Analysis of Soil Water Erosion Risk Using Machine Learning Techniques – A Case Study of Ourika Watershed in Morocco
More details
Hide details
1
Geosciences Geotourism Natural Hasards and Remote Sensing Laboratory, Department of Geology, Faculty of Sciences Semlalia, Cadi Ayyad University, BP 2390, 40 000, Marrakech, Morocco
2
The Chamber of Agriculture of Marrakech, Sa, Morocco
Corresponding author
Ilham L' Arfouni
Department of Geology, Faculty of Sciences Semlalia, Cadi Ayyad University, BP 2390, 40 000, Marrakech, Morocco
Ecol. Eng. Environ. Technol. 2024; 10:324-338
KEYWORDS
TOPICS
ABSTRACT
Soil erosion is a major environmental problem with detrimental consequences. In this article, we present a detailed study on the analysis of soil water erosion using Machine Learning (ML) techniques in the Oued Ourika watershed. We collected data on various factors that may influence the mechanisms of soil water erosion events. Subsequently, we developed machine learning models to predict the potential for soil water erosion based on these factors. Finally, field studies were conducted compared to the obtained results.
A historical inventory of water erosion has been created through fieldwork, satellite imagery, and historical water erosion events. Models were constructed using the training data, and their performance and accuracy in predicting susceptibility to water erosion were evaluated using the validation data. This data division allowed for a fair assessment of the models' ability to make accurate predictions.
Using a Geographic Information System (GIS) and programming in the R language, four supervised machine learning algorithms were applied, including K-Nearest Neighbor (KNN), Extreme Gradient Boosting (XGB), Random Forest (RF), and Naive Bayes (NB). The results show that the NB model exhibited the highest accuracy in predicting and evaluating the effectiveness of these algorithms in forecasting susceptibility to water erosion in the study area. Accuracy was assessed using the Area Under the Curve (AUC) metric, with an AUC of 98%. The XGB algorithm had an AUC of 96%, followed by RF with an AUC of 87%, and KNN with an AUC of 84%. Thus, the Naive Bayes model proved to be the best for determining susceptibility to water erosion in the study area.
The analysis of water erosion reveals that 43% of the total area of the Oued Ourika watershed is exposed to a high to very high risk of erosion in the Ourika region. These findings can assist regional and local authorities in reducing the risk of water erosion and implementing effective measures to prevent potential damages. The goal is to protect the communities and infrastructure located along the course of the Ourika.
Overall analysis of natural disasters, the accuracy of the results heavily depends on the availability and quality of data, which must encompass an adequate number of parameters.