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

Area of Science:

• Geospatial analysis
• Disaster risk reduction
• Population dynamics

Background:

• Flooding is a major global hazard, yet population dynamics in repeatedly flooded zones, especially maladaptation growth, are poorly understood.
• Understanding settlement patterns in hazard-prone areas is crucial for effective disaster risk management and mitigation strategies.

Purpose of the Study:

• To investigate population dynamics and identify predictors of maladaptation hotspots in repeatedly flooded areas of Thailand.
• To challenge the assumption that repeated flooding deters settlement and to provide actionable insights for disaster management.

Main Methods:

• Integrated annual population estimates (LandScan, 2018-2024) with multi-year flood records and topographic variables for 1159 spatial units.
• Employed trajectory classification, Mann-Whitney U tests, and Random Forest with SHAP analysis to identify population growth patterns and key predictors of maladaptation.

Main Results:

• Repeatedly flooded areas showed lower median population growth than non-repeatedly flooded areas, but 27.37% were identified as maladaptation hotspots.
• Flood frequency was the dominant predictor (0.690 importance), followed by Digital Elevation Model (DEM) and distance to streams.
• Non-linear thresholds were identified: hotspot probability increased sharply with ≥3 flood events and DEM < 145 m, indicating lagged population responses.

Conclusions:

• Repeated flooding does not uniformly deter settlement, with significant maladaptation occurring in specific high-risk zones.
• Actionable thresholds for flood frequency and elevation can inform localized early warning systems, zoning, and relocation assistance.
• The geospatial and machine learning methodology offers a replicable approach for disaster research and smart analytics.