Identification of driving factors of algal growth in the South-to-North Water Diversion Project by Transformer-based deep learning

Abstract

Accurate and credible identification of the drivers of algal growth is essential for sustainable utilization and scientific management of freshwater. In this study, we developed a deep learning-based Transformer model, named Bloomformer-1, for end-to-end identification of the drivers of algal growth without the needing extensive a priori knowledge or prior experiments. The Middle Route of the South-to-North Water Diversion Project (MRP) was used as the study site to demonstrate that Bloomformer-1 exhibited more robust performance (with the highest R², 0.80 to 0.94, and the lowest RMSE, 0.22–0.43 ​μg/L) compared to four widely used traditional machine learning models, namely extra trees regression (ETR), gradient boosting regression tree (GBRT), support vector regression (SVR), and multiple linear regression (MLR). In addition, Bloomformer-1 had higher interpretability (including higher transferability and understandability) than the four traditional machine learning models, which meant that it was trustworthy and the results could be directly applied to real scenarios. Finally, it was determined that total phosphorus (TP) was the most important driver for the MRP, especially in Henan section of the canal, although total nitrogen (TN) had the highest effect on algal growth in the Hebei section. Based on these results, phosphorus loading controlling in the whole MRP was proposed as an algal control strategy.