Smart Geospatial Analytics for Maladaptation Hotspot Detection: Integrating Trajectory Classification, Random Forest, and SHAP

Information from the abstract

Flooding is among the most frequent and damaging natural hazards globally, yet the population dynamics of repeatedly flooded areas remain poorly understood, particularly the phenomenon of maladaptation population growth in hazard-prone zones despite repeated flood exposure. This study integrated annual population estimates (LandScan, 2018–2024), multi-year flood records (2018, 2021, 2022), and topographic variables (TWI, slope, DEM, distance to streams) across 1159 spatial units in a flood-prone region of Thailand. We employed trajectory classification to identify late_growth pixels (population increase >5% only after 2022 floods), Mann–Whitney U tests to compare growth rates, and Random Forest with SHAP analysis to identify predictors of maladaptation hotspots. Repeatedly flooded areas (≥2 flood events out of 3 observation years) exhibited significantly lower median growth than non-repeatedly flooded areas (−0.386 vs. −0.200; p = 0.0092). However, 52 out of 190 repeatedly flooded pixels (27.37%) were classified as late_growth and were designated as maladaptation hotspots. Random Forest identified flood_freq as the dominant predictor (importance = 0.690), followed by DEM (0.117) and distance to streams (0.077). SHAP analysis revealed non-linear thresholds: hotspot probability increases sharply when flood_freq ≥ 3 and DEM < 145 m. No significant difference in pulse_2022 was observed (p = 0.4545), indicating lagged population responses and identifying the reconstruction period as a critical intervention window. These findings challenge the assumption that repeated flooding uniformly deters settlement and provide actionable thresholds for localized early warning, zoning restrictions, and targeted relocation assistance. The methodological pipeline—trajectory classification combined with interpretable machine learning—offers a replicable approach for smart geospatial analytics in disaster research.

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: Flood Risk Assessment and Management · Disaster Management and Resilience · Landslides and related hazards