Spatial prediction of soil depth using environmental covariates by quantile regression forest model.

Area of Science:

• Environmental Science
• Soil Science
• Geospatial Analysis

Background:

• Accurate soil depth prediction is crucial for land resource management, crop, nutrient, and ecosystem modeling.
• Understanding spatial soil depth distribution is vital for effective agricultural and environmental planning.

Purpose of the Study:

• To assess the spatial distribution of soil depth over 160,205 km² of Andhra Pradesh, India.
• To compare the predictive accuracy of quantile regression forest (QRF) with ordinary kriging for soil depth mapping.

Main Methods:

• Utilized 2854 soil datasets for calibration and validation (80:20 ratio).
• Employed 20 covariables including Landsat imagery, terrain datasets, and bioclimatic factors.
• Applied quantile regression forest (QRF) for spatial prediction of soil depth.

Main Results:

• Precipitation, multi-resolution index of valley bottom flatness (MrVBF), mean diurnal range, isothermality, and elevation were key predictors.
• QRF model achieved a R² of 42%, ME of -1.81 cm, and RMSE of 34 cm.
• QRF outperformed ordinary kriging (R² of 32%, ME of -0.14 cm, RMSE of 37 cm).

Conclusions:

• The QRF model demonstrated superior accuracy in predicting soil depth compared to ordinary kriging.
• Soil depth is spatially dynamic and significantly influenced by terrain and environmental covariates.
• Future improvements can be achieved by incorporating high-density bioclimatic and high-resolution terrain variables.