Fatores naturais e antrópicos que influenciam na presença de nitrato no Aquífero Serra Geral Livre no perímetro urbano de Campo Grande - MS

• Ano de defesa: 2024
• Autor(a) principal: Diniz, Felipe Gavioli
• Orientador(a): Gabas, Sandra Garcia
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Mato Grosso do Sul
• País: Brasil
• Palavras-chave em Português: Águas Subterrâneas; Aprendizado de Máquina; Hidrogeologia
• Link de acesso: https://repositorio.ufms.br/handle/123456789/11446

The objective of this work was to analyze which anthropogenic and natural factors influence the presence of nitrate in groundwater in urban areas using machine learning. The research site was the urban perimeter of the municipality of Campo Grande, capital of the state of Mato Grosso do Sul, which is located in the northern portion of the municipality. The methodology involved the use of the QGIS software for the production of maps and the Python language together with the Scikit-learn, NumPy, SciPy and Matplotlib libraries for data analysis and development of machine learning models. The following spatial variables were considered as influencing the nitrate level: distance to the water and sewage network, urban area, population density per neighborhood, slope, hypsometry, Normalized Difference Vegetation Index (NDVI), geology, water level depth, geotechnical characteristics and proximity to drainage systems. Nine statistical models from the classifier group and seven from the regressor group were adopted to analyze the data collected from 68 wells in 2018. The predictive results showed that the urban area and slope were the variables with the greatest influence on the nitrate value in both models. On the other hand, geology was the variable with the least influence in the classifier models and the distance from the water network in the regressor models. The results showed that the classifier models outperformed the regressor models, with five models presenting a result with the training group ranging from 0.60 to 0.80 and the test group between 0.40 and 0.60, highlighting the MLP Classifier model with the best performance. The other classifier models presented overfitting, a problem that also affected the performance of the regressor models, with two models presenting the same problem. Furthermore, five regressor models had a negative result and only the Logistic Regressor model revealed an acceptable result with the training group at 0.60 and with a value close to this in the test group. By estimating the distribution of nitrate, it was found that the highest concentrations occur in areas with greater slopes and high population density. On the other hand, the lowest concentrations are found in regions where there is no record of sewage networks, with less slope, low population density and where the groundwater level is deeper.