This study introduces the Socio-Economic Asymmetry Index (SEA) to identify agricultural areas where socioeconomic vulnerability outweighs physical flood exposure. The framework was tested at 63,010 sampling points in Thailand's Upper Chi River Basin using satellite and land-use data, six composite indices, Random Forest, and SHAP. About one quarter of the study area fell into high or very high asymmetry classes. The result is a spatial modelling tool for prioritising field assessment and support; it does not establish which factors causally produce flood losses.
Key findings
- About 25% of the study area was classified as high or very high asymmetry using the reported SEA threshold above 0.478.
- ACI had the strongest positive correlation with SEA (r=0.808), followed by NELI (r=0.788) and flood frequency (r=0.786), while CERI was negatively correlated (r=-0.707). These correlations are not causal evidence.
- With flood frequency as the target, an SEA cutoff above 0.608 produced AUC 0.988, sensitivity 0.972, and specificity 0.962. This was an in-region assessment, not independent validation in a new basin.
- K-means grouped points as High-Asymmetry 9.5%, Moderate-Asymmetry 37.4%, Low-Asymmetry 16.2%, and Resilient 36.8%, providing profiles for subsequent field investigation.
Why this matters globally
SEA offers a common way to identify places that hazard-only maps may overlook because households or farms have limited capacity to cope. If replicated across basins and linked to observed losses, the framework could inform resource allocation, disaster insurance, and climate adaptation in low- and middle-income settings. Strong in-region discrimination alone does not establish global validity.
Thai researcher contribution
Patiwat Littidej and Jiradech Majandang are affiliated with Mahasarakham University, and Benjamabhorn Pumhirunroj with Sakon Nakhon Rajabhat University, placing Thai researchers at the centre of the index development and Upper Chi application. Task-level roles are not inferred beyond the authorship and affiliation evidence displayed by the publisher.
Limitations to consider
Development and assessment occurred in one basin, so the 0.478 and 0.608 thresholds may not transfer across places or years. Composite indices depend on proxy quality, normalisation, and the ratio construction. Flood frequency is not the same as validation against household or crop losses, and SHAP importance and correlations do not establish causality. External and temporal validation with uncertainty analysis remains necessary.
Verify the original sources
SymmetryOriginal article in Symmetry↗DOI: 10.3390/sym18101675