Integrated assessment of dissolved oxygen dynamics using optimized M-K trend detection and ridge regression in the...

The accurate prediction and comprehensive analysis of dissolved oxygen (DO) are essential for maintaining water quality and ecosystem health in watershed environments. In this study, machine learning techniques-specifically the Random Forest algorithm-were integrated with M-K trend analysis and Ridge regression modeling to enhance the analysis of DO dynamics in watersheds. Monitoring data collected in 2022 from 12 stations in the middle and lower reaches of the Yellow River Basin (YRB), comprising nine water quality indicators, were used to assess the spatial and temporal characteristics of water quality. The relative importance of eight indicators to DO was evaluated using Spearman correlation analysis and Random Forest importance metrics. The five most influential factors affecting DO were identified as pH, water temperature (WT), turbidity (TUR), total nitrogen (TN), and the permanganate index (CODMn). Based on these key variables, DO predictions were generated with high accuracy (R² = 0.90). Furthermore, these significant indicators were utilized to refine the M-K trend and Ridge regression models, enabling more rational identification of key drivers influencing DO. Organic matter loading and water temperature (WT) emerged as the primary factors influencing spatiotemporal variations in DO. Additional regression analysis indicated that CODMn and TN were the most significant contributors to DO variation, with TN primarily originating from domestic wastewater. This study offers new insights into DO pollution management in the YRB and provides guidance for developing targeted strategies to improve DO conditions in watershed ecosystems.