Modelagem de fluxos de CO2 em sistemas pastoris do bioma Pampa com dados de sensoriamento remoto e random forest

• Ano de defesa: 2023
• Autor(a) principal: Bremm, Tiago
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• dARK ID: ark:/26339/001300001b3z8
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Meteorologia; UFSM; Programa de Pós-Graduação em Meteorologia; Centro de Ciências Naturais e Exatas
• Palavras-chave em Português: Fluxo de CO2; GPP; RECO; MOD17; Machine learning; Eddy covariance; CO2 flux; CNPQ::CIENCIAS EXATAS E DA TERRA::GEOCIENCIAS::METEOROLOGIA
• Link de acesso: http://repositorio.ufsm.br/handle/1/33314

Recent reports from the Intergovernmental Panel on Climate Change (IPCC) indicate that current greenhouse gas concentrations in the atmosphere have reached their highest levels in the past 800,000 years, intensifying global warming and causing an increase in extreme weather events. The economic sectors that contribute significantly to the increase in greenhouse gases, particularly carbon dioxide (CO2), are agriculture and livestock. These sectors are strongly present in states such as Rio Grande do Sul (RS), mainly due to local climatic and geographical conditions. The southern half of the state, belonging to the Pampa biome, characterized by lowlying vegetation, has been used for livestock for centuries, but is under strong pressure to convert to agriculture, which has already been consolidated in the northern half of the state since 1950. Understanding CO2 exchanges in native Pampa grasslands becomes essential to support public policies for mitigation and maintenance of the natural ecosystem. In this study, the net exchange of CO2 (NEE) between native grassland and the atmosphere was measured using eddy covariance towers, using the turbulent vortex covariance (EC) technique. The NEE obtained was partitioned into the gross primary production (GPP) and ecosystem respiration (RECO) components. The three components were used to obtain regional models of fluxes between the surface and the atmosphere. The data were obtained from three eddy covariance towers on different grazing managements in native Pampa field: one in the center of RS (SMA, in Santa Maria) and two in the south of RS (ACR and ACD, both in Aceguá - ACE). The results of this work are presented in the form of two articles. In the first, we estimated GPP using the MOD17 algorithm for the SMA and ACE sites, exploring the different parameterizations of the tabulated parameters for terrestrial biomes, BPLUT, (savanna and grasses) and input meteorological data (reanalysis and measured surface data). The results showed that the model underestimates measured GPP. The simulation with calibration of the maximum light use efficiency parameter (Ԑmax) seasonally obtained the best results, with a significant decrease in underestimations. In the second article, we present a unified methodology to estimate NEE, GPP and RECO through the use of machine learning (Random Forest - RF), in conjunction with satellite data. The results showed that, even with few years of data for RF model training, it was possible to estimate NEE, GPP and RECO with good accuracy (R 2 > 0.59, R 2 > 0.74 and R 2 > 0.65, respectively) and with underestimation less than 16% for all sites and all components, except NEE in ACR, probably due to the cattle management being more intense than the others. This methodology presented lower GPP underestimation than that estimated by the MOD17 model, and improvements can be made, including a variable that represents cattle management. In this way, the methodology using RF can become an important tool for assessing CO2 exchanges and for feasibility studies of carbon credit projects due to its predictive potential and easy acquisition of the variables needed in the modeling.