The Global Land Data Assimilation System (GLDAS) is an important data source for global water cycle research. Using ground-based measurements over continental China, the monthly scale forcing data (precipitation and air temperature) during 1979–2010 and model outputs (runoff, water storage, and evapotranspiration) during 2002–10 of GLDAS models [focusing on GLDAS, version 1 (GLDAS-1)/Noah and GLDAS, version 2 (GLDAS-2)/Noah] are evaluated. Results show that GLDAS-1 has serious discontinuity issues in its forcing data, with large precipitation errors in 1996 and large temperature errors during 2000–05. While the bias correction of the GLDAS-2 precipitation data greatly improves temporal continuity and reduces the biases, it makes GLDAS-2 precipitation less correlated with observed precipitation and makes it have larger mean absolute errors than GLDAS-1 precipitation for most months over the year. GLDAS-2 temperature data are superior to GLDAS-1 temperature data temporally and spatially. The results also show that the change rates of terrestrial water storage (TWS) data by GLDAS and the Gravity Recovery and Climate Experiment (GRACE) do not match well in most areas of China, and both GLDAS-1 and GLDAS-2 are not very capable of capturing the seasonal variation in monthly TWS change observed by GRACE. Runoff is underestimated in the exorheic basins over China, and runoff simulations of GLDAS-2 are much more accurate than those of GLDAS-1 for two of the three major river basins of China investigated in this study. Evapotranspiration is overestimated in the exorheic basins in China by both GLDAS-1 and GLDAS-2, whereas the overestimation of evapotranspiration by GLDAS-2 is less than that by GLDAS-1.

Extreme hydro-meteorological events have become the focus of more and more studies in the last decade. Due to the complexity of the spatial pattern of changes in precipitation processes, it is still hard to establish a clear view of how precipitation has changed and how it will change in the future. In the present study, changes in extreme precipitation and streamflow processes in the Dongjiang River Basin in southern China are investigated with several nonparametric methods, including one method (Mann-Kendall test) for detecting trend, and three methods (Kolmogorov–Smirnov test, Levene's test and quantile test) for detecting changes in probability distribution. It was shown that little change is observed in annual extreme precipitation in terms of various indices, but some significant changes are found in the precipitation processes on a monthly basis, which indicates that when detecting climate changes, besides annual indices, seasonal variations in extreme events should be considered as well. Despite of little change in annual extreme precipitation series, significant changes are detected in several annual extreme flood flow and low-flow series, mainly at the stations along the main channel of Dongjiang River, which are affected significantly by the operation of several major reservoirs. To assess the reliability of the results, the power of three non-parametric methods are assessed by Monte Carlo simulation. The simulation results show that, while all three methods work well for detecting changes in two groups of data with large sample size (e.g., over 200 points in each group) and large differences in distribution parameters (e.g., over 100% increase of scale parameter in Gamma distribution), none of them are powerful enough for small data sets (e.g., less than 100 points) and small distribution parameter difference (e.g., 50% increase of scale parameter in Gamma distribution). The result of the present study raises the concern of the robustness of statistical change-detection methods, shows the necessity of combined use of different methods including both exploratory and quantitative statistical methods, and emphasizes the need of physically sound explanation when applying statistical test methods for detecting changes.

Conventional streamflow models operate under the assumption of constant variance or season-dependent variances (e.g. ARMA (AutoRegressive Moving Average) models for deseasonalized streamflow series and PARMA (Periodic AutoRegressive Moving Average) models for seasonal streamflow series). However, with McLeod-Li test and Engle's Lagrange Multiplier test, clear evidences are found for the existence of autoregressive conditional heteroskedasticity (i.e. the ARCH (AutoRegressive Conditional Heteroskedasticity) effect), a nonlinear phenomenon of the variance behaviour, in the residual series from linear models fitted to daily and monthly streamflow processes of the upper Yellow River, China. It is shown that the major cause of the ARCH effect is the seasonal variation in variance of the residual series. However, while the seasonal variation in variance can fully explain the ARCH effect for monthly streamflow, it is only a partial explanation for daily flow. It is also shown that while the periodic autoregressive moving average model is adequate in modelling monthly flows, no model is adequate in modelling daily streamflow processes because none of the conventional time series models takes the seasonal variation in variance, as well as the ARCH effect in the residuals, into account. Therefore, an ARMA-GARCH (Generalized AutoRegressive Conditional Heteroskedasticity) error model is proposed to capture the ARCH effect present in daily streamflow series, as well as to preserve seasonal variation in variance in the residuals. The ARMA-GARCH error model combines an ARMA model for modelling the mean behaviour and a GARCH model for modelling the variance behaviour of the residuals from the ARMA model. Since the GARCH model is not followed widely in statistical hydrology, the work can be a useful addition in terms of statistical modelling of daily streamflow processes for the hydrological community.