Spatiotemporal variations and overflow risk analysis of the Salt Lake in the Hoh Xil Region using machine learning methods

Abstract

Global warming is inducing dramatic changes in fluvial geomorphology and reshaping the hydrological connections between rivers and lakes. The water level and area of the Salt Lake have increased rapidly since the outburst of the Zonag Lake in the Hoh Xil region of the Qinghai–Tibet Plateau in 2011, threatening the downstream infrastructure. However, fewer studies have focused on its spatiotemporal variation and overflow risk over long time series. Here, we used three machine learning algorithms: Classification and Regression Trees (CART), Random Forest (RF), and Support Vector Machine (SVM) to extract the area of the Salt Lake for a long time series, analyzed its spatiotemporal variation from 1973 to 2021, and finally assessed the overflow risk. The Kappa coefficient (KAPPA) and the overall accuracy (OA) were used to evaluate the performance of the models. The results showed that Random Forest performs superior in lake extraction (KAPPA = 0.98, overall accuracy = 0.99), followed by Classification and Regression Trees and Support Vector Machine. normalized difference water index is the relatively important feature variable in both RF and CART. Before the outburst event, the area change of the Salt Lake was consistent with the variation in precipitation; after that, it showed a remarkable area increase (circa 350%) in all orientations, and the main direction was the southeast. Without the construction of the emergency drainage channel, the simulation result indicated that the earliest and latest times of the Salt Lake overflow event are predicted to occur in 2020 and 2031, respectively. The results of this paper not only demonstrate that RF is more suitable for water extraction and help understand the water system reorganization event.