A nonlinear hybrid model to assess the impacts of climate variability and human activities on runoff at different time scales

Understanding the contributions of potential drivers on runoff is essential for the sustainable management of water resources; however, the impacts of climate variability and human activities on runoff at inter-annual and inter-decadal scales have rarely been assessed quantitatively. To achieve this goal, this study develops a nonlinear hybrid model, which integrates extreme-point symmetric mode decomposition (ESMD), back propagation artificial neural networks (BPANN) and weights connection method. ESMD allows to separate the times series of drivers and runoff into different time scales. BPANN is then used to simulate the relation between the drivers and runoff at each time scale separately. Weights connection method is employed to quantify the impacts of climate variability and human activities on runoff. The performance of this proposed model is compared with multiple linear regression (MLR). The mountainous area of the Hotan River Basin is selected as case study area. Results reveal that runoff exhibits significant fluctuations at inter-annual (2 and 9 years) and inter-decadal (14 years) scales. Climate variables are responsible for 81% of the runoff variations, while human activities account for 8%. The nonlinear hybrid model substantially outperforms MLR in all performance measures. We attribute this improvement to the ability of the proposed model to represent nonlinear relations and to simulate the association between drivers and runoff at different time scales. For instance, water vapor affects runoff positively at the inter-annual time scale but negatively at the inter-decadal time scale. Such opposing relations cannot be represented by MLR or many other, more traditional methods.