Precipitation Seasonality Over the Indian Subcontinent: an Evaluation of Gauge, Reanalyses, and Satellite Retrievals

APRIL 2015 R A N A E T A L . 631

Precipitation Seasonality over the Indian Subcontinent: An Evaluation of Gauge, Reanalyses, and Satellite Retrievals

SAPNA RANA,JAMES MCGREGOR, AND JAMES RENWICK School of Geography, Environment and Earth Sciences, Victoria University of Wellington, Wellington, New Zealand

(Manuscript received 29 May 2014, in ﬁnal form 5 October 2014)

ABSTRACT

This paper evaluates the seasonal (winter, premonsoon, monsoon, and postmonsoon) performance of seven precipitation products from three different sources: gridded station data, satellite-derived data, and reanalyses products over the Indian subcontinent for a period of 10 years (1997/98–2006/07). The evaluated precipitation products are the Asian Precipitation–Highly-Resolved Observational Data Integration Towards Evaluation of the Water Resources (APHRODITE), the Climate Prediction Center unified (CPC-uni), the Global Pre- cipitation Climatology Project (GPCP), the Tropical Rainfall Measuring Mission (TRMM) post-real-time research products (3B42-V6 and 3B42-V7), the Climate Forecast System Reanalysis (CFSR), and the European Centre for Medium-Range Weather Forecasts (ECMWF) interim reanalysis (ERA-Interim). Several verification measures are employed to assess the accuracy of the data. All datasets capture the large-scale characteristics of the seasonal mean precipitation distribution, albeit with pronounced seasonal and/or regional differences. Compared to APHRODITE, the gauge-only (CPC-uni) and the satellite-derived precipitation products (GPCP, 3B42-V6, and 3B42-V7) capture the summer monsoon rainfall variability better than CFSR and ERA-Interim. Similar conclusions are drawn for the postmonsoon season, with the exception of 3B42-V7, which underestimates postmonsoon precipitation. Over mountainous regions, 3B42-V7 shows an appreciable improvement over 3B42-V6 and other gauge-based precipitation products. Significantly large biases/errors occur during the winter months, which are likely related to the uncertainty in observations that artificially inflate the existing error in reanalyses and satellite retrievals.

1. Introduction regions of the world. In addition, it is well known that gauge measurements are also prone to several sources of There is growing evidence that the statistical charac- systematic and random error (Sevruk 1985; Kuligowski teristics of precipitation are changing at several places 1997). Given the importance of gauge-based precip- globally (Brunetti et al. 2001; Goswami et al. 2006; itation estimates as ‘‘ground truth’’ for other products, Trenberth 2011; Wang and Ding 2006; Yao et al. 2012). significant progress has been made to develop and con- To understand the extent of these changes, accurate and struct gridded global and regional gauge-based pre- timely knowledge of the space–time variability of precipitation analyses. cipitation is essential. Rain gauges provide direct and Satellite-derived precipitation products have been de- accurate point measurement of precipitation over land veloped that fill some of the data gaps by providing more and are often the most used and trusted source of in- spatially homogeneous and temporally complete cover- formation for hydrological studies. However, one of the age over large areas of ocean and land (Yilmaz et al. 2005; major shortcomings of gauge measurements is that they Dinku et al. 2007; Kucera et al. 2013). However, the ac- do not provide complete areal coverage and are often curacy of these products is limited, largely because they sparse or nonexistent in remote or politically unstable are derived from other observables (i.e., cloud-top re- flectance or thermal radiance; Richards and Arkin 1981; Petty and Krajewski 1996). Satellite products can also Corresponding author address: Sapna Rana, School of Geogra- phy, Environment and Earth Sciences, Victoria University of suffer from various discontinuities between different Wellington, Kelburn Parade, Wellington 6012, New Zealand. platforms and instruments and are not available before E-mail: [email protected] the 1970s. Recently, various ‘‘merged’’ satellite and gauge

DOI: 10.1175/JHM-D-14-0106.1

Ó 2015 American Meteorological Society Unauthenticated | Downloaded 10/07/21 05:09 PM UTC 632 JOURNAL OF HYDROMETEOROLOGY VOLUME 16

TABLE 1. Main characteristics of the precipitation products examined in this study.

Datasets Spatial Spatial Temporal used Product Format Unit domain resolution domain 2 APHRODITE APHRO_V1101 NetCDF mm day 1 Land 158S–558N, 0.58 (lat–lon) Daily (1951–2007) (Monsoon Asia) 608–1508E 2 CPC-uni v1.0 Binary mm day 1 Land–global 0.58 (lat–lon) Daily (1979–2010) 2 GPCP v1.2 Binary mm day 1 Global 18 (lat–lon) Daily (Oct 1996–2010) 2 3B42-V6 (Research Binary mm (3 h) 1 Tropics 508S–508N 0.258 (lat–lon) 3 hourly (1998–2010) product) 2 3B42-V7 (Research Binary mm (3 h) 1 Tropics 508S–508N 0.258 (lat–lon) 3 hourly (1998–2010) product) ERA-Interim Reanalysis Gridded m Global T255 (;79 km) Daily (1979–2010) binary 1 1.5831.58 (GRIB1) 2 2 CFSR Reanalysis Gridded kg m 2 s 1 Global T382 (;38 km) Daily (1979–2010) binary 2 0.313830.3138 (GRIB2) analyses have been assembled that maximize (and mini- Annamalai 2012). Far less effort has focused on pre- and mize) the relative benefits (and shortcomings) of each postmonsoon precipitation in spite of its importance to data type (New et al. 2001). water resources, agriculture, and livelihoods in the north- Spatially homogeneous and temporally continuous western/Himalayan and southern peninsular regions of the precipitation estimates are also available from the reanalysis subcontinent (Immerzeel et al. 2009; Bookhagen and systems that use advanced data assimilation techniques Burbank 2010). The magnitude and mechanistic explana- to merge estimates from global circulation models with tion of precipitation and its variability during these seasons observations. Reanalysis products are widely used in and/or regions remains unclear and poorly investigated climate research, even though the products are known to (Kripalani and Kumar 2004; Sen Roy 2006; Rajeevan et al. have biases, particularly in the estimation of precipitation 2012; Kar and Rana 2013). (Ma et al. 2009; Lorenz and Kunstmann 2012). In this study, we investigate the seasonal (winter, In spite of their shortcomings, satellite-derived and premonsoon, monsoon, and postmonsoon) precipitation reanalysis precipitation products offer an exciting oppor- climate and variability over the Indian subcontinent, tunity to better understand the characteristics and vari- using seven recent, high-quality, well-documented pre- ability of precipitation throughout the globe. However, cipitation products from three different sources: gridded before this potential can be fully realized, we need to station data, satellite-derived data, and reanalyses. The validate these datasets and gain a better understanding of study employs several verification methods to determine their inherent biases (Bosilovich et al. 2013; Turk et al. consistency and reliability associated with each pre- 2008). Intercomparisons of satellite-derived datasets and cipitation product, in terms of availability, resolution, reanalyses with gauge observations have been used to accuracy, retrieval technique, and quality control. assess the reliability of precipitation products at individual The main goals of this study are to 1) provide an in- sites and for specific regions (Bosilovich et al. 2008; Ebert tercomparison of the various precipitation datasets over et al. 2007; Joshi et al. 2012; Kidd et al. 2012; Ma et al. 2009; our study region, 2) provide insights into data reliability Palazzi et al. 2013; Rahman et al. 2009; Shah and Mishra and usability for a region that contains subregions that 2014; Shen et al. 2010). They showed that the performance are not well provisioned by rain gauges, and 3) examine of reanalyses and satellite-based precipitation estimates precipitation and its variability for all seasons through- vary for different regions and for different precipitation out the year. regimes. In general, reanalyses showed greatest skill dur- The results are mainly presented in the form of maps ing winter seasons while satellite estimates were best over that highlight the key characteristics and regional differ- wet regions and for warm seasons. ences associated with particular precipitation products. Previous studies of precipitation over the Indian sub- The paper is organized as follows. After the introduction continent have focused on the genesis, dynamics, di- in section 1, section 2 describes the datasets used, with agnostics, and predictability of precipitation and have put a brief description of methodology and choice of refer- the most emphasis on the Indian summer monsoon rainfall ence dataset; and section 3 outlines the results and dis- (ISMR; Krishnamurthy and Shukla 2000; Goswami and cussion, followed by the summary and conclusions in Ajaya Mohan 2001; Mishra et al. 2012; Turner and section 4.

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC APRIL 2015 R A N A E T A L . 633

2. Precipitation datasets product can be obtained online (http://precip.gsfc. nasa.gov/gpcp_daily_comb.html). The main characteristics of the datasets used in this 4) The Tropical Rainfall Measuring Mission (TRMM) is study are shown in Table 1. The basic selection criteria are a joint mission between the Japan Aerospace Explo- based on spatial coverage over the study region, spatial ration Agency (JAXA) and the U.S. National Aero- resolution, and temporal availability. nautics and Space Administration (NASA) designed 1) Asian Precipitation–Highly-Resolved Observational to monitor precipitation and its variability within the Data Integration Towards Evaluation of the Water tropics. The TRMM Multisatellite Precipitation Anal- Resources (APHRODITE) was developed by a con- ysis (TMPA) algorithm merges a variety of satellite- sortium of the Research Institute for Humanity and based observations and ground observations to yield Nature (RIHN), Japan, and the Meteorological Re- high spatiotemporal resolution and quasi-global quan- search Institute of the Japan Meteorological Agency titative precipitation estimation (QPE) products. The (MRI-JMA). APHRODITE is made up of high- 3-hourly TMPA precipitation estimates are available as quality daily precipitation products of varying resolu- a real-time version (3B42-RT) and a gauge-adjusted tion (0.258,0.58) for several Asian subregions. Here, post-real-time research version (3B42). There are two we use the latest version of daily precipitation data generations of 3B42 products: version 6 (3B42-V6), (APHRO_V1101) at 0.58 latitude–longitude resolu- which is available to June 2012; and version 7 (3B42- tion for the Asian monsoon domain (158S–558N, 608– V7), which was retrospectively processed using the 1508E) during the period 1951–2007 (available from updated algorithm back to 1998 and released in May www.chikyu.ac.jp/precip/products/). The product 2012 and is ongoing. The newly released 3B42-V7 does not discriminate between rain and snow but in- incorporates several important changes over 3B42-V6, corporates an improved quality-control method and including use of Global Precipitation Climatology orographic correction of precipitation. Details on the Centre (GPCC) analyses and additional satellite data gridding procedure and reliability of APHRODITE (Huffman and Bolvin 2014). For this study, we use both daily precipitation data can be found in Hamada et al. the research versions (hereafter referred to as 3B42-V6 (2011) and Yatagai et al. (2009, 2012). and 3B42-V7) available through the NASA interface 2) Climate Prediction Center uniﬁed (CPC-uni) is a global (mirador.gsfc.nasa.gov). Both datasets have a spatial gauge-based daily precipitation product from the (temporal) resolution of 0.25830.258 (3 h) over lati- National Oceanic and Atmospheric Administration tudes 508N–508S(Huffman et al. 2007, 2011). (NOAA) Climate Prediction Center (CPC). Gauge 5) The Climate Forecast System Reanalysis (CFSR) is reports from over 30 000 stations are collected from a third-generation reanalysis product developed at multiple sources, including Global Telecommunica- the National Centers for Environmental Prediction tions System (GTS), Cooperative Observer (COOP) (NCEP; Saha et al. 2010). The advantages of CFSR network, and other national and international agencies, relative to previous NCEP reanalyses include (i) higher and quality control is performed. CPC-uni uses optimal horizontal (;38 km, T382) and vertical (64 sigma- interpolation (OI) with orographic consideration to pressure hybrid levels) resolution, (ii) the guess represent the area-averaged values of precipitation forecast is generated from a coupled atmosphere– over the grid boxes (Chen et al. 2008). Here, we used land–ocean–ice system, and (iii) historical satellite the CPC-uni, version 1.0 (v1.0), global land data at radiance measurements are assimilated. In addition, a0.58 latitude–longitude spatial resolution available theCFSRisforcedwithobservedestimatesof from 1979 to the present. evolving greenhouse gas concentrations, aerosols, 3) The Global Precipitation Climatology Project (GPCP), and solar variations (http://cfs.ncep.noaa.gov/cfsr/). sponsored by the World Climate Research Programme This study uses daily precipitation at ;0.3138 Gaussian and Global Energy and Water Cycle Experiment, latitude–longitude, available for the period of 1979– provides global precipitation products based on satel- 2010 from the National Center for Atmospheric lite and gauge information at daily (Huffman et al. Research (NCAR). 2001), pentad (Xie et al. 2003), and monthly (Adler 6) The European Centre for Medium-Range Weather et al. 2003) time scales. The GPCP 1-Degree Daily Forecasts (ECMWF) interim reanalysis (ERA- (1DD), version 1.2 (v1.2), dataset covers the period Interim) is the most recent global atmospheric re- from October 1996 up to the present and is a compan- analysis produced by ECMWF covering the period ion to the GPCP, version 2.2, satellite–gauge (SG) from 1979 onward (Dee et al. 2011). ERA-Interim combination. Information on individual components uses four-dimensional variational data assimilation used as an input in the GPCP-1DD precipitation (4D-Var), a revised humidity analysis, variational

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC 634 JOURNAL OF HYDROMETEOROLOGY VOLUME 16

FIG. 1. Location and topography (m) of the Indian subcontinent.

bias correction for satellite data, and other improve- subcontinent’s climate, such as surface wind/surface ments in data handling. ERA-Interim precipitation pressure, temperature, and precipitation (Shea and estimates are produced by the forecast model based Sontakke 1995). A common overlap period of 10 years on temperature and humidity information derived (1997/98–2006/07) was chosen to enable a consistent from assimilated observations. The dataset is freely comparison. available at a global daily resolution of ;1.58 regular Several statistical methods are employed to analyze Gaussian grid. the precipitation products compared to the reference APHRODITE data. Maps are generated for seasonal a. Data processing and method and monthly mean climatology, standard deviation (std No standard format or grid resolution exists among the dev), and anomaly correlation coefficients (ACC). Per- various precipitation products (Table 1). To facilitate di- centage precipitation difference (PPD), defined as 100 rect comparison of precipitation datasets, a common 0.583 [(product 2 reference data)/reference data], and per- 0.58 regular latitude–longitude grid was chosen and bi- centage root-mean-square error (PRMSE), defined as linear interpolation was used. CFSR and ERA-Interim 100(RMSE/reference data), are computed to measure were extracted from the Gaussian grid gridded binary the biases in all examined products, relative to the sea- (GRIB) format and remapped to the (0.5830.58) regular sonal mean of the reference dataset. To examine the grid. Seasonal averages were calculated from daily pre- spatial–temporal variability of precipitation in each cipitation values and extracted for the region of interest, dataset, empirical orthogonal function (EOF) analysis is the Indian subcontinent (58–388N, 608–1008E; Fig. 1). Only applied. This technique seeks to reduce the dimensionality land points are considered here and the seasons are de- of a dataset by finding a new set of variables that capture fined as winter [December–February (DJF)], premonsoon most of the observed variance in the data through a linear [March–May (MAM)], monsoon [June–September combination of original variables (Wilks 1995). The (JJAS)], and postmonsoon [October–November (ON)]. derived pattern of variability in space is known as the These seasonal designations best segment the large-scale EOF, and in time it is known as principal component annual variations of the major components of the (PC). EOF analysis has been widely used in the field of

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC APRIL 2015 R A N A E T A L . 635

FIG. 2. Distribution of rain gauge stations used by APHRODITE for the year 1998: GTS stations (blue), precompiled datasets (black), and data individually collected by the APHRODITE project team (red). climate research for many years (Ding and Wang These studies, together with the validation work carried 2005; Mishra et al. 2012; Zhang et al. 2013). out by Yatagai and Xie (2006) and Yatagai et al. (2009), give us a sound basis to consider APHRODITE as a reli- b. Why APHRODITE as the reference? able reference dataset for studying precipitation variations In the present study, we have considered APHRODITE over the Indian subcontinent and other regions, where the as our primary surface reference (i.e., benchmark) dataset. dataset contains a dense network of rain gauges. Figure 2 Yatagai et al. (2008, 2009) recommended APHRODITE shows the distribution map of rain gauge stations used by as a reliable dataset for studying rainfall variations APHRODITE for the year 1998 (Yatagai et al. 2009). As over Asia and suggested that it can be considered as seen in Fig. 2, the dataset benefits from a dense network of observations and ground truth around the Himalayas, the rain gauges over a large part of India and Nepal, giving Middle East, and central Asian regions. Rajeevan and greater confidence for diagnostic studies and validation Bhate (2009), in a study over the Indian region, compared work. However, over the northwestern region of the the India Meteorological Department (IMD) rain gauge subcontinent (including northwestern India, Pakistan, data (comprising 6000 stations) against the APHRODITE_ and Afghanistan), the eastern part of peninsular India, V0804 rainfall product (;2000 stations) and found that and the Tibetan Plateau, the distribution of gauge sta- APHRODITE data could capture the large-scale features tions is sparse, which may limit the accuracy of observa- of the ISMR, exhibiting strong correlations (.0.6) with tions and might introduce uncertainties in gauge-based IMD data over most of India. For monitoring large-scale precipitation products. precipitation variations over East Asia, Sohn et al. (2012) quantified the reliability of six precipitation datasets over the region considering APHRODITE as the ground 3. Results and discussion truth. Andermann et al. (2011) compared gridded pre- a. Seasonal to interannual variability cipitation products along the complex relief zone of the Himalayan front and concluded that for all seasons the The Indian subcontinent exhibits a wide range of sea- APHRODITE dataset gives the best precipitation esti- sonal precipitation regimes across different regions. To mates with minimal difference from independent ground visualize this distribution of total annual precipitation, observations. However, the authors noted that the lack daily precipitation values for the time period from De- of gauge stations at higher elevations of the study region cember 1997 to December 2007 are segmented into four limits the accuracy of APHRODITE. Recently, Prakash seasons (DJF, MAM, JJAS, and ON) and spatial fields of et al. (2015) performed a seasonal (JJAS) intercomparison seasonal precipitation distribution are generated for all of six observational rainfall datasets against IMD gridded datasets (Fig. 3). The most pronounced spatial pre- data and found that APHRODITE and GPCC are the two cipitation pattern represents the southwest monsoon best performers in terms of statistical skill scores, with (JJAS) rainfall that contributes ;70%–75% of the total APHRODITE’s root-mean-square error (39.31%) being annual precipitation across the entire domain. Rainfall lower than that of GPCC (42.68%) for all-India JJAS during the southwest monsoon is a result of both the sea- rainfall. sonally changing thermal contrast between land and ocean

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC 636 JOURNAL OF HYDROMETEOROLOGY VOLUME 16

21 FIG. 3. Spatial distribution of seasonal mean precipitation (mm day ) over the Indian subcontinent for each precipitation product for the period 1997/98–2006/07.

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC APRIL 2015 R A N A E T A L . 637 and the seasonal migration of the intertropical conver- 3B42-V6, CPC-uni, and GPCP. According to Huffman gence zone (ITCZ). The highest climatological mean for and Bolvin (2014), the largest difference between 3B42- JJAS rainfall occurs over the western coast of peninsular V7 and 3B42-V6 occurs in mountainous regions owing to India and over the northeastern region; both are associated the major changes in 3B42-V7’s Precipitation Radar al- with orographic lifting and intense convection (Pattanaik gorithm, which uses a higher-resolution elevation map and Rajeevan 2010). based on Shuttle Radar Topography Mission data with The JJAS rainfall gradually decreases northwestward 30-arc-s spacing (SRTM30) over India, Tibet, and South from east-central India, with areas along the southeast- America (TRMM Precipitation Radar Team 2011). The ern coast of peninsular India being comparatively dry. It newer product (3B42-V7) displays the rain-shadow re- is during the postmonsoon season that large parts of the gion of the Western Ghats reasonably well but fails to southeastern coast of peninsular India and neighboring capture the distribution pattern of postmonsoon rainfall Sri Lanka receive a significant amount of rain due to the along the eastern coast of peninsular India. Moreover, retreating southwest monsoonal winds, also known as 3B42-V7 exhibits slightly higher precipitation over the the northeast monsoon. With the retreat of the south- core monsoon zone and the northeastern region of the west monsoon, surface pressure and the upper wind cir- subcontinent, including Myanmar. Reasons for over- culation patterns change rapidly and a trough of low estimation in 3B42-V7 could be bias in the full usage of pressure is established over the southern Bay of Bengal, GPCC gauge data, which, according to Prakash et al. 2 resulting in a northeasterly wind flow across the sub- (2015), overestimates (;5mmday 1) the monsoon rain- continent (Srinivasan and Ramamurthy 1973). During fall over northeastern India and along the Western Ghats this time, the northern–northwestern part of the sub- compared to IMD gauge data, or because of the occur- continent is generally dry but receives a significant rence of heavy convective rainfall, which may cause signal amount of precipitation during winter (DJF) and pre- attenuation and might lead to overestimation of near- monsoon (MAM) months because of the occurrence of surface rainfall (Amitai et al. 2004). midlatitude disturbances and premonsoon thunderstorms In contrast to CPC-uni and the satellite-derived pre- (TSs), respectively. Winter precipitation is usually limited cipitation products (GPCP, 3B42-V6, and 3B42-V7), the in spatial extent and total amount and is largely confined two reanalysis products (CFSR and ERA-Interim) ap- to the northwestern region of the Indian subcontinent pear to reproduce high seasonal precipitation values and follows a general decline from west to east (Sen Roy over regions of pronounced precipitation activity. For 2009). An interesting feature observed during the pre- example, high precipitation totals are observed in the monsoon months is the shift in rainfall activity to the northwestern and northeastern regions of the sub- northeastern region of the subcontinent as a result of the continent for DJF and MAM. In addition, regions of TS activity before the advancement of the southwest spuriously high rainfall are also seen during JJAS, par- monsoon over the region (Mahanta et al. 2013). ticularly over the northeastern region and the core Figure 3 shows that all products are capable of co- monsoon zone. Monsoon precipitation in these regions herently reproducing the key features of the seasonal and is mainly associated with the presence of a moist con- regional precipitation distribution reasonably well and vective regime related to the monsoon flow and also to are roughly in agreement with the APHRODITE data- the propagation of low pressure systems from the Bay of set. However, significant regional differences can be ob- Bengal along the core monsoon zone (Rajeevan et al. served in all precipitation products, especially along the 2010). Therefore, overestimation of precipitation in topographically induced high rainfall regional features these regions, especially by reanalysis datasets, could be (e.g., over northern Pakistan, the foothills of the Hima- associated with overestimation of moisture fields and layas, the extreme northeast, and along the Western hence precipitable water (Trenberth et al. 2011; Shah Ghats). Comparing APHRODITE with other products, and Mishra. 2014). CFSR tends to overestimate seasonal we find that CPC-uni, GPCP, and 3B42-V6 capture the precipitation in the southern part of peninsular India seasonal distribution pattern of DJF and MAM pre- during MAM, JJAS, and ON and to underestimate JJAS cipitation, but the datasets fail to capture the maxima of precipitation over the semiarid region of northwestern orographic precipitation over the complex topography of India (including Pakistan), consistent with Shah and northwestern India. In addition, the products slightly Mishra (2014). underestimate (overestimate) the rainfall amounts over The reanalyses illustrate the strong influence of orog- the windward (leeward) side of the western coast of India raphy on precipitation. While ERA-Interim shows during the JJAS season. a smooth spatial distribution of precipitation, CFSR re- In mountainous regions, 3B42-V7 exhibits slightly produces finer-scale regional patterns, which may be due higher precipitation and hence performs better than to the improved representation of the orographically

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC 638 JOURNAL OF HYDROMETEOROLOGY VOLUME 16

21 FIG. 4. Mean monthly precipitation cycle (mm day ) averaged over the Indian subcontinent for each precipitation product for the period 1997/98–2006/07. influenced precipitation and the high spatial resolution of and southern peninsular India. In comparison to this dataset (Saha et al. 2010). A close comparison of APHRODITE, the overall magnitude of rainfall vari- precipitation products illustrates that 3B42-V6 and CPC- ability is fairly well reproduced by all datasets for all uni display the highest degree of spatial agreement to seasons and/or regions, with the exception of CFSR. 2 APHRODITE’s (JJAS and ON) seasonal climatology. During JJAS, CFSR displays large (.18 mm day 1)in- For the entire subcontinent, the seasonal JJAS (ON) terannual variability over north-central India and north- domain-averaged mean is 4.29 (1.31) for APHRODITE, eastern regions of the subcontinent, where CFSR produces 3.98 (1.22) for CPC-uni, 5.2 (1.84) for GPCP, 4.35 (1.36) excess precipitation totals (Fig. 3). Misra et al. (2012) for 3B42-V6, 5.9 (1.11) for 3B42-V7, 7.1 (2.03) for CFSR, commented that CFSR displays the strongest interannual 2 and 5.8 (1.89) mm day 1 for ERA-Interim. variation of precipitable water over central India (with The mean monthly precipitation cycles over the re- nearly 11% difference between wet and dry years), which gion for the period 1997/98–2006/07 are compared in they associated with the anomalous oceanic sources of Fig. 4. All products capture the unimodal distribution of moisture from the adjacent Bay of Bengal and Arabian the annual cycle fairly well, albeit with different ampli- Sea. In the JJAS (std dev) pattern, an important feature in 2 tude. With respect to the APHRODITE, CPC-uni ERA-Interim is the region of suppressed (,9mmday 1) (GPCP) tends to slightly underestimate (overestimate) rainfall variability along the western coast of India and 2 the monthly precipitation amount, while CFSR and comparatively high (16 mm day 1) variability over the ERA-Interim significantly overestimate mean monthly northeastern region. precipitation from May to December. The 3B42-V7 The results obtained here are consistent with the find- closely agrees with APHRODITE’s averages for the ing of Shah and Mishra (2014), who reported that both winter and premonsoon months, but overestimates CFSR and ERA-Interim overestimated monsoon rainfall (slightly underestimates) the JJAS (ON) precipitation variability over eastern regions compared to IMD gauge amounts over the core monsoon zone (eastern coast) of observations. For the monsoon season, the domain- India. The results seen here are a clear manifestation of averaged std dev is APHRODITE 5 10.4, CPC-uni 5 the spatial climatology patterns depicted in Fig. 3. 10.2, GPCP 5 9.6, 3B42-V6 5 11.2, 3B42-V7 5 11.4, 2 To represent the interannual variability of pre- CFSR 5 17.3, and ERA-Interim 5 11.3 mm day 1.In cipitation at each grid point, we generate interannual std comparison to GPCP, the relative superiority of CPC- dev plots for each season (Fig. 5). All datasets reveal large uni, 3B42-V6, and 3B42-V7 is evident, as these products std dev values over regions that receive large amounts of show close agreement to APHRODITE’s precipitation seasonal precipitation. For example, the western coast of variability and can capture the pattern of large std dev peninsular India and the northeastern region exhibit values over the Western Ghats (eastern coast) of India large std dev values during JJAS, while southeastern during JJAS (ON). Rahman et al. (2009) and Joshi et al. peninsular India displays high interannual variability (2012) also reported that in comparison to IMD gauge during ON. Similarly for DJF and MAM, comparatively data, 3B42-V6 performs better than GPCP for the mon- 2 high (.5mmday 1) std dev values appear over the soon period over the Indian region. Among the reanalysis northwestern region and in some parts of northeastern datasets, the performance of ERA-Interim looks

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC APRIL 2015 R A N A E T A L . 639

21 FIG. 5. Std dev of seasonal mean precipitation (mm day ) over the Indian subcontinent for each precipitation product for the period 1997/98–2006/07.

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC 640 JOURNAL OF HYDROMETEOROLOGY VOLUME 16 somewhat better than CFSR in terms of both domain- However, for the satellite-derived precipitation products, averaged mean and std dev for the JJAS period. For DJF the large bias seems to gradually decrease from the cold and MAM, CPC-uni and GPCP do not provide any useful winter months (DJF) to the warm summer season information with regard to the interannual variability of (JJAS). This result closely agrees with the conclusions seasonal precipitation over the northwestern region of reported by Ebert et al. (2007) that satellite algorithms the Indian subcontinent and along the foothills of the perform best during summers, when rainfall is more Himalayas. Despite the similar std dev distribution pat- convective, than in winters when it is more synoptically tern for both 3B42-V6 and 3B42-V7, differences are no- forced. Apparently, the occurrence of comparatively table over mountainous regions, where 3B42-V7 more smaller positive bias for the satellite-derived pre- successfully captures the interannual variability of the cipitation products could be related to the difficulty of in DJF and MAM precipitation as seen in APHRODITE. situ station data, satellite estimates, and their combina- The domain-averaged std dev values for DJF are tions in detecting the snow component of precipitation APHRODITE 5 3.4, CPC-uni 5 2.9, GPCP 5 3.1, (Rasmussen et al. 2012). 3B42-V6 5 3.1, 3B42-V7 5 3.6, CFSR 5 4.3, and All datasets (except for the gauge-only CPC-uni) 2 ERA-Interim 5 4.3 mm day 1. display distinctly large winter PPDs over the northern flank of the subcontinent and the Tibetan Plateau. This b. Error assessment implies an additional source of uncertainty (i.e., gauge To explore the skill of the precipitation products on measurements) might be contributing toward the exist- a regional basis, a relative bias or PPD (relative to ing errors in the reanalysis estimates and satellite re- APHRODITE) is computed on a seasonal scale. Figure 6 trievals. According to Lorenz and Kunstmann (2012), indicates that CPC-uni exhibits a general tendency to gauge undercatch error for solid precipitation, which is underestimate seasonal precipitation along the foothills especially large in high latitudes or mountainous regions of the Himalayas, the Tibetan Plateau, and Myanmar in during the winter season, can lead to an underestimation all seasons. The poor performance of the CPC-uni in- of the true precipitation of up to 50%. Moreover, sam- dicates that the total number of active rain gauges and pling errors between 15% and 100% could be expected their spatial distribution used to construct the CPC-uni for sparsely gauged regions with less than three gauges dataset is not enough to capture the seasonal pre- per 2.5832.58 grid cell (Rudolf and Rubel 2005). cipitation pattern and interannual variability over the Therefore, depending on the environmental conditions above-mentioned regions. According to Gebregiorgis (season) and geographic location, it is likely that the and Hossain (2015), APHRODITE uses an extensive uncertainties and inhomogeneities in observational/ network of gauges from various partner organizations, reference data are artificially inflating the existing biases which is more than CPC-uni. in all datasets. Furthermore, a comparison of the spatial Large PPDs appear over the data-sparse and topo- and domain-averaged PPDs indicates that 3B42-V6 and graphically complex Tibetan Plateau and the Hindu 3B42-V7 outperform GPCP in all seasons. This can be Kush–Karakoram mountain ranges in the northwestern associated first to the fact that, in regions with low gauge region of the subcontinent. The domain-averaged PPDs density, the GPCP analysis is influenced more by the are maximum (minimum) for the winter (monsoon) gauges (Gruber et al. 2000), while 3B42 estimates are season with values of 13% (4%), 142% (53%), 43% influenced more by the satellite measurements, and (25%), 72% (38%), 243% (41%), and 234% (84%) for second, to the coarser resolution of GPCP, which CPC-uni, GPCP, 3B42-V6, 3B42-V7, CFSR, and ERA- spreads any errors over larger geographical areas. CFSR Interim, respectively. Similarly for MAM (ON), the displays large negative PPDs over Afghanistan and domain-averaged PPDs are 2% (4%), 63% (68%), 33% Pakistan and some parts of northwestern India during (28%), 40% (45%), 79% (103%), and 123% (114%), JJAS. Misra et al. (2012) found that during the monsoon respectively. The large positive bias over the above- season the cross-equatorial and the southwesterly flow of mentioned regions appears to be consistent in the monsoonal (950 hPa) winds is weak in CFSR, which, ac- reanalysis datasets throughout the year. The large posi- cording to our understanding, could result in the weak- tive bias in CFSR and ERA-Interim could be related to 1) ening of the monsoon flow and hence less advection of the inability of the models to resolve orographic pre- moisture to the far northwestern regions (including cipitation in complex terrain with large elevation changes Pakistan), as seen in Fig. 3. (Ma et al. 2009), 2) differences in the land–atmosphere Spatial patterns of ACC are generated for the period interaction and land surface model schemes (Misra et al. from 1997/98–2006/07 in Fig. 7 to analyze how the ex- 2012), and 3) the fact that reanalysis products estimate amined products relate to the seasonal anomalies in total precipitation (rainfall plus snow; Palazzi et al. 2013). APHRODITE data at each grid point. Results indicate

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC APRIL 2015 R A N A E T A L . 641

FIG. 6. PPD with respect to APHRODITE seasonal mean precipitation over the Indian subcontinent for the period 1997/98–2006/07.

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC 642 JOURNAL OF HYDROMETEOROLOGY VOLUME 16

FIG. 7. ACC of seasonal mean precipitation over the Indian subcontinent compared to APHRODITE for the period 1997/98–2006/07.

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC APRIL 2015 R A N A E T A L . 643 that for a season, the ACCs are significantly high (.0.632 the season, PRMSEs and PPDs are large and distinct over at 95% significance level) over regions exhibiting less the topographically complex and data-sparse northwest- interannual variability and vice versa. Reasonably high ern region of the Indian subcontinent, Myanmar, and (low) ACCs are observed during ON (JJAS) over most of the Tibetan Plateau. Conversely, the spatial spread of the regions and for all products. In comparison to other PRMSE is homogeneous and generally smaller over low- precipitation products, CPC-uni displays relatively high elevation plain areas where moderate to dense gauge and significant ACC values (.0.65) along the western stations are available. For the subcontinent as a whole, (eastern) coast of peninsular India during JJAS (ON), the domain-averaged PRMSEs for the winter season are suggesting that the two gauge products have similar much greater in the reanalysis datasets (CFSR 5 335%, anomaly pattern. The domain-averaged ACC for CPC- ERA-Interim 5 311%), than for CPC (104%), 3B42-V6 uni and APHRODITE is found to be DJF 5 0.36, (143%), 3B42-V7 (161%), and GPCP (209%). Similarly MAM 5 0.57, JJAS 5 0.47, and ON 5 0.63. In addition, for the JJAS season, the PRMSEs for CPC-uni, GPCP, 3B42-V6 and 3B42-V7 also show a close agreement to 3B42-V6, 3B42-V7, CFSR, and ERA-Interim are APHRODITE during MAM, JJAS, and ON, indicating found to be 61%, 78%, 62%, 65%, 100%, and 171%, a comparably good anomaly correlation in regions ex- respectively. For the pre- and postmonsoon seasons, hibiting high precipitation variability (e.g., Western the PRMSE is ,90% for CPC-uni, 3B42-V6, and Ghats and northeastern region including Myanmar). The 3B42-V7; and .150% for CFSR, ERA-Interim, and domain-averaged ACCs for 3B42-V6 (3B42-V7) during GPCP. Apparently for the monsoon season, the PRMSE DJF, MAM, JJAS, and ON are found to be 0.21 (0.21), spatial spread and domain-averaged score is a minimum 0.54 (0.57), 0.46 (0.48) and 0.67 (0.68), respectively. Un- in CPC-uni over the Indian region, while among the re- like other precipitation products, 3B42-V6 and 3B42-V7 analysis products, ERA-Interim appears to be slightly exhibit the lowest domain-averaged ACCs (i.e., 0.21) and better than CFSR in terms of the spatial spread of display a different anomaly correlation pattern during PRMSEs over the Indian region alone. Additionally, the DJF season. The diagonal correlation pattern with 3B42-V6 and 3B42-V7 seem to outperform GPCP again, positive–negative–positive ACC values is similar to the with regard to the spatial representation of comparatively high–low–high regional precipitation seasonality pat- smaller errors over the Western Ghats, the foothills of the tern generated in Fig. 3. However, the negative ACC Himalayas, and Myanmar during the monsoon season. values in 3B42-V6 and 3B42-V7 are of concern, indi- The extreme spatial PRMSEs observed during the ‘‘cold’’ cating in some cases the reversed anomalies of winter winter season mentioned above illustrate important lim- precipitation over the Indian subcontinent. itations in all gridded precipitation products examined In general, it appears that all individual precipitation here (including the APHRODITE dataset). Based on the products have a relatively low and seasonally variable results obtained in Figs. 6 and 8, together with the well- ACC spatial pattern over the northwestern region of sub- known fact that gauge observations are subject to several continent, the entire Himalayan belt, the Tibetan Plateau, sources of systematic and random errors, it is likely that and Myanmar. The two reanalyses products more or less an additional source of error could be associated with the represent similar characteristic features in their ACC pat- gauge input data (especially when and where gauge sta- terns and produce poor correlations during MAM tions are sparse and/or snow makes up a large portion of (CFSR 5 0.35, ERA-Interim 5 0.42) and JJAS (CFSR 5 the total precipitation). 0.29, ERA-Interim 5 0.30) as compared to other pre- c. EOF analysis cipitation products. This result perhaps can be attributed entirely to the strong overestimation of precipitation To explore the structure of large-scale coherence (Figs. 3, 4) and high interannual variability (Fig. 5)over within the precipitation datasets, we use EOF analysis on the core monsoon zone and the northeastern region of the the standardized seasonal precipitation anomalies. Re- Indian subcontinent, which largely reduces the skill of the sults are shown in Fig. 9. Here we have focused on the first reanalyses products over these regions. leading mode (EOF1), which accounts for the largest Figure 8 shows the seasonal PRMSE with respect to percentage of the total precipitation variance, repre- the climatological mean of the APHRODITE dataset. sented at the bottom left of each spatial plot. While in- The PRMSE ranges from zero to infinity, with lower terpreting the EOF patterns, it must be kept in mind that values indicating closer agreement with APHRODITE. the sign of the patterns is arbitrary (the temporal mean of The maximum (minimum) error/uncertainty in pre- their PC time series is zero). The spatial pattern of the cipitation estimates exists during the winter (monsoon) leading EOF during DJF reveals a large-scale pattern of period. More or less consistent with the PPD results, the negative loading over most of the Indian subcontinent spatial comparison of PRMSE reveals that, regardless of with some positive loading over the southern tip of

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC 644 JOURNAL OF HYDROMETEOROLOGY VOLUME 16

FIG. 8. PRMSE with respect to APHRODITE seasonal mean precipitation over the Indian subcontinent for the period 1997/98–2006/07.

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC APRIL 2015 R A N A E T A L . 645

FIG. 9. Leading EOF of standardized seasonal mean precipitation over the Indian subcontinent for the period 1997/98–2006/07.

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC 646 JOURNAL OF HYDROMETEOROLOGY VOLUME 16 peninsular India, north of the Tibetan Plateau, and over generated from CPC-uni and the satellite-derived prod- some parts of Myanmar. For APHRODITE, the first ucts. The corresponding MAM-PC1 time series (Fig. 10b) winter EOF explains 53% of the total variance, which is is similar to the DJF-PC1 sequence and exhibits the in- comparable to that of GPCP (50%). The loading patterns terannual variability present in the premonsoon domain- of EOF1 are very much similar for 3B42-V6 and 3B42-V7 averaged precipitation time series (figure not shown). All in all seasons, albeit with differences in the percentage of datasets capture the 3-yr (1998–2001) extended winter variance explained (always higher for 3B42-V7). For and premonsoon drought that lasted over Afghanistan, DJF, the EOF pattern obtained from CFSR and ERA- Iran, and Pakistan. According to Barlow et al. (2002),the Interim resembles APHRODITE’s spatial pattern more prolonged drought was linked to the cold phase of ENSO closely than the one obtained from GPCP, 3B42-V6, and (La Niña) and the unusually warm pool region in the 3B42-V7. western Pacific Ocean. Kar and Rana (2013) used EOF analysis to investigate The JJAS-EOF1 patterns (Fig. 9) are characterized by the impact of the Asian jet on interannual variability of uniform variability over the entire Indian subcontinent. In winter precipitation over northwestern India. The au- general, for all precipitation products, the JJAS pattern thors found that the PC1 and PC2 time series corre- reveals positive loading over a large part of the sub- sponding to EOF1 and EOF2 of DJF zonal winds continent, with small areas of negative loading over central at 200 hPa represented a pattern similar to El Niño– India and in the northeastern region of the subcontinent. Southern Oscillation (ENSO) and the North Atlantic Mishra et al. (2012) reported that the region, for which the Oscillation (NAO)–Arctic Oscillation (AO), respec- JJAS monsoon rainfall oscillates in phase with the leading tively, with a statistically significant correlation between PC1 time series, includes a large part of the Indian region the PC time series and winter precipitation. Syed et al. along with Pakistan, most of the Arabian Sea, and the Bay (2009) also found that both ENSO and NAO have a sig- of Bengal. The percentage of variance explained by the nificant influence on winter precipitation over southwest- leading EOFs for the satellite-derived products (GPCP 5 central Asia (including northern Pakistan, Afghanistan, 31%, 3B42-V6 5 35%, and 3B42-V7 5 37%) are close to and Tajikistan), where winter precipitation increases one another but higher than the variance obtained for the (decreases) during the warm (cold) ENSO phase and two reanalyses products (27%) and APHRODITE (26%) positive (negative) NAO phase. The time coefficient of dataset. Surprisingly, the APHRODITE JJAS-EOF1 our leading EOF, that is, the PC1 of the standardized DJF pattern differs in its spatial structure when compared to precipitation anomalies, is shown in Fig. 10a.Consistent other products. The expected pattern [as reported by with the above findings, we found that the DJF-PC1 time Krishnamurthy and Shukla (2000), Guhathakurta and series obtained for all precipitation products corresponds Rajeevan 2008, Joshi et al. (2012),andMishra et al. (2012)] well with the DJF–Niño-3.4 index obtained from the CPC is quite well captured by CPC-uni and the satellite-derived (www.cpc.ncep.noaa.gov/products/analysis_monitoring/ products, exhibiting a uniform structure of positive loading ensostuff/ensoyears.shtml) for the period 1997/98–2006/ (0.02–0.08) over a large part of the Indian region and some 07. The correlation between the DJF-PC1 time series and pockets of negative loading over central and northeastern the Niño-3.4 index is found to be statistically significant India. We further investigated the EOF pattern obtained and greater than 0.7 for all datasets used here. for APHRODITE using a longer time series (1979–2007) In MAM, we find that the small area of positive loading and found that the leading EOF showed a close re- over the northwestern region of the subcontinent seen in semblance to the expected pattern. Therefore, it can be DJF is persistent and more prominent during MAM for suggested that the observed difference in APHRODITE’s all precipitation products. The loading represents a see- EOF1 pattern is likely due to the short time of analysis saw-like pattern of positive (negative) loading over the considered here. Most studies cited above show that the northwestern region (the rest of the subcontinent), which PC1 time series of JJAS rainfall is strongly correlated with shows similarities with the seasonal mean precipitation the all-India-averaged rainfall index; which suggests that distribution pattern over the subcontinent (Fig. 3). The EOF1 is a good representation of the dominant mode of percentage variance explained by the precipitation the JJAS precipitation variation (flood and drought products is either comparable to or higher than that of years). When the standardized JJAS-PC1 series (Fig. 10c) APHRODITE (37%). However, the structure of loading are correlated with their respective seasonal precipitation pattern is slightly different for the reanalysis datasets, mean time series, the maximum correlation is found for with a broader region of positive loading over the CPC-uni (0.98) and minimum for APHRODITE (0.38). northeastern domain, which is not very prominent in any Apparently for the APHRODITE dataset, the relatively other dataset. Therefore, during MAM the structure of poor correlation relates to the inability of the dataset to EOF1 from APHRODITE most closely resembles those accurately capture the leading JJAS-EOF1 pattern.

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC APRIL 2015 R A N A E T A L . 647

FIG. 10. PC time series for the leading EOFs of standardized seasonal mean precipitation for (a) winter, (b) premonsoon, (c) monsoon, and (d) postmonsoon over the Indian subcontinent for each precipitation product for the period 1997/98–2006/07.

Additionally for GPCP, 3B42-V6, and 3B42-V7 the positive loading over a large region. Kumar et al. (2007) correlation is found to be .0.75, while for CFSR and found that only in recent years (1976–2000) has there ERA-Interim it is 0.68 and 0.5, respectively. been an increased relationship between ENSO and the For the postmonsoon season, the spatial structure of northeast monsoon rainfall occurring over the southern EOF1 is similar to that in MAM but with a pronounced peninsular region of India. Here, the PC1 time series northwest–east and northwest–southeast gradient of (Fig. 10d) is found to be significantly (95%) correlated positive–negative–positive loading over the subcontinent. with the corresponding seasonal precipitation time series The region of positive loading over southern peninsular over the entire subcontinent. However, we did not find India is relatively small for all products except for CFSR a significant correlation between the ENSO index and the and ERA-Interim. Nair et al. (2013) used EOF analysis of PC1 time series, which could possibly be related to the eight global circulation models and found that during shorter time series of the examined products considered October–December the southern peninsular zone re- here or due to the problem of domain dependency of mains spatially correlated to the first EOF mode with EOF solutions (Horel 1981; Richman 1986).

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC 648 JOURNAL OF HYDROMETEOROLOGY VOLUME 16

4. Summary and conclusions the entire subcontinent and Indian region, respectively. The objective of this study was to perform an inter- Among the satellite-derived precipitation products, the comparison of precipitation products over the Indian two post-real-time research products of 3B42 (V6 and subcontinent and to provide useful insights on the reli- V7) outperformed GPCP not only for the monsoon ability and usability of precipitation datasets, over regions period, but also throughout the year. On the contrary, that are not well provisioned by rain gauges. To this end, CFSR and ERA-Interim exhibited a high positive bias seven precipitation products (APHRODITE, CPC-uni, over the northeastern region of the subcontinent and the GPCP, 3B42-V6, 3B42-V7, CFSR, and ERA-Interim) northern flanks of the Tibetan Plateau. were examined in regard to the seasonal (DJF, MAM, Similar conclusions can be drawn for the postmonsoon JJAS, and ON) precipitation climate and variability for the season (ON), where reanalyses (especially CFSR) period 1997/98–2006/07. overestimate the seasonal precipitation totals over parts of peninsular India, the northeastern region of the sub- a. General comments continent, and the Tibetan Plateau. The 3B42-V6 All products performed reasonably well in capturing the product displays the most reasonable precipitation total, main characteristic features of seasonal precipitation var- closer to APHRODITE, while 3B42-V7 underestimates iability, particularly over low-elevation plain areas where the seasonal precipitation amount in comparison to APHRODITE contains a moderate to dense network of APHRODITE and other examined products. Results rain gauge stations. Referring to the gauge distribution show that the gauge-based CPC-uni largely under- map in Fig. 2, there are certain regions/areas where the estimates precipitation along the foothills of the Hima- rain gauge network is dense and perhaps more reliable to layas, the Tibetan Plateau, and Myanmar in all seasons. intercompare and ascertain the accuracy and biases in the To improve the quality of the product, both in accuracy proxy datasets. Hence, over these regions, the data user and reliability over these regions, use of additional rain can expect to have greater confidence in the accuracy and gauge stations is highly desirable. certainty of the results presented here. Likewise, for the premonsoon season (MAM), the Depending on the environmental conditions (season) satellite-derived precipitation products perform better than and geographic location (complex terrain), gauge mea- the reanalyses estimates. Relative to the APHRODITE surements are subject to several sources of errors that dataset, CFSR and ERA-Interim consistently over- largely limit the accuracy of gauge measurements. Over estimate precipitation over the topographically complex regions where the spatial coverage of gauge observation is and data-sparse Tibetan Plateau and the Hindu Kush– poor, proxy datasets can provide a potential source of in- Karakoram mountain ranges in the northwestern region formation, but their accuracy and reliability can be con- of the subcontinent. Consequently, this could account for sidered as examined only. Our results show that over the the fact that reanalysis datasets overestimate more in re- topographically complex and data-sparse regions, pro- gions with complex terrain than in flatter regions. Im- nounced regional and seasonal differences exist between pressively, with regard to the seasonal and interannual the examined products and the APHRODITE dataset. variability of premonsoon precipitation over the moun- Based on our relative comparisons (PPD and PRMSE), it tainous regions, the newly released 3B42-V7 appears to may be concluded, though with lesser confidence that in provide useful information in comparison to CPC-uni, gauge-sparse regions, a potential source of uncertainty in GPCP, and 3B42-V6. all gauge-based precipitation products is likely associated For the winter season (DJF), all precipitation products with the uncertainties and inhomogeneities in gauge display less agreement with respect to APHRODITE, measurements itself. illustrating important limitations in all the gridded precipitation, including our reference dataset. Based on the b. Our interpretation of seasonal comparisons aforementioned reasons, a major source of uncertainty, For the monsoon season (JJAS), results indicate that particularly during the winter months over the data- the gauge-only (CPC-uni) and satellite-derived pre- sparse regions, is likely the representativeness and cipitation products (GPCP, 3B42-V6, and 3B42-V7) undercatch error in gauge measurements that tends to capture the JJAS rainfall variability better than the artificially inflate the existing errors in satellite and reanalyses (CFSR and ERA-Interim). The 3B42-V6 reanalyses precipitation estimates. In spite of the large displayed a close agreement to APHRODITE, with winter biases in reanalyses (CFSR and ERA-Interim) regard to spatial and domain-averaged statistics for the and the newly released 3B42-V7, we are fairly confident entire Indian subcontinent. Additionally, 3B42-V7 and that these precipitation products could be a useful source CPC-uni demonstrated a reasonable performance over of information over the data-sparse regions. However,

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC APRIL 2015 R A N A E T A L . 649 commenting on the accuracy and robustness of these ——, J. Kennedy, D. Dee, R. Allan, and A. O’Neill, 2013: On the products requires further investigation. reprocessing and reanalysis of observations for climate. Cli- Nonetheless, the best approach to fill in these gaps is mate Science for Serving Society: Research, Modeling and Prediction Priorities, G. R. Asrar and J. W. Hurrell, Eds., to recognize the relative strength and weaknesses of Springer, 51–71. each precipitation product and to extract maximum Brunetti, M., M. Colacino, M. Mugeri, and T. Nanni, 2001: Trends possible information for their constructive applicability in the daily intensity of precipitation in Italy from 1951 to 1996. and usability. We anticipate that the outcomes of this Int. J. Climatol., 21, 299–316, doi:10.1002/joc.613. study will benefit data users, model developers, and re- Chen, M., W. Shi, P. Xie, V. B. S. Silva, V. E. Kousky, R. W. Higgins, and J. E. Janowiak, 2008: Assessing objective tech- searchers of hydrology and climatology to understand niques for gauge-based analyses of global daily precipitation. the relevant seasonal characteristics of each pre- J. Geophys. Res., 113, D04110, doi:10.1029/2007JD009132. cipitation product used in this study. Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: In closing, we would like to mention that future work Configuration and performance of the data assimilation system. will address the spatiotemporal variations of winter pre- Quart. J. Roy. Meteor. Soc., 137, 553–597, doi:10.1002/qj.828. Ding, Q. H., and B. Wang, 2005: Circumglobal teleconnection in cipitation in the northwestern region of the Indian sub- the Northern Hemisphere summer. J. Climate, 18, 3483–3505, continent using additional datasets and variables. doi:10.1175/JCLI3473.1. Dinku, T., P. Ceccato, E. Grover-Kopec, M. Lemma, S. J. Connor, and C. F. Ropelewski, 2007: Validation of satellite rainfall Acknowledgments. The authors would like to extend products over East Africa’s complex topography. Int. J. Remote their sincere thanks to the Victoria University of Wel- Sens., 28, 1503–1526, doi:10.1080/01431160600954688. lington (VUW), New Zealand, for supporting this re- Ebert, E. E., J. E. Janowiak, and C. Kidd, 2007: Comparison of search. The lead author acknowledges VUW for the near-real-time precipitation estimates from satellite obser- Doctoral Scholarship Award, and the institutions/teams vations and numerical models. Bull. Amer. Meteor. Soc., 88, 47–64, doi:10.1175/BAMS-88-1-47. responsible for the creation and publication of the Gebregiorgis, A. S., and F. Hossain, 2015: How well can we estimate APHRODITE, CPC-uni, GPCP, TRMM, CFSR, and error variance of satellite precipitation data across the world? ERA-Interim data archives. We would also like to thank Atmos. Res., 154, 39–59, doi:10.1016/j.atmosres.2014.11.005. Dr. Joe Turk and the two anonymous reviewers for their Goswami, B. N., and R. S. Ajaya Mohan, 2001: Intraseasonal constructive comments. The lead author also extends oscillations and interannual variability of the Indian summer monsoon. J. Climate, 14, 1180–1198, doi:10.1175/ special thanks to Dr. Ankita Singh of the Indian In- 1520-0442(2001)014,1180:IOAIVO.2.0.CO;2. stitute of Technology Bhubaneswar for her inputs re- ——, V. Venugopal, D. Sengupta, M. S. Madhusoodanan, and P. K. garding statistical techniques. Xavier, 2006: Increasing trend of extreme rain events over India in a warming environment. Science, 314, 1442–1445, doi:10.1126/ science.1132027. REFERENCES Gruber, A., X. Su, M. Kanamitsu, and J. Schemm, 2000: The comparison of two merged rain gauge–satellite precipitation Adler, R. F., and Coauthors, 2003: The Version-2 Global Precipitation datasets. Bull. Amer. Meteor. Soc., 81, 2631–2644, doi:10.1175/ Climatology Project (GPCP) Monthly Precipitation Analysis 1520-0477(2000)081,2631:TCOTMR.2.3.CO;2. (1979–present). J. Hydrometeor., 4, 1147–1167, doi:10.1175/ Guhathakurta, P., and M. Rajeevan, 2008: Trends in the rainfall 1525-7541(2003)004,1147:TVGPCP.2.0.CO;2. pattern over India. Int. J. Climatol., 28, 1453–1469, doi:10.1002/ Amitai, E., X. Llort, L. Liao, and R. Meneghini, 2004: A framework joc.1640. for global verification of space-borne radar estimates of pre- Hamada, A., O. Arakawa, and A. Yatagai, 2011: An automated cipitation based on rain type classification. 2nd TRMM Int. quality control method for daily rain-gauge data. Global Environ. Science Conf., Nara, Japan, JAXA/NICT/NASA, 4 pp. Res., 15 (2), 183–192. [Available online at www.airies.or.jp/attach. [Available online at www.eorc.jaxa.jp/TRMM/museum/event/ php/6a6f75726e616c5f31352d32656e67/save/0/0/15_2-12.pdf.] 2ndTISC/HP/Extended%20Abstract/5.4_AMITAI_Eyal.pdf.] Horel, J. D., 1981: A rotated principal component analysis of the Andermann, C., S. Bonnet, and R. Gloaguen, 2011: Evaluation of interannual variability of the Northern Hemisphere 500 mb precipitation datasets along the Himalayan front. Geochem. height field. Mon. Wea. Rev., 109, 2080–2092, doi:10.1175/ Geophys. Geosyst., 12, Q07023, doi:10.1029/2011GC003513. 1520-0493(1981)109,2080:ARPCAO.2.0.CO;2. Barlow, M., H. Cullen, and L. Bradfield, 2002: Drought in central Huffman, G. J., and D. T. Bolvin, 2014: TRMM and other data and southwest Asia: La Niña, the warm pool, and Indian precipitation dataset documentation. NASA GSFC Rep., Ocean precipitation. J. Climate, 15, 697–700, doi:10.1175/ 42 pp. [Available online at ftp://precip.gsfc.nasa.gov/pub/ 1520-0442(2002)015,0697:DICASA.2.0.CO;2. trmmdocs/3B42_3B43_doc.pdf.] Bookhagen, B., and D. W. Burbank, 2010: Toward a complete ——, M. Morrissey, D. T. Bolvin, S. Curtis, R. Joyce, B. McGavock, Himalayan hydrological budget: Spatiotemporal distribution and J. Susskind, 2001: Global precipitation at one-degree of snowmelt and rainfall and their impact on river discharge. daily resolution from multisatellite observations. J. Hydro- J. Geophys. Res., 115, F03019, doi:10.1029/2009JF001426. meteor., 2, 36–50, doi:10.1175/1525-7541(2001)002,0036: Bosilovich, M. G., J. Chen, F. R. Robertson, and R. F. Adler, 2008: GPAODD.2.0.CO;2. Evaluation of global precipitation in reanalyses. J. Appl. Meteor. ——, and Coauthors, 2007: The TRMM Multi-satellite Precipi- Climatol, 47, 2279–2299, doi:10.1175/2008JAMC1921.1. tation Analysis: Quasi-global, multiyear, combined-sensor

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC 650 JOURNAL OF HYDROMETEOROLOGY VOLUME 16

precipitation estimates at fine scale. J. Hydrometeor., 8, 38–55, Nair, A., N. Acharya, A. Singh, U. C. Mohanty, and T. C. Panda, doi:10.1175/JHM560.1. 2013: On the predictability of northeast monsoon rainfall over ——, D. T. Bolvin, E. J. Nelkin, and R. F. Adler, 2011: Highlights of south peninsular India in general circulation models. Pure Appl. Version 7 TRMM Multi-satellite Precipitation Analysis Geophys., 170, 1945–1967, doi:10.1007/s00024-012-0633-y. (TMPA). Proc. 5th Int. Precipitation Working Group Work- New, M., M. Todd, M. Hulme, and P. Jones, 2001: Precipitation shop, Hamburg, Germany, Max-Planck-Institut für Meteor- measurements and trends in the twentieth century. Int. ologie, 109–110. [Available online at www.mpimet.mpg.de/ J. Climatol., 21, 1889–1922, doi:10.1002/joc.680. fileadmin/publikationen/Reports/WEB_BzE_100.pdf.] Palazzi, E., J. V. Hardenberg, and A. Provenzale, 2013: Pre- Immerzeel, W. W., P. Droogers, S. M. de Jong, and M. F. P. cipitation in the Hindu-Kush Karakoram Himalaya: Obser- Bierkens, 2009: Large-scale monitoring of snow cover and vations and future scenarios. J. Geophys. Res. Atmos., 118, 85– runoff simulation in Himalayan river basins using remote 100, doi:10.1029/2012JD018697. sensing. Remote Sens. Environ., 113, 40–49, doi:10.1016/ Pattanaik, D. R., and M. Rajeevan, 2010: Variability of extreme j.rse.2008.08.010. rainfall events over India during southwest monsoon season. Joshi, M. K., A. Rai, and A. Pandey, 2012: Validation of TMPA Meteor. Appl., 17, 88–104, doi:10.1002/met.164. and GPCP 1DD against the ground truth rain-gauge data for Petty, G. W., and W. Krajewski, 1996: Satellite rainfall estimation Indian region. Int. J. Climatol., 33, 2633–2648, doi:10.1002/ over land. Hydrol. Sci. J., 41, 433–451, doi:10.1080/ joc.3612. 02626669609491519. Kar, S. C., and S. Rana, 2013: Interannual variability of winter Prakash, S., and Coauthors, 2015: Seasonal intercomparison of precipitation over northwest India and adjoining region: Im- observational rainfall datasets over India during the south- pact of global forcings. Theor. Appl. Climatol., 116, 609–623, west monsoon season. Int. J. Climatol., doi:10.1002/joc.4129, doi:10.1007/s00704-013-0968-z. in press. Kidd, C., P. Bauer, J. Turk, G. J. Huffman, R. Joyce, K.-L. Hsu, and Rahman, S. H., D. Sengupta, and M. Ravichandran, 2009: Vari- D. Braithwaite, 2012: Intercomparison of high-resolution ability of Indian summer monsoon rainfall in daily data from precipitation products over northwest Europe. J. Hydrome- gauge and satellite. J. Geophys. Res., 114, D17113, doi:10.1029/ teor., 13, 67–83, doi:10.1175/JHM-D-11-042.1. 2008JD011694. Kripalani, R. H., and P. Kumar, 2004: Northeast monsoon vari- Rajeevan, M., and J. Bhate, 2009: A high resolution daily gridded ability over south peninsular India vis-à-vis the Indian Ocean rainfall dataset (1971–2005) for mesoscale meteorological dipole mode. Int. J. Climatol., 24, 1267–1282, doi:10.1002/ studies. Curr. Sci., 96, 558–562. joc.1071. ——, S. Gadgil, and J. Bhate, 2010: Active and break spells of Krishnamurthy, V., and J. Shukla, 2000: Intraseasonal and inter- the Indian summer monsoon. J. Earth Syst. Sci., 119, 229–247, annual variability of rainfall over India. J. Climate, 13, 4366– doi:10.1007/s12040-010-0019-4. 4377, doi:10.1175/1520-0442(2000)013,0001:IAIVOR.2.0.CO;2. ——, C. K. Unnikrishnan, J. Bhate, N. K. Kumar, and P. P. Sreekala, Kucera, P. A., and Coauthors, 2013: Precipitation from space: 2012: Northeast monsoon over India: Variability and pre- Advancing Earth system science. Bull. Amer. Meteor. Soc., 94, diction. Meteor. Appl., 19, 226–236, doi:10.1002/met.1322. 365–375, doi:10.1175/BAMS-D-11-00171.1. Rasmussen, R., and Coauthors, 2012: How well are we measuring Kuligowski, R. J., 1997: An overview of National Weather Service snow? The NOAA/FAA/NCAR winter precipitation test quantitative precipitation estimates. TDL Office Note 97-4, bed. Bull. Amer. Meteor. Soc., 93, 811–829, doi:10.1175/ NOAA, Silver Spring, MD, 27 pp. [Available online at www. BAMS-D-11-00052.1. nws.noaa.gov/im/pub/tdl97-4.pdf.] Richards, F., and P. A. Arkin, 1981: On the relationship between Kumar, P., K. Rupa Kumar, M. Rajeevan, and A. K. Sahai, 2007: satellite-observed cloud cover and precipitation. Mon. Wea. On the recent strengthening of the relationship between Rev., 109, 1081–1093, doi:10.1175/1520-0493(1981)109,1081: ENSO and northeast monsoon rainfall over South Asia. Cli- OTRBSO.2.0.CO;2. mate Dyn, 28, 649–660, doi:10.1007/s00382-006-0210-0. Richman, M. B., 1986: Rotation of principal components. J. Cli- Lorenz, C., and H. Kunstmann, 2012: The hydrological cycle in matol., 6, 293–335, doi:10.1002/joc.3370060305. three state-of-the-art reanalyses: Intercomparison and perfor- Rudolf, B., and F. Rubel, 2005: Global precipitation. Observed mance analysis. J. Hydrometeor., 13, 1397–1420, doi:10.1175/ Global Climate, M. Hantel, Ed., Landolt-Börnstein: Numeri- JHM-D-11-088.1. cal Data and Functional Relationships in Science and Tech- Ma, L., T. Zhang, W. O. Frauenfeld, B. Ye, D. Yang, and D. Qin, nology, Vol. 6, Springer-Verlag, 11.01–11.53. 2009: Evaluation of precipitation from the ERA-40, NCEP-1, Saha, S., and Coauthors, 2010: The NCEP Climate Forecast System and NCEP-2 reanalyses and CMAP-1, CMAP-2, and GPCP-2 Reanalysis. Bull. Amer. Meteor. Soc., 91, 1015–1057, doi:10.1175/ with ground-based measurements in China. J. Geophys. Res., 2010BAMS3001.1. 114, D09105, doi:10.1029/2008JD011178. Sen Roy, S. S., 2006: The impacts of ENSO, PDO, and local SSTs Mahanta, R., D. Sama, and A. Choudhury, 2013: Heavy rainfall on winter precipitation in India. Phys. Geogr., 27, 464–474, occurrences over northeast India. Int. J. Climatol., 33, 1456– doi:10.2747/0272-3646.27.5.464. 1469, doi:10.1002/joc.3526. ——, 2009: A spatial analysis of extreme hourly precipitation Mishra, V., B. V. Smoliak, D. P. Lettenmaier, and J. M. Wallace, patterns in India. Int. J. Climatol., 29, 345–355, doi:10.1002/ 2012: A prominent pattern of year-to-year variability in Indian joc.1763. summer monsoon rainfall. Proc. Natl. Acad. Sci. USA, 109, Sevruk, B., 1985: Correction of precipitation measurements: Sum- 7213–7217, doi:10.1073/pnas.1119150109. mary report. Proc. Workshop on the Correction of Precipitation Misra, V., P. Pantina, S. C. Chan, and S. DiNapoli, 2012: A com- Measurements, Zurich, Switzerland, ETH/IAHS/WMO, 13–23. parative study of the Indian monsoon hydroclimate and its Shah, R., and V. Mishra, 2014: Evaluation of the reanalysis prod- variations in three reanalyses. Climate Dyn., 39, 1149–1168, ucts for the monsoon season droughts in India. J. Hydrome- doi:10.1007/s00382-012-1319-y. teor., 15, 1575–1591, doi:10.1175/JHM-D-13-0103.1.

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC APRIL 2015 R A N A E T A L . 651

Shea, D. J., and N. A. Sontakke, 1995: The annual cycle of pre- Wang, B., and Q. H. Ding, 2006: Changes in global monsoon precipitation over the Indian subcontinent: Daily, monthly and cipitation over the past 56 years. Geophys. Res. Lett., 33, seasonal statistics. NCAR Tech. Note NCAR/TN-4011STR, L06711, doi:10.1029/2005GL025347. 168 pp., doi:10.5065/D65M63M8. Wilks, D. S., 1995: Statistical Methods in Atmospheric Sciences. In- Shen, Y., A. Xiong, Y. Wang, and P. Xie, 2010: Performance of ternational Geophysics Series, Vol. 59, Academic Press, 464 pp. high-resolution satellite precipitation products over China. Xie, P., and Coauthors, 2003: GPCP pentad precipitation analyses: J. Geophys. Res., 115, D02114, doi:10.1029/2009JD012097. An experimental dataset based on gauge observations and sat- Sohn, S. J., C. Y. Tam, K. Ashok, and J. B. Ahn, 2012: Quantifying ellite estimates. J. Climate, 16, 2197–2214, doi:10.1175/2769.1. the reliability of precipitation datasets for monitoring large- Yao, T., and Coauthors, 2012: Different glacier status with atmo- scale East Asian precipitation variations. Int. J. Climatol., 32, spheric circulations in Tibetan Plateau and surroundings. Nat. 1520–1526, doi:10.1002/joc.2380. Climate Change, 2, 663–667, doi:10.1038/nclimate1580. Srinivasan, V., and K. Ramamurthy, 1973: Northeast monsoon. Yatagai, A., and P. Xie, 2006: Utilization of a rain-gauge-based Forecasting manual part IV: Comprehensive articles on se- daily precipitation dataset over Asia for validation of pre- lected topics, FMU Rep. IV-18.4, IMD, Pune, India. [Avail- cipitation derived from TRMM/PR and JRA25. Remote able online at www.imdpune.gov.in/weather_forecasting/ Sensing and Modeling of the Atmosphere, Oceans, and In- Forecasting_Mannuals/IMD_IV-18.4.pdf.] teractions, T. N. Krishnamurti, B. N. Goswami, and T. Iwasaki, Syed, F. S., F. Giorgi, J. S. Pal, and K. Keay, 2009: Regional climate Eds., International Society for Optical Engineering (SPIE model simulation of winter climate over central–Southwest Proceedings, Vol. 0604), 64040M, doi:10.1117/12.723829. Asia, with emphasis on NAO and ENSO effects. Int. J. Cli- ——, ——, and P. Alpert, 2008: Development of a daily gridded matol., 30, 220–235, doi:10.1002/joc.1887. precipitation dataset for the Middle East. Adv. Geosci., 12, Trenberth, K. E., 2011: Changes in precipitation with climate 165–170, doi:10.5194/adgeo-12-165-2008. change. Climate Res., 47, 123–138, doi:10.3354/cr00953. ——, O. Arakawa, K. Kamiguchi, H. Kawamoto, M. I. Nodzu, and ——,J.T.Fasullo,andJ.Mackaro,2011:Atmosphericmoisture A. Hamada, 2009: A 44-year daily gridded precipitation dataset transports from ocean to land and global energy ﬂows in reanalyses. for Asia based on a dense network of rain gauges. SOLA, 5, J. Climate, 24, 4907–4924, doi:10.1175/2011JCLI4171.1. 137–140, doi:10.2151/sola.2009-035. TRMM Precipitation Radar Team, 2011: Tropical Rainfall Measur- ——, K. Kamiguchi, O. Arakawa, A. Hamada, N. Yasutomi, and ing Mission (TRMM) Precipitation Radar algorithm instruction A. Kitoh, 2012: APHRODITE: Constructing a long-term daily manual for version 7. Tech. Rep., JAXA/NASA, 170 pp. gridded precipitation dataset for Asia based on a dense net- [Available online at www.eorc.jaxa.jp/TRMM/documents/ work of rain gauges. Bull. Amer. Meteor. Soc., 93, 1401–1415, PR_algorithm_product_information/pr_manual/PR_Instruction_ doi:10.1175/BAMS-D-11-00122.1. Manual_V7_L1.pdf.] Yilmaz, K., T. S. Hogue, K.-L. Hsu, S. Sorooshian, H. V. Gupta, and Turk, F. J., P. A. Arkin, E. E. Ebert, and M. R. P. Sapiano, 2008: T. Wagener, 2005: Evaluation of rain gauge, radar and satellite- Evaluating high-resolution precipitation products. Bull. Amer. based precipitation estimates with emphasis on hydrologic Meteor. Soc., 89, 1911–1916, doi:10.1175/2008BAMS2652.1. forecasting. J. Hydrometeor., 6, 497–517, doi:10.1175/JHM431.1. Turner, A. G., and H. Annamalai, 2012: Climate change and the Zhang, Q., H. Kornich, and K. Holmgren, 2013: How well do re- South Asian summer monsoon. Nat. Climate Change, 2, 587– analyses represent the southern African precipitation? Climate 595, doi:10.1038/nclimate1495. Dyn., 40, 951–962, doi:10.1007/s00382-012-1423-z.

Unauthenticated | Downloaded 10/07/21 05:09 PM UTC