Comparison of NDVI data derived from multiple sensors for treespecies discrimination in pure forest stands: A case study in Türkiye

This study examines the NDVI-based spectral separability of tree species in Turkey's pure forest stands by comparing three satellite sensors: Sentinel-2, Landsat 8/9, and MODIS. NDVI datasets were pre-processed using cloud–shadow masking and median composite techniques, followed by Welch ANOVA, Games–Howell post-hoc, Receiver Operating Characteristic (ROC), and Normalized NDVI Conversion Factor (NNCF) analyses. NDVI variations across species, stand age and canopy closure classes were evaluated, indicating that NDVI reflects not only photosynthetic activity but also ecological traits such as LAI, crown closure, and species-level biochemical properties. Results show that NDVI variance is mainly driven by sensor differences (57.6%), followed by species (9.5%) and canopy closure (4.5%). ROC analysis revealed that MODIS offers the highest discriminative power (AUC ≈ 0.58), while Beech (0.732), Hornbeam (0.719), and Sessile oak (0.718) achieved the strongest species-level separability. The calculated NNCF coefficients (MODIS/Landsat = 2.12 ± 0.74; Sentinel-2/Landsat = 1.81 ± 1.14) demonstrate systematic, sensor-dependent NDVI biases. Overall, the study shows that NDVI is sensitive to structural forest attributes such as species, age and canopy closure, and that the combined ROC–NNCF approach provides an effective framework for harmonizing NDVI data across sensors, offering valuable contributions to national-scale species classification and sensor standardization efforts.