The impacts of climate change on the water environment have aroused widespread concern. With global warming, mountainous basins are facing serious water supply situations. However, there are limited meteorological stations on mountains, which thus creates a challenge in terms of accurate simulation of streamflow and water resources. To solve this problem, this study developed a method to model streamflow in data-scarce mountainous basins. Selecting the two head waters originating in the Tienshan mountains, Aksu and Kaidu Rivers, we firstly reconstructed precipitation and temperature dynamics based on Earth system data products, and then integrated the radial basis function artificial neural network and complete ensemble empirical mode decomposition with adaptive noise to model streamflow. Comparison with the observed streamflow according to hydrological stations indicated that the proposed approach was highly accurate. The modeling results showed that the El-Niño Southern Oscillation, temperature, precipitation, and the North Atlantic Oscillation are the main factors driving streamflow, and the streamflow decreased in both the Aksu River and Kaidu River between 2000 and 2017.

Based on a temperature anomaly time series from 16 international exchange stations in Xinjiang from 1957 to 2012, the multi‐scale characteristics of temperature variability were analysed using the ensemble empirical mode decomposition (EEMD) method. Regional differences in variation trends and change‐points were also preliminarily discussed. The results indicated that in the past 50+ years, the overall temperature in Xinjiang has exhibited a significant nonlinear upward trend, and its changes have clearly exhibited an inter‐annual scale (quasi‐3 and quasi‐6‐year) and inter‐decadal scale (quasi‐10 and quasi‐30‐year). The variance contribution rates of each component demonstrated that the inter‐annual change had a strong influence on the overall temperature change in Xinjiang, and the reconstructed inter‐annual variation trend could describe the fluctuation state of the original temperature anomaly during the study period. The reconstructed inter‐decadal variability revealed that the climate mode in Xinjiang had a significant transformation before and after 1995, namely the temperature anomaly shift from a negative phase to a positive one. Furthermore, there were regional differences in the nonlinear changes and change‐points of temperature. At the same time, the results also suggested that the EEMD method can effectively reveal variations in long‐term temperature sequences at different time scales and can be used for the complex diagnosis of nonlinear and non‐stationary signal changes.

Selecting the Yarkand River as a typical representative of an inland river in northwest China, We identified the variation pattern of hydro-climatic process based on the hydrological and meteorological data during the period of 1957 ~ 2008, and constructed an integrated model to simulate the change of annual runoff (AR) with annual average temperature (AAT) and annual precipitation (AP) by combining wavelet analysis (WA) and artificial neural network (ANN) at different time scale. The results showed that the pattern of hydro-climatic process is scale-dependent in time. At 16-year and 32-year time scale, AR presents a monotonically increasing trend with the similar trend of AAT and AP. But at 2-year, 4-year, and 8-year time scale, AR exhibits a nonlinear variation with fluctuations of AAT and AP. The back propagation artificial neural network based on wavelet decomposition (BPANNBWD) well simulated the change of AR with AAT and AP at the all five time scales. Compared to the traditional statistics model, the simulation effect of BPANNBWD is better than that of multiple linear regression (MLR) at every time scale. The results also revealed the fact that the simulation effect at a larger time scale (e.g. 16-year or 32-year scale) is better than that at a smaller time scale (e.g. 2-year or 4-year scale).

Based on the observed data during the period from 1958 to 2012 in Xinjiang, China, we investigated the temporal and spatial complexity of precipitation distribution by using an integrative approach combining the phase space reconstruction (PSR), correlation dimension (CD), variogram, and cokriging interpolation. The CD values showed that the precipitation dynamic is a complex and chaotic system, and its complexity decreases along with the increase of temporal scale. To describe the precipitation dynamics, it needs at least four independent variables at daily scale, and at least three independent variables at monthly scale, whereas at least two independent variables are needed at seasonal and annual scales. The spatial variation of CD value at daily and monthly scales is described by the exponential variogram model, whereas that at seasonal and annual scale needs to be respectively described by the spherical and Gaussian variogram model. The higher CD values mainly distribute on complex landform such as mountain areas, whereas the lower CD values mainly distribute on the flat landform such as basin area, which indicate that the spatial complexity of precipitation distribution is derived from the complex landform.

Fluctuant and complicated hydrological processes can result in the uncertainty of runoff forecasting. Thus, it is necessary to apply the multi-method integrated modeling approaches to simulate runoff. Integrating the ensemble empirical mode decomposition (EEMD), the back-propagation artificial neural network (BPANN) and the nonlinear regression equation, we put forward a hybrid model to simulate the annual runoff (AR) of the Kaidu River in northwest China. We also validate the simulated effects by using the coefficient of determination (R2) and the Akaike information criterion (AIC) based on the observed data from 1960 to 2012 at the Dashankou hydrological station. The average absolute and relative errors show the high simulation accuracy of the hybrid model. R2 and AIC both illustrate that the hybrid model has a much better performance than the single BPANN. The hybrid model and integrated approach elicited by this study can be applied to simulate the annual runoff of similar rivers in northwest China.