Driving Factors and Trend Prediction for Annual Sediment Transport in the Upper and Middle Reaches of the Yellow River from 2001 to 2020

Abstract

The Yellow River has long been known for having low water and abundant sediment. The amount of sediment transported in the upper and middle reaches of the Yellow River (UMRYR) has changed significantly in recent years, resulting in an obvious imbalance in the spatiotemporal distribution of the water resources in the Yellow River Basin (YRB). The changes in the sediment transport in the Yellow River significantly affect ecological security and socioeconomic development in the YRB. In this study, the Google Earth Engine (GEE) platform was used to obtain the potential driving factors influencing the five main gauge stations in the UMRYR: vegetation, soil moisture, population, precipitation, land types, etc. The data on the annual sediment transport (AST) were from the River Sediment Bulletin of China (2001~2020). Linear regression and the Mann–Kendall test were used to study the temporal variation in the AST. The first-order difference was determined from the original data to remove the autocorrelation, and it met the requirement of sample independence. The factors without collinearity were used for the driving force analysis using linear regression (linear model) and random forest regression (nonlinear model). We used the selected driving factors to establish the linear regression, the random forest model for predicting the AST, and cross-validation for verifying the prediction accuracy. Furthermore, the prediction outcomes were compared with the simplest ARIMA time-series model (control model). Our findings showed that the changing trend and the mutation of the AST were different in the UMRYR during the past 20 years. However, after the first-order difference of the AST, the amount of interannual variation in the annual sediment transport (ΔAST) was almost unchanged in the UMRYR. The five driving factors were chosen to establish the prediction models of linear regression and random forest regression, respectively. Compared with the control model, ARIMA, the prediction accuracy of the random forest model was the highest.