5. Discussion

Precipitation is the key variable input for hydrological modeling and the main source of error in simulation results (Duan et al ., 2019a; Ruan et al ., 2017). Currently, many satellite and reanalysis precipitation products have been widely used in hydrological simulation in areas lacking observation (Bhatta et al ., 2019; Bitew and Gebremichael, 2010; Tang et al ., 2019). However, the quality of precipitation products is obviously different in different zones, owing to different climatic regions, seasonal cycles, and land types (Wang et al ., 2020). Before further use of satellite and reanalysis of precipitation products, it is necessary to evaluate the quality of these precipitation products. The error in satellite precipitation products usually stems from the weak relationship between precipitation rate and remote sensing signals (Bitew and Gebremichael, 2010), satellite revisit time (B et al ., 2013), and retrieval algorithm (Yan et al ., 2020). For reanalysis of precipitation products, the uncertainties and errors mainly come from data sources, interpolation algorithms, and data assimilation systems (Zhu et al ., 2015).
According to the results, the accuracy of TRMM, IMERG, CMADS, and CFSR in the warm season is higher than that in the cold season, and IMERG has the best performance, followed by CMADS, CFSR, and TRMM. This is mainly because snowfall is the main precipitation in winter in the YRSR (Huet al ., 2011). Although most satellite retrieval algorithms perform quite well in rainfall estimation, the accuracy of snowfall estimation is still not high, especially on snow-covered or frozen land surfaces (Cai et al ., 2015; Noh et al ., 2009; Villariniet al ., 2009). Alijanian et al . (2017) and Condom et al . (2011) reported that TRMM precipitation products cannot express the spatio-temporal variability of precipitation over high-altitude, complex terrain. IMERG, as a newer generation of TRMM precipitation products, are more sensitive to the capture of solid precipitation events (Yanget al ., 2020), due to the more advanced GPM microwave Imager sensor and the Dual-frequency Precipitation Radar onboard the GPM satellites. Most scientific literatures reported that IMERG has a better performance in alpine basin (Mou and Santo, 2018; Yuan et al ., 2015). In addition, high-latitude regions respond strongly to climate change, which poses a huge challenge to satellite precipitation observations (Mark et al ., 2016). Tian and Peters-Lidard (2010) reported that the satellite precipitation products have large uncertainty in high latitudes (beyond ± 40°). By contrast, the reanalysis precipitation products is less affected by high-latitude and high-altitude (Beck et al ., 2017; Serreze et al ., 2005; Yong et al ., 2014). At the grid scale, among the four precipitation products, CMADS has the best performance for precipitation observation (Fig. 6 ), and CFSR has the best performance for precipitation events (Fig. 7 ).
Due to the scarcity of in-situ gauged observation stations in the alpine basin, it is not comprehensive to evaluate the performance of precipitation products based on statistical methods. Hydrological simulation verification is a supplementary method for the evaluation of precipitation products (Deng et al ., 2019; Guoqiang et al ., 2015). Using GO from even sparse in-situ gauged observation stations resulted in better performance in runoff simulation than using all four precipitation products, which is consistent with the previous research results (Yuan et al ., 2015). Among the four precipitation products, IMERG has the best performance in runoff simulation, followed by CMADS, CFSR, and TRMM. TRMM seriously overestimated runoff simulation with NSE values of -1.86 and -11.93 at TNH and JM, respectively. This is mainly due to the poor quality of precipitation products in near real time (Tekeli and Fouli, 2016). In general, the simulation results of runoff at TNH are better than those at JM. There are two main reasons: one is the high altitude in the basin above JM and the large snow-covered, which will increase the microwave reflectivity on the land surface, thereby mask the drop in microwave signal due to scattering (Harpold et al ., 2017; Yong et al ., 2014); Second, in-situ gauged observation stations in the YRSR are mostly distributed downstream, and there are only two precipitation stations in the basin above JM (Fig. 1 ). Compared with the model driven by a single precipitation dataset, the model driven by the combination of GO and satellite or reanalysis precipitation products has better performance, especially Scenario S10 performed the best in all scenarios (R 2 = 0.77, NSE = 0.72 at TNH;R 2 = 0.53, NSE = 0.48 at JM). This is probably because CMADS precipitation products consider more gauge-based precipitation than CFSR and IMERG (Meng et al ., 2016). These results can provide reference data, and research ideas, for improved hydrological modeling of alpine basins.