Year: 2026
Type: article
Abstract: Rainfall-induced debris flows are among the most destructive geological hazards in mountainous regions, and accurate early warning is critical for reducing disaster risks. This study proposes an early warning framework for rainfall-induced debris flows by integrating a bidirectional long short-term memory (BiLSTM) network with interferometric synthetic aperture radar (InSAR) technology. First, multi-source data, including InSAR-derived surface deformation measurements, rainfall observations, and historical debris flow records, are collected and preprocessed. Rainfall anomalies are removed using the 3σ criterion, while spatial filtering and phase unwrapping are applied to reduce noise in InSAR deformation data. Temporal synchronization is subsequently performed to establish consistent relationships among heterogeneous datasets. Based on the processed data, key predictive features, including average deformation rate, maximum surface displacement, rainfall intensity, rainfall accumulation, and debris flow indicators, are extracted. To improve the robustness of the warning model, an arithmetic mean filtering layer is introduced before the BiLSTM input layer to suppress short-term fluctuations and enhance long-term deformation trends. Bayesian optimization is employed to automatically determine optimal model parameters, improving the model’s capability to capture temporal characteristics associated with debris flow initiation. Furthermore, debris flow warning levels are classified into four categories (low, moderate, high, and very high risk) based on predicted occurrence probabilities using percentile analysis and receiver operating characteristic (ROC) curve evaluation. The proposed framework is validated using a typical mountainous region in southwestern China. Results demonstrate that the proposed model achieves stable predictive performance, with an F1-score exceeding 0.85 across different periods and a recall rate of 0.92. The model also achieves a Kappa coefficient greater than 0.85, indicating strong agreement between predicted and observed debris flow events. Compared with existing approaches, the proposed method effectively integrates surface deformation information and rainfall characteristics, providing an accurate and reliable tool for rainfall-induced debris flow early warning and emergency risk management.
Source: Frontiers in Earth Science
Language: English
Subfield: Management, Monitoring, Policy and Law
Field: Environmental Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $3,801
OpenAlex ID: W7204196357