Evaluation of Sentinel-2 and PlanetScope Image Fusion for Tree Species Identification in Wanagama Tropical Forest, Indonesia

Remote sensing-based tree species classification requires a combination of high spatial resolution and rich spectral information. Sentinel-2 offers advantages in spectral diversity and spectral consistency, but is limited by its spatial resolution of 10-20 meters. In contrast, PlanetScope provides finer spatial resolution (3.3 meters) and high revisit frequency, yet is often criticized for spectral inconsistency across satellites and potential radiometric noise. This study aims to evaluate the fusion of both sensors to improve species classification accuracy in the Wanagama Educational Forest, Gunung Kidul, Yogyakarta, by leveraging the spectral strength of Sentinel-2 and the spatial resolution of PlanetScope. Image fusion was carried out using the Gram-Schmidt method with two main schemes: (1) spectral band matching from Sentinel-2 pansharpened with first band (GS-PCA 1) of PCA extraction from PlanetScope RGB bands, and (2) second band (GS-PCA 2) of PCA extraction from PlanetScope RGB bands followed by pansharpening with Sentinel-2. Spectral validation was conducted using 700 random samples. The highest correlation was observed in the GS-PCA 1 approach (R = 0.68) against PlanetScope, also showed medium correlation with Sentinel 2 (R = 0.33), indicating that the generated fused data relates to both sources. Further classification was performed using 404 samples model and 151 ground truth with three parametric algorithms: Maximum Likelihood, Minimum Distance to Mean, and Mahalanobis Distance. The highest accuracy was achieved using GS-PCA 1 method under the Maximum Likelihood classifier, with an overall accuracy of 26.96, outperforming Sentinel-2 (24.35) and PlanetScope (23.48). Although the accuracy remains moderate, this approach demonstrates the potential of multisensor fusion for tree species classification in tropical forests. © ACRS 2025.All rights reserved.