@article{fdi:010081344,
  title = {{T}he real potential of current passive satellite data to map aboveground biomass in tropical forests},
  author = {{J}ha, {N}. and {T}ripathi, {N}. {K}. and {B}arbier, {N}icolas and {V}irdis, {S}. {G}. {P}. and {C}hanthorn, {W}. and {V}iennois, {G}. and {B}rockelman, {W}. {Y}. and {N}athalang, {A}. and {T}ongsima, {S}. and {S}asaki, {N}. and {P}{\'e}lissier, {R}apha{\¨e}l and {R}{\'e}jou-{M}{\'e}chain, {M}axime},
  editor = {},
  language = {{ENG}},
  abstract = {{F}orest biomass estimation at large scale is challenging and generally entails large uncertainty in tropical regions. {W}ith their wall-to-wall coverage ability, passive remote sensing signals are frequently used to extrapolate field estimates of forest aboveground biomass ({AGB}). {H}owever, studies often use limited reference data and/or flawed validation schemes and thus report unreliable extrapolation error estimates. {H}ere, we compared the ability of three medium- to high-resolution passive satellite sensors, {L}andsat-8 ({L}8), {S}entinel-2{B} ({S}2) and {W}orldview-3 ({WV}3), to map {AGB} in a forest landscape of {T}hailand. {W}e used a large airborne {L}i{DAR}-derived {AGB} dataset as a reference to train and validate a random forest algorithm and conducted robust error assessments and variable selection using spatialized cross-validations. {O}ur results indicate that the selected predictors strongly varied among the three sensors and between analyses were restricted to low (<= 200 {M}g ha(-1)) and high (>200 {M}g ha(-1)) {AGB} areas. {WV}3 and {S}2 data outperformed {L}8 data to extrapolate {AGB} ({RMSE} of 68 and 72 against 84 {M}g ha(-1), respectively) due to the inclusion of the red-edge band and, probably, to their higher spatial and spectral resolution. {S}ensitivity to large {AGB} values was higher for {WV}3 than for {S}2 and {L}8 with saturation point of 247 {M}g ha(-1) against 204 and 192 {M}g ha(-1). {AGB} values above these saturation points remained poorly predictable, especially for {L}8, indicating that several tropical forest {AGB} maps should be interpreted with extreme caution. {H}owever, predicted gradients of lower {AGB} values (<= 200 {M}g ha(-1)), i.e., in early forest successional stages, were fairly consistent among sensors (r > 0.70), even if the mean absolute difference between estimates was large when {AGB} predictions were extrapolated out of the calibration area at regional level (34%). {W}e finally showed that calibrating the model only within the sensitivity {AGB} domain (e.g., <200 {M}g ha(-1)) minimizes the risk of induced bias for estimating small {AGB} values. {T}hese results provide important benchmarks for interpreting previously published maps and to improve future validation schemes.},
  keywords = {{B}iomass mapping ; forest carbon ; passive satellite sensor ; random forest ; sensor saturation limit ; {THAILANDE} ; {ZONE} {TROPICALE} ; {KHAO} {YAI} {PARC} {NATIONAL}},
  booktitle = {},
  journal = {{R}emote {S}ensing in {E}cology and {C}onservation},
  volume = {[{E}arly access]},
  numero = {},
  pages = {[17 p.]},
  year = {2021},
  DOI = {10.1002/rse2.203},
  URL = {https://www.documentation.ird.fr/hor/fdi:010081344},
}