Evaluation of TRMM precipitation and its application to distributed hydrological model in Naqu River Basin of the Tibetan Plateau

The Tibetan Plateau (TP), with an average altitude of over 4,000 m above sea level, is referred as the roof of the world (Royden et al. 2008). With more than 36,000 glaciers having a total volume of over 4,000 km³ located in the TP (Yao et al. 2007), the TP is the headwater area of several major rivers (e.g. Nujiang-Salween, Yellow River, Yangtze River, and Lancang-Mekong River) in Asia. Nevertheless, research on hydrological processes of the region is rare, and limited by the sparse gauge network in the TP, especially around the source of Nujiang-Salween. On the one hand, the extremely high altitude and low population density result in a low density gauge network, which is a challenge for hydrological simulation due to the lack of accurate spatial distribution of meteorological elements (Wu et al. 2015). On the other hand, the observed streamflow of Nujiang-Salween is scarce in the region. Therefore, hydrological research is seriously limited by the inadequate hydro-meteorological data and cannot support water resource management in Nujiang-Salween.

Distributed hydrological models have been used to investigate the hydrological processes and support water resource management due to the ability to describe spatial heterogeneity, which consists of the underlying heterogeneity and meteorological heterogeneity (Li et al. 2012). The former can be represented by the spatial distribution of a digital elevation model (DEM), soil type, and land use. The latter includes the spatial distribution of precipitation, temperature, wind speed, and other meteorological elements. Nevertheless, the spatial distribution of meteorological elements is uncertain and inaccurate due to the influence of complex topography and the low density of the meteorological gauge network, especially the uneven spatial distribution of precipitation, which is the dominant meteorological element in driving distributed hydrological models (Gourley & Vieux 2006; Taesombat & Sriwongsitanon 2009) and usually obtained from rain gauges by spatial interpolation such as the Thiessen polygon method (Tabios & Salas 1985; Li et al. 2012).

A number of quasi-global satellite-based precipitation products have been developed and applied to improve the accuracy (ACC) of spatial precipitation distribution in hydrological modelling (Stisen & Sandholt 2010). Among them, the Tropical Rainfall Measuring Mission (TRMM) from the National Aeronautics and Space Administration (NASA) and the Japan Aerospace Exploratory Agency is an active instrument dedicated to the measurement of precipitation from a satellite platform conjointly with a radiometer (Kirstetter et al. 2013) that generates a series of widely used precipitation products with a near-global coverage and adequate spatial-temporal resolution from 1997 to present (Huffman et al. 2007, 2010). The most popular TRMM products are the post-real-time research products, which have been globally applied in driving hydrological models. The earlier post-real-time research product (TRMM 3B42V6) has already been used in hydrological simulation in Africa (Hughes et al. 2006; Li et al. 2013), Latin American (Collischonn et al. 2008; Su et al. 2008; Tobin & Bennett 2009), and South Asia (Finsen et al. 2014). With the upgrade of TRMM 3B42V6 to version 7 (TRMM 3B42V7) in May 2012, TRMM 3B42V7 has been compared with the TRMM 3B42V6 and gauge precipitation in hydrological simulation in South Asia (Xue et al. 2013), South American (Zulkafli et al. 2014), and Africa (Adjei et al. 2015). In contrast, researches on the application of TRMM products in driving hydrological models started late and advanced fast in China. TRMM 3B42V6 has been applied in driving hydrological modelling in the Xinjiang River, Poyang Lake Basin (Li et al. 2012), Laohahe Basin (Yong et al. 2012), the source of the Yellow River (Meng et al. 2014), and the Xiangjiang River Basin (Xu et al. 2015). Wang et al. (2015) compared TRMM 3B42V7 with other satellite-based precipitation products in hydrological modelling of the Lhasa and Gongbo Basins, two tributaries of the Yarlung Zangbo (Upper Brahmaputra) River. Zhang et al. (2016) investigated the reliability of satellite products, including TRMM 3B43V7, in the water balance of the Yangtze River Basin. The results of the aforementioned researches are generally consistent and suggest that TRMM products can be an efficient source of precipitation data for hydrological modelling, but need correction, and while unsuitable for daily streamflow simulation achieve better performance in monthly streamflow simulation (Xu et al. 2015). The performance of TRMM 3B43V7 in hydrological modelling produces a significant improvement on TRMM 3B43V6 (Xue et al. 2013; Zulkafli et al. 2014). However, the new TRMM 3B42V7 has rarely been applied to hydrological modelling in China. Besides, hydrological research on the source of the Nujiang-Salween River is scarce due to the low density of gauge station networks and the inadequate hydro-meteorological data. Therefore, the objectives of this paper are designed to: (1) evaluate TRMM 3B42V7 (TRMM precipitation hereinafter) in comparison with gauge precipitation in the Naqu River Basin; and (2) compare the application of TRMM precipitation and gauge precipitation in driving distributed hydrological models in the Naqu River Basin.

The meteorological data, including daily precipitation, temperature (average, maximum and minimum), sunshine duration, wind speed, and relative humidity of eight meteorological stations (Figure 1) from 1998 to 2013, were originally acquired from the China Meteorological Data Sharing Service System (http://cdc.nmic.cn/home.do). In order to extend the daily meteorological data of stations over the basin, particularly the precipitation that is the most spatially variable meteorological factor, it is vital to select suitable interpolation methods, which can be divided into two classes: geo-statistical methods and geometric methods. The geo-statistical methods (e.g. Kriging, co-Kriging) use the structure of spatial correlation from observed data to estimate the spatial distribution of precipitation (Tabios and Salas 1985; Das et al. 2008; Hofstra et al. 2008; Ashiq et al. 2010), and are widely and successfully used in spatial interpolation of monthly and annual precipitation (Lloyd 2005; Ninyerola et al. 2007). However, these methods are inadequate to be applied in spatial interpolation of daily precipitation, especially in sparsely gauged regions (Buytaert et al. 2006; Ly et al. 2011; Castro et al. 2015). The Inverse Distance Weighting (IDW) interpolation is the most popular geometric method and has been applied at different spatial-temporal scales due to its simplicity (Kurtzman et al. 2009; Chen et al. 2010; Hwang et al. 2012). IDW interpolation constructs the spatial distribution of precipitation based on the distance weighting relation between observed precipitation gauges and unknown points. Therefore, IDW interpolation is sensitive to the number of gauges used and the exponential function of distance; it is difficult to identify an optimal number of neighboring gauges to be used and the exponential function of distance in IDW interpolation (Babak and Deutsch 2009), not to mention in sparsely gauged regions. As a result, the simple and popular Thiessen polygon method was applied to extend the daily meteorological data of eight stations over the basin, which was divided into eight polygons controlled by eight stations. Then the polygons were converted into grids with a resolution of 1.0 km to match the resolution of the distributed hydrological model. Consequently, the daily meteorological data (gauge precipitation) of grids within the polygon are the same and equal to the daily meteorological data (gauge precipitation) of the controlling station.

TRMM 3B42V7 3-hourly temporal, 0.25 ° × 0.25 ° spatial scale product ranging between 50 °N-50 °S is applied in this study. It is a post-real-time and accessible online product from the Goddard Earth Sciences Data and Information Services Centre of NASA, and is dependent on two different types of sensors, namely microwave and infrared radiation (Huffman et al. 2007; Chen et al. 2013a, 2013b, 2013c). The sources of passive microwave satellite precipitation estimates include the TRMM Microwave Imager, Special Sensor Microwave Imager/Sounder, Advanced Microwave Scanning Radiometer-EOS, Advanced Microwave Scanning Sounding Unit-B, and the Microwave Humidity Sounder. In addition, the Global Precipitation Climatology Centre gauge analysis and Climate Assessment and Monitoring System are incorporated into the TRMM 3B42V7 (Chen et al. 2013a, 2013b, 2013c). Daily TRMM precipitation from 1998 to 2013 was accumulated from the original TRMM 3B42V7 in the Naqu River Basin. On the one hand, the current downscaling methods concentrate on the monthly and annual TRMM precipitation and are not suitable for downscaling of daily TRMM precipitation, on account of the temporal-spatial variation of daily precipitation (Jia et al. 2011; Duan & Bastiaanssen 2013; Mahmud et al. 2015). Therefore, in order to match the resolution of the distributed hydrological model, the daily TRMM precipitation with 0.25 ° × 0.25 ° resolution was resampled into grids at a 1.0 km resolution by nearest neighbor method, which will not change the values of original TRMM precipitation. On the other hand, in order to evaluate the TRMM precipitation in comparison with gauge precipitation at station and basin scales, the daily TRMM precipitation and gauge precipitation of eight stations were derived from the TRMM precipitation resampled grids and gauge precipitation grids from the Thiessen polygon method, which cover the corresponding station gauges with the same resolution of 1.0 km, and so the station scale represents the single grid scale with a resolution of 1.0 km in this study. The areal TRMM precipitation and gauge precipitation over the basin was averaged from the daily TRMM precipitation resampled grids and gauge precipitation grids in the Naqu River basin (basin scale), respectively. Then, monthly and annual TRMM precipitation and gauge precipitation at station and basin scales were accumulated by the daily TRMM precipitation and gauge precipitation, accordingly.

The Water and Energy Transfer Processes (WEP) model (a distributed hydrological model) was driven by the daily gauge precipitation and TRMM precipitation from 1998 to 2013, yet daily TRMM precipitation underperforms monthly TRMM precipitation in hydrological simulation (Li et al. 2012; Meng et al. 2014; Xu et al. 2015). Moreover, the daily streamflow of Jiayuqiao station (the outlet of the Naqu River basin) in flood season (generally from early June to the end of September, but not a fixed duration) during 1998–2008 was sourced from the Water Conservancy Information Center in the Ministry of Water Resources of the People's Republic of China. Meanwhile, the monthly streamflow of Jiayuqiao station from 1998 to 2008 was provided by the Water Conservancy Bureau of the Tibet Autonomous Region. Although the daily and the monthly streamflows were derived from different sources, they are consistent at monthly scale in flood season. As is seen, the daily observed streamflow is incomplete and only adequate in flood season (June to September) during 1998–2008, and the monthly observed streamflow is adequate from January 1998 to December 2008. Therefore, the monthly streamflow is used to calibrate and validate the empirical parameters in the distributed hydrological model, and the daily streamflow in flood season was also used to verify the performance of the WEP model at daily scale.

The WEP model is a distributed physically-based model that has been widely used in different river basins in Japan, Korea and China (Jia et al. 2009). The WEP model selection in this study is attributed to its detailed energy transfer processes, which is meaningful for hydrological simulation in high latitude and intense radiation basins (Tanaka et al. 2003). In the WEP model, evapotranspiration and latent heat flux are computed by combining the Penman-Monteith equation (Monteith & Unsworth 2014) with the force-restore method (Hu & Islam 1995) instead of the potential value method. Infiltration during heavy rains is simulated by a generalized Green-Ampt model (Jia & Tamai 1998), whereas soil moisture movement during the remaining periods is obtained by the Richards model. Two-dimension groundwater flow is simulated to consider the interactions between grid cells, meanwhile river flow routing and overland flow routing are conducted using the kinematic wave method. Moreover, for the energy processes short-wave radiation is based on observation or deduced from sunshine duration, long-wave radiation is calculated according to temperatures, latent and sensible fluxes computed by the aerodynamic method, and surface temperature is solved by the force-restore method. Much more information about WEP models can been seen in previous publications (Jia & Tamai 1998; Jia et al. 2001, 2009).

Statistical results of daily TRMM precipitation and gauge precipitation at station and basin scales during 1998–2013 are listed in Table 4. As can be seen, the mean daily TRMM precipitation is more than the mean daily gauge precipitation in most stations, except for stations No. 55293 and No. 56018. As a result, daily areal mean precipitation derived from TRMM precipitation is slightly more than that from gauge precipitation over the basin. This implies that TRMM precipitation overestimates precipitation in the Naqu River Basin, which is consistent with the findings from another study in the TP (Chen et al. 2013a, 2013b, 2013c). Meanwhile, maximum daily TRMM precipitation is noticeably more than maximum daily gauge precipitation, particularly at station scale. The possible reason is that as the gauge precipitation merely refers to a specific point it is more random and uncertain, and could easily miss the rainstorm center in comparison with TRMM precipitation, which is derived from the average value of the region with a range of 0.25 ° × 0.25 °. Moreover, the RMSE and BIAS corroborate that there is a clear difference between daily gauge precipitation and TRMM precipitation at station and basin scales.

As a distributed physically-based model, most of the physical parameters of the WEP model can be obtained from the physical inputs including the DEM, land use and soil database. But there are still several empirical parameters such as Manning roughness for the overland flow routing calculation, Manning roughness for the river flow routing calculation, the hydraulic conductivity between the surface river and groundwater, and hydraulic conductivity in the groundwater aquifer, which need to be calibrated by observed streamflow. Therefore, the empirical parameters (Table 6) of the WEP model were automatically optimized by the PEST with the monthly streamflow during 1998–2003, and the monthly streamflow in 2004–2008 was used to validate the model.

The water balance component is an important indicator to test the validity of precipitation data (Li et al. 2012), and the water balance components of the Naqu River Basin derived from the gauge precipitation-based model and corrected TRMM precipitation-based model are illustrated in Table 9. The results show that the average annual precipitation over all of the Naqu river basin obtained from corrected TRMM precipitation is slightly larger than that from gauge precipitation during 1998–2013. As a result, the areal runoff derived from the TRMM precipitation based model is larger than that derived from the gauge precipitation-based model, especially for surface runoff. However, the total areal precipitation of gauge precipitation is approximated to corrected TRMM precipitation, and the two models were calibrated by the same observed streamflow. Although the different empirical parameters also have an influence on the simulated streamflow, the difference in empirical parameters is originally derived from the difference in precipitation data. Consequently, it can be inferred that the difference in water balance components in the two models is essentially due to the difference between the spatial distribution of gauge precipitation and corrected TRMM precipitation at spatial scale.

This paper compares daily and monthly TRMM precipitation with the corresponding gauge precipitation from 1998 to 2013 at station and basin scales in the Naqu River Basin. The statistic results show that TRMM precipitation overestimates the precipitation with a BIAS of 0.2 over the basin, and the poor performance of contingency table indices displays a mismatch between the daily TRMM precipitation and gauge precipitation. However, the percentage distribution of daily TRMM precipitation in different classes and their contributions are generally consistent with that of daily gauge precipitation. Moreover, the scatter plots of monthly TRMM precipitation and gauge precipitation show a good linear relation, especially at basin scale.

Based on the good linear relation, the daily areal TRMM precipitation is corrected and applied to driving the WEP model in comparison with daily gauge precipitation. The models driven by the two precipitation data were calibrated and validated by monthly streamflow from 1998 to 2008. The daily hydrological simulations show that daily simulated streamflow driven by gauge precipitation and corrected TRMM precipitation are generally consistent with the observed streamflow in flood season, and daily gauge precipitation outperforms daily corrected TRMM precipitation. Furthermore, both gauge precipitation and corrected TRMM precipitation perform well in monthly hydrological simulation with NSCE larger than 0.85. The streamflow BIASs derived from gauge precipitation display regular fluctuations and are inconsistent in different months. Additionally, the water balance component of sub-basins from the corrected TRMM precipitation-based model is more reasonably distributed than that from the gauge precipitation-based model at spatial scale.

As mentioned above, the distributed hydrological model driven by corrected TRMM precipitation presents a better understanding of the hydrological cycle in the Naqu River Basin, which is meaningful in supporting water resource management and economic development there. Furthermore, it can be concluded that TRMM precipitation is a potential and effective input for the distributed hydrological model, attributed to its advantage of spatial distribution of precipitation in sparse gauge regions, where the spatial interpolation method is limited by the extremely low density gauge network. Therefore, TRMM precipitation provides an important approach to simulating the hydrological cycle and reducing the hydrological uncertainty caused by the spatial heterogeneity of precipitation in ungauged or poorly gauged basins such as the Naqu River Basin. Nevertheless, there is still a demand for the algorithms of TRMM precipitation estimation in ACC and spatial-temporal resolution (Li et al. 2009). Moreover, hydrologists should focus on the suitable method to correct TRMM precipitation according to regional precipitation characteristics, given the variable effects of climatic and underlying conditions on the spatial distribution of precipitation.

This work was supported by the Major Research plan of the National Natural Science Foundation of China (Grant No. 91547209), the General Program of the National Natural Science Foundation of China (Grant No. 41571037), and the Open Research Foundation from State Key Laboratory of Simulation and Regulation of Water Cycle in River Basin (Grant No. 2015ZY02). The authors give their sincere thanks to Professor Yangwen Jia at the IWHR for the support on the WEP model. Their sincere appreciation is also given to the anonymous reviewers for their detailed, strict and important comments on this paper.