TY - JOUR

T1 - Quantifying Post-Fire Changes in the Aboveground Biomass of an Amazonian Forest Based on Field and Remote Sensing Data

AU - Pontes-Lopes, Aline

AU - Dalagnol, Ricardo

AU - Dutra, Andeise Cerqueira

AU - de Jesus Silva, Camila Valéria

AU - de Alencastro Graça, Paulo Maurício Lima

AU - de Oliveira e Cruz de Aragão, Luiz Eduardo

PY - 2022/3/23

Y1 - 2022/3/23

N2 - Fire is a major forest degradation component in the Amazon forests. Therefore, it is important to improve our understanding of how the post-fire canopy structure changes cascade through the spectral signals registered by medium-resolution satellite sensors over time. We contrasted accumulated yearly temporal changes in forest aboveground biomass (AGB), measured in permanent plots, and in traditional spectral indices derived from Landsat-8 images. We tested if the spectral indices can improve Random Forest (RF) models of post-fire AGB losses based on pre-fire AGB, proxied by AGB data from immediately after a fire. The delta normalized burned ratio, non-photosynthetic vegetation, and green vegetation (ΔNBR, ΔNPV, and ΔGV, respectively), relative to pre-fire data, were good proxies of canopy damage through tree mortality, even though small and medium trees were the most affected tree size. Among all tested predictors, pre-fire AGB had the highest RF model importance to predicting AGB within one year after fire. However, spectral indices significantly improved AGB loss estimates by 24% and model accuracy by 16% within two years after a fire, with ΔGV as the most important predictor, followed by ΔNBR and ΔNPV. Up to two years after a fire, this study indicates the potential of structural and spectral-based spatial data for integrating complex post-fire ecological processes and improving carbon emission estimates by forest fires in the Amazon.

AB - Fire is a major forest degradation component in the Amazon forests. Therefore, it is important to improve our understanding of how the post-fire canopy structure changes cascade through the spectral signals registered by medium-resolution satellite sensors over time. We contrasted accumulated yearly temporal changes in forest aboveground biomass (AGB), measured in permanent plots, and in traditional spectral indices derived from Landsat-8 images. We tested if the spectral indices can improve Random Forest (RF) models of post-fire AGB losses based on pre-fire AGB, proxied by AGB data from immediately after a fire. The delta normalized burned ratio, non-photosynthetic vegetation, and green vegetation (ΔNBR, ΔNPV, and ΔGV, respectively), relative to pre-fire data, were good proxies of canopy damage through tree mortality, even though small and medium trees were the most affected tree size. Among all tested predictors, pre-fire AGB had the highest RF model importance to predicting AGB within one year after fire. However, spectral indices significantly improved AGB loss estimates by 24% and model accuracy by 16% within two years after a fire, with ΔGV as the most important predictor, followed by ΔNBR and ΔNPV. Up to two years after a fire, this study indicates the potential of structural and spectral-based spatial data for integrating complex post-fire ecological processes and improving carbon emission estimates by forest fires in the Amazon.

KW - forest fire

KW - degradation

KW - biomass

KW - change detection

KW - Landsat-8

KW - Google Earth Engine

U2 - 10.3390/rs14071545

DO - 10.3390/rs14071545

M3 - Journal article

VL - 14

JO - Remote Sensing

JF - Remote Sensing

SN - 2072-4292

IS - 7

M1 - e1545

ER -