Simulation of Pollution Load at Basin Scale Based on LSTM-BP Spatiotemporal Combination Model

Abstract

Accurate simulation of pollution load at basin scale is very important for controlling pollution. Although data-driven models are increasingly popular in water environment studies, they are not extensively utilized in the simulation of pollution load at basin scale. In this paper, we developed a data-driven model based on Long-Short Term Memory (LSTM)-Back Propagation (BP) spatiotemporal combination. The model comprises several time simulators based on LSTM and a spatial combiner based on BP. The time series of the daily pollution load in the Zhouhe River basin during the period from 2006 to 2017 were simulated using the developed model, the BP model, the LSTM model and the Soil and Water Assessment Tool (SWAT) model, independently. Results showed that the spatial correlation (i.e., Pearson’s correlation coefficient is larger than 0.5) supports using a single model to simulate the pollution load at all sub-basins, rather than using independent models for each sub-basin. Comparison of the LSTM-BP spatiotemporal combination model with the BP, LSTM and SWAT models showed that the performance of the LSTM model is better than that of the BP model and the LSTM model can obtain comparable performance with the SWAT model in most cases, whereas the performance of the LSTM-BP spatiotemporal combination model is much better than that of the LSTM and SWAT models. Although the variation of the simulated pollution load with the LSTM-BP model is high under different hydrological periods and precipitation intensities, the LSTM-BP model can track the temporal variation trend of pollution load accurately (i.e., the RMSE is 6.27, NSE is 0.86 and BIAS is 19.46 for the NH3 load and the RMSE is 20.27, NSE is 0.71 and BIAS 36.87 is for the TN load). The results of this study demonstrate the applicability of data-driven models, especially the LSTM-BP model, in the simulation of pollution load at basin scale.

A Combined Model for Water Quality Prediction Based on VMD-TCN-ARIMA Optimized by WSWOA

“…Nevertheless, the performance of such a model exhibits large fluctuations and poor generalization to different water quality characteristics. For example, the LSTM-BP [17] and LSTM-CNN [18] combined prediction model can effectively extract changes in specific water quality characteristics, but it has poor generalization and cannot obtain the same effect for other water quality characteristics. The other category is a model based on signal decomposition and feature extraction.…”