Forest type mapping using PRISMA imagery in the Kheyrud forest

Mapping forest types is essential for sustainable management, however, field methods are time-consuming and costly. Therefore, Modern remote sensing, especially hyperspectral imagery like PRISMA with 243 spectral bands, is valuable for mapping Hyrcanian mixed forests. However, its accuracy and capabilities must be assessed. In this study, the capability of PRISMA data was assessed using SVM, SAM, and RF classification approaches to create forest type maps for the Gorazbon district of Kheyrud Forest. A ground truth map consisting of 131 field sample plots, each measuring 45 × 45 m, was generated using Kuchler's abundance method based on the 100 trees with the largest diameters. Five forest types were identified: pure beech, mixed-beech, beech-hornbeam, hornbeam-beech, and mixed hornbeam. Geometric correction, smoothing and noise filtering was applied, then, Illumination correction was performed. Anomalies were detected using the RX algorithm. The minimum noise fraction (MNF) transformation and pixel purity index (PPI) were applied to the remaining 103 spectral bands and then used in the classification process. Additionally, indices such as NDVI, ReNDVI, MTCI, NDIred, and RTVI were incorporated into the classification process. The results showed that the SVM and RF algorithms achieved overall accuracies of 53.09% and 43.19%, with kappa coefficients of 0.38 and 0.25, respectively. The best combination of input data was derived from the spectral bands obtained through the MNF transformation. Based on the findings, high noise in PRISMA imagery and the similar spectral behavior of forest types in this region hindered the species discrimination and classification performance.