Análise da variabilidade agrometeorológica e espectral associada ao ciclo da soja e estimativa da produtividade com imagens de satélites

• Ano de defesa: 2021
• Autor(a) principal: Gasparin, Priscila Pigatto
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual do Oeste do Paraná; Medianeira; Brasil; Programa de Pós-Graduação em Engenharia Agrícola; UNIOESTE
• Palavras-chave em Português: Análise multivariada; Análise de componentes principais; Análise por agrupamento; Multivariate analysis; Principal components analysis; Cluster analysis; CNPQ::CIENCIAS AGRARIAS; Ciências Agrárias
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/29593

Soy is one of the main agricultural products and of great importance for the Brazilian economy, as it is the raw material for the production of food, fuel and industrial applications. However, agricultural production is highly influenced by climate variability that causes both positive and negative impacts on crop productivity. Therefore, studies on these variables are relevant, as well as on the dynamics of the crop, through vegetation indices, in a global and regional context, in order to obtain better results in agricultural activities. On that account, the general objective of this research was to analyze the spatial and temporal variabilities of soy bean and estimate the yield of this crop by using satellite images in the state of Paraná. For this purpose, ten-year metrics associated with agrometeorological variables (VAs) andimproved vegetation index (EVI) during the soybean cycle were evaluated, based on a timeseries of ECMWF images (European Center for Medium Range Weather Forecasts) and the MODIS sensor (Moderate Resolution Imaging Spectroradiometer), in different agricultural scenarios, corresponding to the 2011/2012, 2013/2014 and 2015/2016 crop years, which were identified with low, medium and high rainfall rates. These indices were evaluated from 2000 to 2016. The dissertation is divided into three articles: in the first and second articles, multivariate techniques were used in order to regionalize the state of Paraná. In the first article, factorial and cluster analyses were used, whilst in the second one multivariate techniques that also consider the spatial dependence of the location of virtual stations (EV), called MULTISPATI-PCA, were applied. And finally, in the third article, soybean productivity in the state of Paraná was estimated by the methods of ordinary least squares (OLS) and geographically weighted regression (GWR), from which the results of these models were compared, in order to obtain a model with the best accuracy and precision possible. Ingeneral, the results displayed the formation of similar agroclimatic and spectral regions, obtaining Group 1 (southern meso regions) with the lowest preferences for the three agricultural scenarios and Group 2 (western meso region) with low precipitation and water balance values for the 2011/2012 crop year, the opposite occurring for the 2015/2016 crop year. In addition, the MULTISPATI-PCA method showed linear variations that were morecontiguous than the characteristics by classical ACP and well-defined groupings. The western mesoregion must be highlighted when it comes to the low precipitation scenario, because it presented low exclusion values and water balance during the stages of fullflowering to grain filling (R2 - R5), causing a low productivity, the opposite occurring for thehigh precipitation scenario. Finally, the GWR model showed better accuracy and precision inestimating soybean yield, when compared to the OLS model, demonstrating a spatial andtemporal heterogeneity between yield and the metrics analyzed in the model. With theseresults, it is possible to define the most suitable strategies for the cultivation of soy bean in the state of Paraná, as which allow to help both farmers and the agencies responsible for crop planning in decision making, providing the best results for productivity.