Quantifying and monitoring tropical forest mortality with passive and active optical remote sensing

• Ano de defesa: 2020
• Autor(a) principal: Ricardo Dal'Agnol da Silva
• Orientador(a): Luiz Eduardo de Oliveira e Cruz Aragão, Lênio Soares Galvão
• Banca de defesa: Veraldo Liesenberg, Bruce Walker Nelson
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Instituto Nacional de Pesquisas Espaciais (INPE)
• Programa de Pós-Graduação: Programa de Pós-Graduação do INPE em Sensoriamento Remoto
• País: BR
• Link de acesso: http://urlib.net/sid.inpe.br/mtc-m21c/2020/03.05.13.27

Tree mortality is a key process in the global carbon cycle generally linked to climatic feedbacks and accelerated by human-induced disturbances in the Amazon. Remote sensing can complement ground observations of tree mortality to support Amazon-wide detection. However, different from temperate forests, tree mortality detection over tropical forests is challenging because of the high heterogeneity in forest structure and biodiversity. It requires the development of new methods with multiple data sources to address challenges such as the detection of vegetation-specific mortality at the landscape scale; the quantification of individual tree mortality related to logging at the local scale; and the characterization of gap dynamics as a proxy for tree mortality, potentially related to natural and anthropogenic activities, and up-scaling estimates from local to regional scales. Here, the objective was to develop and validate novel approaches for the detection and monitoring of tropical forest mortality, using Moderate Resolution Imaging Spectroradiometer (MODIS), Very High Resolution (VHR) and airborne Light Detection And Ranging (LiDAR) data acquired over the Amazon region. For the vegetation-specific approach at the landscape scale, MODIS data processed by the Multi-Angle Implementation of Atmospheric Correction (MAIAC) algorithm was used to map the bamboo die-off in the southwest Amazon and to test whether it enhanced fire occurrence. At the individual tree level, multi-temporal VHR data from the WorldView-2 and GeoEye-1 satellites were used to evaluate the detection of canopy tree loss from selective logging at the Jamari National Forest. Finally, to explore the use of gaps as a proxy for tree mortality, five multi-temporal LiDAR datasets, and 610 single-date flight lines were considered to provide a systematic assessment of gaps and tree mortality, and explore their relationships with environmental and climate drivers. Results at the landscape scale, using MODIS (MAIAC) data, showed automatic detection of historical bamboo die-off (accuracy of 79%) and mapping of 15.5 million ha of bamboo-dominated forests. The bamboo-fire hypothesis was not supported, because the bamboo die-off areas did not show higher fire probability than the other areas. However, the fire occurrence was mostly associated with ignition sources from land use, suggesting a bamboo-human-fire association. At the local scale, individual tree losses from logging were successfully detected using VHR satellite imagery and a random forest (RF) model with 64% accuracy. In addition, large-gap openings associated with the tallest trees were more successfully detected by VHR data. At the local scale, LiDAR-gaps delineated using the relative height method, represented at least 50% of the tree mortality. The mortality of shorter trees at the canopy level (<25 m) was more successfully detected than the mortality of taller emergent trees (>25 m). Higher gap fractions (proxy for mortality) were associated with increased water deficit, soil fertility, and the occurrence of degraded and flooded forests. The Amazon-wide tree mortality map showed higher tree mortality rates in the west and southeast regions than in the central-east and north regions. This pattern was consistent with field-based observations. Overall, the findings highlighted the feasibility and importance of using passive and active optical remote sensing for detecting different processes of tropical forest mortality over a broad scale in the Amazon region.