Optimization and attribution analysis of annual runoff simulation models in the upper reaches of the Heihe River, northwestern China

Objective The primary objective of this study is to conduct an in-depth investigation into the impact of human activities and climate change on the annual runoff in the upper reaches of the Heihe River of northwestern China, with the aim of providing scientific support for regional water resource conservation and planning.

Method This study employed a comprehensive approach involving the Mann-Kendall non-parametric statistical test, Pettitt test, and sliding t-test methods to assess the trends in the annual runoff series in the upper reaches of the Heihe River from 1954 to 2020. The objective was to identify abrupt change points in the annual runoff series and delineate the reference period and analysis period. Building upon this foundation, we employed the BP neural network model, the grey time series model, and the multivariate linear regression model to simulate the annual runoff variations during the reference period. We then selected the model with the best simulation performance. Subsequently, utilizing the selected model and runoff attribution methods, we quantitatively analyzed the driving mechanisms of human activities and climate change factors on the annual runoff variations.

Result Trend analysis revealed that the annual runoff series experienced abrupt changes around 1982 and 2006. Consequently, the annual runoff series in the upper reaches of the Heihe River can be divided into three phases: 1954–1982 (reference period), 1982–2006 (analysis period 1), and 2006–2020 (analysis period 2). In the simulation of the annual runoff series during the reference period, the BP neural network model exhibited a clear advantage over the other two models in three parameters during the validation period: relative error (0.79%), Nash-Sutcliffe efficiency coefficient (0.84), and goodness of fit (0.84). Utilizing the neural network model for annual runoff attribution analysis, it was determined that human activities led to an average decrease of 7.56 × 10⁸ m³ in annual runoff between 1982 and 2020. However, during the period of 2006 to 2020, the adverse contribution of human activities in the upper reaches of the Heihe River to runoff decreased by approximately 18.00% compared with the period from 1982 to 2006. A detailed analysis of the impact of climate change on annual runoff revealed that between 2006 and 2020, precipitation and evapotranspiration contributed approximately 11.00% and 8.00% more, respectively, to annual runoff compared with the period from 1954 to 1982.

Conclusion The BP neural network model demonstrates a strong performance in simulating the annual runoff series of the upper reaches of the Heihe River, achieving a simulation accuracy of 94.23% with a maximum error of only 1.36%. The annual runoff series in the upper Heihe River Basin exhibited trend transitions in 1982 and 2006. Increased human activities after 1982 lead to a reduction in annual runoff, while the comprehensive river basin management measures implemented after 2006 result in a mitigation of the negative impacts of human activities on annual runoff. Regarding the influence of climate change during the period from 1982 to 2020, evapotranspiration and precipitation contribute 46.57% and 53.43%, respectively to runoff.