Abstract

Accurate estimation of total ecosystem carbon requires integrated assessment of above-ground biomass (AGB) and soil organic carbon (SOC), yet most remote sensing approaches focus on individual components. This study develops and validates multi-sensor remote sensing frameworks for total carbon estimation across 347 validation sites spanning forests (134 sites), grasslands (98 sites), croplands (85 sites), and shrublands (30 sites) in 28 countries. We integrated optical (Sentinel-2, Landsat-8), radar (Sentinel-1, PALSAR-2), and LiDAR data using machine learning algorithms including random forest, support vector machines, and deep neural networks. Results demonstrate that integrated models achieve superior accuracy for total carbon estimation (R² = 0.87, RMSE = 23.4 Mg C ha⁻¹) compared to single-component approaches (R² = 0.64-0.72). Multi-sensor fusion improved AGB estimation accuracy by 34% (RMSE = 18.7 vs 28.3 Mg ha⁻¹) and SOC estimation by 28% (RMSE = 8.9 vs 12.4 Mg C ha⁻¹) compared to single-sensor methods. Deep learning models showed strongest performance for complex forest ecosystems (R² = 0.91), while random forest algorithms excelled in agricultural landscapes (R² = 0.84). Optical-radar fusion proved most effective for AGB estimation, while thermal infrared and topographic variables were critical for SOC prediction. Validation across biomes revealed consistent performance with total carbon estimates ranging from 45.2 ± 12.8 Mg C ha⁻¹ in grasslands to 287.6 ± 68.4 Mg C ha⁻¹ in mature forests. Temporal analysis using 5-year time series demonstrated carbon change detection capability with 89% accuracy for changes >10 Mg C ha⁻¹. Spatial scaling analysis indicates potential for global carbon mapping with uncertainty <15% for 90% of terrestrial areas. However, challenges remain for areas with persistent cloud cover, complex topography, and sparse ground truth data. Economic analysis reveals cost savings of 67% compared to field-based carbon assessment while providing complete spatial coverage. These findings demonstrate that integrated remote sensing approaches enable accurate, cost-effective total carbon estimation essential for climate change monitoring, carbon market verification, and ecosystem management across multiple scales.