Using advanced AI to elucidate the factors behind algae growth in the north-south diversion project

Accurate and robust identification of factors contributing to algal growth is essential for the sustainable use and scientific management of freshwater resources. As scientific research evolves from small datasets to large datasets, the shortcomings of traditional machine learning become more apparent, and deep learning, which excels at processing large amounts of data, is gaining more attention.

Deep learning has occasionally been used to predict chlorophyll-a (Chl-a) time series, but it has rarely been used to identify key factors for algal growth.

To address this gap, a cross-border team of researchers from China, Germany, and the Netherlands developed Bloomformer, a deep-learning-based Transformer model designed to identify algal growth drivers end-to-end. Developed -1.

“Deep learning models have less operational transparency than traditional machine learning, but have significant performance advantages,” said Jing Qian, the first author of the paper. “The development of Bloomformer-1 aims to create a win-win situation in terms of interpretability and performance, allowing us to transparently and accurately identify the drivers of algal growth.”

Qian, a doctoral student at the Karlsruhe Institute of Technology in Germany, conducted this research as a co-development doctoral student at the Institute of Hydrology in China.

The middle route of China’s national large-scale project, the North-South Water Distribution Project (MRP), was selected as the research site to demonstrate the superior performance of Bloomformer-1. It is compared to four widely used traditional machine learning models, Extra Tree Regression (ETR), Gradient Boosted Regression Tree (GBRT), Support Vector Regression (SVR), and Multiple Linear Regression (MLR), The highest R2 (0.80 to 0.94) was obtained. ) and the lowest RMSE (0.22–0.43 μg/L).

“Bloomformer-1 employs a multi-headed self-attention mechanism that collects and learns dynamic contextual information by comparing each token in the input sequence to other tokens, thus fully compiling all field sampling data. This is one of the reasons for its excellent performance,” said co-author Stefan Nora of the University of Potsdam.

The research results are Water biology and securityIt was revealed that total phosphorus (TP) was the most important factor affecting MRP, especially in the Henan section, while total nitrogen (TN) had the greatest impact on algal growth in the Hebei section. rice field.

“Phosphorus control and reduction is an important strategy to control algae growth and maintain stable MRP water quality, while nitrogen management in Hebei region is also noteworthy,” said Yonghong, Institute of Hydrobiology, Chinese Academy of Sciences. Mr Bi said. Corresponding author of the study. “Furthermore, the promotion and application of Bloomformer-1 in other waters will be an important future challenge.”