinguish from SWA, CC Mediterranean-like climate region has pluvial and nivo-pluvial regime catchments, being the high elevation basins (nivo-pluvial basins), the focus of investigation in this chapter which seek to assess climate variability covering both kind of regimes (pluvial and nivo-pluvial).

As noted in Chapter 2, runoff projections in the high elevation catchments of CC indicate reductions over the next decades (Fuenzalida et al., 2007), however the lack of long term records hampers a deep analysis of the observed trends. Understanding current changes in runoff requires analysis of long records in order to capture the natural temporal and spatial variability of runoff (Field, 2014).

In this chapter the use of tree rings as a proxy to reconstruct hydroclimatic variability in a high elevation catchment in CC is proposed, based on the observation that tree growth depends on water availability and changes in temperature, which are the same variables that drive runoff generation in the region. Therefore, tree ring chronologies were used to reconstruct 300 years of pluvial (April to September) and melting (October to March) season runoff in the upper part of Biobío river and analysis of the natural variability of runoff at different time scales was developed. The description of the data used in this investigation is presented in Chapter 3, where decreases in mean annual and melting season runoff and increases in pluvial season runoff are shown (Figure 3.11).

It is important to note that the hydrology of the upper and the lower part of the river are considerably different. The high elevation part of the Biobío river is characterized by an absence 99

of a dry season with 30% of the total mean annual precipitation (2276 mm) falling during the spring-summer half of the year (October to March, see Figure 3.9a). Also, the upper and the lower Biobío river have different runoff seasonality. The former has a nival-pluvial regime of runoff which is noticeable from the seasonal variation curve that has a double peak, whereas the lower part of the river has a single peak Figure 3.9b). A better temporal characterization of the upper part of Biobío catchment is crucial in order to understand current changes in water availability in the region.

The methodology used to pursue the aims of the study is detailed in section 6.2 of the thesis, whereas the analyses of the tree ring reconstructions are presented in section 6.3 followed by the principal findings and conclusions in section 6.4.

6.2 Methodology

The time series of inflows to Laja and Ralco are highly correlated at the monthly and seasonal scales and they represent the same seasonal variation. Therefore, one time series representing the temporal variability of the entire region was constructed by taking the area weighted average of both the Laja and Ralco inflows. The characteristics of the individual stations and the combined series are presented in Table 6.1.

TABLE 6.1 CATCHMENTS AND STREAMFLOW CHARACTERISTICS Streamflow Gauge Period Mean Mean annual Area Approximate Elevation runoff (m3s1) 0° isotherm (Km2) (masl) (masl) Inflows to Laja Dam 1960/61-2015/16 1740 55.9 967 1200-1500

Inflows to Ralco Dam 1960/61-2015/16 1461 261.2 5115 1200-1500

Combined series, upper part 1960/61-2015/16 1505 228.5 - - of the Biobío river

Correlations between all the tree ring chronologies (annual resolution) and annual and seasonal runoff of the upper Biobío River were calculated in order to find the periods that best correlate with the growth of the tree rings and to select the best predictors for the reconstruction. These correlations are presented in Table 6.2. Despite the low variance they explain by themselves it is important to remark that after the calculation of the principal components and the combination of them into the multiple regression model, the explained variance of runoff increases substantially. A negative lag of 1 or 2 years was allowed in the runoff time series to account for the possibility of a physiological carry-over process already documented in these species which

100

is related to the response of the roots at deeper layers to moisture variability (Muñoz et al., 2014). Once chronologies and lagged chronologies with better correlations with seasonal streamflow were identified the principal components (PCs) were calculated and used (with their lag) as predictors. This is a common dendrochronology method already used and validated in other Chilean catchments (Muñoz et al., 2016). 12 chronologies were selected to be used in this analysis, 6 for the pluvial season reconstruction and 6 for the summer season reconstruction. Although 5 of the 12 chronologies used in the study are located in Argentina (west of the dotted line in Figure 3.13), there is evidence that streamflow of rivers located between ~30° and 37° on both sides of the Andes have a strong common signal and similar temporal variation, related to precipitation, snowpack and snowmelt variability in the Andes (Masiokas et al., 2006; Masiokas et al., 2015).

TABLE 6.2 CORRELATIONS BETWEEN THE TREE RING CHRONOLOGIES AND SEASONAL RUNOFF OF THE UPPER PART OF BIOBÍO RIVER Negative Lag Coefficient Negative Lag Coefficient of (year) (year) of correlation correlation Code melting season pluvial runoff season runoff LON 0.26 0 RAL 0.24 0 CAV -0.27 1 PIN -0.42 2 CHE 0.29 2 RAH -0.24 1 NAL 0.26 2 NIT -0.33 2 RIK -0.29 1 CAP 0.21 0 PAR -0.39 1 COL -0.22 2

In order to reconstruct runoff, a linear model using multiple linear regression (MLR) was calibrated, where the predictand is runoff, and the predictors are the principal components of the tree ring chronologies following the equation presented below:

푄푟푒푐,𝑖 = 푏^0 + 푏^1푥𝑖,1 + ⋯ + 푏^퐾푥𝑖,퐾

푄푟푒푐,𝑖: Reconstructed mean seasonal runoff of year i.

101

th 푥𝑖,푛: Value of the n predictor in year i, principal component of the tree ring chronologies.

푏^0: Regression constant

th 푏^푛: Coefficient of the n predictor

K: Total number of predictors

From the six principal components calculated using the six different chronologies, the stepwise methodology was used to select 4 PCs for each model (melting season and pluvial season runoff). This involves choosing which predictors to include in the model by entering them one by one while evaluating the residual variance, which is the variance not accounted by the variables already considered in the models.

Only 42 years of gauged runoff data matches the period sampled by the chronologies (1960/61 to 2001/02). Given the limited overlap between runoff and the chronologies, the whole runoff record was used to calibrate the reconstruction model. Robustness of the model was tested using the leave-one-out (Michaelsen, 1987; Meko, 1997) cross-validation methodology. This involves dividing the predictand and predictors time series into two groups, one containing one year of observations and the other the rest of data. Then, a multiple linear regression model is calibrated using the N-1 (N is the length of the observed time series) values of the time series and the model is evaluated using the observation that was left out. This method is repeated N times and it is evaluated using the reduction of error (RE) and the root mean square error (RMSE; Wilks (2011)).

The calibrated model was evaluated using the coefficient of determination, that measures

2 the percentage of explained variance, and the adjusted coefficient of determination (R adj), which is corrected by the number of parameters in the equation used to model runoff. The reduction of error (RE), the F value of the regression and the Durbin-Watson test were also assessed. The robustness of the model was evaluated during the cross-validation using the RE and the RMSE of the validation set, which is composed of all the predicted values from the leave-one-out cross validation process (Wilks, 2011). The Durbin-Watson statistic (d) tests the first order autocorrelation of the residuals of the reconstruction model (Ostrom, 1990). The autocorrelation of the residuals was also tested through the Portmanteau test (Ostrom, 1990), where a lack of autocorrelation means that the predictors are consistent and homoscedastic.

The calibrated model was then used to reconstruct streamflow in the upper Biobío catchment. Responses of runoff at different frequencies were tested using spectral analysis. Welch's averaged power spectrum (Welch, 1967) was used, which is an approximation for continuous spectral analysis and calculates the average power spectrum of the linearly detrended time series at a given frequency band. Wavelet analysis (Torrence and Compo, 1998) was also

102

performed in order to study the different oscillation modes of the time series and their variation in time. Spectral analysis of the linearly detrended PDO and SAM reconstructions were also conducted in order to compare their oscillation modes with the streamflow variability. The association between large scale circulation features and runoff at different time scales was assessed using simple correlations and a z mean test analysis (Wilks, 2011).

To evaluate whether red noise or white noise is the most appropriate model to test significance of the spectral analysis of the time series, the following Yevjevich (1972) equation, which assesses if the autocorrelation of the time series is significantly different from zero, was used:

1 − 푟2 푆 = 푟 푁

Where r is the first order autocorrelation and N is the length of the reconstruction. According to Yevjevich (1972)’s equation, r is significantly different from zero when the two limits given by r ± 3Sr are of the same sign as r.

The reconstruction was also analysed for drought, defined by the length of the period in which runoff was below the long term median. Extreme droughts are a fundamental factor to consider in the design of water infrastructure and in the planning of water resources. However, the short length of instrumental records impedes an adequate analysis and a complete observation of those events. The Salas et al. (2015) definition of drought has been used in this project. A drought event is declared when the mean seasonal volume of runoff for more than 5 years lies below the median of the mean seasonal volume of the whole record.

6.3 Results

Observed monthly runoff data were used to construct time series for pluvial season runoff (April-September) and for melting season runoff (October-March). The comparison of the seasonal variation between the lower and the upper part of Biobío river is plotted in Figure 3.9b and the seasonal and annual runoff of the area weighted time series for the upper Biobío river are shown in Figure 3.11. The upper part of Biobío River presents a nival-pluvial runoff regime whereas the lower part of the catchment is characterized by a pluvial regime.

According to Figure 3.11, reductions have been observed for the melting season and annual (April-March) runoff, but in the pluvial season an increase has been detected. According to Carrasco et al. (2008) the increases in pluvial season runoff are related to the increased ELA, due to changes in temperature and precipitation described in the Introduction, which increases the pluvial area of the catchment in winter at the expense of the snow accumulation area. However, these changes in the runoff are not statistically significant at a 95% of level of significance, using

103

the Mann-Kendall test (Wilks, 2011), which might be due to the limited temporal extent of the records. Extending the time series through tree ring reconstruction is required to further analyse trends and runoff variability for both the melting and pluvial seasons.

6.3.1 Reconstruction

According to coefficients of correlation between the 2, 1 and 0 years negative lagged chronologies and seasonal runoff presented in Table 6.2, a good relationship between the proxies and runoff was found in the melting season, when trees undergo a major growth pattern. Also, a good relationship was observed in the pluvial season, when precipitation is maximum and the largest contribution to the accumulation of snow is produced which controls the water availability and the tree ring growth during the following months. It is decided to make the reconstruction independent and comparable with an earlier reconstruction of the lower Biobío river (Muñoz et al., 2016), so different chronologies were used in the model of the melting season runoff of the upper part of the catchment. However, the pluvial season runoff reconstruction is not strictly independent from the lower part of the river because it uses 2 of the 11 tree ring chronologies used in the lower Biobío reconstruction.

The calibration and validation statistics for the melting season runoff and pluvial season runoff reconstruction models are presented in Table 6.3. Both models explain around 50% of the variance of the observed runoff, which is exemplary for a model using proxies in the Southern Hemisphere and indicates that the trees are a good indicator of the climatic and physical characteristics of the catchment. Both models present positive values of the error reduction statistic (RE) and significant values of the F statistics, with p values less than 0.001. The RMSE for the calibration and the validation period are small when considering the variance of the time series. The Durbin-Watson statistics obtained are close to 2 which is indicative of a lack of autocorrelation in the residuals, which is consistent with results from the Portmanteau test. The residuals are also normally distributed.

TABLE 6.3 STATISTICS OF THE RECONSTRUCTIONS MODELS OF THE UPPER PART OF BIOBÍO RIVER

2 2 Variable Period R R adj F Pf RE RMSE RMSE DW Validation Q melting season 1960- 0.52 0.49 13.79 0.0000 0.41 0.77 0.81 2.48 2002 Q pluvial season 1960- 0.52 0.49 7.00 0.0001 0.43 0.74 0.77 2.25 2002 Figures 6.1a and 6.1b present the melting and pluvial season observed runoff and the reconstructed melting and pluvial season runoff for the calibration period (1960/61 to 2001/02) respectively. The reconstructed time series underestimate the variance of the observations (around 50% less in both cases). However, the models are good at simulating La Niña years (1998/99 and 104

1962/63), which were important periods of drought in Chile. The reconstructions fit the observations well and replicate the observed changes in both melting season and pluvial season runoff.

FIGURE 6.1 (A) COMPARISON OF MELTING SEASON RUNOFF RECONSTRUCTION MODEL AND OBSERVATIONS DURING THE CALIBRATION PERIOD (1960-2002). THE RECONSTRUCTION HAS AN R2 OF 0.52 AND A RMSE OF 0.77 REGARDING THE OBSERVED DATA. 4.(B) COMPARISON OF PLUVIAL SEASON RUNOFF RECONSTRUCTION MODEL AND OBSERVATIONS DURING THE CALIBRATION PERIOD (1960-2002). THE RECONSTRUCTION HAS AN R2 OF 0.52 AND A RMSE OF 0.77 REGARDING THE OBSERVED DATA The entire reconstructions obtained using the multiple linear regression models and a 20- year spline fitted to both time series in order to see multidecadal variation are presented in Figures 6.2a and 6.2b for the melting season runoff and pluvial season runoff respectively.

105

FIGURE 6.2 (A) MELTING SEASON RUNOFF RECONSTRUCTION OF THE UPPER PART OF BIOBÍO RIVER (M3SEC-1). THE LIGHT GREY SHADING PRESENTS A +/- 2 STANDARD ERROR BAND. THE LIGHT BLUE LINE IS THE HISTORICAL MEDIAN OF THE RECONSTRUCTED TIME SERIES AND THE DARK BLACK LINE CORRESPONDS TO THE 20-YEARS FITTED SPLINE. THE FITTED LINEAR TRENDS INDICATE INCREASES OF 0.038 M3S-1 CONSIDERING THE PERIOD 1739-2002 AND REDUCTIONS OF 0.092 M3S-1 SINCE 1850. 5. (B) PLUVIAL SEASON RUNOFF RECONSTRUCTION OF THE UPPER PART OF BIOBÍO RIVER (M3SEC-1). THE LIGHT GREY SHADING PRESENTS A +/- 2 STANDARD ERROR BAND. THE LIGHT BLUE LINE IS THE HISTORICAL MEDIAN OF THE RECONSTRUCTED TIME SERIES AND THE DARK BLACK LINE CORRESPONDS TO THE 20-YEARS FITTED SPLINE. THE FITTED LINEAR TRENDS INDICATE REDUCTIONS OF 0.026 M3S-1 CONSIDERING THE PERIOD 1739-2002 AND INCREASES OF ABOUT 0.051 M3S-1 SINCE 1850. THE DARK LINE REPRESENTS THE 20 YEARS FITTED SPLINE TO THE TIME SERIES. The melting season reconstruction spans the period between 1739 and 2002 and the pluvial season goes back to 1666. The models reconstruct almost 300 years of data and allow us to complement the 56 years of observations. Regarding extremes, the reconstructions show periods that are considerably wetter and drier than the extreme years observed in the instrumental records. However, it is important to remark that in the case of the multidecadal variability of the pluvial season runoff the last decades represent lower runoff volumes compared to the previous years of reconstruction.

The observed changes over the matching 42 years spanned by the chronologies were replicated by the reconstructions with increases in the pluvial season runoff starting around 1960/61 and decreases in the melting season runoff starting around 1960/61. However, Mann- Kendall trend tests were not statistically significant in either series over the whole reconstruction period, between 1666 and 2002 for the pluvial season runoff and between 1739 and 2002 for the melting season runoff. The 20-year fitted spline shows clear multidecadal fluctuations. The decreases/increases evidenced in the last 42 years (1960/61-2001/2) in the melting/pluvial season runoff fit into a negative phase in the case of melting season runoff and into a positive phase in

106

the case of pluvial season runoff. Further analysis regarding cycles of variability was performed in order to address the multidecadal phases identified in the reconstructions.

6.3.2 Links between runoff reconstructions and large scale climatic features

Analysis of the main drivers of interannual variability of melting season and pluvial season observed runoff in the upper and the lower part of the river was performed by calculating linear correlations between runoff and observed large scale climatic features, and the statistically significant correlations are presented in Table 6.4. It was found that the SOI, PDO, TPI (tripole index of PDO; Henley et al. (2015)) and annual runoff in the upper Biobío river are highly correlated. These forcings explain about 30% of the interannual variability of the upper part of the river, but they have much less influence in the lower part of the catchment, with no statistical significance in the PDO effect on runoff generation in this area. In terms of winter runoff of the upper Biobío River, winter SOI and TPI explain about 30% of the variability. It was found that ENSO and the PDO are the main drivers of both seasonal and annual runoff and that the SAM exerts a major influence on the runoff during the melting season in the upper part of Biobío river.

TABLE 6.4 STATISTICALLY SIGNIFICANT CORRELATIONS BETWEEN OBSERVED SEASONAL RUNOFF IN THE UPPER AND THE LOWER PART OF BIOBÍO RIVER AND CLIMATIC FORCINGS (1960-2000)

ENSO ENSO SAM SAM SOI SOI PDO TPI 3.4 3.4 Marshall Annual Winter Li summer summer winter summer summer summer

Q annual 0.50 0.52 -0.55 -0.60 - - 0.32 0.53

Upper Upper Q pluvial 0.52 0.54 -0.54 -0.60 - - 0.33 0.55 season

Biobío river Biobío Q melting - - - - -0.35 -0.37 - - season

Q annual 0.35 0.39 -0.35 -0.40 - - - 0.32 Q pluvial Lower Lower 0.34 0.34 -0.35 -0.39 - - - 0.32 season

Biobío river Biobío Q melting - 0.37 - - -0.36 -0.39 - - season The link between runoff and large scale climatic features on a multidecadal scale was further analysed. An 11 year and 20 year moving average of the melting and pluvial season reconstructions were used respectively to analyse the multidecadal variability of the time series and compared them with the large scale climatic features that drive climate in the region. The comparisons are presented in Figures 6.3 and 6.4 for melting season and pluvial season runoff respectively and the correlations are presented in Table 6.5. It is important to mention that those correlations are not statistically significant because of the large autocorrelation of the moving

107

average series decreases the degrees of freedom of the data. According to Figure 6.3, melting season runoff has a positive correlation with the two reconstructions of SAM (Villalba et al., 2012), which were called A and B, at the multidecadal scale (r=0.48) and a substantial negative correlation was observed with PDO (r=-0.52). It indicates above average runoff when the SAM is positive and PDO is in its negative phase. In Figure 6.4, the pluvial season reconstruction has a positive correlation with the PDO reconstructions developed by Shen et al. (2006), and the two reconstructions developed by D'arrigo et al. (2001), named A and B, with correlations around 0.5, 0.7 and 0.9 respectively. These positive correlations indicate that above average pluvial season runoff at the multidecadal scale coincides with a positive phase of the PDO, and the opposite in the negative phase of the PDO. The correlations calculated for both reconstructions, melting season and pluvial season runoff are larger considering the period between 1850 till the end of the reconstruction (around 2002) than considering the years previous to 1850 (Table 6.5), which is coincident with the time when greenhouse gases started to increase in the atmosphere (Houghton, 1996) and a shift was observed in the PDO (Shen et al., 2006).

FIGURE 6.3 11 YEARS MOVING AVERAGE OF THE NORMALIZED RECONSTRUCTED UPPER BIOBIO MELTING SEASON RUNOFF AND THE SAM AND THE PDO RECONSTRUCTIONS. 108

FIGURE 6.4 20 YEARS MOVING AVERAGE OF THE NORMALIZED UPPER BIOBIO RECONSTRUCTED PLUVIAL SEASON RUNOFF AND THE PDO RECONSTRUCTIONS

TABLE 6.5 CORRELATIONS BETWEEN THE 11 YEAR (FOR THE SAM) OR 20 YEAR (FOR THE PDO) MOVING AVERAGE OF SEASONAL RUNOFF RECONSTRUCTIONS AND CLIMATIC FORCING RECONSTRUCTIONS

Melting season Pluvial season Melting Pluvial runoff runoff season runoff season runoff since 1850 till since 1850 (1739-2002) (1666-2002) 2002 till 2002 PDO Shen et al. (2006) 0.43 0.52 PDO D'Arrigo et al. (2001) A 0.59 0.73 PDO D'Arrigo et al. (2001) B -0.42 0.76 -0.52 0.88 SAM Villalba et al. (2012) A 0.33 - 0.48 SAM Villalba et al. (2012) B 0.33 - 0.48 To assess the importance of SAM or PDO phase on the reconstructed runoff a z test of difference in mean was used to test the significance of the difference between the 11-year moving average of melting season runoff and different phases of large scale climatic features. The mean

109

of the 11-year moving average of the SAM and PDO time series were used to stratify the melting season runoff reconstruction into years with a positive phase in the SAM and years with a negative phase in the SAM. The same procedure was used for PDO. For the three climatic forcings reconstructions, SAM Villalba A, SAM Villalba B (Villalba et al., 2012) and PDO (D'arrigo et al., 2001), the mean runoff was found to be significantly different between phases of the climatic forcings at the 95% of significance level (Table 6.6). This result reinforces the findings presented in Figure 6.3, which showed the relationship between SAM and PDO forcings and the multidecadal variability of melting season runoff. A similar analysis was developed for pluvial season runoff using the 20-year moving average of reconstructed runoff (Table 6.6). For all the three PDO reconstructions, the Shen et al. (2006), the A and B reconstructions of D'arrigo et al. (2001), runoff was found to be significantly different for both the positive and negative phases of the PDO. This result is concordant with the relation observed between PDO and the multidecadal variability of pluvial season runoff in Figure 6.4.

TABLE 6.6 Z TEST OF MEAN OF SEASONAL RECONSTRUCTED RUNOFF (M3S-1) FROM 1850 TILL 2002 SAM Villalba et al. SAM Villalba et al. (2012) B PDO Darrigo et al. (2001) B (2012) A Positive Negative Positive Negative Positive Negative

Phase Phase Phase Phase Phase Phase Q melting 215.7 201.3 215.7 201.3 203.9 217.8 season PDO Shen et al. (2006) PDO Darrigo et al. (2001) A PDO Darrigo et al. (2001) B Positive Negative Positive Negative Positive Negative Phase Phase Phase Phase Phase Phase Q pluvial 267.9 252.4 271.4 245.7 276.9 244.3 season An additional exploratory analysis was performed to investigate the worrying megadrought that has affected CC since 2010 (Boisier et al., 2016) in the long term context of the reconstructions developed here. The tree rings chronologies have been dated until 2002, thus to extend the reconstructions to 2016 the variance correction methodology described by Juckes et al. (2007) was adopted in combination with the observed runoff data. The time series of the extended reconstructions along with the 11 years moving average and 20-years moving average are presented in Figures 6.5 and 6.6 for melting season runoff and pluvial season runoff respectively. According to these results, the period that covers the megadrought (from 2010) is the fourth and the second driest on record for melting season and pluvial season runoff since 1739 and 1666 respectively.

110

FIGURE 6.5 EXTENDED MELTING SEASON RUNOFF TIME SERIES USING THE GAUGED RUNOFF UNTIL 2015 AND THE FITTED 11 YEARS MOVING AVERAGE OF THE TIME SERIES OF THE UPPER BIOBÍO RIVER

FIGURE 6.6 EXTENDED PLUVIAL SEASON RUNOFF TIME SERIES USING THE GAUGED RUNOFF UNTIL 2015 AND THE FITTED 20 YEARS MOVING AVERAGE OF THE TIME SERIES OF THE UPPER BIOBÍO RIVER

6.3.3 Spectral analysis of reconstructions

Wavelet analyses to investigate the relationship between variance and frequency over time in the time series are presented in Figures 6.7a and 6.8a for melting season and pluvial season runoff respectively. The Welch’s power spectrum analysis of the detrended reconstructed melting season runoff and pluvial season runoff along with the 95% and 5% percentiles of the power

111

spectrum of fitted white noise for the melting season runoff and the pluvial season runoff respectively are presented in Figures 6.7b and 6.8b. White noise was used to test significance of the runoff spectral analysis due to the very low autocorrelation of the reconstructions (𝜌=0.039 for melting season runoff and 𝜌 =-0.023 for pluvial season runoff) which according to the Yevjevich (1972) test is not statistically significantly different from zero.

The wavelet analysis for melting season runoff (6.7b) shows an increase in the power spectrum at around 2-6 years period and at around 20 years. The ENSO signal (2-6 years) is strong during the whole record, alternating with periods of stronger activity which were related to El Niño or La Niña years, like the period around 1750, ~1790, ~1820, ~1915, and the period between 1955 till 1960. In contrast, the 20-year cycle was strongest around 1870. That is in concordance with a strengthening of the main mode of variability of the PDO signal, the bidecadal (Mantua and Hare, 2002; Zhu and Yang, 2003) that has been observed in both the Shen et al. (2006) and Biondi et al. (2001) PDO reconstructions. According to the Welch’s power spectrum results, two important peaks that are above the 95% confidence level for the melting season runoff were found, one at around 20 years and another that fluctuates between 2 to 6 years. The first peak is associated with the PDO and the second with interannual variability modulated by the ENSO signal.

In terms of pluvial season reconstruction, the wavelet analysis shows a clear ENSO signal over 2-6 years which has been stronger during the periods around 1670, ~1750, ~1820 and especially after 1950. A 20-year cycle is also indicated, presenting a stronger signal after 1850 that can be related to the PDO bidecadal influence. Finally, a 75-year cycle is observed in the wavelet analysis which has been stronger since the beginning of the 19th century but that lies beyond the cone of significance after the beginnings of the 20th century. This is concordant with the pentadecadal cycle of PDO found in the Shen et al. (2006) and Biondi et al. (2001), which according to the authors has been stronger since 1850. According to the Welch’s power spectrum analysis, three main cycles were found that are above the 95% confidence level that are also indicated consistently in the wavelet analysis. As manifested by the peaks in the spectral analysis, one at around 75 years, another at around 20 years which is related to the PDO and the third one at around 2-6 years that can be associated to ENSO. The first peak correlates with a pentadecadal cycle of PDO (Shen et al., 2006) or the 88 years of the Gleissberg solar cycle. Given the limited length of reconstruction relative to the cycle length of around 75 years, these associations are speculative.

112

FIGURE 6.7(A) WAVELET POWER SPECTRUM OF THE NORMALIZED AND DETRENDED RECONSTRUCTED HIGH ELEVATION PART OF BIOBÍO RIVER MELTING SEASON RUNOFF (1739-2002). 6.7 (B) WELCH SPECTRAL ANALYSIS OF THE NORMALIZED AND DETRENDED RECONSTRUCTED HIGH ELEVATION PART OF BIOBÍO RIVER MELTING SEASON RUNOFF (1739-2002)

113

FIGURE 6.8(A) WAVELET POWER SPECTRUM OF THE NORMALIZED AND DETRENDED RECONSTRUCTED HIGH ELEVATION PART OF BIOBÍO RIVER PLUVIAL SEASON RUNOFF (1666-2002). 6.8.(B) WELCH SPECTRAL ANALYSIS OF THE NORMALIZED AND DETRENDED RECONSTRUCTED HIGH ELEVATION PART OF BIOBÍO RIVER PLUVIAL SEASON RUNOFF (1666-2002)

6.3.4 Drought analysis of the reconstructions

The reductions in annual and summer runoff evidenced in the instrumental records during the last 56 years are very concerning as drought events can put societies and ecosystems into water stress. Despite the importance of the ongoing megadrought, this event has not been analysed in this section as the tree ring chronologies have only been dated to 2002. Adhering to the methodology of Salas et al. (2015), droughts were defined according to their length, as periods when there were more than 5 consecutive years below the median of the season inflows to the dams. The events found in both reconstructions are presented in Table 6.7.

114

TABLE 6.7 DROUGHTS OBSERVED IN BOTH MELTING SEASON AND PLUVIAL SEASON RUNOFF RECONSTRUCTIONS AND PHASE IN THE CLIMATIC FORCINGS RECONSTRUCTIONS Melting season runoff PDO SAM 1751-1755 Neg Phase 1783-1787 Neg Phase 1893-1898 Pos Phase - 1967-1971 Pos Phase - Pluvial season runoff PDO SAM 1690-1695 Neg Phase - 1744-1748 Neg Phase - 1827-1832 Neg Phase - 1845-1849 Neg Phase - 1869-1873 Neg Phase - 1887-1893 Neg Phase - Few 5-year drought events were identified in the time series, which can be explained by the low autocorrelation of the reconstructions. Peel et al. (2004) analysed drought events in more than 1000 catchments around the world. They defined drought as a run of years with runoff below the median and they found that catchments with lower persistence (small autocorrelation) showed shorter drought lengths. Their results are consistent with the results of this thesis, where just 4 events were found for melting season runoff (𝜌=0.039) and 6 for pluvial season runoff (𝜌 =- 0.023).

The drought events observed during the pluvial season are associated and coincident with negative phases of the PDO at the multidecadal scale. Regarding the melting season reconstruction, before the mid-19th century droughts events were found to be coincident with the negative phase of the SAM and after 1850 these events are associated with positive phases of the PDO. This might be associated with the shift in the cyclicity of the PDO around 1850 described by Shen et al. (2006) and Biondi et al. (2001). Although one of the five drought events occurred during the observed period (1967-1971), more severe droughts have been identified in the reconstructed time series in the past than in the gauged period.

6.3.5 Comparison with the lower Biobío catchment reconstruction

A comparison of the higher and lower Biobío river reconstructions was developed to increase the understanding of the climate variability in the high elevation part of CC. The maximum and minimum values of streamflow for the melting and pluvial season reconstructions of the higher Biobío catchment were compared with the reconstruction of the lower Biobío river.

Analysis of the extreme wet and dry periods in the reconstructions was performed, then the five wettest and the 5 driest years in the reconstructions were ranked, the five years moving

115

average and the 10-year moving average of the time series were also calculated and ranked and the results are presented in Table 6.8.

TABLE 6.8 COMPARISON OF HIGHEST AND LOWEST N YEAR MOVING AVERAGE OF THE RECONSTRUCTED LOWER AND UPPER PART OF BIOBÍO RIVER (1739-2002) AND THE GAUGED RUNOFF (1960-2002). PERIOD 1-10 ARE THE N YEARS MOVING AVERAGE, RANK 1-5 ARE THE EXTREME RECONSTRUCTED EVENTS AND OBSERVED ARE THE EXTREME GAUGED EVENTS. BOLD VALUES INDICATE COINCIDENT PERIODS-YEARS AMONG THE DIFFERENT RECONSTRUCTIONS.

Wet period Drought period Upper Upper Upper Upper Lower Lower Period Rank Biobío Biobío Biobío Biobío Biobío Biobío melting pluvial melting pluvial 1 1951 1866 1919 1998 1955 1821 2 1950 1959 1944 1764 1754 1768 1 3 1940 1865 1959 1994 1809 1964 4 1919 1966 1754 1893 1748 1962 5 1945 1915 1951 1993 1783 1960

Observed 1950 1984 1993 1998 1998 1998 1 1949-1953 1864-1868 1916-1920 1994-1998 1751-1755 1960-1964 2 1944-1948 1963-1967 1933-1937 1862-1866 1783-1787 1889-1893 5 3 1917-1921 1743-1747 1949-1953 1743-1747 1934-1938 1995-1999 4 1937-1941 1957-1961 1927-1931 1983-1897 1848-1852 1744-1748 5 1928-1932 1778-1782 1833-1837 1965-1969 1997-2001 1817-1821 Observed 1950-1954 1963-1967 1991-1995 1995-1999 1998-2002 1960-1964 1 1944-1953 1975-1984 1912-1921 1990-1999 1748-1757 1960-1969 2 1913-1922 1861-1870 1718-1727 1891-1900 1934-1943 1885-1894 10 3 1932-1941 1957-1966 1928-1937 1813-1822 1783-1792 1744-1753 4 1780-1789 1914-1923 1944-1953 1743-1752 1848-1857 1863-1872 5 1955-1964 1838-1847 1704-1713 1825-1834 1808-1817 1768-1777 Observed 1944-1953 1963-1972 1977-1986 1978-1987 1989-1998 1967-1976 The first observed results are that considering the whole record of reconstruction in both parts of the catchment (lower and upper) and for the pluvial and melting seasons, there are at least 5 higher or lower years than the extreme years observed in the instrumental record. It was also noticed that the period between 1995 and 2001 is the driest period during the observed record and one of the driest spells in all the reconstructions.

When the single extreme years (lowest and highest runoff years observed in the reconstructions) were compared, the 1959 year was found to be one of the wettest years in both pluvial and melting season reconstructions. The year 1959 is part of a positive phase in PDO which leads to above average precipitation in this area. The years 1919 and 1951 are also within

116

the highest runoff years in the upper and the lower part of Biobío river, which are associated with weak and moderate El Niño phase respectively and part of a positive phase in the PDO.

There are no coincident extreme dry years in the three reconstructions. For melting season runoff reconstruction, all the driest years were not part of the period in which observations are available. Three of them were part of the negative cycle of PDO (1748, 1754 and 1783). For pluvial season runoff, it was observed that 3 of the 5 driest years were contained during the period of instrumental records: 1960, 1962 and 1964, which correspond to La Niña years.

For the 5-year moving average it was found that both reconstructions (pluvial season of upper Biobío river and lower Biobío river) are in phase in the period between 1916 and 1921 with above average runoff volumes: this corresponds to a strong El Niño year in 1917 followed by neutral years. They are also in phase in the periods between 1949-1953 and 1927-1931. The three aforementioned periods are part of positive phases of PDO. Regarding droughts, the period 1995- 2001 was part of the five driest years in all the reconstructions of the Biobío River, which might be explained by two La Niña years (1998/99 and 2000/01) during this period. Moreover, the pluvial season of the upper Biobio and the low elevation Biobio river are in phase during the negative phase of PDO in the period 1744-1748.

For the 10-year moving average, the period 1914-1924 is one of the wettest in the three reconstructions (low and high part of the catchment), coincident with a PDO positive phase. Furthermore, the period between 1955-1965 is one of the wettest in the lower part and the upper part of the Biobio river during the melting season, which is associated with a positive phase in the SAM. Finally, the period between 1944 and 1953 is ranked among the wettest in both, the pluvial season of the upper Biobío river and the lower part of the river, coincident with a positive phase of PDO. Regarding the driest years, the period 1744-1757 appears in both pluvial and melting season reconstructions. That spell is coincident with the beginning of a negative phase of PDO. Regarding the lower part of the Biobío river, three of the 5 driest periods (1990-1991, 1891-1900 and 1825-1834) correspond to positive phases in the SAM.

6.4 Summary

In this chapter the runoff records of the upper Biobío River, a mountainous catchment known for its scarcity of data, were extended through tree ring reconstruction in order to address one of the main gaps in the study of the CC hydrology. As indicated in Chapter 2, runoff analyses in the high elevation part of CC are affected by a lack of well distributed and long records of reliable observed data. As mentioned before, extended regional runoff dataset is fundamental to improve our understanding of current changes in the hydrology of the region and the analysis of future projections. Then, the results presented in this chapter aims to improve our knowledge

117

about the natural variability of runoff generation in mountainous nivo-pluvial catchments of CC. In summary, the main conclusions are:

- Two tree ring reconstructions that explain around 50% of the variance of pluvial and melting season runoff in the upper Biobío River were developed, which reinforces the result that tree rings are a useful proxy for hydroclimatologic variables in the high elevation parts of central Chile. - Over 300 years of reconstruction indicate that according to the Mann-Kendall test, the reductions in runoff during the melting season and the increases in runoff during the pluvial season observed in the instrumental records are not part of long term trends. - Important multidecadal variability was found in the reconstructed time series, evidenced by the 20-year moving average and through spectral analysis, which is not apparent in the shorter instrumental records. - Three main cycles of variability were found in the pluvial season reconstruction. The first represents the interannual variability of the series at around 2-8 years associated with the ENSO signal. The second is linked to the PDO signal at around 20 years, and the third at around 75 years is representative of the multidecadal variability of the pluvial season runoff. - Regarding melting season runoff, two main cycles of variability were found: the interannual variability, or 2-8 years cycle associated with the ENSO signal, and a 20 years cycle which might be associated with the PDO signal. - Regarding hydrological interannual variability, several studies have reported the important relationship between seasonal and annual runoff and the ENSO variability in different regions of World and in particular in South America (Hastenrath, 1990; Mechoso and Iribarren, 1992; Marengo, 1995; Gutiérrez and Dracup, 2001; Souza Filho and Lall, 2003; Córdoba-Machado et al., 2016), reinforcing the findings presented in this chapter. - Substantial but non-statistical significant correlations between the 11 and the 20-year moving average time series of melting season and pluvial season runoff, and the SAM and PDO reconstructions were found. These correlations were larger after 1850 than prior to 1850. This and the results of the z test of means for both reconstructions are indicative of the influence of the large scale climatic features in the long-term cycles of variability of runoff in the region. The PDO influences the pluvial season runoff multidecadal variability, whereas prior to 1850 the SAM had a major influence on the multidecadal variability of the melting season runoff, and after 1850 the PDO appears to be the major influence.

118

- Regarding droughts, in total 4 events were observed in the melting season runoff reconstruction and 6 in the pluvial season runoff reconstruction. Just one of those events occurs during the years when observations are available in the summer season runoff, indicating that more severe events have occurred in the catchment in the past as part of the natural variability of climate. - The pluvial season runoff drought events can be associated with a negative phase of PDO for pluvial runoff and to a positive phase of PDO for melting season runoff. It was also found that the multidecadal cyclicity of the SAM influences the multidecadal variability of melting season runoff. - Differences were found between the lower and higher elevation parts of the Biobío River in terms of extreme dry and wet years. This might be explained by the differences in the climatic forcings that drive interannual variability in both regions of the Biobío River. - The SAM has been reported to play a major role in runoff generation in the lower elevation part of Biobío River and ENSO dominates the interannual variability of the upper part of the Biobío River.

119

CHAPTER 7. CONCLUSIONS AND FUTURE WORK

With temperature/precipitation expected to continue rising/decreasing respectively during the 21st century in the extratropical regions of the Southern Hemisphere, thus endangering water availability, an improved understanding of runoff projections, and the uncertainties around them is required. This thesis has contributed to this understanding through analysis of runoff variability at seasonal, interannual and multidecadal time scales to improve our understanding of how runoff has changed in the last three centuries and how much is it projected to change.

The analyses have centred in SWA and CC, both areas have a temperate climate and have presented reductions in precipitation since the mid-70s which have caused reductions in runoff. These regions are also similar in that the SAM is one of the main drivers of interannual and multidecadal climate and runoff variability.

Four aims were described in Chapter 2 which seek to address the main gaps presented in the literature review of Chapter 3. The first aim is to quantify the impact of within-GCM uncertainties on runoff projection in SWA catchments which was presented in Chapter 4. The second and third aims were to quantify the impact of between-GCM uncertainties on runoff projections in SWA and CC, and compare the between and within-GCM uncertainties on runoff projections on SWA which were presented in Chapter 5. Finally, the fourth aim was to analyse the runoff variability at lower frequencies in a high elevation catchment of CC (Biobío river) using a multi-century reconstruction of runoff, which is presented in Chapter 6 of this thesis.

The analyses presented in this thesis are based on some key assumptions described below. First, following the conclusions of previous work which have explored the uncertainties on streamflow projections, it was considered the largest source of uncertainties arises from the GCMs used in the runoff simulations, so results and conclusions presented in this thesis do not consider the uncertainties due to the downscaling methodology or the hydrological model used in the simulations. But they focus in, to the author knowledges, the component that largely contributes to the total uncertainty. Also, this thesis was conducted using the only dataset that has addressed the true within-GCM uncertainty, sampling a group of the parameters that account for the physics of the atmosphere and the ocean in GCMs. However, CPDN does not sample all the physical interactions that occur when running a GCM or do not explicitly explore some relevant processes to local climate such as the ozone influence in the climate of the SH. Despite its limitations, CPDN is the only and the first source of data that allow a thorough study of the uncertainties in the projections of streamflow during the 21st century.

120

The projections of runoff presented in this thesis were produced using a hydrological model that was calibrated against observed climatological data. Then, it was assumed that current conditions in every studied catchment are valid and will remain constant during all the period modelled, which might not be necessarily true. This limitation is justified because the focus of the thesis is the analysis of the impact of the largest source of uncertainty in runoff projections, the GCMs uncertainties.

Further limitations of the results presented in this thesis arise from the resolutions of the GCM data, which in the case of CPDN is very coarse (Giorgi regions) and in the case of CMIP5 data has a resolution of 1.5° latitude by 1.5° longitude. This is particularly problematic in CC catchments, which have small contributing areas, large slopes and where the hydrologic regime of rivers changes rapidly over short distances having a snowmelt dominating regime in the upper elevations of the country and rainfall dominating regime in the coastal region (~150 km of distance among them). Regional climatic models have the potential to contribute valuable knowledge to streamflow projections in this region. The main difficulty of producing regional climatic projections with perturbed physics is they are computationally expensive. A quantile- quantile bias corrections methodology was used in this work to address the resolution issues.

Finally, regarding the tree ring streamflow reconstruction of runoff developed in Chapter 6 of this thesis, it was assumed that there is a linear relationship between the climatic proxies (tree rings) and runoff. However, there is evidence in the literature that the influence of the climatic variables in runoff generation can be nonlinear (Kwak et al, 2016), which was not explored in this thesis as it would increase the uncertainty in the model and the results obtained.

Keeping in mind the main assumptions adopted to perform the analyses presented in this thesis, the main conclusions of the three results chapters are described in the following sections, which encompass the analysis of runoff variability at different time scales in SWA and CC catchments.

7.1 Within-GCM uncertainties in runoff projections in SWA catchments.

The first aim of the thesis was achieved through publication of the first analysis of the within-GCM uncertainties on runoff projections in SWA catchments.

Despite the extended research on the natural variability and projections of runoff already performed in SWA, a gap was found that is addressed in Chapter 4 of this thesis. This is the impact on runoff projections of the within-GCM uncertainties. This is crucial to be studied as

121

current hydrological assessments may be underestimating the uncertainty of future water availability in SWA.

According to the results presented in Chapter 3, reductions of about 4.3 mm year-1, 0.1 mm year-1 and 0.57 mm year-1 have been observed during the gauged period at Donnelly, Helena and Denmark catchments respectively. Within-GCM uncertainties around runoff projections were estimated using a multi-thousand member ensemble of GCM runs with perturbed physics from the CPDN dataset. The main conclusion of Chapter 4 is that the parameters specified within-GCM uncertainties of precipitation and temperature projections for the second half of this century are very large. These uncertainties translate into larger uncertainties in runoff projections with an average hydrological sensitivity among the three catchments (Donnelly, Helena and Denmark) of around 2.5, similar to the hydrological sensitivity reported in the literature (Chiew, 2006; Jones et al., 2006). It is important to emphasize that within-GCM uncertainty may be underestimated as only a single GCM with perturbed physics has been considered. Furthermore, the impact on within-GCM uncertainties from different emissions scenarios, downscaling/bias correction techniques, short observations lengths, the ongoing debate regarding the causative order and interplay between the increases in temperature and PET, the non-stationary relation between rainfall and runoff in space and time (Kiem et al., 2016), different hydrological models structures or vegetation response to CO2 enrichment has not been considered.

Plausible projections in annual precipitation indicate a reduction between the years 2050– 2080 compared to 1970–2000 that range between 0% and 40% for the Donnelly and Denmark Rivers and between 0% and 50% for the dry Helena catchment. The rainfall reductions cause runoff decreases that range between 10% and 80% over the same period. On average, a reduction of around 22% in annual and winter precipitation drives reductions of about 50% in runoff.

The results presented in Chapter 4 also indicate that the runoff during summer and spring months is more sensitive to changes in precipitation. This is related to the increases in temperature and reductions in precipitation during these months which increases the runoff sensitivity to changes in precipitation, a result that has also been discussed by Chiew (2006). The largest reductions in precipitation and runoff occur during winter months, when approximately 80% of annual precipitation falls. Runoff sensitivity to changes in precipitation was highest in the driest catchment, Helena with a sensitivity of 2.7, which is particularly concerning as the Helena catchment provides inflows to Perth water supplies, the largest city in Western Australia.

Results regarding uncertainties in runoff projections due to the variation of single parameters in the GCM physics parameterisations indicate that among the 5 parameters analysed, Rhcrit is the most important, as it produces the largest differences in the median of projections, with a difference of 9% considering the different plausible values the parameter can take. Because

122

the focus of the present assessment is the quantification of the total within-GCM uncertainty rather than the single-parameter uncertainties on runoff projections, only five parameters rather than combination of parameters were here analysed. However, the analysis of the impact of uncertainties in single parameters on runoff projections is recommended as an interesting future work which would give insights into selection of models for hydrological assessments.

A comparison between the true within-GCM uncertainties in runoff projections, i.e. using the CPDN ensemble of GCM runs with perturbed physics and a statistical approach which uses stochastic generation of data developed by Peel et al. (2015), shows that the range between the 25th and the 75th percentiles in the histogram of rainfall and runoff projections is almost double for the CPDN data compared with the stochastic generation of data. This result highlights the conclusion that current hydrological assessments that address uncertainties in runoff projections using statistical approaches are underestimating the uncertainties of runoff projections.

Finally, because only uncertainties due to GCMs have been considered in this assessment, it was noted that uncertainties in runoff projections are underestimated due to not considering uncertainties in emissions scenarios, downscaling techniques, hydrologic models, hydrologic model parameter uncertainty, and uncertainties in input and output data. However, the results indicate the importance of taking into account uncertainties in climate projections to make decisions regarding the water management of future water resources.

7.2 Comparison of the between-GCM and within-GCM uncertainties on projected runoff in southwest Western Australia.

The second aim of the thesis seeks to study the differences in the impact of the between- GCM and the within-GCM uncertainties on SWA catchment runoff projections. This aim is fulfilled in Chapter 5 of this thesis. The between-GCM uncertainties were quantified using the plausible range of rainfall and temperature projections between the 5th and the 95th percentiles informed by the CMIP5 ensemble of GCMs, and the modelled runoff using the PERM model fed by the CPDN data over the same three SWA catchments presented in Chapter 4 of this thesis.

On average, reductions of about 34% and 47% in mean annual runoff for the period 2050- 2080 relative to 1970-2000 are projected under the RCP4.5 and the RCP8.5 scenarios for SWA catchments respectively. The CMIP5 simulations of annual and seasonal precipitation and temperature in the region reinforce the findings that warmer and drier conditions are expected to occur in Mediterranean-like catchments of the Southern Hemisphere during the 21st century (Prosser, 2011; Teng et al., 2012b; Prudhomme et al., 2014).

123

The results indicate that both the within and the between-GCM uncertainties in the precipitation and runoff projections in SWA are very large, with larger reductions when considering the scenario with the larger GHGs emissions, i.e, the RCP8.5 scenario. On average, the between-GCM uncertainties on runoff projections under the RCP 8.5 scenario among the three SWA catchments is about 58% whereas the within-GCM uncertainties on runoff projections considering the same period is around 53%.

The results presented in Chapter 5 indicate that the within and between-GCM uncertainties for SWA catchment precipitation and runoff simulations are very similar, especially when considering the RCP8.5 scenario in the CMIP5 GCMs ensemble. Also, the drier the catchment the larger the differences, such as for Helena river, a catchment where the within and the between-GCM uncertainties of runoff projections are about 65% and 72% respectively. But for wetter catchments, such as Donnelly river, the Kolmogorov-Smirnov test shows that the CPDN and CMIP5 RCP8.5 scenario projections are from the same distribution.

The most important conclusion of this section is that despite the CMIP5 ensemble not being a perturbed physics sample of parameterisations, but instead being an ensemble of opportunity, the runoff projections and the uncertainty around them are very similar to the within- GCM uncertainty obtained from the CPDN, i.e. the ensemble of perturbed physics in the SWA region. This can be explained because some of the GCMs in CMIP5 have multiple runs using different initial conditions, giving some insights into within-GCM uncertainty as well. Considering the CMIP5 ensemble is much easier to access than the CPDN for all regions of the world, this study recommends the CMIP5 ensemble be used in hydrological assessments. Because GCM runs do not seek to replicate the observed climatological variables at a certain point of time or space, and to avoid an underestimation of the uncertainties associated with runoff projections, it is recommended to use the whole range of CMIP5 ensemble members in hydrologic assessments, rather than a subset of them chosen by goodness of fit.

7.3 Analyses of the between-GCM uncertainties on runoff projections in CC catchments.

The third aim of the project is to investigate the between-GCM uncertainties in runoff projections in CC catchments and to assess whether it is possible to extend the results of the comparison of runoff uncertainties performed in SWA to CC.

As in SWA, the quantification of the between-GCM uncertainties in CC catchments was performed as the range between the 5th and the 95th percentile of the rainfall and runoff simulations using the CMIP5 GCMs ensemble in the three CC catchments described in Chapter 3. According

124

to the time series plotted in Figure 3.10 and the fitted linear curves to the mean annual runoff, reductions of about 1.9 mm year-1, 0.9 mm year-1 and 2.2 mm year-1 for Cauquenes, Cato and Lumaco respectively have been observed, which are projected to continue during the years to come.

To the author knowledge this thesis provides the first quantification and analysis of GCM uncertainty on runoff projections in CC. The results presented in Chapter 5 show that on average among the three CC catchments, rainfall reductions of about 13% and 24% are projected to lead to reductions in mean annual runoff of about 21% and 37% under the RCP4.5 and RCP8.5 scenarios respectively for the period between 2050-2080 relative to 1970-2000.

Regarding uncertainties, on average among the three CC catchments, the between-GCM uncertainties on runoff projections are of about 55% and 51% under the RCP4.5 and the RCP8.5 scenarios respectively. Mean annual precipitation and runoff are projected to continue decreasing during the 21st century in CC, but with large uncertainties, with some simulations indicating even increases in runoff.

Further investigation of the runoff projections for the period between 2050-2080 and 1970-2000 were performed in CC and SWA catchments under the RCP4.5 scenario by dividing the ensemble into groups of models with and without dynamic chemistry. According to the results, the dichotomy in projections for the second half of the 21st century in CC catchments might be related to differences in the rate of stratospheric ozone recovery represented in the GCMs. Whereas the dichotomy in SWA runoff projections can’t be attributed to the stratospheric ozone definition within a GCM. This result reinforces the importance of an accurate simulation of stratospheric ozone chemistry in GCMs as the interplay between recovery of the ozone layer and increases in GHG influences climatic variability in the Southern Hemisphere, particularly for CC. Despite that, it is possible that the differences between the models with or without dynamic chemistry could arise from other processes represented in the models which should be explored in future work along with the impact of different representations of the dynamic chemistry within the models.

7.4 Study of the runoff variability at lower frequencies in a high elevation catchment of CC using a multi-century reconstruction of runoff.

The fourth aim of the thesis is fulfilled in Chapter 6 of the thesis which seeks to improve the understanding of runoff generation in mountainous catchments of CC, analysing runoff

125

variability at different time scales through an extension of the runoff records of a high elevation catchment using tree ring chronologies.

The upper part of Biobío river was reconstructed in this thesis using annually resolved tree ring chronologies of A.araucaria and A.chilensis sampled at different latitudes of CC, to extend the melting season and the pluvial season runoff of the catchment. According to the gauged data, the upper part of Biobío river has a nival-pluvial regime of runoff with different hydrological characteristics compared to the low elevation catchments in the region. This catchment is particularly important because it provides the inflows to two dams, the Ralco and Laja dams, which supply water for irrigation purposes and where 50% of Chilean hydroelectric energy is generated (Lara et al., 2003). The fitted linear trends to the gauged high elevation Biobío river time series indicate reductions of 0.65 m3s-1 for annual runoff, an increase of about 0.07 m3s-1 for pluvial season runoff and a reduction of about 1.4 m3s-1 for melting season runoff, which are not statistically significant, and which were investigated using the reconstruction of runoff.

In Chapter 6, two tree ring reconstructions of about 300 years that explain around 50% of the variance of observed pluvial and melting season runoff in the upper Biobío River were successfully calibrated. Thus tree rings were demonstrated to be a useful proxy for hydroclimatologic variables in the high elevation parts of central Chile.

The 20 and the 11 years moving average of the melting season and the pluvial season runoff respectively show that both time series have important multidecadal variability. The Mann- Kendall test results do not reveal any significant trend considering the 300 years of data in both time series. Thus, the observed decreases/increases in gauged melting season/pluvial season runoff are not part of long term trends and they might be part of multidecadal cycles of variability of runoff driven by climatic features. This conclusion was reinforced by the spectral analysis of the time series, which show three main cycles of variability in the pluvial season runoff and two in the melting season runoff time series. The pluvial season runoff has a statistically significant mode of variability at around 2-8 years related to the ENSO signal, the second at about 20 years related to the PDO interdecadal variability and the third at about 75 years which according to the investigation can be associated to the PDO multidecadal variability. Regarding melting season runoff, the first significant mode of variability is about 2-8 years which is associated with ENSO and the second is about 20 years cycle associated with the PDO variability.

Further analyses of the coefficient of correlation between the 11 and the 20-year moving average time series of melting season and pluvial season runoff and the SAM and PDO reconstructions were developed which indicate that although large correlations are observed, they are not statistically significant. A z test of means indicated that the PDO has an important influence over the pluvial season runoff. Regarding melting season runoff, the SAM has a major

126

influence on the multidecadal variability of the melting season runoff before 1850 whereas the PDO has a larger influence over the melting season runoff variability after 1850.

The drought analyses indicated that in the case of the pluvial season runoff, drought events occur during negative phases of the PDO, whereas for melting season runoff, drought events are associated with a positive phase of PDO and a negative phase of SAM.

Finally, there are significant differences in the runoff generation of the lower and the upper part of the Biobío river, enhancing the importance of extending the mountainous catchment records in order to have a better understanding of the variability at different time scales. According to the results presented in Chapter 6, the SAM has been reported to play a major role in runoff generation in the lower elevation part of Biobío River and ENSO dominates the interannual variability of the upper part of the Biobío River.

7.5 Future Work

This investigation has assessed the impact of the within and the between-GCM uncertainties in runoff projections and the runoff variability in Mediterranean like catchments of the Southern Hemisphere, but a number of knowledge gaps remain. The conclusions presented in this thesis suggest some interesting future work in the hydroclimatology field and in the study of runoff projections to follow and complement this research.

As indicated in Chapter 2 of the thesis the largest source of uncertainties in runoff projection arises from the Global Climate Model (GCM) used in the simulation of the climatic variables used as input to produce runoff simulations (Chiew et al., 2008; Prudhomme and Davies, 2008; Ardoin-Bardin et al., 2009; Chiew et al., 2009; Xu et al., 2011; Teng et al., 2012b; Lafaysse et al., 2014), which have been analysed in this thesis. However, there are also uncertainties associated with the downscalling methodology that is used to translate the GCM outputs to the regional scale of the runoff generation, and due to the hydrological model used to simulate runoff. Bosshard et al (2013) quantified the uncertainty on runoff projections in the alpine Rhine river, concluding that the total uncertainty on streamflow projections depends upon the interactions of the chain components rather than on a single component. Thereby, further analysis of the uncertainties on runoff projections in Mediterranean like catchments considering the cascade effect of uncertainties due to poor observations, different downscaling techniques, and the parameterisations used by the hydrological models could be performed. This analysis could be performed using the Integrated Bayesian Uncertainty Estimator framework, which allows us to distinguish between the different sources of uncertainty and consider the interactions between them in a probabilistic manner (Ajami et al, 2007). Special attention should be paid when using the Bayesian uncertainty approach, regarding the assumptions needed to be made to assign

127

probability distributions to the different sources of uncertainties (input data, models) and assumption of error models, which are currently poorly understood and requires further research (Kavetski et al., 2006; Kuczera et al., 2006).

The results presented in Chapter 5 indicate that the specification of the rate of ozone depletion in the past, and the rate of ozone recovery in the following decades, is particularly important in CC catchments, where a dichotomy is observed in the projections of runoff when considering different ozone specifications within the GCMs. Further research should be performed which analyses the impact on precipitation simulations in CC and in SWA of the specification of ozone in the models. In addition to addressing this issue from the perspective of GCM modelling, this analysis should also focus on the development of recommendations to be made when performing runoff simulations to be used by water managers.

Regarding the study of the impact of the within-GCM uncertainties on runoff projections, future investigation should include other regions of the World such as CC. Currently CPDN data have been released in Giorgi regions which are too coarse for areas like CC, where the scarp topography and the influences of different climatic forcings have led to the occurrence of a dichotomy in the temperature trends, with increases observed in the coast and decreases in the high altitudes. Running the CPDN project at a regional scale, using a better resolution in the output to produce runoff projections in CC, would be a useful tool for water managers in the country, especially under current drought conditions.

Furthermore, the study of the impact of the within-GCM uncertainties on runoff projections should target regional analyses in both areas, SWA and CC. The weather@home initiative is a novel ongoing experiment which seeks to contribute to a better understanding of the local impact of climate change. In this experiment, a regional model is embedded in a Global Climate Model in order to generate simulations of the climate of the last decades and to run projections of climate for the following decades. Interesting results have been obtained in the analyses of the Australian extreme temperature in 2013 (Lewis, 2013), but so far there is a lack of analyses that cover the uncertainties in regional projections of streamflow in Australia and CC. It would be useful to employ the methodology developed in this thesis to extend the understanding of runoff projections at a regional scale into the two regions of study, CC and SWA.

Moreover, decision-making under climate change scenario is still an active area of research, out of the scope of the present thesis, but that should be undertaken in the near future. Considerable research has assessed the ability of certain tools to be used in the decision-making process under uncertain climatic conditions such as model-based decision support systems, or decision frameworks like the eco-engineering decision scaling (White et al., 2015; Poff et al., 2016). However, to date, climatic projections and the associated uncertainties around them are

128

not included in Water Management plans in any of the regions studied in this assessment. Thus, specific recommendations regarding the integration of climatic projections in Water Management plans are suggested. These recommendations should include analyses of the risk associated with the inclusion of medium-term and long-term projections of runoff and the uncertainties around them in order to assess and manage water availability.

Also, it is necessary to increase the network of meteorological stations in the mountainous regions of Chile. First, because this is the area where most of the Chilean rivers originate; and second, because the 50 years of observations that are available in some of the catchments are not enough to fully understand the natural variability of runoff, in particular the multidecadal variability, as was found in this thesis. A gridded dataset of observed data should be produced in CC, similar to the AWAP data in Australia, which incorporates daily modelled based on gauged data of precipitation and temperature. This would be a useful tool to be used in hydrological assessments in the region. This dataset can be produced using gauged data complemented by satellite data and reanalyses datasets.

Finally, future work in the field of dendrochronological reconstructions should target a recompilation of current precipitation, temperature and streamflow reconstructions in Chile, focusing in the comparison of the climatic variability considering regions with different climate and analysing the influence of the large scale climatic features such as ENSO, PDO and SAM on the climatic variables. Also, special attention should be paid to the analysis of drought, which would help to understand the occurrence of the current Megadrought that has affected CC since 2010. To do this, it is fundamental to work through a methodology that allows us to extend the proxies such as tree rings to present time, as currently they cover only to 2002. The study of the performance of nonlinear reconstructions models is also suggested. Previous investigation has indicated that linear models are not always able to simulate the concomitant interaction between rainfall and the large-scale ocean-atmospheric climate processes, and between rainfall and runoff (Tozer and Kiem, 2017). The inclusion of non-linear models in CC reconstructions and in CC and SWA hydrologic models used to generate projections of water resources, would help to improve the representation of nonlinearities in the relationship between the climatic variables and runoff.

129

CHAPTER 8. REFERENCES

Adler, R.F. et al., 2003. The version-2 global precipitation climatology project (GPCP) monthly precipitation analysis (1979-present). Journal of Hydrometeorology, 4(6): 1147-1167. DOI:10.1175/1525-7541(2003)004<1147:TVGPCP>2.0.CO;2

Alavian, V. et al., 2009. Water and climate change: understanding the risks and making climate- smart investment decisions. World Bank: Washington, DC.

Alexander, L.V., Arblaster, J.M., 2009. Assessing trends in observed and modelled climate extremes over Australia in relation to future projections. International Journal of Climatology, 29(3): 417-435. DOI:10.1002/joc.1730

Alexander, L.V. et al., 2007. Trends in Australia’s climate means and extremes: a global context. Australian Meteorological Magazine, 56(1): 1-18.

Allan, R.J., 1988. El Niño southern oscillation influences in the Australasian region. Progress in Physical Geography, 12(3): 313-348.

Allan, R.J., Haylock, M.R., 1993. Circulation features associated with the winter rainfall decrease in southwestern Australia. Journal of Climate, 6(7): 1356-1367.

Allen, K. et al., 2015. Preliminary December–January inflow and streamflow reconstructions from tree rings for western Tasmania, southeastern Australia. Water Resources Research, 51(7): 5487-5503.

Andréassian, V., Perrin, C., Michel, C., 2004. Impact of imperfect potential evapotranspiration knowledge on the efficiency and parameters of watershed models. Journal of Hydrology, 286(1): 19-35.

Ansell, T.J., Reason, C.J.C., Smith, I.N., Keay, K., 2000. Evidence for decadal Variability in Southern Australian Rainfall and Relationships with regional pressure and sea surface temperature. International Journal of Climatology, 20.

Arblaster, J.M., Meehl, G.A., Karoly, D.J., 2011. Future climate change in the Southern Hemisphere: Competing effects of ozone and greenhouse gases. Geophysical Research Letters, 38(2). DOI:10.1029/2010gl045384

Ardoin-Bardin, S. et al., 2009. Using general circulation model outputs to assess impacts of climate change on runoff for large hydrological catchments in West Africa. Hydrological Sciences Journal, 54(1): 77-89. DOI:10.1623/hysj.54.1.77

Arora, V.K. et al., 2011. Carbon emission limits required to satisfy future representative concentration pathways of greenhouse gases. Geophysical Research Letters, 38(5): n/a-n/a. DOI:10.1029/2010gl046270

Ashcroft, L., Gergis, J., Karoly, D.J., 2016. Long-term stationarity of El Niño–Southern Oscillation teleconnections in southeastern Australia. Climate dynamics, 46(9-10): 2991-3006.

130

ASTER, 2009. ASTER global DEM validation summary report. ASTER GDEM Validation Team: METI, NASA and USGS in cooperation with NGA and other collaborators. Accessed January, 20: 2013.

Banco Mundial, 2011. CHILE. Diagnóstico de la gestión de los recursos hídricos.

Bao, Q. et al., 2013. The Flexible Global Ocean-Atmosphere-Land system model, Spectral Version 2: FGOALS-s2. Advances in Atmospheric Sciences, 30(3): 561-576. DOI:10.1007/s00376-012-2113-9

Barnes, E.A., Barnes, N.W., Polvani, L.M., 2014. Delayed Southern Hemisphere climate change induced by stratospheric ozone recovery, as projected by the CMIP5 models. Journal of Climate, 27(2): 852-867.

Barria, P., Peel, M.C., Walsh, K.J., Muñoz, A., 2018. The first 300‐year streamflow reconstruction of a high‐elevation river in Chile using tree rings. International Journal of Climatology, 38(1): 436-451.

Barria, P., Walsh, K.J., Peel, M.C., Karoly, D., 2015. Uncertainties in runoff projections in southwestern Australian catchments using a global climate model with perturbed physics. Journal of Hydrology, 529: 184-199.

Barros, V. et al., 2014. Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part B: Regional Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change.

Bates, B., Kundzewicz, Z.W., Wu, S., Palutikof, J., 2008. climate change and Water: technical Paper vi. Intergovernmental Panel on Climate Change (IPCC).

Bates, B.C., Chandler, R.E., Charles, S.P., Campbell, E.P., 2010. Assessment of apparent nonstationarity in time series of annual inflow, daily precipitation, and atmospheric circulation indices: A case study from southwest Western Australia. Water Resources Research, 46(3): n/a- n/a. DOI:10.1029/2010wr009509

Biondi, F., Gershunov, A., Cayan, D.R., 2001. North Pacific decadal climate variability since 1661. Journal of Climate, 14(1): 5-10.

Boisier, J.P., Aceituno, P., 2006. Changes in surface and upper-air temperature along the arid coast of northern Chile, Anales del XIV Congresso Brasileiro de Meteorología (Florianopolis 27 nov-1 dec 2006).

Boisier, J.P., Rondanelli, R., Garreaud, R.D., Muñoz, F., 2016. Anthropogenic and natural contributions to the Southeast Pacific precipitation decline and recent megadrought in central Chile. Geophysical Research Letters, 43(1): 413-421. DOI:10.1002/2015gl067265

Bosshard, T. et al., 2013. Quantifying uncertainty sources in an ensemble of hydrological climate-impact projections. Water Resources Research, 49(3): 1523-1536. DOI:10.1029/2011wr011533

131

Brohan, P., Kennedy, J.J., Harris, I., Tett, S.F.B., Jones, P.D., 2006. Uncertainty estimates in regional and global observed temperature changes: A new data set from 1850. Journal of Geophysical Research, 111: 2156-2202. DOI:10.1029/2005JD006548

Cai, W., Cowan, T., 2006. SAM and regional rainfall in IPCC AR4 models: Can anthropogenic forcing account for southwest Western Australian winter rainfall reduction? Geophysical Research Letters, 33(24).

Carrasco, J.F., Osorio, R., Casassa, G., 2008. Secular trend of the equilibrium-line altitude on the western side of the southern Andes, derived from radiosonde and surface observations. Journal of Glaciology, 54(186): 538-550.

Charles, S. et al., 2010a. Climate analyses for south-west Western Australia. A report to the Australian Government from the CSIRO South-West Western Australia Sustainable Yields Project. CSIRO, Australia.

Charles, S. et al., 2010b. Climate analyses for south-west Western Australia. A report to the Australian Government from the CSIRO South-West Western Australia Sustainable Yields Project. CSIRO, Australia.

Charles, S.P., Bari, M.A., Kitsios, A., Bates, B.C., 2007. Effect of GCM bias on downscaled precipitation and runoff projections for the Serpentine catchment, Western Australia. International Journal of Climatology, 27(12): 1673-1690. DOI:10.1002/joc.1508

Chen, J., Brissette, F.P., Poulin, A., Leconte, R., 2011. Overall uncertainty study of the hydrological impacts of climate change for a Canadian watershed. Water Resources Research, 47(12).

Chiew, F. et al., 2008. Modelling Runoff and Climate Change Impact on Runoff in 178 Catchments in the Murray-Darling Basin Using Sacramento and SIMHYD Rainfall-Runoff Models Proceedings of Water Down Under 2008, 2008: 10.

Chiew, F.H., 2006. Estimation of rainfall elasticity of streamflow in Australia. Hydrological Sciences Journal, 51(4): 613-625.

Chiew, F.H.S., Teng, J., Vaze, J., Kirono, D.G.C., 2009. Influence of global climate model selection on runoff impact assessment. Journal of Hydrology, 379(1-2): 172-180. DOI:10.1016/j.jhydrol.2009.10.004

Climate Commission, 2011. The critical decade: Western Australia climate change impacts. In: Efficiency, D.o.C.C.a.E. (Ed.). Commonwealth of Australia 2011, Canberra, ACT, Commonwealth of Australia.

Collins, W.J. et al., 2011. Development and evaluation of an Earth-System model – HadGEM2. Geoscientific Model Development, 4(4): 1051-1075. DOI:10.5194/gmd-4-1051-2011

Córdoba-Machado, S., Palomino-Lemus, R., Gámiz-Fortis, S.R., Castro-Díez, Y., Esteban- Parra, M.J., 2016. Seasonal streamflow prediction in Colombia using atmospheric and oceanic patterns. Journal of Hydrology, 538: 1-12.

132

Cortés, G., Vargas, X., McPhee, J., 2011. Climatic sensitivity of streamflow timing in the extratropical western Andes Cordillera. Journal of Hydrology, 405(1-2): 93-109. DOI:10.1016/j.jhydrol.2011.05.013

Cullen, L.E., Grierson, P.F., 2009. Multi-decadal scale variability in autumn-winter rainfall in south-western Australia since 1655 AD as reconstructed from tree rings of Callitris columellaris. Climate Dynamics, 33(2-3): 433-444.

CSIRO, BOM., 2015. Climate Change in Australia Information for Australia’s Natural Resource Management Regions: Technical Report. CSIRO and Bureau of Meteorology, Australia.

CSIRO., BOM., 2009. Surface water yields in south-west Western Australia. A report to the Australian Government from the CSIRO South-West Western Australia Sustainable Yields Project. CSIRO Water for a Healthy Country Flagship. Australia.

D'arrigo, R., Villalba, R., Wiles, G., 2001. Tree-ring estimates of Pacific decadal climate variability. Climate Dynamics, 18(3-4): 219-224.

Delworth, T.L., Zeng, F., 2014. Regional rainfall decline in Australia attributed to anthropogenic greenhouse gases and ozone levels. Nature Geoscience, 7(8): 583-587.

Demaria, E., Maurer, E., Thrasher, B., Vicuña, S., Meza, F., 2013. Climate change impacts on an alpine watershed in Chile: Do new model projections change the story? Journal of Hydrology, 502: 128-138.

Dix, M. et al., 2013. The ACCESS coupled model: documentation of core CMIP5 simulations and initial results. Aust. Meteorol. Oceanogr. J, 63(1): 83-99.

Dobler, C., Hagemann, S., Wilby, R.L., Stötter, J., 2012. Quantifying different sources of uncertainty in hydrological projections in an Alpine watershed. Hydrology and Earth System Sciences, 16(11): 4343-4360. DOI:10.5194/hess-16-4343-2012

Donner, L.J. et al., 2011. The Dynamical Core, Physical Parameterizations, and Basic Simulation Characteristics of the Atmospheric Component AM3 of the GFDL Global Coupled Model CM3. Journal of Climate, 24(13): 3484-3519. DOI:10.1175/2011jcli3955.1

Dooge, J., Bruen, M., Parmentier, B., 1999. A simple model for estimating the sensitivity of runoff to long-term changes in precipitation without a change in vegetation. Advances in Water Resources, 23(2): 153-163.

Dooge, J.C., 1992. Sensitivity of runoff to climate change: A Hortonian approach. Bulletin of the American Meteorological Society, 73(12): 2013-2024.

Dufresne, J.L. et al., 2013. Climate change projections using the IPSL-CM5 Earth System Model: from CMIP3 to CMIP5. Climate Dynamics, 40(9-10): 2123-2165. DOI:10.1007/s00382- 012-1636-1

Dunne, J.P. et al., 2012. GFDL’s ESM2 Global Coupled Climate–Carbon Earth System Models. Part I: Physical Formulation and Baseline Simulation Characteristics. Journal of Climate, 25(19): 6646-6665. DOI:10.1175/jcli-d-11-00560.1

133

Efron, B., Tibshirani, R.J., 1993. An introduction to the bootstrap, vol. 57 of Monographs on Statistics and Applied Probability. New York: Chapman and Hall.

Efstratiadis, A., Koutsoyiannis, D., 2010. One decade of multi-objective calibration approaches in hydrological modelling: a review. Hydrological Sciences Journal–Journal Des Sciences Hydrologiques, 55(1): 58-78.

Evans, J.P., Ji, F., 2012. Choosing GCMs. NARCliM Technical Note 1. NARCliM Consortium. Sydney, Australia: 7.

Eyring, V. et al., 2013. Long‐term ozone changes and associated climate impacts in CMIP5 simulations. Journal of Geophysical Research: Atmospheres, 118(10): 5029-5060.

Falvey, M., Garreaud, R.D., 2009a. Regional cooling in a warming world: Recent temperature trends in the southeast Pacific and along the west coast of subtropical South America (1979– 2006). Journal of Geophysical Research: Atmospheres, 114(D4).

Falvey, M., Garreaud, R.D., 2009b. Regional cooling in a warming world: Recent temperature trends in the southeast Pacific and along the west coast of subtropical South America (1979– 2006). Journal of Geophysical Research, 114(D4). DOI:10.1029/2008jd010519

Frederiksen, C.S., Frederiksen, J.S., Sisson, J.M., Osbrough, S.L., 2011. Changes and projections in Australian winter rainfall and circulation: anthropogenic forcing and internal variability. International Journal of Climate Change: Impacts & Responses, 2(3).

Frederiksen, J.S., Frederiksen, C.S., 2006. Decadal changes in Southern Hemisphere winter cyclogenesis. CSIRO Marine and Atmospheric Research.

Frederiksen, J.S., Frederiksen, C.S., 2007. Interdecadal changes in southern hemisphere winter storm track modes. Tellus A, 59(5): 599-617. DOI:10.1111/j.1600-0870.2007.00264.x

Frederiksen, J.S., Frederiksen, C.S., 2011. Twentieth century winter changes in Southern Hemisphere synoptic weather modes. Advances in Meteorology, 2011.

Fritts, H., 2012. Tree rings and climate. Elsevier.

Fuenzalida, H. et al., 2007. Study on climate variability for Chile during the 21st century.

Gallant, A.J., Gergis, J., 2011. An experimental streamflow reconstruction for the River Murray, Australia, 1783–1988. Water Resources Research, 47(12).

Garreaud, R., 2007. Precipitation and Circulation Covariability in the Extratropics. Journal of Climate, 20(18): 4789-4797. DOI:10.1175/jcli4257.1

Garreaud, R.D., 2011. Cambio Climático: Bases Físicas e Impactos en Chile. Revista Tierra Adentro- INIA, 93.

Garreaud, R.D., Vuille, M., Compagnucci, R., Marengo, J., 2009. Present-day south american climate. Palaeogeography, Palaeoclimatology, Palaeoecology, 281(3): 180-195.

Gent, P.R. et al., 2011. The Community Climate System Model Version 4. Journal of Climate, 24(19): 4973-4991. DOI:10.1175/2011jcli4083.1

134

Gillett, N., Kell, T., Jones, P., 2006. Regional climate impacts of the Southern Annular Mode. Geophysical Research Letters, 33(23).

Gillett, N.P., Fyfe, J.C., Parker, D.E., 2013. Attribution of observed sea level pressure trends to greenhouse gas, aerosol, and ozone changes. Geophysical Research Letters, 40(10): 2302-2306.

Giorgetta, M.A. et al., 2013. Climate and carbon cycle changes from 1850 to 2100 in MPI-ESM simulations for the Coupled Model Intercomparison Project phase 5. Journal of Advances in Modeling Earth Systems, 5(3): 572-597. DOI:10.1002/jame.20038

Giorgi, F. et al., 2001. Emerging patterns of simulated regional climatic changes for the 21st century due to anthropogenic forcings. Geophysical research letters, 28(17): 3317-3320.

Gong, D., Wang, S., 1999. Definition of Antarctic oscillation index. Geophysical Research Letters, 26(4): 459-462.

Gordon, C. et al., 2000. The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments. Climate dynamics, 16(2- 3): 147-168.

Gutiérrez, F., Dracup, J., 2001. An analysis of the feasibility of long-range streamflow forecasting for Colombia using El Nino–Southern Oscillation indicators. Journal of Hydrology, 246(1): 181-196.

Hastenrath, S., 1990. Diagnostics and prediction of anomalous river discharge in northern South America. Journal of Climate, 3(10): 1080-1096.

Hawkins, E., Sutton, R., 2009. The potential to narrow uncertainty in regional climate predictions. Bulletin of the American Meteorological Society, 90(8): 1095.

Hawkins, E., Sutton, R., 2011. The potential to narrow uncertainty in projections of regional precipitation change. Climate Dynamics, 37(1-2): 407-418.

Haylock, R.J.A.a.M.R., 1993. Circulation features associated with the winter rainfall decrease in southwest australia. Journal of Climate, 6.

Hazeleger, W. et al., 2011. EC-Earth V2.2: description and validation of a new seamless earth system prediction model. Climate Dynamics, 39(11): 2611-2629. DOI:10.1007/s00382-011- 1228-5

Hendon, H.H., Thompson, D.W.J., Wheeler, M.C., 2007. Australian Rainfall and Surface Temperature Variations Associated with the Southern Hemisphere Annular Mode. Journal of Climate, 20(11): 2452-2467. DOI:10.1175/jcli4134.1

Henley, B.J. et al., 2015. A tripole index for the interdecadal Pacific oscillation. Climate Dynamics, 45(11-12): 3077-3090.

Hennessy, K. et al., 2007. Australia and New Zealand. In ‘Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change’. Eds ML Parry, OF Canziani, JP Palutikof, PJ van der Linden, CE Hanson) pp 507–540. Cambridge University Press: Cambridge, UK. 135

Ho, M., Kiem, A., Verdon-Kidd, D., 2012. The Southern Annular Mode: a comparison of indices. Hydrology and Earth System Sciences, 16(3): 967.

Hope, P. et al., 2015a. Southern and South-Western Flatlands Cluster Report. Climate Change in Australia Projections for Australia’s Natural Resource Management Regions: Cluster Reports, eds. Ekström, M. et al., CSIRO and Bureau of Meteorology, Australia.

Hope, P. et al., 2015b. Seasonal and regional signature of the projected southern Australian rainfall reduction. AUSTRALIAN METEOROLOGICAL AND OCEANOGRAPHIC JOURNAL, 65(1): 54-71.

Hope, P.K., Drosdowsky, W., Nicholls, N., 2006. Shifts in the synoptic systems influencing southwest Western Australia. Climate Dynamics, 26(7-8): 751-764.

Hughes, J., Petrone, K., Silberstein, R., 2012. Drought, groundwater storage and stream flow decline in southwestern Australia. Geophysical Research Letters, 39(3).

INE, 2015. Compendio Estadístico 2015.

IOCI, 2002. Climate variability and change in south west Western Australia, report, 34 pp. IOCI Panel.

IOCI, 2012. Western Australia's Weather and Climate: A Synthesis of Indian Ocean Climate Initiative Stage 3 Research. CSIRO and BoM, Australia.

Irving, D.B., Whetton, P., Moise, A.F., 2012. Climate projections for Australia: a first glance at CMIP5. Australian Meteorological and Oceanographic Journal, 62(4): 211-225.

Isdale, P.J., Stewart, B.J., Tickle, K.S., Lough, J. M., 1998. Palaeohydrological variation in a tropical river catchment: a reconstruction using fluorescent bands in corals of the Great Barrier Reef, Australia. The Holocene, 8(1): 1-8.

Iversen, T. et al., 2013. The Norwegian earth system model, NorESM1-M—Part 2: climate response and scenario projections. Geosci. Model Dev, 6(2): 389-415.

Ji, D. et al., 2014. Description and basic evaluation of Beijing Normal University Earth System Model (BNU-ESM) version 1. Geoscientific Model Development, 7(5): 2039-2064. DOI:10.5194/gmd-7-2039-2014

Jones, D.A., Wang, W., Fawcett, R., 2009. High-quality spatial climate data-sets for Australia. Australian Meteorological and Oceanographic Journal, 58(4): 233.

Jones, D.A., Weymouth, G., 1997. An Australian monthly rainfall dataset. Bureau of Meteorology.

Jones, P.D., 1994. Hemispheric surface air temperature variations: a reanalysis and an update to 1993. Journal of Climate, 7(11): 1794-1802. DOI:10.1175/1520- 0442(1994)007<1794:HSATVA>2.0.CO;2

Jones, R.N., Chiew, F.H., Boughton, W.C., Zhang, L., 2006. Estimating the sensitivity of mean annual runoff to climate change using selected hydrological models. Advances in Water Resources, 29(10): 1419-1429.

136

Juckes, M.N. et al., 2007. Millennial temperature reconstruction intercomparison and evaluation. Climate of the Past, 3(4): 591-609.

Jun, M., Knutti, R., Nychka, D.W., 2008. Spatial analysis to quantify numerical model bias and dependence: how many climate models are there? Journal of the American Statistical Association, 103(483): 934-947.

Kang, S.M., Polvani, L., Fyfe, J., Sigmond, M., 2011. Impact of polar ozone depletion on subtropical precipitation. Science, 332(6032): 951-954.

Kavetski, D., Kuczera, G., Franks, S.W., 2006. Bayesian analysis of input uncertainty in hydrological modeling: 2. Application. Water Resources Research, 42(3).

Kiem, A.S., Franks, S.W., 2001. On the identification of ENSO-induced rainfall and runoff variability: a comparison of methods and indices. Hydrological sciences journal, 46(5): 715- 727.

Kiem, A.S. et al., 2008. Future hydroclimatology of the Mekong River basin simulated using the high‐resolution Japan Meteorological Agency (JMA) AGCM. Hydrological processes, 22(9): 1382-1394.

Kiem, A.S. et al., 2016. Natural hazards in Australia: droughts. Climatic Change, 139(1): 37-54.

Kiem, A.S., Verdon‐Kidd, D.C., 2011. Steps toward “useful” hydroclimatic scenarios for water resource management in the Murray‐Darling Basin. Water Resources Research, 47(12).

King, A.D., Alexander, L.V., Donat, M.G., 2013. The efficacy of using gridded data to examine extreme rainfall characteristics: a case study for Australia. International Journal of Climatology, 33(10): 2376-2387.

Knutti, R., Masson, D., Gettelman, A., 2013. Climate model genealogy: Generation CMIP5 and how we got there. Geophysical Research Letters, 40(6): 1194-1199.

Kuczera, G., Kavetski, D., Franks, S., Thyer, M., 2006. Towards a Bayesian total error analysis of conceptual rainfall-runoff models: Characterising model error using storm-dependent parameters. Journal of Hydrology, 331(1): 161-177.

Kundzewicz, Z.W. et al., 2008. The implications of projected climate change for freshwater resources and their management. Hydrological Sciences Journal, 53(1): 3-10. DOI:10.1623/hysj.53.1.3

Lafaysse, M., Hingray, B., Mezghani, A., Gailhard, J., Terray, L., 2014. Internal variability and model uncertainty components in future hydrometeorological projections: The Alpine Durance basin. Water Resources Research, 50(4): 3317-3341.

Lara, A. et al., 2015. Reconstructing streamflow variation of the Baker River from tree-rings in Northern Patagonia since 1765. Journal of Hydrology, 529: 511-523.

Lara, A., Soto, D., Armesto, J., Donoso, P., Wernli, C., 2003. Componentes científicos clave para una política nacional sobre usos, servicios y conservación de los bosques nativos chilenos.

137

Lara, A. et al., 2005. The potential use of tree-rings to reconstruct streamflow and estuarine salinity in the Valdivian Rainforest eco-region, Chile. Dendrochronologia, 22(3): 155-161.

Lara, A., Villalba, R., Urrutia, R., 2008. A 400-year tree-ring record of the Puelo River summer–fall streamflow in the Valdivian Rainforest eco-region, Chile. Climatic Change, 86(3- 4): 331-356.

Le Quesne, C., Aravena, J., Álvarez, M., Fernández, J., Roig, F., 2000. Dendrocronología de Austrocedrus chilensis (Cupressaceae) en Chile central. Dendrocronología en América Latina. Mendoza, Argentina. EDIUNC: 159-175.

Lee, S., Feldstein, S.B., 2013. Detecting ozone-and greenhouse gas–driven wind trends with observational data. Science, 339(6119): 563-567.

Li, L. et al., 2013. The flexible global ocean-atmosphere-land system model, Grid-point Version 2: FGOALS-g2. Advances in Atmospheric Sciences, 30(3): 543-560. DOI:10.1007/s00376-012- 2140-6

Lim, E.P. et al., 2016. The impact of the Southern Annular Mode on future changes in Southern Hemisphere rainfall. Geophysical Research Letters, 43(13): 7160-7167.

Magrin, G. et al., 2014. Central and South America. Climate change: 1499-1566.

Mantua, N.J., Hare, S.R., 2002. The Pacific decadal oscillation. Journal of oceanography, 58(1): 35-44.

Marengo, J.A., 1995. Variations and change in South American streamflow. Climatic Change, 31(1): 99-117.

Martin, G. et al., 2011. The HadGEM2 family of Met Office Unified Model climate configurations, Geosci. Model Dev., 4, 723–757, doi: 10.5194. gmd-4-723-2011.

Masiokas, M. et al., 2015. Reconstructing glacier mass balances in the Central Andes of Chile and Argentina using local and regional hydro-climatic data. The Cryosphere Discussions, 9(5): 4949-4980.

Masiokas, M.H., Villalba, R., Luckman, B.H., Le Quesne, C., Aravena, J.C., 2006. Snowpack variations in the central Andes of Argentina and Chile, 1951-2005: Large-scale atmospheric influences and implications for water resources in the region. Journal of Climate, 19(24): 6334- 6352.

Masson, D., Knutti, R., 2011. Climate model genealogy. Geophysical Research Letters, 38(8).

McGowan, H.A., Marx, S.K., Denholm, J., Soderholm, J., Kamber, B.S., 2009. Reconstructing annual inflows to the headwater catchments of the Murray River, Australia, using the Pacific Decadal Oscillation. Geophysical Research Letters, 36(6).

McMahon, T., Peel, M., Karoly, D., 2015. Assessment of precipitation and temperature data from CMIP3 global climate models for hydrologic simulation. Hydrology and Earth System Sciences, 19(1): 361-377.

138

McMillan, H., Freer, J., Pappenberger, F., Krueger, T., Clark, M., 2010. Impacts of uncertain river flow data on rainfall‐runoff model calibration and discharge predictions. Hydrological Processes, 24(10): 1270-1284.

Mechoso, C.R., Iribarren, G.P., 1992. Streamflow in southeastern South America and the southern oscillation. Journal of climate, 5(12): 1535-1539.

Meehl, G. et al., 2007. The WCRP CMIP3 multi-model dataset: A new era in climate change research. Bulletin of the American Meteorological Society, 88: 1383-1394.

Meehl, G.A. et al., 2012. Climate system response to external forcings and climate change projections in CCSM4. Journal of Climate, 25(11): 3661-3683.

Meko, D., 1997. Dendroclimatic reconstruction with time varying predictor subsets of tree indices. Journal of Climate, 10(4): 687-696.

Mendoza, P.A. et al., 2015. Effects of hydrologic model choice and calibration on the portrayal of climate change impacts. Journal of Hydrometeorology, 16(2): 762-780.

Meneghini, B., Simmonds, I., Smith, I.N., 2007. Association between Australian rainfall and the Southern Annular Mode. International Journal of Climatology, 27(1): 109-121. DOI:10.1002/joc.1370

Meteorological Office, H.C., 2001. The Hadley Centre regional climate modelling system.

Michaelsen, J., 1987. Cross-validation in statistical climate forecast models. Journal of Climate and Applied Meteorology, 26(11): 1589-1600.

Milly, P.C., Dunne, K.A., Vecchia, A.V., 2005. Global pattern of trends in streamflow and water availability in a changing climate. Nature, 438(7066): 347-50. DOI:10.1038/nature04312

Mizukami, N. et al., 2016. Implications of the methodological choices for hydrologic portrayals of climate change over the contiguous United States: statistically downscaled forcing data and hydrologic models. Journal of Hydrometeorology, 17(1): 73-98.

Moise, A.F., Hudson, D.A., 2008. Probabilistic predictions of climate change for Australia and southern Africa using the reliability ensemble average of IPCC CMIP3 model simulations. Journal of Geophysical Research, 113(D15). DOI:10.1029/2007jd009250

MOP, 2012. Estrategia Nacional de Recursos Hídricos 2012-2025.

Moss, R.H. et al., 2010. The next generation of scenarios for climate change research and assessment. Nature, 463(7282): 747-756.

Mote, P., Brekke, L., Duffy, P.B., Maurer, E., 2011. Guidelines for constructing climate scenarios. Eos, 92(31): 257-258.

Muñoz, A.A. et al., 2014. Patterns and drivers of Araucaria araucana forest growth along a biophysical gradient in the northern Patagonian Andes: linking tree rings with satellite observations of soil moisture. Austral Ecology, 39(2): 158-169.

139

Muñoz, A.A. et al., 2016. Streamflow variability in the Chilean Temperate-Mediterranean climate transition (35° S–42° S) during the last 400 years inferred from tree-ring records. Climate Dynamics: 1-16.

Nakicenovic, N. et al., 2000. Special report on emissions scenarios: a special report of Working Group III of the Intergovernmental Panel on Climate Change, Pacific Northwest National Laboratory, Richland, WA (US), Environmental Molecular Sciences Laboratory (US).

Nan, S., Li, J., 2003. The relationship between the summer precipitation in the Yangtze River valley and the boreal spring Southern Hemisphere annular mode. Geophysical Research Letters, 30(24).

Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models part I—A discussion of principles. Journal of hydrology, 10(3): 282-290.

Neely, R., Marsh, D., Smith, K.L., Davis, S., Polvani, L.M., 2014. Biases in southern hemisphere climate trends induced by coarsely specifying the temporal resolution of stratospheric ozone. Geophysical Research Letters, 41(23): 8602-8610.

Nicholls, N., Drosdowsky, W., Lavery, B., 1997. Australian rainfall variability and change. Weather, 52(3): 66-72.

Nicholls, N., Lavery, B., 1992. Australian rainfall trends during the twentieth century. International Journal of Climatology, 12(2): 153-163. DOI:10.1002/joc.3370120204

Nohara, D., Kitoh, A., Hosaka, M., Oki, T., 2006. Impact of climate change on river discharge projected by multimodel ensemble. Journal of Hydrometeorology, 7(5): 1076-1089.

Ostrom, C.W., 1990. Time series analysis: Regression techniques, 9. Sage.

Parker, W.S., 2013a. Ensemble modeling, uncertainty and robust predictions. Wiley Interdisciplinary Reviews: Climate Change, 4(3): 213-223.

Parker, W.S., 2013b. Ensemble modeling, uncertainty and robust predictions. Wiley Interdisciplinary Reviews: Climate Change, 4(3): 213-223 %@ 1757-7799.

Peel, M., Srikanthan, R., McMahon, T., Karoly, D., 2015. Approximating uncertainty of annual runoff and reservoir yield using stochastic replicates of global climate model data. Hydrology and Earth System Sciences, 19(4): 1615-1639.

Peel, M.C., Blöschl, G., 2011. Hydrological modelling in a changing world. Progress in Physical Geography, 35(2): 249-261.

Peel, M.C., Chiew, F.H., Western, A.W., McMahon, T.A., 2000. Extension of unimpaired monthly streamflow data and regionalisation of parameter values to estimate streamflow in ungauged catchments. Australian Natural Resources Atlas website.

Peel, M.C., Finlayson, B.L., McMahon, T.A., 2007. Updated world map of the K¨ oppen-Geiger climate classification. Hydrology and Earth System Sciences Discussions, 4: 439–473.

140

Peel, M.C., Pegram, G.G., Mcmahon, T.A., 2004. Global analysis of runs of annual precipitation and runoff equal to or below the median: run length. International Journal of Climatology, 24(7): 807-822.

Pennell, C., Reichler, T., 2011. On the effective number of climate models. Journal of Climate, 24(9): 2358-2367.

Petrone, K.C., Hughes, J.D., Van Niel, T.G., Silberstein, R.P., 2010. Streamflow decline in southwestern Australia, 1950-2008. Geophysical Research Letters, 37(11): n/a-n/a. DOI:10.1029/2010gl043102

Poff, N.L. et al., 2016. Sustainable water management under future uncertainty with eco- engineering decision scaling. Nature Climate Change, 6(1): 25-34.

Polvani, L.M., Waugh, D.W., Correa, G.J., Son, S.-W., 2011a. Stratospheric ozone depletion: The main driver of twentieth-century atmospheric circulation changes in the Southern Hemisphere. Journal of Climate, 24(3): 795-812.

Polvani, L.M., Waugh, D.W., Correa, G.J.P., Son, S.-W., 2011b. Stratospheric Ozone Depletion: The Main Driver of Twentieth-Century Atmospheric Circulation Changes in the Southern Hemisphere. Journal of Climate, 24(3): 795-812. DOI:10.1175/2010jcli3772.1

Pook, M.J., Risbey, J.S., McIntosh, P.C., 2012. The synoptic climatology of cool-season rainfall in the central wheatbelt of Western Australia. Monthly Weather Review, 140(1): 28-43.

Power, S., Casey, T., Folland, C., Colman, A., Mehta, V., 1999a. Inter-decadal modulation of the impact of ENSO on Australia. Climate Dynamics, 15(5): 319-324.

Power, S., Colman, R., 2006. Multi-year predictability in a coupled general circulation model. Climate Dynamics, 26(2-3): 247-272.

Power, S., Haylock, M., Colman, R., Wang, X., 2006. The predictability of interdecadal changes in ENSO activity and ENSO teleconnections. Journal of Climate, 19(19): 4755-4771.

Power, S. et al., 1999b. Decadal climate variability in Australia during the twentieth century. International Journal of Climatology, 19(2): 169-184.

Preston, B.L., Jones, R.N., 2008. Evaluating sources of uncertainty in Australian runoff projections. Advances in Water Resources, 31(5): 758-775. DOI:10.1016/j.advwatres.2008.01.006

Prosser, I.P., 2011. Water: science and solutions for Australia. CSIRO.

Prudhomme, C., Davies, H., 2008. Assessing uncertainties in climate change impact analyses on the river flow regimes in the UK. Part 1: baseline climate. Climatic Change, 93(1-2): 177-195. DOI:10.1007/s10584-008-9464-3

Prudhomme, C. et al., 2014. Hydrological droughts in the 21st century, hotspots and uncertainties from a global multimodel ensemble experiment. Proceedings of the National Academy of Sciences, 111(9): 3262-3267.

141

Purich, A., Son, S.-W., 2012. Impact of Antarctic ozone depletion and recovery on Southern Hemisphere precipitation, evaporation, and extreme changes. Journal of Climate, 25(9): 3145- 3154.

Qiao, F. et al., 2013. Development and evaluation of an Earth System Model with surface gravity waves. Journal of Geophysical Research: Oceans, 118(9): 4514-4524. DOI:10.1002/jgrc.20327

Quintana, J., Aceituno, P., 2012. Changes in the rainfall regime along the extratropical west coast of South America (Chile): 30-43º S. Atmósfera, 25(1): 1-22.

Räisänen, J., 2007. How reliable are climate models? Tellus A, 59(1): 2-29. DOI:10.1111/j.1600-0870.2006.00211.x

Randall, D. et al., 2007. Climate models and their evaluation. In ‘Climate Change 2007: the Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change’.(Eds Solomon S, Qin D, Manning M, Chen Z, Marquis M, Averyt KB, Tignor M, Miller HL) pp. 589–662. Cambridge University Press: Cambridge, UK.

Raut, B.A., Jakob, C., Reeder, M.J., 2014. Rainfall Changes over Southwestern Australia and Their Relationship to the Southern Annular Mode and ENSO. Journal of Climate, 27(15): 5801- 5814. DOI:10.1175/jcli-d-13-00773.1

Reisinger, A. et al., 2014. Australasia. Climate change: 1371-1438.

Risbey, J.S., Pook, M.J., McIntosh, P.C., Wheeler, M.C., Hendon, H.H., 2009. On the Remote Drivers of Rainfall Variability in Australia. Monthly Weather Review, 137(10): 3233-3253. DOI:10.1175/2009mwr2861.1

Rotstayn, L.D. et al., 2012. Aerosol- and greenhouse gas-induced changes in summer rainfall and circulation in the Australasian region: a study using single-forcing climate simulations. Atmospheric Chemistry and Physics, 12(14): 6377-6404. DOI:10.5194/acp-12-6377-2012

Rougier, J., 2007. Probabilistic inference for future climate using an ensemble of climate model evaluations. Climatic Change, 81(3-4): 247-264.

Rowlands, D.J. et al., 2012. Broad range of 2050 warming from an observationally constrained large climate model ensemble. Nature Geoscience, 5(4): 256-260.

Rubio, E., McPhee, J., 2010. Patterns of spatial and temporal variability in streamflow records in south central Chile in the period 1952–2003. Water Resources Research, 46(5).

Saft, M., Peel, M.C., Western, A.W., Perraud, J.M., Zhang, L., 2016. Bias in streamflow projections due to climate‐induced shifts in catchment response. Geophysical Research Letters.

Salas, J.D., Tarawneh, Z., Biondi, F., 2015. A hydrological record extension model for reconstructing streamflows from tree‐ring chronologies. Hydrological Processes, 29(4): 544- 556.

Samaniego, J. et al., 2009. La economía del cambio climático en Chile: síntesis.

142

Sankarasubramanian, A., Vogel, R.M., Limbrunner, J.F., 2001. Climate elasticity of streamflow in the United States. Water Resources Research, 37(6): 1771-1781.

Sansó, B., Forest, C., 2009. Statistical calibration of climate system properties. Journal of the Royal Statistical Society: Series C (Applied Statistics), 58(4): 485-503.

Schmidt, G.A. et al., 2006. Present-day atmospheric simulations using GISS ModelE: Comparison to in situ, satellite, and reanalysis data. Journal of Climate, 19(2): 153-192.

Shen, C., Wang, W.C., Gong, W., Hao, Z., 2006. A Pacific Decadal Oscillation record since 1470 AD reconstructed from proxy data of summer rainfall over eastern China. Geophysical Research Letters, 33(3).

Silberstein, R.P. et al., 2012. Climate change and runoff in south-western Australia. Journal of Hydrology, 475: 441-455. DOI:10.1016/j.jhydrol.2012.02.009

Souza Filho, F.A., Lall, U., 2003. Seasonal to interannual ensemble streamflow forecasts for Ceara, Brazil: Applications of a multivariate, semiparametric algorithm. Water Resources Research, 39(11).

Stainforth, D.A. et al., 2005. Uncertainty in predictions of the climate response to rising levels of greenhouse gases. Nature, 433(7024): 403-406.

Steinschneider, S., Polebitski, A., Brown, C., Letcher, B.H., 2012. Toward a statistical framework to quantify the uncertainties of hydrologic response under climate change. Water Resources Research, 48(11): n/a-n/a. DOI:10.1029/2011wr011318

Stocker, T. et al., 2013. Technical summary, Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, pp. 33-115.

Taylor, K.E., Stouffer, R.J., Meehl, G.A., 2012. An overview of CMIP5 and the experiment design. Bulletin of the American Meteorological Society, 93(4): 485.

Teng, J., Chiew, F.H.S., Vaze, J., Marvanek, S., Kirono, D.G.C., 2012a. Estimation of Climate Change Impact on Mean Annual Runoff across Continental Australia Using Budyko and Fu Equations and Hydrological Models. Journal of Hydrometeorology, 13(3): 1094-1106. DOI:10.1175/jhm-d-11-097.1

Teng, J., Vaze, J., Chiew, F.H.S., Wang, B., Perraud, J.-M., 2012b. Estimating the Relative Uncertainties Sourced from GCMs and Hydrological Models in Modeling Climate Change Impact on Runoff. Journal of Hydrometeorology, 13(1): 122-139. DOI:10.1175/jhm-d-11-058.1

Themeßl, M.J., Gobiet, A., Heinrich, G., 2011. Empirical-statistical downscaling and error correction of regional climate models and its impact on the climate change signal. Clim. Change, 112: 449-468. DOI:10.1007/s10584-011-0224-4

Themeßl, M.J., Gobiet, A., Heinrich, G., 2012. Empirical-statistical downscaling and error correction of regional climate models and its impact on the climate change signal. Climatic Change, 112(2): 449-468.

143

Thompson, D.W. et al., 2011. Signatures of the Antarctic ozone hole in Southern Hemisphere surface climate change. Nature Geoscience, 4(11): 741-749.

Timbal, B., 2004. Southwest Australia past and future rainfall trends. Climate Research, 26(3): 233-249.

Timbal, B., Drosdowsky, W., 2013. The relationship between the decline of Southeastern Australian rainfall and the strengthening of the subtropical ridge. International Journal of Climatology, 33(4): 1021-1034.

Torrence, C., Compo, G.P., 1998. A practical guide to wavelet analysis. Bulletin of the American Meteorological society, 79(1): 61-78.

Tozer, C., Kiem, A., 2017. Large‐scale ocean‐atmospheric processes and seasonal rainfall variability in South Australia: accounting for non‐linearity and establishing the hierarchy of influence. International Journal of Climatology.

Tozer, C., Kiem, A., Verdon-Kidd, D., 2012. On the uncertainties associated with using gridded rainfall data as a proxy for observed. Hydrology and Earth System Sciences, 16(5): 1481.

Trenberth, K.E., 1997. The definition of el nino. Bulletin of the American Meteorological Society, 78(12): 2771-2777.

Trenberth, K.E., Hurrell, J.W., 1994. Decadal atmosphere-ocean variations in the Pacific. Climate Dynamics, 9(6): 303-319.

Troup, A., 1965. The ‘southern oscillation’. Quarterly Journal of the Royal Meteorological Society, 91(390): 490-506.

Uhlenbrook, S., Seibert, J., Leibundgut, C., Rodhe, A., 1999. Prediction uncertainty of conceptual rainfall-runoff models caused by problems in identifying model parameters and structure. Hydrological Sciences Journal, 44(5): 779-797.

Urrutia, R.B. et al., 2011. Multicentury tree ring reconstruction of annual streamflow for the Maule River watershed in south central Chile. Water Resources Research, 47(6).

Van Vuuren, D.P. et al., 2011. The representative concentration pathways: an overview. Climatic change, 109: 5-31.

Vance, T., Roberts, J., Plummer, C., Kiem, A., Van Ommen, T., 2015. Interdecadal Pacific variability and eastern Australian megadroughts over the last millennium. Geophysical Research Letters, 42(1): 129-137.

Verdon-Kidd, D., Kiem, A., 2014. Synchronicity of historical dry spells in the Southern Hemisphere. Hydrology and Earth System Sciences, 18(6): 2257.

Verdon-Kidd, D., Kiem, A., Moran, R., 2014. Links between the Big Dry in Australia and hemispheric multi-decadal climate variability. Hydrology and Earth System Sciences, 18(6): 2235.

144

Vichi, M. et al., 2011. Global and regional ocean carbon uptake and climate change: sensitivity to a substantial mitigation scenario. Climate Dynamics, 37(9-10): 1929-1947. DOI:10.1007/s00382-011-1079-0

Vicuña, S., Garreaud, R., McPhee, J., 2010. Climate change impacts on the hydrology of a snowmelt driven basin in semiarid Chile. Climatic Change, 105(3-4): 469-488. DOI:10.1007/s10584-010-9888-4

Villalba, R. et al., 2012. Unusual Southern Hemisphere tree growth patterns induced by changes in the Southern Annular Mode. Nature Geoscience, 5(11): 793-798.

Villalba, R., Veblen, T.T., 1997. Regional patterns of tree population age structures in northern Patagonia: climatic and disturbance influences. Journal of Ecology: 113-124.

Vogel, R.M., Sankarasubramanian, A., 2003. Validation of a watershed model without calibration. Water resources research, 39(10).

Voldoire, A. et al., 2013. The CNRM-CM5. 1 global climate model: description and basic evaluation. Climate Dynamics, 40(9-10): 2091-2121.

Volodin, E., Dianskii, N., Gusev, A., 2010. Simulating present-day climate with the INMCM4. 0 coupled model of the atmospheric and oceanic general circulations. Izvestiya, Atmospheric and Oceanic Physics, 46(4): 414-431.

Wang, G., Cai, W., Purich, A., 2014. Trends in Southern Hemisphere wind‐driven circulation in CMIP5 models over the 21st century: Ozone recovery versus greenhouse forcing. Journal of Geophysical Research: Oceans, 119(5): 2974-2986.

Wang, G., Hendon, H.H., 2007. Sensitivity of Australian rainfall to inter–El Nino variations. Journal of climate, 20(16): 4211-4226.

Watanabe, S. et al., 2011. MIROC-ESM 2010: model description and basic results of CMIP5- 20c3m experiments. Geoscientific Model Development, 4(4): 845-872. DOI:10.5194/gmd-4- 845-2011

Welch, P.D., 1967. The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging over short, modiﬁed periodograms. IEEE Transactions on audio and electroacoustics, 15(2): 70-73.

Westra, S., Renard, B., Thyer, M., 2015. The ENSO–precipitation teleconnection and its modulation by the interdecadal Pacific oscillation. Journal of Climate, 28(12): 4753-4773.

White, D.D., Wutich, A.Y., Larson, K.L., Lant, T., 2015. Water management decision makers' evaluations of uncertainty in a decision support system: the case of WaterSim in the Decision Theater. Journal of Environmental Planning and Management, 58(4): 616-630.

Wilks, D.S., 2011. Statistical methods in the atmospheric sciences, 100. Academic press.

Wolter, K., Timlin, M.S., 1993. Monitoring ENSO in COADS with a seasonally adjusted principal component index, Proc. of the 17th Climate Diagnostics Workshop.

145

Wolter, K., Timlin, M.S., 1998. Measuring the strength of ENSO events: How does 1997/98 rank? Weather, 53(9): 315-324.

Wu, T., 2012. A mass-flux cumulus parameterization scheme for large-scale models: Description and test with observations. Climate dynamics, 38(3-4): 725-744.

Xu, H., Taylor, R.G., Xu, Y., 2011. Quantifying uncertainty in the impacts of climate change on river discharge in sub-catchments of the Yangtze and Yellow River Basins, China. Hydrology and Earth System Sciences, 15(1): 333-344. DOI:10.5194/hess-15-333-2011

Yang, X.-Y., Huang, R.X., Wang, D.X., 2007. Decadal changes of wind stress over the Southern Ocean associated with Antarctic ozone depletion. Journal of Climate, 20(14): 3395- 3410.

Yevjevich, V., 1972. Probability and statistics in hydrology. Water resources publications Fort Collins, CO.

Yukimoto, S. et al., 2012. A New Global Climate Model of the Meteorological Research Institute: MRI-CGCM3 ^|^mdash;Model Description and Basic Performance^|^mdash. Journal of the Meteorological Society of Japan, 90A(0): 23-64. DOI:10.2151/jmsj.2012-A02

Zazulie, N., Rusticucci, M., Raga, G.B., 2017. Regional climate of the subtropical central Andes using high-resolution CMIP5 models—part I: past performance (1980–2005). Climate Dynamics: 1-21.

Zhang, D., Li, X., Liang, Y., 2003. Supplement of yearly charts of dryness/wetness in China for the last 500-year period, 1993–2000. J Appl Meteorol Sci, 14: 379-389.

Zhang, Y. et al., 2013. Collation of Australian modeller’s streamflow dataset for 780 unregulated Australian catchments. Water for a Healthy Country National Research Flagship, 115pp. Catchment Management.

Zhao, Y. et al., 2016. Detailed dynamic land cover mapping of Chile: Accuracy improvement by integrating multi-temporal data. Remote Sensing of Environment, 183: 170-185.

Zhu, Y., Yang, X., 2003. Relationships between Pacific Decadal Oscillation (PDO) and climate variabilities in China. Acta Meteorologica Sinica, 61(6): 641-654.

146

APPENDICES

APPENDIX 1 DONNELLY AT STRICKLAND OBSERVED TIME SERIES Annual Annual Annual temperature Years precipitation runoff (°C) (mm) (mm) 1961 15.42 1088.4 212.3 1962 15.48 1026.2 174.6 1963 15.41 1233.4 322.8 1964 14.33 1163.8 270.9 1965 14.98 1170 254.1 1966 14.63 879.7 163.4 1967 15.35 1052.2 256 1968 13.83 990.7 201.1 1969 15.07 665.9 63.4 1970 15.23 1022.6 179.6 1971 14.22 1024.4 172.4 1972 15.74 767.9 100.7 1973 14.85 1121.1 231.2 1974 15.21 1022 213.8 1975 15.19 913 135 1976 15.72 974.5 85 1977 15.68 843.8 97.9 1978 15.91 997 178.2 1979 15.28 838.9 94.5 1980 15.13 918.3 123.6 1981 14.81 1055.3 197.7 1982 15.06 785.8 83.3 1983 15.95 934.8 168.9 1984 15.05 1008.3 166.9 1985 15.49 890.1 109.2 1986 14.63 776.2 76.6 1987 15.28 694.9 40.1 1988 15.99 1119.4 258.2 1989 15.37 916.3 100 1990 14.86 984.8 149.3 1991 15.54 997.2 164.7 1992 15.25 1014.6 155.6

147

APPENDIX 2 HELENA AT NGANGAGURINGURING OBSERVED TIME SERIES Annual Annual Annual temperature Years precipitation runoff (°C) (mm) (mm) 1973 16.59 779.63 8.53 1974 16.71 830.80 24.70 1975 16.69 654.49 7.81 1976 17.31 642.44 2.09 1977 17.47 550.74 1.74 1978 17.56 648.47 5.30 1979 16.84 522.16 1.66 1980 16.92 578.30 1.70 1981 16.63 761.30 17.51 1982 16.93 604.68 3.08 1983 17.56 802.37 12.73 1984 16.56 672.81 3.18 1985 17.39 540.85 2.56 1986 16.19 655.99 6.76 1987 17.04 583.61 2.63 1988 17.49 716.71 5.93 1989 16.89 554.90 2.38 1990 16.45 736.21 4.96 1991 17.33 722.17 5.76 1992 16.91 763.19 5.62 1993 16.61 599.52 2.37 1994 17.58 453.19 2.50 1995 17.15 765.62 8.03 1996 17.39 821.12 17.87 1997 17.23 590.62 3.59 1998 17.26 622.56 3.99 1999 17.50 801.95 7.85 2000 17.12 634.31 5.15

148

APPENDIX 3 DENMARK AT KOMPUP OBSERVED TIME SERIES Annual Annual Annual temperature Years precipitation runoff (°C) (mm) (mm) 1961 15.80 967.75 116.72 1962 15.66 822.80 43.97 1963 15.67 903.57 64.48 1964 14.63 953.46 91.35 1965 15.29 994.18 45.23 1966 14.95 762.29 50.02 1967 15.51 859.68 104.58 1968 14.22 809.92 73.20 1969 15.30 663.26 19.19 1970 15.14 880.44 49.40 1971 14.53 1014.92 78.21 1972 15.93 635.27 22.17 1973 15.01 915.87 61.85 1974 15.29 753.67 30.00 1975 15.23 777.90 41.69 1976 15.38 888.67 41.66 1977 15.47 821.71 80.34 1978 15.81 951.90 159.07 1979 15.26 858.78 68.17 1980 15.18 782.71 44.22 1981 14.89 807.73 80.77 1982 15.31 724.54 18.08 1983 16.10 735.08 25.70 1984 15.19 844.19 88.14 1985 15.59 788.05 38.81 1986 14.74 678.70 19.92 1987 15.36 694.52 9.47 1988 15.60 1005.98 160.81 1989 15.45 925.88 51.15 1990 14.95 855.95 62.40 1991 15.69 838.06 56.73 1992 15.11 940.00 78.57 1993 14.73 829.57 59.99 1994 15.81 661.95 28.15 1995 15.41 849.13 34.16 1996 15.39 918.41 62.00 1997 15.63 718.45 34.34 1998 15.30 884.83 59.07 1999 15.66 841.21 44.17 2000 15.52 826.16 23.59

149

APPENDIX 4 CAUQUENES AT CAUQUENES OBSERVED TIME SERIES Annual Annual Annual temperature Years precipitation runoff (°C) (mm) (mm) 1966 8.40 855.50 414.27 1967 8.23 645.50 294.29 1968 8.70 408.50 45.67 1969 8.33 680.00 439.74 1970 8.22 573.80 475.26 1971 8.08 782.70 413.06 1972 8.73 1279.50 1371.15 1973 7.88 464.50 294.54 1974 7.97 668.70 271.43 1975 8.24 700.80 526.03 1976 8.33 610.40 278.87 1977 9.08 913.30 603.00 1978 8.99 609.80 342.94 1979 8.56 654.00 295.88 1980 9.13 1015.10 645.42 1981 9.15 724.70 500.86 1982 9.22 1061.00 988.00 1983 8.77 592.50 316.82 1984 8.48 989.50 621.54 1985 8.85 511.00 210.13 1986 9.01 1001.90 621.92 1987 9.38 837.50 705.53 1988 8.63 531.10 440.77 1989 8.98 492.50 254.69 1990 8.61 541.70 93.85 1991 8.88 732.60 414.13 1992 8.91 1151.40 767.07 1993 9.18 752.50 365.38 1994 9.21 475.80 266.03 1995 8.83 679.60 399.26 1996 9.27 560.60 176.08 1997 9.73 899.80 646.42 1998 9.03 270.90 65.86 1999 9.04 599.50 395.64 2000 8.89 753.30 562.93

150

APPENDIX 5 CATO AT PUENTE CATO OBSERVED TIME SERIES Annual Annual Annual temperature Years precipitation runoff (°C) (mm) (mm) 1966 8.40 1930.40 594.00 1967 8.23 1204.00 357.11 1968 8.70 1004.30 104.51 1969 8.33 1790.70 611.82 1970 8.22 1283.50 432.10 1971 8.08 1556.80 498.60 1972 8.73 2315.00 951.89 1973 7.88 1384.90 363.23 1974 7.97 1444.70 384.95 1975 8.24 1782.00 529.37 1976 8.33 1375.50 283.71 1977 9.08 1939.90 593.25 1978 8.99 1781.70 486.17 1979 8.56 1843.50 507.70 1980 9.13 2232.00 719.40 1981 9.15 1590.90 514.90 1982 9.22 2425.60 846.81 1983 8.77 1305.50 392.24 1984 8.48 1879.00 544.00 1985 8.85 1605.60 399.80 1986 9.01 2134.50 671.00 1987 9.38 1666.90 434.48 1988 8.63 1331.90 349.00 1989 8.98 1136.00 333.20 1990 8.61 1340.60 253.00 1991 8.88 1883.40 498.80 1992 8.91 1973.10 643.30 1993 9.18 2016.70 596.50 1994 9.21 1727.50 423.60 1995 8.83 1669.30 499.20 1996 9.27 1130.90 212.30 1997 9.73 2025.90 609.10 1998 9.03 738.90 152.40 1999 9.04 1395.50 308.60 2000 8.89 2144.00 579.20 2001 8.58 2175.30 805.50 2002 8.30 2501.90 717.50 2003 8.98 1372.20 311.40 2004 9.11 1644.80 425.80 2005 8.92 2073.50 539.60

151

APPENDIX 6 LUMACO AT LUMACO OBSERVED TIME SERIES Annual Annual Annual temperature Years precipitation runoff (°C) (mm) (mm) 1966 9.72 1188.45 707.98 1967 9.55 1124.92 694.97 1968 10.02 974.32 225.46 1969 9.65 1111.89 733.08 1970 9.54 1044.16 587.43 1971 9.40 1048.37 602.66 1972 10.05 1490.92 891.42 1973 9.20 1082.01 545.08 1974 9.29 955.82 476.58 1975 9.56 1054.44 768.48 1976 9.65 873.96 419.04 1977 10.40 1415.21 953.58 1978 10.31 1295.24 892.02 1979 9.88 1095.41 529.37 1980 10.45 1466.78 1040.27 1981 10.47 1137.55 655.69 1982 10.54 1367.76 853.00 1983 10.09 934.20 494.58 1984 9.80 1231.03 757.05 1985 10.17 997.05 593.43 1986 10.33 1187.77 656.92 1987 10.70 951.18 496.69 1988 9.95 932.02 351.11 1989 10.30 993.58 448.91 1990 9.93 963.49 617.06 1991 10.20 1124.09 627.18 1992 10.23 1266.20 750.53 1993 10.50 1305.83 1013.01 1994 10.53 959.67 576.16 1995 10.15 1068.07 752.92 1996 10.59 833.74 210.02 1997 11.05 1456.68 767.52 1998 10.35 441.47 212.08 1999 10.36 823.04 451.77 2000 10.21 696.58 618.79 2001 9.90 454.33 711.61 2002 9.62 1203.41 588.69 2003 10.30 686.49 527.63 2004 10.43 1159.32 557.26 2005 10.24 732.67 745.58

152

APPENDIX 7 INFLOWS TO RALCO DAM OBSERVED TIME SERIES Annual Annual Annual temperatura Years precipitation runoff (°C) (mm) (mm) 1966 6.26 2823.17 1825.89 1967 6.12 1817.52 1540.32 1968 6.37 1548.52 831.76 1969 5.72 2342.92 1770.79 1970 5.91 1871.39 1379.68 1971 5.35 2108.48 1637.50 1972 6.09 3142.08 2188.79 1973 5.55 1841.13 1353.54 1974 5.64 1972.25 1205.48 1975 5.77 2401.25 1627.33 1976 5.70 1990.68 1348.07 1977 6.71 2730.48 1991.10 1978 6.49 2504.94 1802.31 1979 6.18 2655.03 1611.79 1980 6.27 2979.28 2070.90 1981 6.35 2149.49 1823.72 1982 6.50 3325.05 2054.28 1983 6.20 1770.03 1429.63 1984 5.93 2535.67 1889.10 1985 6.44 2250.34 1475.97 1986 6.15 2922.62 1810.87 1987 6.62 2257.96 1480.10 1988 5.82 1799.98 1160.71 1989 6.26 1553.29 1150.63 1990 6.09 1873.02 1369.93 1991 6.04 2634.89 1574.79 1992 6.19 2652.55 1873.07 1993 6.35 2684.89 2440.87 1994 6.53 2356.47 1987.39 1995 5.93 2152.82 1872.73 1996 6.19 1534.87 833.17 1997 6.88 2775.07 1914.47 1998 6.49 1007.87 612.04 1999 6.51 1791.77 1221.33 2000 5.99 2815.19 1598.86

153

APPENDIX 8 INFLOWS TO LAJA DAM OBSERVED TIME SERIES Annual Annual Annual temperature Years precipitation runoff (°C) (mm) (mm) 1966 4.86 2778.39 1951.99 1967 4.72 1789.27 1749.89 1968 4.97 1402.43 989.48 1969 4.32 2379.60 2045.43 1970 4.51 1779.64 1530.26 1971 3.95 2166.04 1912.03 1972 4.69 3179.68 2679.32 1973 4.15 1573.34 1477.92 1974 4.24 1938.46 1465.48 1975 4.37 2183.62 1981.58 1976 4.30 2213.14 1614.18 1977 5.31 3056.82 2176.48 1978 5.09 3005.78 2149.31 1979 4.78 3128.11 1933.98 1980 4.87 3353.42 2423.91 1981 4.95 2256.78 1851.48 1982 5.10 3643.51 2447.16 1983 4.80 1752.78 1660.13 1984 4.53 2175.72 1942.97 1985 5.04 2118.55 1734.66 1986 4.75 3001.83 2178.18 1987 5.22 2059.01 1790.36 1988 4.42 1697.25 1491.82 1989 4.86 1532.41 1222.97 1990 4.69 1979.04 1430.23 1991 4.64 2683.56 1948.69 1992 4.79 2796.72 2056.44 1993 4.95 2854.81 2524.01 1994 5.13 2087.15 1934.52 1995 4.53 1942.60 1859.28 1996 4.79 1405.39 996.82 1997 5.48 2175.22 2269.89 1998 5.09 971.35 751.68 1999 5.11 2257.37 1541.66 2000 4.59 2677.39 2108.86

154

APPENDIX 9 DONNELLY SEASONAL TRENDS

155

APPENDIX 10 HELENA SEASONAL TRENDS

156

APPENDIX 11 DENMARK SEASONAL TRENDS

157

APPENDIX 12 CAUQUENES SEASONAL TRENDS

158

APPENDIX 13 CATO SEASONAL TRENDS

159

APPENDIX 14 LUMACO SEASONAL TRENDS

160

APPENDIX 15 LAJA SEASONAL TRENDS

161

APPENDIX 16 RALCO SEASONAL TRENDS

162

APPENDIX 17 LARGE SCALE CLIMATIC FEATURES RECONSTRUCTIONS PDO Shen et al. PDO D'Arrigo et al. PDO D'Arrigo SAM Villalba a) SAM Villalba Year (2006) a)(2001) et al. b (2001) (2012) b) (2012)

1470 -1.4246 -4.5923 -3.2039

1471 -0.8716 -4.514 -3.1539

1472 -0.5924 -5.8013 -3.9762

1473 -1.755 -5.3403 -3.6817

1474 -0.7191 -5.5578 -3.8206

1475 -0.6007 -4.9228 -3.4151

1476 -0.266 -4.8706 -3.3817

1477 -1.1693 -3.8703 -2.7428

1478 -1.3882 -3.8094 -2.7039

1479 -0.0778 -5.1142 -3.5373

1480 -0.079 -2.9483 -2.1539

1481 0.1526 -4.0008 -2.8261

1482 -0.8509 -4.4444 -3.1095

1483 0.4157 -2.7831 -2.0483

1484 0.4843 -4.3748 -3.065

1485 0.111 -4.8358 -3.3595

1486 -0.2059 -2.8581 -2.0963

1487 0.0771 -2.3475 -1.7701

1488 0.5369 -3.0379 -2.2111

1489 -0.1998 -3.9522 -2.7951

1490 0.0714 -3.6878 -2.6262

1491 -0.3482 -2.6411 -1.9577

1492 -0.1153 -3.2068 -2.319

1493 0.2918 -2.3983 -1.8026

1494 -0.4221 -1.4439 -1.1929

1495 0.457 -2.6887 -1.988

1496 -0.0298 -3.127 -2.268

1497 -0.4208 -2.5888 -1.9242

1498 -0.0902 -2.8956 -2.1202

1499 -0.867 -2.9108 -2.1299

1500 1.143 -1.6971 -1.3547

1501 -0.4014 -1.7769 -1.4057

1502 -0.8321 -2.0679 -1.5915

1503 0.6446 -2.2898 -1.7332

1504 -0.177 -3.028 -2.2047

1505 0.6113 -3.7796 -2.6848

1506 -2.0044 -2.7627 -2.0353

1507 0.3455 -4.0374 -2.8495

1508 0.8425 -3.3595 -2.4165

1509 0.263 -4.4293 -3.0999

1510 -1.5924 -3.7838 -2.6875

163

1511 -1.079 -1.7977 -1.419

1512 1.682 -2.494 -1.8637

1513 1.067 -1.9222 -1.4984

1514 -0.6402 -3.1286 -2.269

1515 0.2546 -3.0572 -2.2234

1516 -0.2907 -1.4695 -1.2093

1517 -1.1354 -3.4147 -2.4518

1518 -0.9426 -3.0081 -2.1921

1519 -0.5358 -2.8301 -2.0784

1520 0.8511 -2.5439 -1.8956

1521 -0.2658 -2.9914 -2.1814

1522 -0.4794 -3.5056 -2.5098

1523 -0.8976 -2.7822 -2.0478

1524 0.545 -3.6656 -2.612

1525 -1.5796 -4.4754 -3.1293

1526 0.1792 -3.7992 -2.6974

1527 2.0201 -2.3499 -1.7716

1528 0.5035 -2.195 -1.6727

1529 0.9401 -2.7224 -2.0096

1530 0.8194 -1.6835 -1.346

1531 -1.3908 -1.9341 -1.506

1532 -0.2445 -1.8558 -1.456

1533 0.0716 -1.8105 -1.4271

1534 -0.2237 -3.3714 -2.4241

1535 0.3695 -3.8584 -2.7352

1536 -0.3518 -3.9911 -2.8199

1537 0.6821 -3.6046 -2.5731

1538 1.4428 -3.953 -2.7956

1539 -0.7767 -2.711 -2.0023

1540 0.1267 -2.8963 -2.1207

1541 1.2661 -3.7826 -2.6868

1542 1.6084 -4.5064 -3.1491

1543 -0.9006 -4.0677 -2.8689

1544 0.0149 -3.6452 -2.599

1545 0.3119 -2.6622 -1.9711

1546 -2.3305 -2.0696 -1.5926

1547 0.2607 -2.4831 -1.8567

1548 0.2574 -2.0954 -1.6091

1549 -0.019 -1.947 -1.5143

1550 0.3378 -3.5385 -2.5309

1551 -0.2189 -3.3743 -2.426

1552 -0.9682 -1.9438 -1.5122

1553 -1.6286 -2.8569 -2.0955

1554 -1.602 -3.5417 -2.5329 164

1555 -1.364 -2.5108 -1.8744

1556 -1.4018 -3.1996 -2.3144

1557 -0.9752 -3.3193 -2.3908

1558 0.179 -4.5926 -3.2042

1559 0.0441 -4.4544 -3.1159

1560 0.2844 -3.5741 -2.5536

1561 -0.3134 -4.6042 -3.2115

1562 -0.5919 -4.1206 -2.9026

1563 -1.032 -4.5749 -3.1928

1564 1.2726 -3.2546 -2.3495

1565 -0.2746 -2.7785 -2.0454

1566 0.0146 -2.508 -1.8726

1567 -1.1731 -3.8113 -2.7051

1568 1.8668 -3.3545 -2.4133

1569 -1.3024 -3.5798 -2.5572

1570 -0.0146 -2.784 -2.0489

1571 -0.6892 -2.8141 -2.0682

1572 0.2904 -1.849 -1.4517

1573 1.0874 -2.0236 -1.5632

1574 -0.188 -3.7695 -2.6784

1575 -0.9248 -2.8857 -2.1139

1576 -0.442 -1.3246 -1.1168

1577 0.551 -2.9671 -2.1659

1578 -0.4104 -1.8978 -1.4829

1579 -1.3223 -2.3307 -1.7594

1580 0.4654 -2.1509 -1.6445

1581 -0.1983 -2.9718 -2.1689

1582 -0.8563 -2.1878 -1.6681

1583 -1.4375 -3.1251 -2.2668

1584 0.9561 -4.4155 -3.091

1585 0.7524 -3.7109 -2.641

1586 -0.6625 -3.5714 -2.5519

1587 -0.5332 -3.5857 -2.561

1588 -0.138 -3.5312 -2.5262

1589 -0.4114 -3.6873 -2.6259

1590 -0.208 -4.63 -3.228

1591 -0.6097 -4.8909 -3.3947

1592 -1.6901 -3.3015 -2.3795

1593 -1.3347 -3.0747 -2.2346

1594 -0.3999 -3.5438 -2.5343

1595 0.2851 -2.7734 -2.0422

1596 -0.733 -4.5375 -3.1689

1597 -0.1919 -3.8808 -2.7495

1598 -0.5633 -3.1965 -2.3124 165

1599 -0.5638 -1.3564 -1.1371

1600 0.3147 -3.7748 -2.6818

1601 -1.4032 -3.5199 -2.519

1602 0.4625 -2.3062 -1.7437

1603 -0.0482 -1.6265 -1.3096

1604 -0.2447 -2.0626 -1.5881

1605 -0.3427 -2.9795 -2.1738

1606 -0.0925 -3.2482 -2.3454

1607 0.7463 -3.231 -2.3345

1608 0.7528 -1.9648 -1.5257

1609 0.9011 -4.1048 -2.8925

1610 0.6624 -4.1326 -2.9103

1611 0.3185 -2.1812 -1.6639

1612 0.7129 -1.59 -1.2863

1613 -1.4311 -2.3653 -1.7814

1614 0.9599 -2.9612 -2.1621

1615 0.7091 -2.6626 -1.9714

1616 0.3653 -3.1134 -2.2593

1617 0.3465 -3.1587 -2.2882

1618 0.4522 -2.981 -2.1747

1619 1.412 -4.0343 -2.8475

1620 0.0212 -3.3227 -2.393

1621 0.4246 -3.6249 -2.586

1622 0.7036 -2.9031 -2.125

1623 -0.0028 -3.2597 -2.3528

1624 -0.4844 -2.3521 -1.773

1625 -0.3592 -1.8072 -1.425

1626 0.303 -2.1794 -1.6628

1627 0.0144 -1.543 -1.2562

1628 0.1855 -2.4432 -1.8312

1629 -0.0355 -1.411 -1.172

1630 0.1264 -2.3916 -1.7983

1631 -0.8978 -3.1642 -2.2918

1632 -1.1746 -1.2511 -1.0698

1633 -0.8936 -2.6824 -1.984

1634 -0.6554 -4.0642 -2.8666

1635 1.0248 -4.5501 -3.177

1636 -0.5961 -4.1734 -2.9364

1637 0.767 -2.8004 -2.0594

1638 0.1615 -3.3256 -2.3949

1639 0.0972 -2.8737 -2.1062

1640 1.4641 -2.0186 -1.5601

1641 0.4903 -2.4439 -1.8317

1642 -0.3387 -3.4008 -2.4429 166

1643 -0.7409 -2.8571 -2.0956

1644 0.4067 -1.8567 -1.4566

1645 -0.2377 -1.5958 -1.2899

1646 -0.71 -1.6881 -1.3489

1647 -0.5654 -1.9471 -1.5144

1648 -0.69 -3.3909 -2.4366

1649 -1.2606 -2.6188 -1.9434

1650 -1.1468 -3.2161 -2.3249

1651 -0.4428 -1.7628 -1.3966

1652 0.1198 -3.2587 -2.3521

1653 0.4982 -3.5269 -2.5234

1654 -0.3981 -3.7276 -2.6516

1655 -0.5394 -1.5632 -1.2692

1656 1.0044 -2.1322 -1.6326

1657 -0.5216 -1.9831 -1.5374

1658 -1.2496 -3.6275 -2.5877

1659 -0.6701 -3.4311 -2.4622

1660 -1.1094 -3.6614 -2.6094

1661 0.1154 -2.6185 -1.9432

1662 -1.2141 -1.6909 -1.3507

1663 0.4225 -3.1471 -2.2808

1664 0.9525 -1.9509 -1.5168

1665 -0.9182 -1.9237 -1.4994

1666 -0.6755 -1.5506 -1.2611

1667 2.3184 -1.9841 -1.538

1668 -0.5457 -2.461 -1.8426

1669 -0.8504 -2.1085 -1.6175

1670 -0.7856 -2.5726 -1.9139

1671 0.2239 -2.8006 -2.0595

1672 -1.2929 -2.5792 -1.9181

1673 -1.1004 -2.0178 -1.5595

1674 -1.3252 -2.8517 -2.0922

1675 -0.938 -1.798 -1.4191

1676 0.5682 -2.8451 -2.0879

1677 -0.7598 -2.3865 -1.795

1678 -0.3946 -3.0542 -2.2215

1679 0.7303 -3.3636 -2.4191

1680 -0.7428 -2.4003 -1.8038

1681 0.8753 -1.4979 -1.2274

1682 -0.0081 0.2299 -0.1238

1683 0.5761 -1.6214 -1.3063

1684 0.0122 -2.6559 -1.9671

1685 -0.8729 -2.3804 -1.7911

1686 0.5237 -1.4333 -1.1862 167

1687 0.4361 -1.3813 -1.1529

1688 -0.0588 -1.9342 -1.5061

1689 1.5398 -1.9674 -1.5273

1690 -1.4518 -1.7519 -1.3897

1691 -0.084 -1.8043 -1.4231

1692 0.7295 -2.3827 -1.7926

1693 0.0405 -1.0092 -0.9153

1694 -0.0688 -3.1533 -2.2848

1695 -0.0354 -2.9005 -2.1233

1696 -0.367 -1.6168 -1.3034

1697 0.8761 -2.1235 -1.627

1698 -0.4828 -1.9442 -1.5125

1699 -0.6901 -1.6096 -1.2988 1700 -0.1711 0.2146 -2.8389 -2.084 1701 -0.2995 -0.1672 -2.6759 -1.9799 1702 0.2382 -0.4883 -2.8042 -2.0618 1703 0.5799 0.2607 -3.486 -2.4973 1704 -0.8199 -0.277 -3.596 -2.5675 1705 -0.7684 -0.0881 -1.8087 -1.4259 1706 -0.8419 0.5387 -1.8696 -1.4649 1707 0.3978 -0.3202 -2.6716 -1.9772 1708 0.7953 -0.2541 -3.595 -2.567 1709 -1.3689 -0.745 -3.3631 -2.4188 1710 -1.4736 0.5914 -2.3692 -1.7839 1711 0.9221 0.3363 -2.2219 -1.6899 1712 0.8006 0.9336 -1.9691 -1.5284 1713 0.7884 0.5002 -2.103 -1.6139 1714 -0.0368 0.1069 -1.4626 -1.2049 1715 0.1133 -0.1303 -2.1928 -1.6713 1716 1.1373 0.0138 -0.6565 -0.69 1717 -0.2804 0.1495 -1.2396 -1.0625 1718 -0.1856 0.769 -1.5246 -1.2445 1719 -1.7821 0.8237 -0.9925 -0.9046 1720 -1.7715 0.3046 -3.5061 -2.5101 1721 0.1631 0.6899 -2.9346 -2.1451 1722 1.5287 -0.1422 -2.3988 -1.8029 1723 0.9438 0.8376 -2.7222 -2.0094 1724 -0.6007 -0.8493 -2.4931 -1.8631 1725 -0.5511 -0.0375 -2.8611 -2.0982 1726 -0.8188 -0.4278 -3.0039 -2.1894 1727 -0.2901 -0.1475 -2.0409 -1.5743 1728 -0.4113 -0.1632 -2.8322 -2.0797 1729 -0.4324 -0.3941 -2.3257 -1.7562 1730 -0.0661 -0.6755 -1.9709 -1.5295 168

1731 1.1118 -0.5048 -2.0717 -1.594 1732 0.2484 -0.5657 -2.2655 -1.7178 1733 -2.0403 -0.3386 -1.7574 -1.3932 1734 -0.4733 0.2246 -2.038 -1.5724 1735 -0.6154 0.5735 -1.8071 -1.4249 1736 -1.242 0.2579 -2.4292 -1.8223 1737 -0.9991 0.6714 -1.2961 -1.0985 1738 0.2787 0.1345 -1.774 -1.4038 1739 -0.1073 -0.9327 -2.171 -1.6574 1740 0.2959 0.2925 -2.446 -1.833 1741 -1.2424 -0.532 -2.4136 -1.8124 1742 -0.8124 -1.3111 -2.1911 -1.6702 1743 0.6045 -0.9875 -0.5931 -0.6495 1744 0.0747 -1.1283 -1.6823 -1.3452 1745 -0.4191 -0.6958 -1.8136 -1.4291 1746 -0.8942 -0.0825 -1.5239 -1.244 1747 0.0546 0.6762 -1.0469 -0.9394 1748 -0.0815 -0.4642 -1.1611 -1.0123 1749 -0.8853 -0.7148 -3.8407 -2.7238 1750 -1.6281 -0.6263 -3.3565 -2.4146 1751 -0.5386 -0.1207 -1.1706 -1.0184 1752 -0.713 -1.8356 -0.6864 -0.7091 1753 -0.9463 -0.3575 -1.5054 -1.2322 1754 -1.2473 -1.8575 -2.1389 -1.6369 1755 -1.0063 -0.6508 -1.2471 -1.0673 1756 -0.1956 -0.3883 -2.0002 -1.5483 1757 0.0676 -0.4432 -2.1274 -1.6295 1758 -0.4628 -0.0681 -2.6007 -1.9318 1759 0.4575 0.1656 -2.2708 -1.7211 1760 -0.6319 -0.0782 -3.3367 -2.402 1761 -0.7096 0.442 -3.0424 -2.214 1762 -0.7023 0.2266 -1.1795 -1.0241 1763 -0.324 -0.2611 -2.6917 -1.9899 1764 -0.434 -0.4056 -2.5201 -1.8804 1765 -0.4087 0.0735 -2.0713 -1.5937 1766 0.4219 -0.0531 -2.4472 -1.8338 1767 -0.1584 -0.5114 -1.0901 -0.967 1768 0.4379 0.0377 -1.8376 -1.4444 1769 0.1139 0.0578 -2.7771 -2.0445 1770 0.5527 -0.0695 -2.145 -1.6408 1771 -1.1677 0.1445 -2.166 -1.6542 1772 -0.2049 -0.1296 -1.3809 -1.1527 1773 -1.0926 0.1159 -1.6835 -1.346 1774 0.3164 0.4607 -2.2034 -1.6781 169

1775 0.2148 0.4345 -2.4697 -1.8481 1776 -1.0896 0.2466 -2.5797 -1.9184 1777 -0.3993 0.2321 -2.1196 -1.6245 1778 -0.0321 -0.8358 -2.2534 -1.71 1779 -1.4616 -0.8108 -1.2435 -1.065 1780 -0.3699 0.034 -1.3675 -1.1441 1781 -1.4782 -0.8829 -2.79 -2.0527 1782 -0.2281 -0.6094 -2.6092 -1.9372 1783 0.0297 -0.3197 -1.2313 -1.0572 1784 0.7164 -0.072 -2.9016 -2.1241 1785 0.5758 -1.0617 -2.0153 -1.5579 1786 -0.1772 -0.7987 -1.0384 -0.934 1787 -1.0884 -1.0543 -1.0474 -0.9397 1788 0.1713 0.2292 -2.5792 -1.9181 1789 -0.5641 -0.49 -1.8795 -1.4712 1790 0.8089 -0.6736 -1.5795 -3.1098 -2.257 1791 0.6732 0.1622 -0.6864 -3.2375 -2.3386 1792 1.9443 0.6586 0.096 -2.6323 -1.952 1793 -0.5092 0.9027 0.0857 -2.5616 -1.9069 1794 -0.4043 0.6692 -0.3098 -3.1471 -2.2808 1795 -0.6764 0.8864 -0.0851 -2.499 -1.8669 1796 -0.4748 0.69 -0.7514 -2.2596 -1.7139 1797 0.019 -0.2014 -0.8402 -2.8731 -2.1059 1798 -0.5044 0.1734 -0.9028 -2.7566 -2.0314 1799 -1.0972 0.0708 -0.6355 -2.6133 -1.9399 1800 0.2852 -0.1892 -0.3423 -2.5428 -1.8949 1801 -0.6794 -1.5441 -1.2383 -1.1563 -1.0093 1802 -0.0016 -0.0337 0.1934 -0.8837 -0.8351 1803 0.2002 0.0375 0.2673 -1.749 -1.3878 1804 0.529 -0.2977 0.2412 -1.5325 -1.2496 1805 -0.3442 -0.4026 -0.1057 -0.589 -0.6469 1806 -1.0227 -0.5954 -0.4608 -1.4508 -1.1973 1807 -0.0409 -0.2189 -0.3489 -1.4537 -1.1992 1808 0.3954 -0.599 -0.714 -1.7362 -1.3796 1809 -0.5525 -1.3357 -1.0206 -2.6204 -1.9445 1810 0.1397 -0.9488 -1.3919 -1.7618 -1.396 1811 -0.2937 -0.8736 -1.0822 -1.395 -1.1617 1812 -0.3649 -0.4501 -0.998 -2.0145 -1.5574 1813 -0.0883 -1.1997 -0.9381 0.3005 -0.0787 1814 0.8989 -0.4371 -0.2913 -0.9949 -0.9062 1815 -0.0786 -0.8092 -0.4846 -1.4689 -1.2089 1816 -0.4788 0.0926 0.6849 -2.0047 -1.5512 1817 0.5254 -0.7573 -0.7372 -2.2706 -1.721 1818 0.9269 -0.3871 -0.4324 -1.7653 -1.3983 170

1819 -0.5923 -1.1192 -1.3887 -0.5056 -0.5936 1820 0.037 -0.7833 -1.2187 -0.8844 -0.8355 1821 1.0341 -0.845 -0.8523 -0.2821 -0.4508 1822 -0.7067 -0.2557 -0.2197 -1.7503 -1.3887 1823 -0.3281 -1.2144 -0.8186 -1.9398 -1.5097 1824 0.5405 -0.6909 -0.2722 -2.153 -1.6459 1825 0.4054 -0.0945 0.4308 -1.351 -1.1336 1826 -1.6512 0.3806 0.4189 -2.23 -1.6951 1827 -0.0802 -0.0807 -0.1614 -1.0445 -0.9379 1828 -0.5415 0.1749 0.2965 -2.091 -1.6063 1829 -0.5076 0.494 0.8233 -2.743 -2.0228 1830 0.8295 0.2117 -0.1678 -2.2667 -1.7185 1831 -1.5668 0.3008 -0.344 -2.239 -1.7008 1832 0.0761 0.3592 -0.0357 -3.0269 -2.2041 1833 -0.9662 0.5665 0.2717 -3.707 -2.6385 1834 -0.8207 0.655 0.2011 -3.145 -2.2795 1835 -0.1229 0.7152 0.41 -2.4441 -1.8318 1836 -0.0014 0.1664 -0.0196 -1.6629 -1.3328 1837 0.5306 -0.5351 -0.413 -2.9291 -2.1416 1838 -0.2584 0.0741 0.0044 -3.2381 -2.339 1839 -0.4137 0.2318 0.4025 -1.4061 -1.1688 1840 -0.5829 -0.3719 -0.5638 -1.7537 -1.3908 1841 -2.2191 -0.7097 -0.7134 -0.0218 -0.2846 1842 0.0774 -0.475 -0.6161 -1.2939 -1.0971 1843 -1.027 -0.3515 -0.3006 -2.4248 -1.8195 1844 -0.0651 0.0099 0.1538 -2.2811 -1.7277 1845 -0.0339 0.3295 0.3242 -0.8113 -0.7889 1846 0.1268 0.7414 0.1717 -1.8936 -1.4802 1847 -0.0072 0.0792 -0.4724 -1.1596 -1.0113 1848 0.3836 0.8059 0.5676 -1.8135 -1.429 1849 -1.3549 -0.4623 -0.5563 -1.9313 -1.5043 1850 -0.5409 0.6792 0.0889 -1.0593 -0.9473 1851 -0.678 0.2047 -0.0507 -0.4288 -0.5446 1852 0.2365 0.1928 -0.1269 -2.3743 -1.7873 1853 -1.01 0.4274 0.006 -1.0233 -0.9243 1854 -0.5586 -0.5223 -0.9022 -0.8273 -0.7991 1855 -1.5305 -0.3447 -0.5251 -1.4614 -1.2041 1856 0.0109 -0.073 -0.245 -2.3309 -1.7595 1857 -0.8265 -0.5542 -0.9064 -2.8524 -2.0926 1858 -1.0024 -0.0029 -0.1929 -2.254 -1.7104 1859 0.5591 -0.8813 -1.1657 -0.7111 -0.7249 1860 -0.129 -0.5781 -0.7854 -1.0876 -0.9654 1861 0.686 -0.6466 -0.2584 0.1476 -0.1764 1862 0.3149 -1.4419 -1.4847 -0.8563 -0.8176 171

1863 -0.6028 -0.7267 -1.0043 -1.8859 -1.4753 1864 -0.0146 -0.5703 -0.6203 -1.4591 -1.2027 1865 -0.7604 -0.8129 -0.6317 -0.5001 -0.5901 1866 -1.0098 -0.6942 -0.6282 -1.0439 -0.9374 1867 2.148 -0.9659 -0.6015 -0.95 -0.8775 1868 0.3647 -0.3858 -0.1778 -3.2461 -2.3441 1869 -0.3676 0.5139 0.3646 -2.3588 -1.7773 1870 1.1029 0.0055 -0.1151 -2.0701 -1.5929 1871 -0.9216 -0.1822 -0.5845 -1.1869 -1.0288 1872 -0.4332 -0.484 -0.6777 -2.2708 -1.7211 1873 -0.2187 -0.0446 -0.1292 -1.3768 -1.1501 1874 -0.0575 -0.1121 -0.0709 -1.0049 -0.9126 1875 0.5358 0.2379 -0.3024 0.2896 -0.0857 1876 0.8885 -0.0783 -1.0285 -1.4869 -1.2204 1877 0.4048 0.0971 -0.6164 -0.1639 -0.3754 1878 0.663 0.4098 0.0515 -2.3446 -1.7682 1879 -1.5886 -0.4461 -0.3513 -0.9787 -0.8958 1880 0.1783 -0.8713 -0.9321 -1.7586 -1.394 1881 -0.3564 -0.5019 -0.1183 -1.3815 -1.1531 1882 -1.6263 0.1816 -0.0735 -1.3596 -1.1391 1883 -1.0857 0.7077 -0.1856 -1.4781 -1.2148 1884 -0.4345 -0.3176 -0.4164 -2.1899 -1.6695 1885 0.6165 0.5021 0.46 -0.8131 -0.79 1886 -0.8396 -0.6465 -0.6663 -1.5191 -1.241 1887 -0.1309 -0.8785 -1.1488 -1.8767 -1.4694 1888 -0.4016 0.0874 0.0851 -1.7247 -1.3723 1889 0.391 0.3006 0.1658 -1.0275 -0.927 1890 -0.8384 -0.8726 -0.9201 -1.1603 -1.0118 1891 0.4631 -0.0334 -0.0124 -1.0471 -0.9395 1892 -1.2563 -0.3645 -0.2741 -1.4512 -1.1976 1893 -0.8435 -1.367 -1.1526 0.2269 -0.1258 1894 -0.3478 -0.7419 -0.8137 -0.7507 -0.7502 1895 -1.1072 -0.3321 -0.2558 -1.2588 -1.0747 1896 0.7643 -0.3948 -0.6409 -0.5302 -0.6093 1897 -1.4528 -0.0968 -0.4892 -0.2035 -0.4007 1898 -0.3737 -0.0868 -0.3275 -3.0555 -2.2223 1899 0.9969 -1.0041 -0.7815 -2.4745 -1.8513 1900 0.2559 0.1946 0.2345 -2.2889 -1.7327 1901 -0.7327 -0.2811 -0.1373 -2.1673 -1.655 1902 0.7982 -0.2787 0.0133 -2.3548 -1.7747 1903 -1.3816 0.5212 0.5933 -2.9287 -2.1414 1904 -0.4924 -0.2366 -0.5416 -2.1709 -1.6573 1905 -0.6687 0.3258 0.6994 -2.0787 -1.5984 1906 -0.986 0.0798 0.3459 -1.0596 -0.9475 172

1907 -0.447 -0.1294 -0.2582 -1.7019 -1.3577 1908 -1.0692 0.2262 0.2692 -0.8302 -0.801 1909 0.3116 -0.7731 -0.9629 0.159 -0.1691 1910 -0.0556 0.1081 -0.4307 -1.3749 -1.1489 1911 -1.2723 0.1341 0.0299 -1.0346 -0.9315 1912 -0.7372 -0.0905 -0.0847 -1.8814 -1.4724 1913 0.4449 -0.2455 -0.2309 0.0108 -0.2638 1914 0.2538 0.2053 0.0602 -1.0653 -0.9511 1915 -0.8509 0.6539 0.6162 -1.4141 -1.1739 1916 -0.3858 -0.4013 -0.4903 -1.0089 -0.9151 1917 -2.0125 0.094 -0.3807 -0.1737 -0.3817 1918 0.4467 -0.457 -0.5053 -1.7844 -1.4105 1919 0.1708 0.1869 0.2191 -1.3687 -1.1449 1920 -0.2413 -0.5776 -0.6022 -1.2496 -1.0688 1921 -0.4532 -0.1519 0.1811 -1.9462 -1.5138 1922 0.8698 -0.7322 -0.4165 -1.376 -1.1496 1923 -0.1535 0.1426 0.4377 -1.2806 -1.0886 1924 -0.7727 0.5881 0.7406 -0.5562 -0.6259 1925 -0.2286 -0.3254 0.31 -2.0886 -1.6048 1926 0.6174 0.9737 1.0996 -3.1356 -2.2735 1927 0.1622 -0.5474 -0.3116 -1.1968 -1.0351 1928 -0.2941 0.3458 0.5566 -2.3445 -1.7682 1929 0.4428 -0.6856 -0.1732 -3.0167 -2.1976 1930 -0.0893 0.3761 0.5454 -2.4857 -1.8584 1931 1.1199 1.2025 0.993 -2.7549 -2.0304 1932 0.0288 0.23 -0.3658 -1.6816 -1.3448 1933 -0.6746 0.0865 0.1626 -3.2711 -2.3601 1934 0.3596 0.4123 0.4425 -1.8535 -1.4546 1935 1.1192 0.3421 0.6397 -2.1721 -1.6581 1936 1.4532 1.1218 1.2125 -1.791 -1.4147 1937 -0.1314 0.0014 -0.0812 -1.3241 -1.1165 1938 0.109 0.2125 0.1608 -2.7155 -2.0052 1939 -0.0404 -0.3823 -0.1574 -2.5225 -1.8819 1940 0.9392 0.5073 0.4248 -2.624 -1.9467 1941 1.2597 1.0316 1.173 -2.7868 -2.0507 1942 0.936 0.5889 0.3722 -2.3041 -1.7424 1943 -0.0513 0.6074 0.3558 -0.7674 -0.7608 1944 -0.0892 0.1362 0.2432 -1.5363 -1.252 1945 -0.0581 -0.2986 -0.1844 -2.5774 -1.9169 1946 -0.6203 -0.7212 0.1071 -3.4793 -2.493 1947 0.3414 0.0036 0.1732 -1.9516 -1.5172 1948 -0.7052 -0.6723 -0.4418 -2.2703 -1.7208 1949 -0.8395 -0.3165 -0.4081 -1.5417 -1.2554 1950 -1.3505 -0.919 -0.7187 -1.4966 -1.2266 173

1951 0.1391 -0.1312 -0.2331 -3.023 -2.2016 1952 -0.1961 -0.265 -0.343 -1.4537 -1.1992 1953 -0.4507 -0.5143 -0.2779 -1.2053 -1.0406 1954 -1.2436 -1.0872 -0.918 -1.1048 -0.9763 1955 -1.468 -1.2527 -1.2802 -0.7411 -0.7441 1956 -1.5954 -0.5514 -0.7657 -0.8512 -0.8144 1957 -0.5474 0.0024 -0.1895 0.103 -0.2049 1958 1.0356 0.3191 -0.086 0.2021 -0.1416 1959 0.0716 -0.2864 -0.4743 -0.0494 -0.3022 1960 -0.2122 -0.2481 -0.4471 0.6362 0.1357 1961 -0.8575 0.0942 -0.0427 -0.4895 -0.5833 1962 -0.9366 -0.6851 -0.506 1.4274 0.641 1963 -0.6605 -0.0634 0.0827 -0.5196 -0.6026 1964 -1.0742 -0.7925 -0.4915 -2.083 -1.6012 1965 0.2455 -0.4461 -0.1385 -1.0916 -0.968 1966 -0.0151 -0.327 -0.2561 -0.9351 -0.868 1967 -0.1605 -0.1446 -0.5661 0.0603 -0.2322 1968 -0.4274 0.082 -0.4269 -0.4538 -0.5605 1969 -0.0101 -0.1571 -0.5642 -0.3628 -0.5024 1970 -0.4239 -0.3328 -0.7283 0.7523 0.2099 1971 -0.7566 -0.8897 -1.4218 -0.7648 -0.7592 1972 -1.1281 0.1462 -0.7258 -0.1906 -0.3924 1973 -0.2864 -0.0709 -0.4542 0.243 -0.1154 1974 -0.3343 -0.4046 -0.799 -0.525 -0.606 1975 -0.7228 -0.357 -0.2871 -1.0389 -0.9343 1976 -0.4293 -0.2611 -0.3117 -1.1575 -1.01 1977 -0.3864 0.1264 0.2734 -0.9295 -0.8644 1978 0.5923 -0.7105 -0.4575 -0.1329 -0.3556 1979 -0.1859 0.2328 0.7493 -0.2441 -0.4266

1980 0.277 -0.344 -0.4904

1981 0.7716 -0.0064489 -0.2748

1982 0.474 -0.7947 -0.7783

1983 2.2808 -1.1674 -1.0163

1984 0.8282 -1.5869 -1.2843

1985 0.0078 -1.5753 -1.2769

1986 1.1053 -0.9546 -0.8804

1987 1.6284 0.1692 -0.1626

1988 0.5323 -0.19 -0.3921

1989 0.3947 0.5352 0.0712

1990 -0.3847 0.327 -0.0618

1991 -0.0239 -0.9248 -0.8614

1992 0.944 -1.2765 -1.086

1993 1.1111 -0.9323 -0.8662

1994 0.293 0.0211 -0.2572 174

1995 0.875 0.3925 -0.02

1996 0.2616 0.7472 0.2066

1997 1.1826 0.6496 0.1442

1998 0.3769 2.1845 1.1246 1999 3.03 1.6647 2000 0.1633 -0.1664 2001 0.3071 -0.0745 2002 0.7997 0.2401 2003 0.0468 -0.2408

2004 1.5138 0.6963 2005 1.2964 0.5574 2006 1.5051 0.6907

APPENDIX 18 HISTOGRAMS OF ANNUAL CHANGES IN RUNOFF CONSIDERING ALL THE SIMULATIONS OF CPDN AND THE GROUPS OF SIMULATIONS WITH DIFFERENT PERTURBATIONS OF 4 ATMOSPHERIC PARAMETERS. RED LINE REPRESENTS THE MEDIAN OF THE WHOLE ENSEMBLE AND DOTTED RED LINE THE MEDIAN OF THE SIMULATIONS FOR A PARTICULAR PERTURBATION

175

176

177