Integration of LiDAR and hyperspectral data for forest disturbance characterization and aboveground biomass estimation in the Brazilian Amazon

• Ano de defesa: 2020
• Autor(a) principal: Catherine Torres de Almeida
• Orientador(a): Lênio Soares Galvão, Luiz Eduardo Oliveira e Cruz de Aragão
• Banca de defesa: Jean Pierre Henry Balbaud Ometto, Paulo Maurício Lima de Alencastro Graça, Michael Maier Keller
• 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/2020/04.03.20.59

Advancements in remote sensing technologies provide new opportunities to answer complex ecological questions in tropical forests, which play a crucial role on the stability of global biogeochemical cycles and biodiversity. Light Detection And Ranging (LiDAR) and Hyperspectral Imaging (HSI) provide complementary information that can potentially improve the characterization of tropical forests and reduce the uncertainties in estimating greenhouse gas emissions from deforestation and forest degradation. This thesis aims to explore optimal procedures for improving tropical forest disturbance characterization and aboveground biomass (AGB) modeling using integrated LiDAR and HSI data and advanced machine learning algorithms. The study area covered 12 sites distributed across the Brazilian Amazon biome, spanning a variety of environmental and anthropogenic conditions. The methods were divided into three parts: (1) classification of forest disturbance status (Chapter 5); (2) AGB modeling (Chapter 6); and (3) analysis of the AGB variability according to anthropogenic and environmental variables (Chapter 7). Firstly, four classes of forest disturbance (undisturbed forests, disturbed mature forests, and two stages of secondary forests) were identified using Landsat time series between 1984 and 2017. Several LiDAR and HSI metrics obtained over 600 sample plots were then used as input data to three machine learning models for distinguishing those classes. Secondly, georeferenced inventory data from 132 sample plots were used to obtain a reference field AGB. A great number of LiDAR and HSI metrics (45 and 288, respectively) were submitted to a correlation filtering followed by a feature selection procedure (recursive feature elimination) to optimize the performance of six regression models. Finally, the average of AGB predictions from the best multisensor models was calculated over 600 sample plots where field AGB data were not available. A multivariable linear regression model was then used to assess the extent to which the predicted AGB variability was affected by anthropogenic (disturbance type and time) and environmental (annual rainfall, climatic water deficit, and topography) factors in secondary and mature forests. Overall, the results showed that the combination of LiDAR and HSI data improved both the classification of forest disturbances and the estimation of AGB compared to using a single data source. Using multisource remote sensing data was more effective than using advanced machine learning for both classification and regression models. The LiDAR-based upper canopy cover and the HSI-based absorption bands in the nearinfrared (NIR) and shortwave infrared (SWIR) spectral regions were the most influential metrics for characterizing the disturbance status and estimating AGB. Anthropogenic disturbances played the greatest effect on predicted AGB variability, reducing up to 44% the AGB of disturbed mature forests compared to the undisturbed ones. Secondary forests displayed an AGB recovery rate of 4.4 Mg.ha-1.yr-1. Water deficit also affected the variability of AGB in both mature and secondary forests, suggesting a lower recovery potential in water-stressed areas. The results highlight the potential of integrating LiDAR and HSI data for improving our understanding of forest dynamics in the face of increasing anthropogenic global changes.