Spatiotemporal dynamics and anthropologically dominated drivers of chlorophyll-a, TN and TP concentrations in the Pearl River Estuary based on retrieval algorithm and random forest regression

Abstract

Estimation of large-scale and high-precision water quality parameters is critical in explaining the spatiotemporal dynamics and the driving factors of water quality variability, especially in areas with environmental complexity (e.g., crisscrossing waterways, high flood risk in rainy season and seawater invasion). Thus, in this study, a retrieval algorithm was developed to predict chlorophyll-a (Chl-a), total nitrogen (TN) and total phosphorus (TP) concentrations in the Pearl River Estuary (PRE) based on a large amount of in situ measurements and Landsat 8 remote sensing images. Random Forest (RF) machine learning was conducted to identify the relationship between environmental indicators (pH, turbidity, conductivity, total dissolved solids and water temperature), Chl-a, TN and TP. The results showed that the NIR/R Binomial algorithm for Chl-a estimation presented appreciable reliability with R² of 0.7429, root mean square error (RMSE) of 1.2089 and mean absolute percent error (MAPE) of 15.33%. The water quality variation in the PRE showed a characteristic of overall improvement and regional deterioration with average concentrations of 7.28 μg/L, 1.15 mg/L and 0.12 mg/L for Chl-a, TN, and TP respectively. Turbidity and pH were identified as the most important indicators to explain Chl-a (52.86%, 39.91%), TN (52.38%, 40.57%) and TP (55.23%, 40.03%) variation. Agricultural pollution was the main pollution source due to the intensive application of fertilizer and increased field size. Besides, land use patterns (e.g., increasing farmland but decreasing forest) greatly influenced water quality from 2010 to 2020. Moreover, light limitation caused by high turbidity reduced the algae productivity and further lowered the Chl-a concentration. The driving factors for regional water quality variations were anthropologically dominated and supplemented by climate change. This study improved the monitoring accuracy of regional water environment and provided quantitative early warning of water pollution events for environmental practitioners, so as to achieve long-term monitoring, precise pollution management and efficient water resources management.

Keywords: Driving factors; Random forest; Retrieval algorithm; Spatiotemporal dynamics; Water quality estimation; Water quality variation.