A Physics-Informed, Energy-Autonomous LoRaWAN Framework for the Early Warning of Landslides

Information from the abstract

The effectiveness of Landslide Early Warning Systems (LEWS) in resource-constrained tropical regions is often compromised by intermittent connectivity, stochastic environmental noise, and limited training data. To address these vulnerabilities, we aimed to design and field-validate a resilient, energy-autonomous IoT framework. Our methods involved engineering a low-cost (~$200 USD) custom sensor node capable of >14 days of energy autonomy, supported by a dual-mode LoRaWAN telemetry protocol. For data analysis, we developed a Random Forest classifier trained on 148,919 field data points, employing Recursive Exponential Moving Average feature engineering to minimize lag and an Effective Saturation (Se) normalization technique to evaluate heterogeneous soils. Key findings from field validation in Thailand demonstrated robust connectivity under severe Non-Line-of-Sight conditions (SNR down to -5.5 dB), with the baseline model achieving a Critical Event F1-Score of 0.97. As a major improvement and novelty, the integration of physics-informed constraints (Se) resolved performance variance across differing soil profiles, elevating the ultimate F1-score to 1.00. This physics-informed approach demonstrated unprecedented site-specific precision, improving critical recall by 14% and 5% in sandy clay and sandy clay loam environments, respectively, effectively eliminating false alarms while ensuring fail-safe detection.

Why this record is monitored

This record has an Impact Signal of 73/100 based on recency, source, collaboration, and bibliographic signals. It prioritizes monitoring and is not a judgment of research quality.

Related topics: Landslides and related hazards · Energy and Environmental Systems · IoT Networks and Protocols