[Assessing soil pH in Anhui Province based on different features mining methods combined with generalized boosted...

Area of Science:

• Digital Soil Mapping
• Machine Learning in Geosciences
• Environmental Modeling

Background:

• Accurate soil pH prediction is crucial for agricultural and environmental management.
• Traditional digital soil mapping methods often face challenges with high-dimensional environmental data.
• Feature selection and mining are essential for improving model performance and interpretability.

Purpose of the Study:

• To evaluate the effectiveness of feature mining methods combined with generalized boosted regression (GBR) and random forest (RF) models for digital soil mapping.
• To compare the performance of GBR and RF models in predicting soil pH using selected environmental covariates.
• To identify the spatial distribution patterns of soil pH in Anhui Province.

Main Methods:

• Environmental covariates were pre-processed using recursive feature elimination and filtering methods for feature selection.
• Soil pH prediction models were established using Generalized Boosted Regression (GBR) and Random Forest (RF) algorithms.
• Model accuracy was assessed using both training and validation datasets, and spatial mapping was performed.

Main Results:

• Both feature mining methods significantly improved the accuracy and reduced dimensionality for both GBR and RF models.
• GBR models demonstrated higher accuracy and interpretability on the training set compared to RF models.
• Spatial mapping revealed a distinct soil pH gradient in Anhui Province, characterized by 'south acid and north alkali'.

Conclusions:

• Feature mining is a valuable approach for enhancing digital soil mapping accuracy and efficiency.
• GBR models offer strong performance, particularly in training accuracy, but require careful parameter tuning.
• The identified spatial patterns of soil pH provide essential information for regional land management and agricultural practices.