ALSAT Satellite Imagery Enhancement using Generative Models for Forest Mapping

National satellite programs offer strategic autonomy but are often constrained by sensor hardware limitations. Algeria's ALSAT-2B satellite lacks the critical Short-Wave Infrared (SWIR) and Red-Edge spectral bands essential for quantitative forest monitoring, limiting the utility of this sovereign data asset. This research introduces a novel AI-driven framework to digitally enhance ALSAT-2B imagery, unlocking its full potential for national forest management (SDG 13 & 15).

Our methodology employs a synergistic, two-stage process. Stage 1 (Spectral Enrichment): A Generative Adversarial Network (GAN) with a SwinUNet architecture was trained to synthetically reconstruct the five missing spectral bands by learning the relationship between ALSAT-2B and Sentinel-2 data. Stage 2 (Spatial Enhancement): The now nine-band data was fused with ALSAT-2B’s 2.5m panchromatic band using the Gram–Schmidt pansharpening algorithm, enhancing its spatial resolution.

The framework's efficacy was validated through a forest cover classification task. Results showed a dramatic improvement at each stage. The classification F1-Score on the baseline four-band, 10m ALSAT-2B data was 60.28%. After spectral enrichment, the score increased to 70.08%. The final, fully enhanced nine-band, 2.5m product achieved an F1-Score of 74.03%—a total improvement of nearly 14 percentage points.

This research presents a powerful framework for valorizing sovereign satellite data through AI. By successfully generating critical spectral bands and enhancing spatial detail, we can transform existing imagery into a high-value, analysis-ready product, providing Algerian institutions with a tangible, deployable methodology to significantly improve national forest monitoring and management.