Soil Property Predictions for the Upper Colorado River Basin Using Deep Learning

A 2026 data release accompanying a manuscript submitted to Geoderma. It contains code, model outputs, and spatial data for predicting 8 soil properties at 7 depth intervals across the Upper Colorado River Basin. The release includes environmental covariate rasters, prediction maps, and scripts for TensorFlow/Keras models developed in R and Python.

Use Cases

• Benchmarking convolutional neural networks, Vision Transformers, and Masked Autoencoders against random forests for soil property prediction.
• Creating 3D soil property maps (sand, silt, clay, CaCO3, EC, SAR, pH, SOC) at standard depth intervals.
• Analyzing spatial patterns of soil systems using the provided GeoTIFF prediction maps and environmental covariate layers.
• Conducting SHAP variable importance analysis on environmental covariates for soil modeling.

Strengths

• Includes predictions for 8 soil properties across 7 depth intervals, totaling 56 prediction targets.
• Provides full reproducibility with source code in R and Python, model outputs, and training histories.
• Contains spatial data products including GeoTIFF prediction maps and environmental covariate raster layers.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset size is listed as 0.0 B, indicating limited scope or a metadata placeholder.

Provenance

Source

figshare, author Travis Nauman.

Collection Method

Data and code generated for a comparative study of deep learning architectures for soil mapping.

Freshness

Last updated 2026-04-22 06:55:11; freshness should be verified.

Geography

Upper Colorado River Basin.