Título: Risk Analysis of Soil Erosion Using Remote Sensing, GIS, and Machine Learning Models in Imbabura Province, Ecuador

Autor(es): GARRIDO SANCHEZ JOSE FERNANDO, GRANDA GUDIÑO PEDRO DAVID

Fecha de publicación: 27-aug-2024

Resumen: Soil erosion occurs when natural or man-made agents tear away the top layer of the soil, making it extremely difficult to grow vegetation on the site. Wind and water (the two main causes of erosion) easily wash away soil if it is bare. Agriculture is probably the most crucial activity that accelerates soil erosion due to the amount of cultivated land and the number of agricultural practices that disturb the soil. Soil erosion poses a significant environmental challenge, adversely affecting soil fertility through nutrient loss and contributing to sedimentation in aquatic environments. Besides this, water-induced soil erosion stands out as the most severe form of land degradation observed in diverse regions, both locally and globally. This study analyzed the risk of soil erosion in the province of Imbabura, north part of Ecuador, using Machine Learning (ML) models, applying the Revised Universal Soil Loss Equation (RUSLE) Model involves utilizing remote sensing techniques with sufficient resolution spatial–temporal and spectral, thus identifying vulnerable areas according to factors that play an essential role in erosion processes, namely topography (LS), land use or cover management (C), rainfall erosivity (R), erodibility (K) and anti-erosive agricultural practices (P). The map indicating erosion probability zones was generated using Geographic Information Systems (GIS) tools, that most of the study area is located within the zone of low probability of erosion (64.73%), and a small section is located within the zone of high likelihood of erosion (4.17%). These discoveries can provide valuable insights into reducing and preventing soil erosion in Imbabura. So, these findings provide robust evidence for the efficiency of employing machine learning in soil erosion prediction. LightGBM, which utilizes decision tree algorithms, achieved a Classification Accuracy Rate (CAR) of over 75.7% and an area under the receiver operating characteristic (AUC) of more than 0.89, underscoring its effectiveness.

Palabras clave: Soil erosion · Remote sensing · GIS · Soil loss · ML · Land Use Land Cover (LU/LC)

DOI: https://doi.org/10.1007/s42979-024-03150-3

ISSN: 2661-8907

Tipo publicación: Artículo