The study of terrestrial phosphorus inflow (hereafter referred to as phosphorus inflow) fluxes is essential for controlling non-point source (NPS) pollution. The SWAT model was successfully used to simulate phosphorus inflow fluxes in the Dongting Lake area, while a hybrid model (LSTM and SWAT) was developed and validated for predicting the reduction in phosphorus inflow fluxes among rivers based on three typical reduction scenarios: agricultural control, livestock and poultry reduction, and soil and water conservation measures. The results showed that the inflow flux of TP was 3.9×104 t·a-1, with a TP load of 14.4 kg·(hm2·a)-1. The flux exhibited notable seasonal variation, peaking during the flood season and decreasing during the non-flood season, which was strongly correlated with runoff, soil erosion, and cropping density. The Sankou area, with an annual average phosphorus inflow flux of 1.87×104 t, provided the highest risk and required specific regulation for its phosphorus levels, with the river system area and agricultural land emerging as the principal factors contributing to this risk. The LSTM model, using multi-feature input, effectively simulated phosphorus input fluxes at the sub-basin scale, achieving NSE > 0.6. Additionally, the hybrid model demonstrated better performance at the river system scale, with NSE > 0.8, RPE < 10%, and reduced data feature requirements. The study identified soil inflow reduction as a relatively effective approach for reducing phosphorus flux in the Sankou area, highlighting soil as the primary carrier of phosphorus and emphasizing its significance in non-point source pollution control, in addition to the fact that attention should be paid to the problem of livestock and poultry pollution. These findings offer valuable scientific insights and data support for understanding the spatial and temporal patterns of phosphorus input fluxes and devising phosphorus reduction strategies in the Dongting Lake area.
Read full abstract