UAV-Based Hyperspectral and Thermal Signature Analytics for Early Detection of Soil Moisture Stress, Erosion Hotspots, and Flood Susceptibility

Abstract

This study addresses the problem that soil moisture stress, erosion initiation, and localized flood exposure emerge at micro-scales that routine monitoring misses, delaying mitigation. The purpose was to test whether UAV-based hyperspectral and thermal signature analytics can deliver defensible early detection of soil moisture stress and improve erosion hotspot and flood susceptibility mapping using a quantitative, cross-sectional, case-based design. Across cloud and enterprise GIS deployment cases, the sample included 210 spatial zones within the case-study area, with five-point Likert composites for Soil Moisture Stress (SMS), Erosion Hotspot Concern (EHC), and Flood Susceptibility Perception (FSP). Key predictors were a hyperspectral moisture proxy (HSI_MoistureIndex), local thermal anomaly (TH_Anomaly), and terrain and hydrologic controls (slope, flow accumulation, distance to channel, and elevation). The analysis plan used descriptive statistics, Cronbach’s alpha, Pearson correlations, multivariate regression with sensor-fusion model comparison, risk-class summaries, hotspot agreement (Jaccard), and robustness checks. Results indicated moderate-to-high perceived risk (SMS M=3.62, SD=0.71; EHC M=3.48, SD=0.76; FSP M=3.55, SD=0.74) and strong construct reliability (alpha=0.86, 0.83, and 0.88). SMS increased with TH_Anomaly (r=0.58, p<.001) and decreased with HSI_MoistureIndex (r=-0.52, p<.001). Regression confirmed a fusion gain: the fused stress model achieved Adj. R²=0.58 and RMSE=0.44, outperforming hyperspectral-only (Adj. R²=0.42, RMSE=0.56) and thermal-only (Adj. R²=0.46, RMSE=0.53), with significant effects for TH_Anomaly (beta=0.41) and HSI_MoistureIndex (beta=-0.36). Erosion modeling achieved Adj. R²=0.54 with flow accumulation and slope as primary drivers, and flood susceptibility improved from Adj. R²=0.55 (terrain only) to 0.62 when thermal information was added. Spatial validation showed strong agreement for stress versus thermal hotspots (68% overlap; J=0.52) and flood susceptibility versus flow corridors (71% overlap; J=0.55), while sensitivity tests reduced the fused model from Adj. R²=0.58 to 0.49 without TH_Anomaly and to 0.50 without HSI_MoistureIndex. High-risk classes covered 22 to 28% of zones and consistently aligned with Likert means above 4.0. Implications are that sensor-fusion UAV analytics, operationalized through cloud and enterprise workflows, can support earlier irrigation triage, prioritized erosion control in convergent high-slope zones, and more defensible flood preparedness zoning with auditable quantitative evidence.