How can statistical and artificial intelligence approaches predict piping erosion susceptibility?

Area of Science:

• Environmental Science
• Geosciences
• Soil Science

Background:

• Piping erosion poses a significant threat to environmental quality through soil loss.
• Current methods for regional-scale piping erosion susceptibility mapping are limited.
• Understanding the relationship between geo-environmental factors and piping erosion is crucial for effective land management.

Purpose of the Study:

• To develop and validate a novel modeling approach for mapping piping erosion susceptibility.
• To integrate machine learning algorithms with Unmanned Aerial Vehicle (UAV) imagery for enhanced spatial analysis.
• To identify key geo-environmental factors influencing piping erosion in a loess-covered hilly region.

Main Methods:

• Utilized three machine learning algorithms: Mixture Discriminant Analysis (MDA), Flexible Discriminant Analysis (FDA), and Support Vector Machine (SVM).
• Incorporated Unmanned Aerial Vehicle (UAV) imagery for high-resolution data acquisition.
• Employed 22 geo-environmental indices as predictors and 345 identified pipe locations as dependent variables.

Main Results:

• The developed models achieved high accuracy in mapping piping erosion susceptibility.
• Area Under the ROC Curve (AUC) values ranged from 90.32% to 92.45% for the tested algorithms.
• The study successfully generated detailed piping susceptibility maps for the study area.

Conclusions:

• The proposed machine learning-based approach, enhanced by UAV data, is highly effective for regional piping erosion susceptibility mapping.
• The findings provide a valuable tool for land managers to mitigate soil loss and environmental degradation.
• This research contributes to advancing the understanding and prediction of piping erosion in vulnerable landscapes.