Evaluation of Multi-Sensor Satellite Data Fusion for Improved Land Cover Classification in the Guinea Savanna Zone of Nigeria

The Guinea Savanna zone of Nigeria, spanning parts of Kwara, Niger, Kogi, Nassarawa, and Benue States, presents mixed agricultural, woodland, and settlement landscapes that are difficult to resolve accurately using single-sensor remote sensing imagery. This study evaluates multi-sensor data fusion approaches for improving land cover classification accuracy in the Guinea Savanna zone, addressing the research gap in fusion technique performance assessment for this specific vegetation and land use context. Optical data from Sentinel-2, SAR data from Sentinel-1, and texture features from ALOS PALSAR are fused using three approaches, namely principal component substitution, wavelet transform fusion, and stacked layer machine learning, and compared against single-source classification baselines. Random Forest classification is applied consistently across all fusion scenarios for comparability, and accuracy is evaluated using confusion matrix statistics for ten land cover classes. Results demonstrate that stacked Sentinel-1 and Sentinel-2 fusion achieves the highest overall accuracy of 93.7 percent, representing a statistically significant improvement over the single-source Sentinel-2 baseline of 86.4 percent. The greatest accuracy gains are achieved for distinguishing fallow agricultural land from dry season woodland and for separating flooded forest from non-flooded vegetation. The study provides specific guidance on optimal sensor combination strategies for land cover mapping programmes operating across the Guinea Savanna transition zone in Nigeria and West Africa. Keywords: multi-sensor fusion, land cover classification, Guinea Savanna, Random Forest, Sentinel.