Modelagem da suscetibilidade à erosão laminar (sheet erosion) na bacia hidrográfica do rio Xopotó (MG) através de análise multicritério e machine learning
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidade Federal de Viçosa
Abstract
O estudo da erosão do solo constitui uma parte importante do plano de manejo de bacias hidrográficas. Com tais estudos é possível desenvolver diferentes ações que buscam contribuir para o uso e ocupação da terra a partir de uma lógica conservacionista. Nesse contexto, modelos de susceptibilidade à erosão podem gerar informações que apoiem os processos de tomada de decisão. Devido a importância desses modelos e a existência de diferentes métodos que podem ser aplicados para os gerar, objetiva-se no presente estudo aferir, a partir da análise multicritério e Machine Learning, qual o método que gera o modelo de susceptibilidade à erosão laminar na BHRX com maior acurácia. Para tanto, foram utilizados os métodos de Sobreposição Ponderada, Analytic Hierarchy Process (AHP) e Random Forest (RF) com algumas variações, onde o RF foi utilizado de três formas, sendo a primeira com as variáveis de entrada selecionadas de forma manual, a segunda a partir do método Recursive Feature Elimination (RFE) e a terceira com o uso das variáveis selecionadas pelo RFE e o modelo gerado pelo AHP como variável. As variáveis selecionadas manualmente pelos decisores foram: Uso da terra, Declividade, Orientação das vertentes, Altitude e Geomorfologia. As variáveis selecionadas pelo método RFE foram: Altitude, Orientação das vertentes, Declividade, S/ope Lenght, Distância das estradas rurais, Pluviosidade, TWI, Uso da terra e Solos. Dentre os métodos avaliados através da curva ROC, foi visto que o método mais simples, que é a Sobreposição Ponderada obteve melhor resultado nesse contexto. Também, verifica-se que os locais com declividade mais acentuada, vertentes voltadas para o norte e nordeste e com uso de pastagem ou solo exposto são locais que, em todos os modelos apresentaram alta ou altissima susceptibilidade à erosão. Palavras-chave: Erosão do solo. Bacias hidrográficas. Modelos espaciais. Análise multicritério. Machine Learning.
The study of soil erosion 1s an Important part of the watershed management plan. With such studies 1t1s possible to develop different actions that seek to contribute to the use and occupation of the soil from a conservationist logic. In this context, eroston susceptibility models can generate information to support decision-making processes. Due to the importance of these models and the existence of different methods that can be applied to generate them, the objective of this study Is to assess, based on multicriteria analysis and Machine Learning, which method generates the model of susceptibility to laminar erosion m the BHRX with greater accuracy. For this purpose, the Weighted Overlay, Analytic Hierarchy Process (AHP) and Random Forest (RF) methods were used with some variations, where the RF was used 1n three ways, the first with manually selected input variables, the second from the Recursive Feature Elimination (RFE) method and the third with the use of variables selected by RFE and the model generated by AHP as a variable. The variables selected manually by the decision makers were: Land use, Slope, Orientation of the slopes, Altitude and Geomorphological Forms. The variables selected by the RFE method were Altitude, Orientation of slopes, Slope, Slope Lenght, Distance from rural roads, Rainfall, TWI, Land use and Soils. Among the methods evaluated through the ROC curve, it was seen that the simplest method, which 1s Weighted Overlay, obtammed the best result mn this context. Also, it appears that the places with steeper slopes, slopes facing north and northeast and using pasture or exposed soil are places that, in al] models, showed high or very high susceptibility to erosion. Keywords: Soil erosion. Watersheds. Spatial models. Multicriteria analysis. Machine Learning.
The study of soil erosion 1s an Important part of the watershed management plan. With such studies 1t1s possible to develop different actions that seek to contribute to the use and occupation of the soil from a conservationist logic. In this context, eroston susceptibility models can generate information to support decision-making processes. Due to the importance of these models and the existence of different methods that can be applied to generate them, the objective of this study Is to assess, based on multicriteria analysis and Machine Learning, which method generates the model of susceptibility to laminar erosion m the BHRX with greater accuracy. For this purpose, the Weighted Overlay, Analytic Hierarchy Process (AHP) and Random Forest (RF) methods were used with some variations, where the RF was used 1n three ways, the first with manually selected input variables, the second from the Recursive Feature Elimination (RFE) method and the third with the use of variables selected by RFE and the model generated by AHP as a variable. The variables selected manually by the decision makers were: Land use, Slope, Orientation of the slopes, Altitude and Geomorphological Forms. The variables selected by the RFE method were Altitude, Orientation of slopes, Slope, Slope Lenght, Distance from rural roads, Rainfall, TWI, Land use and Soils. Among the methods evaluated through the ROC curve, it was seen that the simplest method, which 1s Weighted Overlay, obtammed the best result mn this context. Also, it appears that the places with steeper slopes, slopes facing north and northeast and using pasture or exposed soil are places that, in al] models, showed high or very high susceptibility to erosion. Keywords: Soil erosion. Watersheds. Spatial models. Multicriteria analysis. Machine Learning.
Description
Citation
SOARES, Wesley Oliveira. Modelagem da suscetibilidade à erosão laminar (sheet erosion) na bacia hidrográfica do Rio Xopotó (MG) através de análise multicritério e machine learning. 2023. 86 f. Dissertação (Mestrado em Geografia) - Universidade Federal de Viçosa, Viçosa. 2023.
