Gully erosion prediction method from geoenvironmental data and supervised machine learning techniques.

Area of Science:

• Geosciences
• Environmental Science
• Machine Learning

Background:

• Predictive models are crucial for forecasting environmental disasters like gully erosion.
• Geoenvironmental data analysis is key to understanding erosion triggers.
• Machine learning offers advanced tools for developing accurate predictive models.

Purpose of the Study:

• To present a novel method for creating gully erosion predictive models.
• To evaluate the effectiveness of machine learning algorithms in geoscientific predictions.
• To compare the performance of different machine learning models for gully erosion forecasting.

Main Methods:

• Utilized geoenvironmental data for model development.
• Applied machine learning techniques to build predictive models.
• Compared four distinct machine learning algorithms for predictive performance.

Main Results:

• The developed method successfully produced models with high predictive ability.
• The random forest algorithm exhibited the highest accuracy and generated realistic erosion maps.
• The technique proved effective in a large-scale study area (40,000 km²).

Conclusions:

• The proposed method is effective for predicting gully erosion.
• The random forest algorithm is a powerful tool for geoscientific predictive modeling.
• This approach can be adapted for predicting other geological processes without programming knowledge.