Modelos agrometeorológicos, espectrais e de inteligência artificial para estimação de produtividade de soja

• Autor(a) principal: Richetti, Jonathan
• Data de Publicação: 2018
• 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/4173
• topic: Aprendizado de Máquina; Sensoriamento Remoto; Modelagem de Cultura; Machine Learning; Remote Sensing; Crop Modeling; CIENCIAS AGRARIAS::ENGENHARIA AGRICOLA

Global and local agricultural issues are often raised due to the steady growth in demand for food and the fluctuations in the commodity prices. Various forms can be used to obtain information on the agricultural production of a particular region. Among the different methods of estimating agricultural production, censuses and surveys are the methods used by official institutes such as IBGE and CONAB. These methods are time consuming and labor intensive, causing the information to be delayed. In order to improve estimates, clear and accurate methods are needed. Therefore, the aim of this thesis is to present methods to estimate the yield and production of soybeans in the state that also can be applied for forecasting of agricultural yield. Thus, this thesis is divided, fundamentally, in three scientific papers; each one approaches a method to estimate soybean yield using remote sensing data and modeling techniques. The first paper presents the use of machine learning algorithms and artificial intelligence in remote sensing data to estimate soybean yield in the state of Paraná. The second one shows how remote sensing data can be used to calibrate the CROPGRO growth model in two farms, one in Brazil and another in the United States. The third one presents an application of the FAO agrometeorological model to estimate soybean yield in the state of Paraná. The first paper presented performances with a mean error (ME) of 3.52 kg ha-1, root mean square error (RMSE) of 373 kg ha-1 and Wilmoott concordance index (dr) of 0.85 when compared with farm data. The second article shows an ME of -3.4 kg ha-1 and R2 of 89% for yield in Paraná, Brazil. In Iowa, USA, a RMSE of 864 kg ha-1, dr of 0.98 for biomass and RMSE, dr of 904 kg ha-1, and dr of 0.89 for pod weight were obtained. This study showed that the determination of leaf area index and light interception values from remote sensing vegetation indexes, such as the Normalized Difference Vegetation Index (NDVI) data can be used for calibration purposes of the model. The third article evaluated the CyMP software to estimate soybean yield in the state of Paraná from 2013 to 2016, obtaining differences starting at 31 kg ha-1 between the estimated yield and reported yield by CONAB.