Deep learning forecast of rainfall-induced shallow landslides

Area of Science:

• Geosciences
• Environmental Science
• Data Science

Background:

• Landslides pose significant threats to human populations and ecosystems globally.
• Projected climate change is expected to exacerbate landslide risks, necessitating improved forecasting methods.
• Current empirical and physically-based models have limitations for large-scale, operational landslide prediction.

Purpose of the Study:

• To develop and evaluate a deep-learning strategy for anticipating rainfall-induced landslides.
• To assess the effectiveness of this new approach for operational forecasting over extensive regions.
• To identify the primary meteorological drivers controlling landslide occurrence and timing.

Main Methods:

• Utilized a deep-learning model to establish a predictive link between rainfall patterns and landslide events.
• Trained and validated the model using 20 years of historical rainfall and landslide data from Italy.
• Focused on shallow landslides triggered by precipitation.

Main Results:

• The deep-learning strategy demonstrated effective anticipation of rainfall-induced landslides across large areas.
• Results confirmed that precipitation is the dominant factor controlling landslide timing and location.
• The model's performance suggests feasibility for operational landslide forecasting.

Conclusions:

• Deep learning offers a promising approach for enhancing the operational forecasting of rainfall-induced landslides.
• Accurate rainfall measurements and meteorological forecasts are key to successful landslide prediction systems.
• This research opens new avenues for proactive landslide risk management strategies.