Tibetan Plateau precipitation as depicted by gauge observations, reanalyses and satellite retrievals

Transcription

1 INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. (2013) Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: /joc.3682 Tibetan Plateau precipitation as depicted by gauge observations, reanalyses and satellite retrievals Kai Tong, a,b Fengge Su, b * Daqing Yang, b,c Leilei Zhang b and Zhenchun Hao a a State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering, Hohai University, Nanjing, China b Key Laboratory of Tibetan Environment Changes and Land Surface Processes, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China c National Hydrology Research Center, Environment Canada, Saskatoon, Canada ABSTRACT: The European Centre for Medium-range Weather Forecasts (ECMWF) reanalysis ERA-40, ERA-Interim, University of Washington (UW) data, APHRODITE s Water Resources (APHRODITE), and Tropical Rainfall Measuring Mission (TRMM) Multisatellite Precipitation Analysis (TMPA) precipitation estimates are compared with each other and with the corrected gauge observations over the Tibetan Plateau (TP) at both basin and plateau scales. The ERA-40 generally can capture the broad spatial and temporal distributions in the gauge-based precipitation estimates over the TP. However, the ERA-40 shows little agreement with the gauge-based precipitation in annual variations for the years before The anticipated improvements in the ERA-Interim precipitation relative to ERA-40 have not been realized in this study. It greatly overestimates the Corrected-China Meteorological Administration (CMA) (by %) and other datasets, although the ERA-Interim has a better correspondence than ERA-40 with the Corrected-CMA data at both annual and monthly scales among the selected basins. All the products can detect the large-scale precipitation regime, including the monsoon-dominated precipitation in summer and the westerly-wind-induced precipitation in winter. The Corrected-CMA and APHRODITE estimates generally show decreasing trends in summer and increasing trends in spring and winter precipitation during at both basin and plateau scales. However, the Corrected-CMA shows larger values in trends and more cases with significance than the APHRODITE, suggesting the effects of the undercatch corrections on the precipitation trends. The use of precipitation derived from current reanalysis projects is less preferable for hydrology analysis than the TP observational data at basin scales. However, using gauge-based precipitation datasets as hydrologic model forcings should be careful in the river basins where gauge station network is spare, such as in the Yarlung zangbo river basin. Satellite products still hold a great potential for providing high-resolution precipitation information in remote regions such as the western TP, although more evaluations are needed on the feasibility of satellite precipitation products on the TP where the topography is complex and rainfall rate is highly variable. Copyright 2013 Royal Meteorological Society KEY WORDS Tibetan Plateau; precipitation; climatology Received 11 July 2012; Revised 11 January 2013; Accepted 16 January Introduction The Tibetan Plateau (TP), with an average elevation of over 4000 m above sea level and an area of approximate 2.5 million km 2, is the highest and most extensive highland in the world. The TP is of considerable importance to the Asian monsoon and global general circulation via mechanical and thermal forcings (Ye and Gao, 1979; Yanai et al., 1992; Webster et al., 1998) due to its unique altitude and horizontal extent. The TP is also the source of major Asian rivers (e.g. the Indus, Ganges, Brahmaputra, Yangtze, and Yellow rivers, Figure 1), which support hundreds of millions of people downstream. Therefore, the TP is called Asian water tower (Ye and Gao, 1979; * Corresponding author: F. Su, Key Laboratory of Tibetan Environment Changes and Land Surface Processes, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China. fgsu@itpcas.ac.cn Immerzeel et al., 2010). Studies have suggested a general warming trend over the TP during recent decades (Lin and Zhao, 1996; Liu and Chen, 2000; Frauenfeld et al., 2005; Wang et al., 2008). This warming trend led to glacier retreat (Yao et al., 2004) and permafrost degradation (Wu and Zhang, 2008, 2010). Changes in climate, glacier, and permafrost greatly affect the hydrological cycle over the TP regions. The recent changes over the TP point to the need to better understand the interactions between climate and hydrology systems. Land surface model is a useful tool in studying the hydrology systems response to climate changes. Precipitation is the most important atmospheric input to the terrestrial hydrologic system. Hydrologic simulations by land surface models are highly sensitive to the accuracy of the precipitation forcing (Gourley and Vieux, 2005). Reliable estimates of precipitation are necessary to assess climate variability/change and to drive hydrology models, but they are unfortunately Copyright 2013 Royal Meteorological Society

2 K. TONG et al. Figure 1. Location and topography of Tibetan Plateau (a), and six source river basins and distribution of meteorological stations in the Tibetan Plateau (b). The sequence numbers 1, 2, 3, 4, 5,and 6 in (b) denote the source regions of the Yellow, Yangtze, Yalong, Lancang (Mekong), Nujiang (Salween) and Yarlung zangbo (Brahmaputra) river basins, respectively. Red points denote meteorological stations and green ones streamflow stations. The contour of 2000 m indicates the boundary of study domain in this work. This figure is available in colour online at wileyonlinelibrary.com/journal/joc sparse or nonexistent in many remote parts of the world. This is especially true in the TP. Due to the high elevation, complex terrain, severe weather, and the inaccessibility, direct meteorological observations do not exist over large portions of the TP, especially in the western part of the plateau (Figure 1(b)). Knowledge on the spatial and temporal characters and variations of precipitation over the TP is thus greatly incomplete. This will hamper understanding the climate variability and the corresponding hydrological response over the TP. Precipitation information and estimates from atmospheric reanalysis with their spatial and temporal continuity provide a potential alternative in regions where conventional in situ precipitation measurements are not readily available. Atmospheric reanalysis projects, based on frozen forecast and data assimilation systems, yield long-term records of atmospheric and surface fields and provide continuous global gridded data which have been widely used (Serreze and Hurst, 2000; Betts, 2004; Serreze et al., 2005). ERA-40 (Uppala et al., 2005) is a reanalysis data set produced by the European Center for Medium-Range Weather Forecasts (ECMWF) (Simmons and Gibson, 2000). A number of evaluations for ERA-40 precipitation have been performed at global or continental scales (Serreze et al., 2005; Su et al., 2006; Voisin et al., 2008; Ma et al., 2009; Su and Lettenmaier, 2009). These studies suggest that the ERA-40 reanalysis data (particularly for the years after 1970) could be a useful resource for depictions of seasonal and monthly variations of precipitation for large river basins (e.g. the Arctic river basins, La Plata basin in South America), while some studies demonstrated the regional limitations of the ERA-40 precipitation (Betts et al., 2005). It is worthy to examine the validity of ERA-40 precipitation estimates in the TP, which is characterized by complex topography.

3 TIBET PRECIPITATION DEPICTED BY GAUGE, REANALYSES, AND SATELLITE DATA ERA-Interim (Simmons et al., 2006; Dee and Uppala, 2009; Dee et al., 2011) is the latest global atmospheric reanalysis produced by the ECMWF. Various difficulties encountered in the production of ERA-40 are revised in ERA-Interim, including the representation of the hydrological cycle, the quality of the stratospheric circulation, and the consistency in time of the reanalysed fields. Relative to the ERA-40 system, ERA-Interim incorporates many important model improvements such as resolution and physics changes, the use of four-dimensional variational (4D-Var) data assimilation, and various other changes in the analysis methodology (Dee et al., 2011). Although atmospheric reanalysis precipitation estimates are widely used and could well capture general precipitation features at continental scales, there are increasing demands for gauge-based precipitation products for the validation of simulation products of numerical models and satellite-based precipitation estimates (Turk et al., 2008). Considerable efforts have been made in developing gridded precipitation datasets based on gauge observations. A number of global monthly gauge-based precipitation (gridded) products have been developed (New et al., 1999, 2000; Willmott and Matsuura, 2001; Chen et al., 2002; Adler et al., 2003; Hijmans et al., 2005; Mitchell and Jones, 2005). However, regional analysis requires higher resolution datasets both in space (10 of kilometres) and time (sub-daily or daily data). Daily gridded precipitation products are especially expected to be very useful to verify high-resolution climate model simulations that include extreme events and to drive hydrological models (Xie et al., 2007). A daily gaugebased precipitation dataset developed by the University of Washington (UW) (Adam and Lettenmaier, 2003) was used to depict the precipitation over the TP in this study. The UW data was derived from the University of Delaware dataset (Willmott and Matsuura, 2001) and corrected using climatologic factors in an attempt to more accurately reflect likely precipitation values. Another suite of gauge-based precipitation products, created by the Asian Precipitation Highly Resolved Observational Data Integration Towards Evaluation of the Water Resources (APHRODITE s Water Resources) Project (APHRODITE) (Yatagai et al., 2009, 2012) were used to characterize the precipitation over the TP. The APHRODITE daily gridded precipitation dataset consists of rain gauge observation data. It is the only long-term continental-scale daily product that contains a dense network of daily rain gauge data for Asia, including the Himalayas and mountainous areas in the Middle East (Yatagai et al., 2009). APHRODITE has been used to investigate the precipitation climatology and evaluate water resources (Bai et al., 2011; Fukutomi et al., 2012; Yen et al., 2011; Raziei et al., 2012), verify model simulations (Kusunoki et al., 2011; Yang and Wang, 2012), and drive hydrologic models (Jaranilla-Sanchez et al., 2011). In situ observations and datasets are always restricted by spatial or temporal sampling limitations; we thus also use satellite precipitation data with high spatiotemporal resolution. The Tropical Rainfall Measuring Mission (TRMM) Multisatellite Precipitation Analysis (TMPA) provides a calibration-based sequential scheme for combining precipitation estimates from multiple satellites, as well as gauge analyses where feasible, at fine scales ( and 3 hourly) (Huffman et al., 2007). We also obtain gauge data from 176 meteorological stations over the TP (Figure 1(b)) from the National Climate Center, China Meteorological Administration (CMA). These precipitation measurements by gauges have been adjusted for wind-induced undercatch, trace amount of precipitation, and wetting loss as described by Ye et al. (2004) (Corrected-CMA data). In this study, we compare the ERA-40, ERA-Interim, UW data, APHRODITE and TMPA precipitation estimates with each other and with Corrected-CMA data over the TP, and investigate the annual, seasonal, inter-annual, and spatial variations and trends of precipitation at both basin and plateau scales. The aim of this work is to: (1) assess the agreements among the selected precipitation data over the TP; (2) understand the spatial and temporal characteristics and variations of precipitation over the TP at both basin and plateau scales; and (3) assess the feasibility of using these products as hydrological model inputs. 2. Dataset and methodology The Tibetan Plateau is bordered on the south by Myanmar, Bhutan and Nepal and by India and Pakistan on the western side (Figure 1(a)). TP has no definite boundary before, in this study we define the boundary based on elevation of 2000 m between E and N (Figure 1(b)). Then Corrected-CMA data was used to interpolate inside of the domain at a fine spatial resolution (1/12 ), so all maps from Corrected-CMA data closely agree with the 2000 m contour. The TP is referred to the area within the boundary hereafter. Figure 1(b) shows the spatial distribution of the 176 meteorological stations of CMA, with most gauges located in the southern and southeastern TP and very few in the middle and west part of the TP. In addition, all the stations are limited within China. Most of these meteorological stations have reliable continuous records since the 1960s. The data from all these stations have undergone quality control procedures to eliminate erroneous and homogenous assessment in CMA. Additional routines to identify potential outliers (e.g. daily precipitation values less than 0 mm) were manually checked (either validated, corrected or removed). The daily corrected precipitation for the 176 stations (Corrected-CMA data) covers the period Both ERA-40 and ERA-Interim monthly precipitation reanalysis data were obtained from the ECMWF website ( The ERA-40 covers a period of September 1957 to August 2002 with a spatial resolution of 2.5 latitude 2.5 longitude grid, while the ERA-Interim has a finer spatial resolution

4 K. TONG et al. of 1.5 latitude 1.5 longitude for the period 1979 to present. The UW daily precipitation product was downloaded from Eemaurer/global_data/. It is a global, gridded dataset for the period The UW data was derived from the University of Delaware dataset (UDel data, Willmott and Matsuura 2001), which was corrected using climatologic factors. The UDel data used Global Historical Climatology Network (GHCN version 2) and Legates and Willmott s (1990) station records. The UDel precipitation data was adjusted for gauge catch biases. Gauge type-specific regression equations from the World Meteorological Organization (WMO) Solid Precipitation Measurement Intercomparison (Goodison et al., 1998) and a methodology described by Legates and Willmotts (1990) were applied to adjust wind-induced undercatch of solid and liquid precipitation, respectively. Application of bias corrections to the Willmott and Matsuura (2001) data resulted in 11.7% increase in global terrestrial mean annual precipitation (Adam and Lettenmaier, 2003). The APHRODITE data are a suite of precipitation products constructed by the APHRODITE s water resources project, The APHRODITE daily (MA V1003R1) gauge-based gridded ( ) precipitation dataset spans the period , covering the whole Asia. This product is based on data collected at stations, which represents times the data made available through the Global Telecommunication System network and used for most daily gridded precipitation products (Yatagai et al., 2009, 2012). It has contributed to various studies, such as evaluation of Asian water resources, diagnosis of climate change, statistical downscaling, and verification of numerical model simulation and high-resolution precipitation estimates using satellites (Yatagai et al., 2009). TMPA is based on the calibration by the TRMM Combined Instrument and TRMM Microwave Imager precipitation products, respectively. The after-real-time product system has been developed as the version 6 algorithm for the TRMM operational product 3B42 (3B42 V6). The TMPA research product (3B42 V6) was used in this study and it covers the latitude band 50 N S at grids for the period from 1998 to the delayed present. The TMPA used two gaugebased datasets for monthly bias adjustment The Global Precipitation Climatology Project (GPCP, Rudolf, 1993) and the Climate Assessment and Monitoring System (CAMS) monthly rain gauge analysis (Xie et al., 1996). To facilitate direct comparisons among the precipitation data sets and keep consistent with our hydrological model setup over the TP, all the gridded data set were regridded to 1/12 1/12 grids using the nearest neighbour method, meaning that the precipitation value of the small grid (1/12 ) is assigned with the value of the big grid (e.g. ERA 2.5 grid) which is nearest to the small one. Therefore, the resulted grids usually preserve the spatial pattern and topographic features of the data at original resolutions. An inverse distance weighting (IDW) interpolation method was used for the Corrected-CMA station data from points to 1/12 1/12 grids. The IDW method enjoys a long history of usage and reliability, due primarily to its simplicity of formulation and its persistent application in operational settings (Kurtzman et al., 2009). IDW assigns weights to neighbouring observed values based on distance to the interpolation location and the interpolated value is the weighted average of the observations. IDW has been demonstrated efficient and reliable in many precipitation interpolation applications (e.g. Nalder and Wein, 1998; Garcia et al., 2008; Ly et al., 2011; Chen and Liu, 2012). Other statistical interpolation methods like multiple linear regression, optimal interpolation or Kriging can perform better, but only if data density is sufficient (Eischeid et al., 2000). With sparse rain gauge networks, IDW is suggested for spatial rainfall estimates (Hsieh et al., 2006). Six source river basins in the TP were selected to evaluate the consistency between different precipitation datasets the upstream of the Lancang (Mekong), Nujiang (Salween), Yarlung zangbo (Brahmaputra), Yellow, Yalong (a tributary of Yangtze), and Yangtze rivers (Figure 1(b)). Table I provides information on the characteristics of the six basins. All these selected basins are located in the southeast of TP where there is relatively dense station coverage (Figure 1(b)). In this study, we arbitrarily choose the Corrected-CMA precipitation data as the reference to evaluate the performance of other datasets and products over the TP. In the evaluation, basin-mean precipitation estimates from the ERA-40, ERA-Interim, UW data, APHRODITE and TMPA were compared with the Corrected-CMA data. Several statistical indices were used to describe the agreements between the reference and other datasets. They are normalized root-mean-square error (Nrmse), correlation coefficient (R 2 ), and bias (BIAS). Nrmse, R 2 and BIAS are defined as follows: Nrmse = 1 n n ( ) 2 Ppi P gi i=1 1 n (1) n P gi i=1 n ( )( ) Pgi P gi Ppi P pi R 2 = i 1 n ( ) 2 (Ppi ) 2 Ppi P gi P gi BIAS = i 1 n ( ) P pi P gi i=1 2 (2) 100% (3) n P gi i=1

5 TIBET PRECIPITATION DEPICTED BY GAUGE, REANALYSES, AND SATELLITE DATA Rivers Hydrological stations Table I. Characteristics of six upstream river basins in the TP. Station location Latitude Longitude Basin average elevation (m) Drainage area of upstream of control stations (km 2 ) Percent of total river basin area (%) Lancang Changdu Nujiang Daojieba Yarlungzangbo Nuxia Yellow Tangnaihai Yalong Yajiang Yangtze Zhimenda where P gi denotes basin-mean annual or monthly Corrected-CMA precipitation data; P pi is basin-mean ERA-40, ERA-Interim, UW data, APHRODITE and TMPA; n is the number of annual or monthly precipitation pairs in the analysis. The Mann Kendall test method (Mann, 1945; Kendall, 1975) was used for detecting the temporal and spatial trends of precipitation over the TP. This method is a nonparametric trend test and it has been widely used in change detection for hydroclimatic and related variables, such as streamflow, temperature, and precipitation (Hirsch et al., 1982; Burn and Hag Elnur, 2002; Yue and Wang, 2002; Xu et al., 2003; Ludwig et al., 2004). In this article, a trend is considered to be statistically significant if it is at the 5% confident level. 3. Result 3.1. Basin-mean precipitation In this section, the basin-mean precipitation estimates from the ERA-40, ERA-Interim, UW data, APHRODITE, and TMPA were compared to each other and with the Corrected-CMA data over the six river basins at annual, seasonal, and monthly time scales. Figure 2 shows the time series of annual basin-mean precipitation from all datasets. The ERA-40 reanalysis is considerably out of phase in annual amount and variation, compared to Corrected-CMA data and other data sets. The correlations with Corrected-CMA data are below 0.2 over all the river basins except for the Yarlung zangbo during the overlap period (Table III). Relative to the Corrected-CMA, the ERA-40 underestimates mean precipitation by 20 32% over the Lancang, Yellow, Yalong, and Yangtze during However, for the Nujiang and Yarlung zangbo, the ERA-40 tends to overestimate the annual precipitation by 7 and 87%, respectively (Table II). Annual precipitation from the ERA-Interim is consistently higher than the other datasets over TP, with % more than the Corrected- CMA means during among the six basins (Table II). However, the ERA-Interim shows much better correspondence with Corrected-CMA data than the ERA- 40 in annual variations, with R 2 of among the basins (Table III). The UW precipitation data show a sudden drop in the early 1990s over all the basins, which are not observed in other datasets (Figure 2). The possible reason for this sharp drop might be the degradation of gauge network used in the Willmott and Matsuura dataset (2001) after 1990 (Adam and Lettenmaier, 2003). Therefore the statistics results in Tables II and III for the UW data only include the period Despite the drop in the 1990s, the UW data show reasonable agreements with Corrected-CMA data during , with R 2 of to among the basins (Table III). Relative to Corrected-CMA, the UW data underestimate annual mean precipitation over all the basins by 7 16% during (Table II), except for Nujiang and Yarlung zangbo where the UW data overestimate the Corrected- CMA by 88% during Due to the sparse precipitation gauge stations within the Yarlung zangbo river basin (Figure 1(b)), it is difficult to judge which dataset is closer to the reality. Over the Nujiang basin, the UW annual precipitation is fairly close to the Corrected- CMA data, with the total bias of 3% and Nrmse of 5.7% for Both the UW and Corrected- CMA data include undercatch adjustments, but they are from different data sources. The UW data are derived from Willmott and Matsuura (2001) gridded dataset, while the Corrected-CMA data are directly from gauge measurements. Different data sources and different approaches used in the undercatch adjustments may contribute to the difference between the two datasets. Relative to Corrected-CMA data, the APHRODITE consistently underestimate annual precipitation 13 25% over the six basins for The smallest bias was found in the Yalong basin and the largest bias was observed in the Yarlung zangbo basin (Table II). These results are not completely unexpected, given the fact that no bias corrections for gauge undercatch have been done in the APHRODITE. In spite of differences, the annual precipitation estimates from the APHRODITE are well correlated with the Corrected-CMA data, with R 2 above 0.9 for the Lancang, Yellow, and Yalong, and over 0.7 for the other three basins. The Nrmse, relative to the Corrected-CMA annual precipitation, is below 26% for all basins (Table III). The overlap between TMPA and Corrected-CMA data covers a short period of The TMPA shows

6 K. TONG et al. Figure 2. Time series of annual basin-averaged precipitation from the ERA-40 ( ), ERA-Interim ( ), UW data ( ), APHRODITE ( ), TMPA ( ) and Corrected-CMA data ( ) over the six upstream river basins in the TP. This figure is available in colour online at wileyonlinelibrary.com/journal/joc comparable skills with the APHRODITE in annual precipitation estimates, with negative bias of 8 23% (Table II) and R 2 above 0.7 (except for the Yangtze with R 2 of 0.6) relative to the Corrected-CMA data (Table III). Figure 3 shows mean annual cycle of precipitation from the six datasets over the basins. It should be noted that the averaging periods are different due to the difference in the overlapping periods with the Corrected-CMA. We assume that the averaging periods are long enough such that they capture sufficiently well the seasonal pattern of the precipitation regime. All basins show similar precipitation regime among the six datasets, with more than 70% of annual total precipitation occurring in summer and autumn, and less than 10% in winter (Table II). It is worth noting that, although the ERA-40 is biased from the other datasets in annual precipitation variations over most of the TP basins (Figure 2), it is able to capture the seasonal precipitation patterns (Figure 3). Consistent with the results in Figure 2, the ERA-Interim precipitation estimates are very higher than the other products for all seasons and all basins, with biases of % relative to the Corrected-CMA data for the Yarlung zangbo. However, the ERA-Interim is capturing the general precipitation regime of all the basins, with 65 75% of annual precipitation in summer and autumn, 22 25% in spring, and 2 6% in winter. In spite of the consistency in seasonal variations, precipitation estimates from the ERA-40, UW data, APHRODITE, and TMPA tend to be lower than the Corrected-CMA estimates for all the seasons over most basins except for the Yarlung zangbo. The ERA-40 has the lowest estimates among the datasets over the Lancang, Yellow, Yalong and Yangtze river basins, with 15 42% lower than the reference during summer and autumn. The APHRODITE underestimates the Corrected- CMA estimates during summer and autumn by 10 27% over all the basins, while the underestimation in the UW summer and autumn precipitation is within 15% for all

7 TIBET PRECIPITATION DEPICTED BY GAUGE, REANALYSES, AND SATELLITE DATA Table II. Annual and seasonal bias of precipitation estimates in the ERA-40, ERA-Interim, UW data, APHRODITE and TMPA relative to the Corrected-CMA data over the six river basins defined in Table I. Mean (mm) Annual Spring Summer Autumn Winter Bias (%) Mean (mm) Bias (%) Mean (mm) Bias (%) Mean (mm) Bias (%) Mean (mm) ERA-40 ( ) Lancang Nujiang Yarlung zangbo Yellow Yalong Yangtze ERA-Interim ( ) Lancang Nujiang Yarlung zangbo Yellow Yalong Yangtze UW data ( ) Lancang Nujiang Yarlung zangbo Yellow Yalong Yangtze APHRODITE ( ) Lancang Nujiang Yarlung zangbo Yellow Yalong Yangtze TMPA ( ) Lancang Nujiang Yarlung zangbo Yellow Yalong Yangtze Bias (%) the basins except for the Nujiang and Yarlung zangbo. The UW data and ERA-40 have comparable estimates over the Yarlung zangbo for all the seasons, with the overestimation of % for the ERA-40 and % for the UW data, relative to the Corrected-CMA data. Figure 4 shows the scatter plots of monthly basinmean precipitation among Corrected-CMA data and other datasets. The R 2 (with the mean annual cycle removed) and Nrmse for each basin are given in Table III. The UW and APHRODITE exhibit best correspondence with Corrected-CMA data, with R 2 above 0.65 for most of the basins; while the scatter plots between the ERA- 40 and the Corrected-CMA show the scatterest with R 2 of among the basins (Table III). Despite the great overestimation throughout the basins, the ERA- Interim is well correlated with the Corrected-CMA for monthly variations, with R 2 of among the basins (Table III). TMPA has a moderate correlation with the Corrected-CMA months with R 2 of among the basins (Table III). Linear regression lines for each pair of datasets are also included in Figure 4. The ERA-Interim considerably overestimates the Corrected-CMA throughout the basins, with Nrmse of % among the basins in months during The linear regression lines between the ERA-Interim and Corrected-CMA data lie on the far left of the 1:1 line. For the basins of Lancang, Yellow, Yalong, and Yangtze, the linear regression lines from the UW, APHRODITE, and TMPA data are much closer to the 1:1 line than those from the ERA-40, although they all lie on the right of the 1:1 line. The UW data have the least Nrmse (13 20%), followed by the APHRODITE (Nrmse of 17 25%) and TMPA (Nrmse of 24 40%), and the ERA-40 has the largest Nrmse (49 55%) relative to the monthly Corrected-CMA estimates (Table III). The ERA-40 monthly precipitation estimates show the best performance over the Nujiang and are even comparable to the performance of the APHRODITE and TMPA in terms of Nrmse (30%), consistent with the seasonal results for the Nujiang in Figure 3. Over the

8 K. TONG et al. Figure 3. Mean annual cycles of precipitation (mm) from the ERA-40 ( ), ERA-Interim ( ), UW data ( ), APHRODITE ( ), TMPA ( ) and Corrected-CMA data ( ) over the Lancang, Nujiang, Yarlung zangbo, Yellow, Yalong and Yangtze river basins. Yarlung zangbo, the ERA-40, ERA-Interim, and UW data considerably overestimate the Corrected-CMA with Nrmse around 120% in ERA-40 and UW data and around 200% in the ERA-Interim. The APHRODITE and TMPA have similar performance over the Yarlung zangbo, with most of the dots lying on the right of the 1:1 line, and Nrmse of 22 36%. In general, the performance of TMPA is slightly worse than the gauged-based products (UW and APHRODITE data), but much better than the reanalyses (ERA-40 and ERA-Interim) estimates, in terms of both Nrmse and R 2 for all the basins (Table III) Spatial fields It is useful to examine the spatial fields of precipitation over the entire TP. Figures 5 and 6 present the spatial distribution of annual and seasonal mean precipitation from all the datasets over the TP. Results from the Corrected- CMA only cover the areas east of E80 in the TP because

9 TIBET PRECIPITATION DEPICTED BY GAUGE, REANALYSES, AND SATELLITE DATA Table III. Summaries of Nrmse and correlation coefficient (R 2 ) relative to the Corrected-CMA data over the six source river basins in the TP. Annual Monthly R 2 Nrmse (%) R 2 Nrmse (%) ERA-40 and Corrected-CMA data ( for the first column, for the second column) Lancang 0.162/ / / /53.4 Nujiang 0.048/ / / /27.4 Yarlung zangbo 0.293/ / / /122.3 Yellow 0.011/ / / /60.2 Yalong 0.006/ / / /49.0 Yangtze 0.192/ / / /58.2 ERA-Interim and Corrected-CMA data ( ) Lancang Nujiang Yarlung zangbo Yellow Yalong Yangtze UW data and Corrected-CMA data ( ) Lancang Nujiang Yarlung zangbo Yellow Yalong Yangtze APHRODITE and Corrected-CMA data ( ) Lancang Nujiang Yarlung zangbo Yellow Yalong Yangtze TMPA and Corrected-CMA data ( ) Lancang Nujiang Yarlung zangbo Yellow Yalong Yangtze the fields west of E80 are not considered reliable due to the scarce gauge stations there (Figure 1(b)). In addition, the stations included in the Corrected-CMA are limited within China, therefore areas outside of China may not be properly reflected, such as the regions along the southern Himalayas (Figures 5 and 6). The general spatial pattern of the annual mean precipitation fields is roughly in agreement among the datasets. Annual precipitation exhibits an east to west gradient, ranging from over mm year 1 in the southeast to less than 100 mm year 1 in the west. The UW data and APHRODITE have consistent spatial variations in annual means over the TP (Figure 5(c) and (d)), and similar to the Corrected-CMA (Figure 5(f)) for the region east of E80. However, apparent disagreements are found over the upstream of the Yarlung zangbo, where mean annual precipitation from the APHRODITE and Corrected-CMA are less than 400 mm due to the rain shadow effect of Himalayas. While the estimates from the UW data reach to 2000 mm over the upstream of the Yarlung zangbo, suggesting that the rain shadow effects of the Himalayas were not captured well in the UW estimates (Figure 5(d)). The satellite-based estimates (TMPA, Figure 5(e)) resemble well the annual spatial variations in both APHRODITE and the Corrected-CMA for the southeastern TP (including all the six selected basins), probably due to the monthly gauge adjustments included in the TMPA (Huffman et al., 2007). However, the westerlywind-induced precipitation signals in the western TP (Wang et al., 2005) are not apparent in the TMPA, as reflected in the UW data and APHRODITE data (Figure 5(c) and (d)). ERA-40 (Figure 5(a)) can roughly follow the largescale spatial patterns of the gauge-based estimates; however, large discrepancies between ERA-40 and the three gauge-based estimates exist. For instance, in the very southeast corner of TP, the ERA-40 underestimates all the gauge-based estimates at least by 200 mm year 1 in annual means. For the Yarlung zangbo basin, the ERA-40 and UW data have similar biases (87 88%) relative to the Corrected-CMA in basin-mean annual precipitation

10 K. TONG et al. Figure 4. Scatter grams of monthly basin-averaged precipitation from each products during the overlap with the Corrected-CMA data [ERA-40 ( ), ERA-Interim ( ), UW data ( ), APHRODITE ( ) and TMPA ( )] over the six selected basins. This figure is available in colour online at wileyonlinelibrary.com/journal/joc (Table II). However, they show different spatial patterns (Figure 5(a) and (c)), suggesting big uncertainties in precipitation estimates for this basin. The ERA-40 estimates show less spatial variations than the other datasets mostly due to its coarse resolution. Although the ERA-Interim (Figure 5(b)) can capture the large-scale spatial patterns of annual precipitation with the highest precipitation predominantly in the southeast of the region, the tendency of great overestimation compared to the other datasets for the entire TP is clearly visible in Figure 5(b). Figure 6 shows maps of seasonal mean precipitation from all the datasets over the TP. All products show similar spatial patterns, with precipitation peaks (>200 mm mon 1 ) in summer over the southeastern TP and the dry season appears in winter (<20 mm mon 1 ) in the middle and southeast of the domain. In summer months, the heavy precipitation in the southeast TP is mainly produced by the southeast monsoon and the monsoon weakens from east to west. In winter and spring, westerly winds bring moisture to the west TP, but the amount is much less than the summer precipitation from the east monsoon. Therefore, summer precipitation mostly dominates the annual spatial pattern. Figure 6 suggests that all the products can detect the large-scale regime, including the westerly-wind-induced precipitation in winter. One exception is the TMPA, where the westerly signals are week (Figure 6). The ERA-Interim shows a persistent tendency of overestimation relative to the other datasets throughout the region and seasons Trend analysis Precipitation is a key driver of river runoff. Understanding precipitation changes at basin scales helps to understand the streamflow response to climate change in the TP. Trends assessment using reanalysis data is dangerous due to changes in the amount and quality of assimilation data (Frauenfeld et al., 2005). In this section, we investigate precipitation changes over both basin and plateau scales from two long-term gauge-based precipitation data (APHRODITE and Corrected-CMA data). Due to the sudden change in precipitation after 1990 (Figure 2), the UW data were not used for the trend analysis. Figure 7 shows the annual and seasonal precipitation trends from the Corrected-CMA and APHRODITE over the six source river basins for the overlap period of Table IV also gives trend results, with significant changes in the bolded numbers. The two datasets show consistent tendency in annual precipitation changes for all the basins except for the Lancang, where the APHRODITE suggests an increase (2.7 mm decade 1 ), while the Corrected-CMA data has a weak decreasing trend ( 0.9 mm decade 1 ). However, significant changes in annual precipitation are only observed in Nujiang from the Corrected-CMA estimates with a trend of

11 TIBET PRECIPITATION DEPICTED BY GAUGE, REANALYSES, AND SATELLITE DATA Figure 5. Spatial patterns of mean annual precipitation (mm) from the ERA-40 ( ), ERA-Interim ( ), UW data ( ), APHRODITE ( ), TMPA ( ) and Corrected-CMA data ( ) over the TP. This figure is available in colour online at wileyonlinelibrary.com/journal/joc 19.3 mm decade 1. Some consistencies are also observed in seasonal changes for the two datasets. Both datasets show increasing trends in winter precipitation for all the basins except for the APHRODITE in the Nujiang and Yarlung zangbo. The increases are statistically significant in most basins (except Yellow) from the Corrected- CMA data (trends of mm decade 1 ) and from the APHRODITE records (trends of mm decade 1 ) for the three basins (Lancang, Yalong, and Yangtze). On the other hand, a consistent decreasing trend in summer precipitation is observed in both datasets for all the basins, except for the Yarlung zangbo and Yangtze from the APHRODITE estimates. However, the changes in summer precipitation are not significant for any basins. Spring precipitation shows increasing trends for all the basins in both datasets (Table IV). The trends are significant for the Lancang, Nujiang, and Yalong in the APHRODITE estimates ( mm decade 1 )and Lancang, Nujiang, Yarlung zangbo, and Yalong in the Corrected-CMA estimates ( mm decade 1 ). The trends in autumn are only significant for the Nujiang in the Corrected-CMA data (7.4 mm decade 1 ), while none of the basins shows significant changes in the APHRODITE records.

12 K. TONG et al. Figure 6. Spatial patterns of seasonal mean (mm mon 1 ) from the ERA-40 ( ), ERA-Interim ( ), UW data ( ), APHRODITE ( ), TMPA ( ) and Corrected-CMA data ( ) over the TP. This figure is available in colour online at wileyonlinelibrary.com/journal/joc Despite the difference in the rates and significance, both datasets suggest, over the six basins in the TP, a decreasing trend in summer precipitation and an increasing trend in spring and winter during Figure 8 shows spatial pattern of annual precipitation trends (mm decade 1 ) for the Corrected-CMA and APHRODITE data during There are extensive regions with an increasing tendency throughout the central, southeastern and northern TP in the Corrected-CMA data (Figure 8(a)). The increasing trends in the southeastern part of the TP (20 50 mm decade 1 ) are generally greater than those in the northern part of the region (0 10 mm decade 1 ). The largest increasing rates ( mm decade 1 ) are located around E95 and N30. Areas with a decreasing tendency (Figure 8(a)) mostly include the eastern edge of TP, the upstream of the Yangtze and Yellow river, Qiandam Basin, and the upstream of the Yarlung zangbo. When averaged over the entire region east of E80, the annual precipitation from the Corrected-CMA data indicate a statistically significant increase by 7.6 mm decade 1 during (Table IV). Trends estimated from the APHRODITE (Figure 8(b)) have similar spatial pattern as those from the Corrected- CMA data for the same domain. There are larger areas with a decreasing tendency especially over the upstream of Yarlung Zangbo, Nujiang, and the Qiandam Basin. The mean precipitation both over the entire TP and over the region east of E80 from the APHRODITE show an increasing trend during but only significant for the region east of E80 with a trend of 7.6 mm decade 1 (Table IV). Figure 9 presents maps of seasonal precipitation trends from the Corrected-CMA and APHRODITE data over the TP during There is a widespread increasing tendency over most regions for the winter and spring Corrected-CMA precipitation (Figure 9(a)). There are very small areas with a decreasing trend, for instance, the eastern and northwestern edges of the domain for spring precipitation. The regional mean precipitation

13 TIBET PRECIPITATION DEPICTED BY GAUGE, REANALYSES, AND SATELLITE DATA Figure 7. Annual and seasonal precipitation trends (mm decade 1 ) for the six source river basins in the TP from the APHRODITE and Corrected- CMA estimates during from the Corrected-CMA exhibits significant increasing trends for both winter (1.5 mm decade 1 ) and spring (4.1 mm decade 1 ) (Table IV). Decreasing trends are prevalent in northern part of the study area for autumn precipitation, and central eastern of the TP including the Yellow, Yangtze, Nujiang, and upstream of the Yarlung zangbo river basin for summer precipitation. Except for a small area in the central TP and upstream of the Yarlung zangbo river basin, most areas still show an increasing tendency. The regional mean precipitation generally show increases in summer and autumn (Table IV), however the trends are very small and not significant ( mm decade 1 ) during the period Roughly consistent with the spatial pattern in the Corrected-CMA, the APHRODITE data (Figure 9(b)) show a widespread increase in the eastern part of the TP for winter and spring seasons and decreasing tendency for summer and autumn seasons. In the western part of the study area, the APHRODITE show large areas with decreasing trends through the year. When averaged over the entire TP, however, the trends are not significant for any season except spring for both the entire TP and the region east of E80 (1.3/2.6 mm decade 1 ) Relationship between precipitation and discharge In the above analysis, we choose the Corrected-CMA data as the reference so as to evaluate the performance of the other precipitation products over the TP. All the precipitation datasets are subject to errors. Precipitation is generally the major driver of river streamflow. Zhang

14 K. TONG et al. Table IV. Annual and seasonal precipitation trends (mm decade 1 ) for the six source river basins and the entire TP during from the APHRODITE and Corrected-CMA data. The bold indicates the trends are statistically significant (α<0.05). Lancang Nujiang Yarlung zangbo Yellow Yalong Yangtze TP a APHRODITE data Annual /4.9 Spring /2.6 Summer /1.4 Autumn /0.3 Winter /0.1 Corrected-CMA data Annual Spring Summer Autumn Winter a The first number indicates the trends for the entire TP and the second one for the region east of E80 within the TP. et al. (2012) provide a broad picture for the hydrologic regimes of the major upstream river basins in the TP and suggest that the runoff of the southeastern basins in the TP is predetermined by the monsoon precipitation. Discharge data provide an opportunity for completely independent validation over the basins governed by monsoon climates, i.e. otherwise unavailable from a simple comparison among precipitation datasets (Pavelsky and Smith, 2006). Here, we further evaluate the precipitation datasets by investigating the correspondence between annual precipitation and annual runoff over the selected basins in the southeastern TP. Figure 10 shows normalized basin-averaged annual precipitation time series from the ERA-40, ERA-Interim, UW data, APHRODITE, Corrected-CMA data and observed discharge at each of the hydrological stations (Figure 1(b) and Table I) for ( for the ERA-Interim and for the UW data). The correlation coefficients R 2 between annual precipitation and discharge are presented in Table V. Due to the short time coverage, the TMPA data (1998 ) were not used for this analysis. The Yalong river was not included in Figure 10 because of the absence of observed discharge data. Among the three gauge-based estimates, the APHRODITE has the highest R 2 for the Nujiang and Yarlung zangbo (0.680 and 0.719, respectively). On the other hand, the APHRODITE has the lowest R 2 for the Yellow and Yangtze basins (0.492 and 0.674, respectively). While the R 2 of APHRODITE is in the middle for the Lancang (0.650) among the three gauge-based estimates. The stronger relationship between the bias-corrected precipitation and annual runoff over the Yellow river may indicate that bias corrections generate more reliable precipitation data (Ye et al., 2012), however this is not always the case for the selected basins in this study. It can be seen that the two corrected products do not always show stronger relationship with the annual discharge than the uncorrected APHRODITE among the basins (Table V). The ERA-40 annual precipitation show poor correspondence with the annual discharge with R 2 of among the basins during (Table V). However, visual inspections on Figure 10 suggest that the correspondence between the ERA-40 precipitation and discharge data seem to be improved after the later 1970s or early 1980s. We therefore re-calculated the R 2 between ERA-40 precipitation and annual discharge for The R 2 did improve with the R 2 values of over all the basins except for the Yellow river (R 2 of 0.001). These results motivate us to re-check the R 2 between the ERA-40 and the Corrected-CMA precipitation and we found that the R 2 greatly improved ( ) when only considering the period (the second column for the ERA-40 in Table III). The improved annual variation in the ERA-40 precipitation after 1979 is probably due to the assimilation of satellite data in ERA-40 begun in late 1978 (Hernandez et al., 2004). In spite of the high bias (up to 156%, Table II) in the ERA-Interim, good correspondence was observed between the ERA-Interim annual precipitation and annual discharge over the Nujiang and Yarlung zangbo during , with R 2 of and 0.788, respectively (Table V). Light improvement was seen in the Yellow river (R 2 of 0.098) in comparison with the ERA-40, while the performance was comparable between the two reanalyses for the Lancang and Yangtze (R 2 of and 0.210, respectively). In summary, the gauge-based precipitation generally show better correspondence with the discharge data than the reanalyses over the selected basins except for the Yarlung zangbo, where the ERA-Interim has the highest R 2. The ERA-40 show apparent improvements in annual variations after 1979 over most of the basins (except for the Yellow) and have comparable performance with the ERA-Interim for the Lancang and Yangtze river basins. The agreement among the precipitation estimates is worst in the Yarlung zangbo among the selected basins. In the Yarlung zangbo, the magnitude of UW precipitation is almost twice as the Corrected-CMA data, while the TMPA and APHRODITE are 22 25% lower than the Corrected-CMA (Figure 2, Table II). Hydrological

15 TIBET PRECIPITATION DEPICTED BY GAUGE, REANALYSES, AND SATELLITE DATA Figure 8. Spatial pattern of annual precipitation trends (mm decade 1 ) from the Corrected-CMA and APHRODITE over the TP during This figure is available in colour online at wileyonlinelibrary.com/journal/joc modelling of streamflow is a useful approach to evaluate precipitation products because the uncertainties in discharge data are much smaller than those in the precipitation data sets (Lammers et al., 2001). We drive a hydrology model Variable Infiltration Capacity (VIC) (Liang et al., 1994, 1996) with both UW and Corrected- CMA precipitation over the Yarlung zangbo river basin and compare the simulated streamflow with the flow observations at the Nuxia station (Table I, Figure 1(b)). The VIC model has been extensively calibrated and evaluated over six source river basins by Zhang et al. (2012). The results show that the VIC model forced by the gridded CMA data exhibits an acceptable performance over the Lancang, Nujiang, Yellow, and Yangtze river basins with the model efficiencies of and annual bias less than 10%; while the VIC simulations considerably underestimate the observed streamflow by 42% over the Yarlung zangbo basin for Zhang et al. (2012) also suggest that the VIC hydrology model performance cannot be significantly improved through model calibration and thus claim that the gridded CMA data for the Yarlung zangbo basin might be largely underestimated. Figure 11 shows the mean monthly precipitation and simulated and observed streamflow for for the Yarlung zangbo basin. The significant difference between the observed and simulated streamflow occurs in monsoon season (May to October). The simulated streamflow driven by the UW data is 100% higher than the observed in annual means during the simulation period, while the flow simulation driven by the Corrected-CMA precipitation is 46% lower than the observation (Figure 11). Given that the precipitation is the major driver of simulated

16 K. TONG et al. Figure 9. Spatial pattern of seasonal mean trends (mm decade 1 ) from the Corrected-CMA and APHRODITE over the TP during This figure is available in colour online at wileyonlinelibrary.com/journal/joc streamflow, precipitation in the UW data is obviously overestimated in the Yarlung zangbo basin, while it is still underestimated in the Corrected-CMA data. This study and the studies of Zhang et al. (2012) both suggest the large uncertainties existing in the current precipitation estimates for the Yarlung zangbo basin. The Corrected-CMA data s underestimation may be due to insufficient meteorological stations in the Yarlung zangbo river basin (Figure 1(b)) and without considering the influence of topography effects on precipitation in the interpolation process from points to grids. Gauge locations usually tend to lie at low elevations relative to the surrounding terrain. Simple interpolation from spare gauge stations may not capture the influence of orographic lifting on precipitation, especially in topographically complex regions like the Tibetan Plateau. Adam et al. (2006) suggest that the correction for orographic effects resulted in a net precipitation increase of 20.2% in orographically influenced regions. The real areal precipitation estimates for the Yarlung zangbo basin is thus left unknown; an objective programme of measurement and analysis would have to be undertaken. 4. Discussion The ERA-40 generally can capture the seasonal variation and the broad spatial distributions in the gauge-based precipitation estimates over the TP, while apparent local uncertainties exist. ERA-40 shows little agreement with the gauge-based precipitation in annual variations over most of the selected source river basins for the years before The correspondence between the ERA- 40 and the Corrected-CMA precipitation and observed discharge greatly improved for the years after Therefore, for depicting the large-scale climatology with the ERA-40 precipitation, the data after 1979 is recommended. Voisio et al. (2008) evaluated the ERA-40 ( ), a satellite-based dataset, and a gridded station-based dataset (which actually was the UW data

17 TIBET PRECIPITATION DEPICTED BY GAUGE, REANALYSES, AND SATELLITE DATA Figure 10. Normalized annual time series of precipitation estimates from the ERA-40, ERA-Interim, UW data, APHRODITE, and Corrected- CMA data and observed discharge for the Lancang, Nujiang, Yarlung zangbo, Yellow, and Yangtze river basins during ( for the ERA-Interim and for the UW data; no discharge data available for the Yalong river). This figure is available in colour online at wileyonlinelibrary.com/journal/joc with orographic corrections, Adam et al., 2006) for the purpose of global hydrological prediction. The ERA-40 precipitation was preferred for use in global hydrological predications by Voisio et al. (2008), although apparent biases over some portions of the global land areas (e.g. in the Himalaya) were recognized. Ma et al. (2009) found that ERA-40 precipitation has big bias related to adjusted observational precipitation around the Qinghai- Tibetan Plateau (maximum percentages of precipitation differences can exceed 1000%), indicating a strong terrain dependence of gridded precipitation data. For the purpose of regional hydrological modelling, especially in the Tibetan Plateau area, the ERA-40 precipitation might not be a good option. ERA-Interim is the latest global atmospheric reanalysis from the ECMWF and incorporates many important model improvements. ERA-Interim shows a significant improvement compared to ERA-40 in the representation of global hydrological cycle (Uppala et al., 2008) and river basin hydrometeorology (Betts et al., 2009). However, the anticipated improvements in ERA-Interim relative to ERA-40 have not been realized in this study. Our results suggest that ERA-Interim obviously overestimates the precipitation over the TP, though it has a higher correlation coefficient than ERA-40 with the Corrected-CMA data at both annual and monthly scales. Study conducted by Wang and Zeng (2012) also found that ERA-Interim precipitation has big bias relative to the station observations in the TP. On the basis of the results of this study, ERA-Interim precipitation is not be a suitable alternative for hydrological model inputs over the TP. Overall, our results suggest that the use of precipitation time series derived from current reanalysis projects is less preferable for hydrology analysis than the TP observational data at the fine spatial scales (e.g. the selected basins scale). Precipitation is one of the least well-simulated parameters in numerical models. This is due to the fact that the spatial scale of the physical mechanisms which generate rainfall is much finer than can be represented by the models. Additionally, surface feedback processes may be important which are also often poorly simulated within models (Diro et al., 2009). The UW data has been widely used in global hydrological simulations (Su et al., 2005; Haddeland et al., 2006; Adam et al., 2009; Biemans et al., 2009) and statistics downscaling of GCMs outputs