Monitoring total suspended solids concentration in Poyang Lake via machine learning and Landsat images

Abstract

Study region: Poyang Lake, Jiangxi Province, China. Study focus: Total suspended solids (TSS) represents an important water quality parameter in inland waters. This study compared the performance of empirical models, semi-analytical models, and four machine learning models. Then, based on field-measured data, the back propagation neural network (BPNN) model with best performance (R2 = 0.89, RMSE = 20.03 mg/L, MAE = 10.8 mg/L, MAPD = 24.89%) was established and applied to Landsat images for retrieval. The spatial-temporal distribution of TSS in Poyang Lake from 1988 to 2022 was obtained. The various effects of driving forces on temporal and spatial variations of TSS concentration in Poyang Lake were analyzed. New hydrological insights for the region: The BPNN model proposed in this study has good performance in the testing dataset and two independent datasets, indicating good generalization ability and spatial portability. The TSS concentration showed an increasing trend over the past 35 years. The seasonal characteristics of TSS are distinct, with higher concentrations in the dry season and obvious variation in summer months. Before 2001, the TSS concentration was highly correlated with precipitation (r = −0.74) and wind speed (r = 0.60). Thereafter, sand mining (r = 0.66) became the main driving force underlying interannual variations in TSS concentration.

Keywords

• Poyang Lake
• Total suspended solids concentration
• Landsat
• Machine learning
• Sand mining