LST-CDPM: A lightweight spatio-temporal conditional denoising diffusion probabilistic model for invasive alien species distribution prediction

The impact of invasive alien species on ecosystems and economies is increasingly severe, making accurate predictions of their distribution crucial for effective management strategies. However, traditional species distribution models struggle to capture complex spatio-temporal dependencies, deep learning methods remain limited in their ability to model environmental conditions, and the denoising networks used in existing diffusion models are not well suited for handling multi-source environmental condition information. To address these issues, this study introduces a Lightweight Spatio-Temporal Conditional Denoising Diffusion Probabilistic Model (LST-CDPM) designed for predicting the distribution of invasive alien species. The model first incorporates a Spatio-Temporal Adaptive Feature Module to capture the spatio-temporal dependencies of species distributions. Second, by introducing a diffusion model framework that continuously injects multi-source environmental conditions at each denoising step, it effectively alleviates the problem of condition-information decay commonly seen in traditional approaches. Finally, an attention mechanism is integrated into the denoising network to enhance selective focus on key features. Together, these designs enable the model to more effectively integrate historical observations, ecological priors, and multi-source environmental conditions, thereby substantially improving prediction accuracy and generalization performance. To support model training and evaluation, two high-quality multi-temporal datasets are constructed. The experimental results demonstrate that the proposed method significantly outperforms existing baseline models across multiple evaluation metrics. This approach not only significantly enhances the accuracy and stability of invasive alien species distribution predictions but also serves as an ecological management tool, providing managers and policymakers with predictive outcomes and decision support. It thereby aids in monitoring, risk assessment, and the formulation of management strategies for invasive alien species, fully demonstrating its clear ecological value and practical applicability. • Proposing a lightweight diffusion model for species distribution prediction. • The STAFM integrates historical dynamics with prior environmental information. • Incorporating CBAM to optimize the denoising network structure. • Constructing two multi-temporal datasets for invasive alien species distribution.