Comparative evaluation of vegetation indices for water and heat stress detection and monitoring across land cover types

Studying variations in vegetation indices (VIs) under plant stress is still limited by the need for validation across scales, from leaf to field, to reliably distinguish different stressors in agricultural systems. This study aimed to analyze the water and heat stress on sunflower, grass, and forest trees using vegetation indices. We assessed the spatial variations in soil water content, soil temperature, and vegetation indices (NDVI and leaf chlorophyll content) across land use types of cropland (sunflower), grassland, and forest. We used a combination of field measurements and remote sensing data. The research sites are situated close to each other on medium to steep slopes. Soil water content (SWC), soil temperature, VIs of NDVI_F, leaf chlorophyll content (CCI_F), and leaf area index (LAI) field measurements were collected at 4 to 5 points along a transect during the vegetation period of 2022. During the study period an extreme drought occurred. Field measurements were implemented using handheld devices. The remote sensing data were obtained from the Sentinel-2 (S2) satellite for NDVI_S2 or green chlorophyll index (GCI_S2). Statistical assessment of the relationships was performed using Pearson correlation, PCA, complemented by RMSE and MAE metrics to quantify data precision. Among the land use types, we found distinct differences in SWC with the higest in grassland (15.39%) and the lowest in forest (9.71%; p < 0.05); in soil physical and chemical properties (e.g., significantly higher SOC and total N content for the forest and grassland site compared to cropland, p < 0.05); and in plant indices . NDVI_F and CCI_F showed significant differences among all land use types with the highest in cropland (NDVI_F = 0.66) or forest (CCI_F = 18.8; p < 0.05). S2 data detected significantly higher VIs for the forest (NDVI_S2 = 0.56 and GCI_S2 = 2.16; p < 0.05). However, no significant differences were observed for the VIs within the same land use. NDVI_F and NDVI_S2 only correlated well for grassland (r = 0.83; p < 0.001). Our data indicated that LAI is the strongest predictor among the VIs, exhibiting good correlations with CCI_F, NDVI_F, NDVI_S2, and GCI_S2 (r = 0.71-0.79; p < 0.001). The strongest relationships were observed in cropland LAI, with correlation coefficients of r = 0.80 for NDVI_S2 and r = 0.96 for GCI_S2.