Snow and glacier are important components in the hydrological cycle of the Tibetan Plateau (TP). Air temperature, as the main driver in freezing and thawing processes, becomes vital for hydrological modelling and prediction in this region. Due to a sparse ground gauging network, spatial density of air temperature measurement is insufficient for hydro‐meteorological studies. Therefore, the aim of this study is to identify the best representative temperature data for hydrological applications from four widely used reanalysis products, including ERA‐Interim, ERA‐5, GLDAS‐2.1 and NCEP‐R2, with reference to in situ measurements and gridded snow depth from the year 2008–2017 over the entire TP. To reduce errors, Bayesian Joint Probability (BJP) approach based on K‐Nearest Neighbour (KNN) classification algorithm (KNN‐BJP) is proposed to post‐process gridded reanalyses. The results indicate that all the reanalysis datasets provide highly correlated but cold biased air temperature. The correlation ecoefficiency is greater than 0.85. The cold biases are near −3 ° C and mainly distributed in the southeastern TP. Bias in daily maximum temperature during Spring is greater than −8 ° C for most stations. ERA‐Interim is found to have the closest agreement with in situ measurements, closely followed by GLDAS‐2.1. KNN‐BJP is found to be effective within a distance smaller than 5°. After post‐processing, the prominent underestimation is efficiently corrected with Bias near 0. RMSE is markedly reduced to be smaller than 2.5 ° C. The post‐processed ERA‐5 and GLDAS‐2.1 are as accurate as ERA‐Interim, but able to provide more detailed information for extreme events due to their finer spatial resolution. Thus, ERA‐5 and GLDAS‐2.1 are more recommended to represent air temperature in the TP. Snow depth as complementary reference data is able to present spatial variance of air temperature. Our study can help alleviate the problem of sparse air temperature data over the TP.

This study applied a GR4J model in the Xiangjiang and Qujiang River basins for rainfall-runoff simulation. Four recurrent neural networks (RNNs)—the Elman recurrent neural network (ERNN), echo state network (ESN), nonlinear autoregressive exogenous inputs neural network (NARX), and long short-term memory (LSTM) network—were applied in predicting discharges. The performances of models were compared and assessed, and the best two RNNs were selected and integrated with the lumped hydrological model GR4J to forecast the discharges; meanwhile, uncertainties of the simulated discharges were estimated. The generalized likelihood uncertainty estimation method was applied to quantify the uncertainties. The results show that the LSTM and NARX better captured the time-series dynamics than the other RNNs. The hybrid models improved the prediction of high, median, and low flows, particularly in reducing the bias of underestimation of high flows in the Xiangjiang River basin. The hybrid models reduced the uncertainty intervals by more than 50% for median and low flows, and increased the cover ratios for observations. The integration of a hydrological model with a recurrent neural network considering long-term dependencies is recommended in discharge forecasting.