Environmental factors affecting Ulva prolifera blooms in the South Yellow Sea: Insights from deep learning models with attention mechanisms

Abstract

Here, we developed a deep learning model with integrated attention mechanisms. We used the model to examine environmental factors influencing Ulva prolifera green tides in the southern Yellow Sea. Analysis of bloom coverage in 2022 and 2023 revealed significant differences in spatial distribution and temporal patterns, with a substantial increase in 2023, when the maximum coverage expanded to 940.06 km² compared to 261.07 km² in 2022. Key environmental factors such as water temperature and salinity were identified as major predictors of bloom formation, showing strong positive correlations with bloom coverage. A novel attentionconvolutional neural networklong short-term memory network model was used to predict bloom occurrences. This model achieved an accuracy of 92.64 % with a root mean square error of 0.2830, demonstrating a significant improvement over the conventional model. Notably, the model captured non-linear interactions that were not apparent in traditional analyses, such as the impact of reduced wind speeds on bloom aggregation and growth in 2023. These findings demonstrate the effectiveness of integrating deep learning techniques with environmental data to enhance predictive accuracy. This approach provides critical insights into the complex environmental interactions driving green tides.