Ensemble-based estimation of fractional vegetation cover in arid and semi-arid regions

Fractional Vegetation Cover (FVC) is a key biophysical parameter for characterizing vegetation dynamics and ecosystem functioning, and is essential for ecological monitoring in arid and semi-arid regions. However, sparse vegetation and strong soil background effects in these regions lead to substantial variability in the applicability and stability of different inversion models, thereby limiting further improvements in FVC estimation accuracy. To address these challenges, this study applied three FVC inversion models—Random Forest (RF), Fully Constrained Least Squares (FCLS), and the Dimidiate Pixel Model (DPM)—to Sentinel-2 data, and further proposed a consistency-constrained ensemble approach. The results show that the three models exhibit pronounced discrepancies in both spatial patterns and value ranges. Specifically, RF performs better in high-coverage areas, whereas FCLS is more sensitive to soil background effects under low-coverage conditions. The proposed ensemble approach effectively integrates the strengths of individual models, improving overall accuracy by approximately 2.1% and reducing estimation uncertainty relative to individual models. These results demonstrate that the proposed approach improves FVC estimation in arid and semi-arid regions and provides a robust framework for large-scale quantitative vegetation monitoring using remote sensing.