The Brazilian Soil Available Water (TAW) acquired by remote sensing and machine learning techniques

• Ano de defesa: 2025
• Autor(a) principal: Vogel, Letícia Guadagnin
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Drenagem do solo; Hydrological Modeling; Modelagem do solo; Modelagem hidrológica; Remote sensing; Saúde do solo; Sensoriamento remoto; Soil Drainage; Soil Health; Soil Modeling
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/11/11140/tde-04042025-163503/

Brazilian agricultural production is a major supplier of good the international market, applying research and technology to improve yield, reduce costs, and preserve natural resources, especially water. Soil available water (AW) relies on soil attributes, such as texture, structure, and mineral composition, and determines plant development. This study aimed to map the total available water of the soil (TAW) across the Brazilian territory using soil data, remote sensing, machine learning, and hydrological models. A georeferenced database with 41,438 soil profiles was used. Soil attributes (clay, silt, sand, and organic matter) were in layers at 1 m deep. We implemented a pedotransfer function (PTF) to calculate soil hydraulic parameters, SWAP, and MFlux hydrological models to calculate field capacity (FC) and permanent wilting point (PWP). We combined these models to calculate the TAW for all soil profiles. We calibrated a Random Forest algorithm with 21 environmental covariates, including topography, spectral characteristics of vegetation, and soil. The predictive model showed high accuracy (R2 = 0.66), highlighting the relevance of spectral and topographic characteristics in soil water retention. The results demonstrated significant variations in TAW among the biomes (Amazon Forest, Caatinga, Cerrado, Atlantic Forest, Pampa, and Pantanal), soil classes (Acrisol/Alisol, Cambisol, Gleysol, Ferralsol, Arenosol, Leptosol, Nitisol, and Plinthosol) and land uses (Forest, Savanna, Mangrove, Floodable, Wetland, Grassland, Herbaceous, Pasture, Soybean, Sugar cane, Rice, Cotton, Coffee, Citrus, Palm, Silviculture), reflecting complex interactions between environmental and pedological factors. The Brazilian semiarid region (143 mm.m-1) and the MATOPIBA region (138 mm m-1) had the lowest TAW values, highlighting the need for adequate management practices and environmental monitoring. The spatial distribution of TAW provides valuable insights for sustainable land management, guiding decisions on irrigation, agricultural practices, and conservation strategies tailored to each biome. The preservation of water resources is essential for maintaining biodiversity, ecosystem services, and agricultural productivity, especially in fragile ecosystems like the Cerrado and the Caatinga. By leveraging advanced geospatial and machine learning techniques, this study contributes to the sustainable development of Brazilian agriculture while addressing key environmental challenges.