Pedotransfer Functions for Field Capacity, Permanent Wilting Point, and Available Water Capacity Based on Random Forest Models for Routine Soil Health Analysis

Summary

This study developed machine learning-based pedotransfer functions to predict three key soil water-holding properties—field capacity, permanent wilting point, and available water capacity—from readily available soil measurements on a dataset of 7,232 samples. Random forest models substantially outperformed traditional linear regression, with full models reducing prediction error by 12.8–16.3% for these properties. The functions provide a cost-effective means of assessing soil physical health in comprehensive soil health evaluations, with novel predictor variables such as permanganate-oxidisable carbon and extractable magnesium improving predictions.

Regional applicability

The study was conducted in the United States using North American soil samples; direct transferability to United Kingdom soils would require validation against UK soil types and climatic conditions. However, the methodology and random forest framework could be adapted for UK soil health assessments if recalibrated with British soil datasets.

Key measures

Root mean square error (RMSE) reduction; field capacity (FC); permanent wilting point (PWP); available water capacity (AWC); predictor variables include texture, soil organic matter, permanganate-oxidisable carbon, soil respiration, wet aggregate stability, and extractable potassium, magnesium, iron, and manganese

Outcomes reported

The study developed and validated random forest pedotransfer functions to predict field capacity, permanent wilting point, and available water capacity from routine soil measurements. Performance was evaluated against traditional linear regression models using an independent validation dataset of 1,406 samples.

Topic tags