<![CDATA[This study presents a comparative evaluation of multispectral UAV-derived vegetation indices (VIs) in rice fields. Seven UAV-derived VIs — the Chlorophyll Index (CI), Chlorophyll Vegetation Index (CVI), Modified Soil-Adjusted Vegetation Index (MSAVI), Green Normalized Difference Vegetation Index (GNDVI), Normalized Difference Red Edge Index (NDRE), Normalized Difference Vegetation Index (NDVI), and Perpendicular Vegetation Index (PVI) were computed from UAV multispectral orthomosaics acquired over a lowland rice field in Purwokerto, Central Java. The data were collected during the late vegetative stage of rice growth. Pearson correlation analysis showed strong positive correlations between the structural indices NDVI, MSAVI, and PVI (r > 0.93) and the chlorophyll-related indices CI and NDRE (r = 0.93), whereas CVI was negatively correlated with NDRE (r = −0.66), indicating that biomass accumulation and chlorophyll status did not always change together. Forty-three field points, classified into rice condition categories, were used to train and test a Random Forest classifier. Both models reached the same overall test accuracy of 63.64% (Kappa = 0.42 and 0.41, respectively), and the cross-validated model reached a cross-validation Kappa of 0.48 at mtry = 7. Variable importance ranked PVI as by far the most influential predictor (importance = 100), followed at a distance by CI (18.4), NDRE (12.9), and CVI (9.5), while MSAVI, NDVI, and GNDVI contributed little. Soil-background-sensitive PVI carried the strongest discriminative signal, chlorophyll-related indices added complementary information, and GNDVI was largely redundant. These findings offer preliminary evidence that PVI, paired with CI or NDRE, is the most relevant index combination for UAV-based rice field assessment.]]>
