Sensoriamento remoto associado a técnicas de mineração de dados para estimativas de produção agrícola

• Autor(a) principal: Becker, Willyan Ronaldo
• Data de Publicação: 2021
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações do UNIOESTE
• Texto Completo: http://tede.unioeste.br/handle/tede/5629
• topic: Análise de agrupamento; Estimativa de produtividade de soja; WOFOST; Dados agrometeorológicos ERA5/Nasa Power; Google Earth Engine; Python; Cluster analysis; Soybean yield estimation; WOFOST; ERA5/Nasa Power agrometeorological data; Google Earth Engine; Python; Sistemas Biológicos e Agroindustriais

The use of orbital remote sensing (RS) techniques has proven to be a valuable tool allowing agricultural monitoring on a regional scale with greater advance and accuracy and lower operating costs than traditional techniques. Given the country's geographical extent, the ability to identify and quantify agricultural areas and their production objectively and quickly, as well as the adequate regionalization of the Paraná crop and its agricultural statistics, are important aspects in the Brazilian agricultural context. The dissertation’s main objective was to map land use and land cover in a hydrographic basin, perform cluster analysis for regional restructuring and estimate soybean productivity in the state of Paraná. The three objectives and their corresponding scientific publications were grouped into three pieces for this objective. The first objective of this research was to develop a methodology for using Landsat-8 images on the Google Earth Engine platform to automatically classify land use and land cover in the São Francisco Verdadeiro hydrographic basin in the western region of Paraná (Paper 1). This approach compared commonly performed classifications with classification based on statistical attributes (median and standard deviation). The dataset of statistical attributes had an overall accuracy of 97.3% and Kappa of 0.9644, indicating that it was a reliable and representative method of mapping land use and land cover. The second objective of the dissertation was to develop a cluster analysis methodology for delimiting regions based on agroclimatic factors, considering the spatial proximity of data on soybean areas in Paraná (Paper 2). The current territorial divisions present many heterogeneities, due in part, to the form of delimitation of the regions and the different evolutions of the municipalities over the years. The findings in paper 2 demonstrated the efficacy of the agglomerative hierarchical grouping method for state regionalization of the investigated attributes. We indicate the state rearrangement under the soybean aspect in 9 groups, using the 'conventional' dataset and 17 groups using the 'soybean mapping' dataset. We highlight the usage of Geosilhouettes to better understand the local agro-geographic organization and find the appropriate distribution of groups with similar qualities and geographical regions. The third objective proposed here was to develop and evaluate the assimilation of RS data from satellite images in the WOFOST culture growth model, in a spatialized form (Paper 3). The premise of this study was that the orbital RS is a viable alternative to apply productivity models, as it provides the necessary spatialization to obtain localized and widely disseminated information. According to the yield maps, soybean yield varies more over time than it does spatially. This is primarily because agrometeorological data such as temperatures, precipitation, and solar radiation are spatially dependent. There is also a spatial dependence of phenological and biophysical data, as farmers follow a similar pattern in terms of sowing and harvest date with the climate determining agricultural activities. In comparison to the field yields, an R² of 0.51, MAE of 657.25 kg ha-1 and RMSE of 762.85 kg ha-1 were obtained. Overall, Google Earth Engine enabled the mapping to be completed quickly and easily. The interactivity and the speed of the platform’s processing were crucial during the learning and application stages. Python was equally essential throughout the dissertation, optimizing several tasks and enabling new approaches. The WOFOST model estimated productivity at the pixel level on a municipal scale, allowing for some reflection and knowledge.