The real potential of current passive satellite data to map aboveground biomass in tropical forests

Forest biomass estimation at large scale is challenging and generally entails uncertainty in tropical regions. With their wall-to-wall coverage ability, passive remote sensing signals are frequently used to extrapolate field estimates of forest aboveground (AGB). However, studies often use limited reference data and/or flawed validation schemes thus report unreliable extrapolation error estimates. Here, we compared the ability three medium- high-resolution satellite sensors, Landsat-8 (L8), Sentinel-2B (S2) Worldview-3 (WV3), map AGB a landscape Thailand. We airborne LiDAR-derived dataset as train validate random algorithm conducted robust assessments variable selection using spatialized cross-validations. Our results indicate that selected predictors strongly varied among sensors between analyses were restricted low (?200 Mg ha?1) high (>200 Mg ha?1) areas. WV3 S2 outperformed L8 (RMSE 68 72 against 84 Mg ha?1, respectively) due inclusion red-edge band and, probably, higher spatial spectral resolution. Sensitivity values was for than with saturation point 247 Mg ha?1 204 192 Mg ha?1. above these points remained poorly predictable, especially L8, indicating several maps should be interpreted extreme caution. predicted gradients lower (?200 Mg ha?1), i.e., early successional stages, fairly consistent (r > 0.70), even if mean absolute difference when predictions extrapolated out calibration area regional level (34%). finally showed calibrating model only within sensitivity domain (e.g., <200 Mg ha?1) minimizes risk induced bias estimating small values. These provide important benchmarks interpreting previously published improve future schemes.