Integrating Sentinel-1 SAR and Sentinel-2 multispectral data for tree species classification and mapping in smallholder planted forests of Tanzania

Abstract The rising demand for timber, alongside declining natural forests, has driven rapid growth of plantation forestry, especially smallholder woodlots in sub-Saharan Africa. Despite their increasing importance for livelihoods, ecosystem services, and climate change mitigation, detailed information on their extent and composition remains scarce. Remote sensing, particularly the integration of radar and optical data, offers a scalable solution for monitoring such heterogeneous landscapes. This study evaluated the potential of Sentinel-1 Synthetic Aperture Radar (SAR), Sentinel-2 multispectral data, and their combination for classifying smallholder plantation tree species in the Southern Highlands of Tanzania using Random Forest (RF) and Support Vector Machine (SVM) algorithms. Sentinel-2 achieved slightly higher accuracy compared to Sentinel 1, achieving overall accuracies of 90.3% and 89.6%, respectively, with Kappa coefficients above 86%. The combined dataset yielded the highest accuracy (90.3%) with RF, demonstrating the complementary strengths of radar and optical data. Although SVM performed lower (81.3%), its accuracy improved substantially with data integration. The classification maps also revealed that smallholder plantations dominate the study area, covering 72.8% of the landscape. These results demonstrate the effectiveness of integrating open-access Sentinel data with machine-learning algorithms for plantation tree species mapping in the Southern Highlands of Tanzania, while acknowledging that broader transferability requires further testing beyond the study area.