Evaluation of multi-sensor image fusion techniques for super-resolution LULC mapping using Sentinel-2 and PlanetScope images

Accurate geospatial data is essential for monitoring land-use change and supporting sustainable territorial management. This study evaluates five image fusion techniques – injection, average interpolation, ratio component substitution, Bayesian, and local mean variance matching – to integrate Sentinel-2 (10 m and 20 m) with PlanetScope (3.1 m) imagery for land use/land cover (LULC) mapping in two Colombian watersheds characterized by high environmental and social complexity. Unlike prior studies, we propose a systematic band-pairing framework based on spectral linear regression that integrates SWIR-1 and SWIR-2 bands to produce super-resolution outputs. The fused datasets were assessed using standard fusion quality metrics and thematic accuracy indicators. Results show that the regression-based pairing improves fusion consistency, while the Bayesian method yields the highest classification accuracy, particularly in heterogeneous landscapes. This work advances current fusion practices by offering a transparent, reproducible methodology for integrating multiscale satellite data, overcoming spectral mismatches in SWIR bands, and enabling fine-scale LULC monitoring. The proposed approach offers a practical, scalable alternative to black-box models, yielding more accurate and timely spatial information for sustainable land management and policy development.