This is the authors' abstract. We don't add key points for this paper.
Fractional vegetation cover (FVC) estimated from remote sensing images is crucial for understanding vegetation composition and spatial distribution. The regression-based unmixing model with synthetic training data offers a promising method for FVC estimation using time-series remote sensing images. However, in practice, noise and missing data due to cloud or shadow contamination often introduce significant uncertainty when conventional machine learning models are used. This paper proposes a novel deep learning model that performs the regression-based unmixing to effectively preserve spectral-temporal information while addressing the challenges of missing data in a time-series of images. Unlike traditional methods, the proposed method uses original reflectance values to generate synthetic training data containing missing values and trains an end-to-end dual-layer Long Short-Term Memory (LSTM) deep learning model for FVC estimation. Applied to Sentinel-2 time series images in the Yarlung Zangbo River Basin, a typical alpine region, the proposed method mapped FVC for four vegetation types – broadleaf trees, needleleaf trees, shrubs, and herbaceous – and analysed their spatial distribution. The results showed that the proposed method achieved mean absolute errors ranging from 6.29% to 12.94%, improving FVC estimation accuracy by 5.30% to 7.56% over traditional random forest models. The spatial distribution of FVC was significantly influenced by terrain factors, with higher FVCs generally associated with lower elevations, gentler slopes, and sunnier aspects. This study exhibits the value of the proposed model for accurate and robust FVC estimation, highlighting its potential for advancing remote sensing applications in vegetation monitoring.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Yang et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: