Early Reforestation Detection in Kenya Using Multi-Temporal Analysis

Reforestation is critical for climate mitigation and biodiversity restoration. Yet, early verification (0–5 years post-planting) remains challenging due to limitations of standard vegetation indices like NDVI and SAVI, which saturate under moderate biomass and are sensitive to soil background. Here we develop a multi-temporal framework integrating full-spectrum Sentinel-2 imagery with machine learning and symbolic regression to create climate-zone–specific models for Kenya. Our approach improves early reforestation detection accuracy by up to +0.21 ROC AUC over NDVI, particularly at +1 and +2 years post-planting. Symbolic regression yields compact, interpretable indices that approach machine learning performance, enhancing transparency. This scalable, open-data method enables earlier, more reliable restoration verification, supporting results-based climate finance and adaptable monitoring across diverse ecological zones.