Attention Convolutional Neural Networks and Long Short‐Term Memory Model: Unveiling Spatiotemporal Dynamics of Ecological Indicators in Yili Mining Area

Abstract

Mining activities disrupt the ecosystems, causing soil erosion and landscape degradation. In this study, fractional vegetation cover (FVC), remote sensing ecological index (RSEI), and land cover (LC) were selected as indicators. The spatiotemporal variation and spatial autocorrelation were revealed by analyzing FVC, RSEI, and LC in the Yili of China. The impacts of climate conditions, human activities, and their interactions were discussed by attention convolutional neural networks (CNN) and Long Short‐Term Memory (LSTM) models. The results showed that (1) The attention CNN‐LSTM model significantly outperformed other models, achieving an accuracy of 0.734 (FVC), 0.721 (RSEI), and 0.978 (LC). (2) The model predicted the FVC and RSEI in 2024 to be 0.580 and 0.563. (3) By integrating an attention mechanism, the proposed model dynamically prioritizes critical spatiotemporal features, significantly enhancing prediction accuracy in imbalanced datasets. The findings highlight the potential of advanced deep learning frameworks for analyzing large‐scale remote sensing data.