Simulation and explanatory analysis of dissolved oxygen dynamics in Lake Ulansuhai, China

Abstract

Study region: The largest lake in China's Yellow River Basin, Ulansuhai. Study focus: After implementing the optimal noise reduction strategies based on wavelet transform for the high-frequency monitoring data, hybrid models coupling random forests, support vector machines, and artificial neural networks were employed to simulate the dissolved oxygen in the lake during both the open-water and ice-covered periods. The optimal hybrid model for each period was selected through extensive analysis, and the shapley additive explanations method was utilized to quantify the contribution of feature variables to the results. New hydrological insight for the region: The results revealed MAE, RMSE, and NSE values of 0.76, 1.01, and 0.92 for WT-RF, and 0.42, 0.55, and 0.84 for WT-SVM. They represent the optimal hybrid models for the open-water and ice-covered periods, respectively. The SHAP analysis showed that water temperature, blue-green algae, pH, turbidity, electrical conductivity, and chlorophyll.a exhibited significant importance in the WT-RF model, in that order. BGA and Chl.a synergistically promoted the increase in DO levels during the open-water period. For the WT-SVM model, Chl.a, BGA, Tur, and Temp exhibited significant contributions to the simulation results in that order. However, only BGA was able to enhance DO levels during the ice-covered period. As seasons change, shifts in key water environment factors affecting DO reveal the complexity of lake ecosystems in a cold and arid region.

Keywords

• Dissolved oxygen
• Seasonal changes
• Wavelet transform
• Machine learning
• Hybrid model
• Ulansuhai Lake