Gupta et al. (2026) Mapping Swiss soil bulk density at 30 m Resolution: Insights from Machine Learning, environmental Covariates, and national data

Identification

• Journal: International Journal of Applied Earth Observation and Geoinformation
• Year: 2026
• Date: 2026-02-01
• Authors: Surya Gupta, Simon Scheper, Christine Alewell
• DOI: 10.1016/j.jag.2026.105112

Research Groups

• Department of Environmental Sciences, University of Basel, Basel, Switzerland
• Dr. Simon Scheper – Research, Teaching, Daehre, Germany

Short Summary

This study generated high-resolution (30 m) soil bulk density maps for Switzerland at four standard depths (0 m, 0.3 m, 0.6 m, 1.0 m) using a Quantile Random Forest model and national soil data, revealing that croplands and the Central Plateau exhibit the highest bulk density, which generally increases with depth.

Objective

• To develop high-resolution (30 m) soil bulk density maps for Switzerland at four standard soil depths (0 m, 0.3 m, 0.6 m, and 1.0 m) by integrating the national Swiss Soil Information System (NABODAT) dataset and environmental covariates using a Quantile Random Forest algorithm.

Study Configuration

• Spatial Scale: Entire territory of Switzerland (41,285 km²) at 30 m spatial resolution, with maps generated for four soil depths (0 m, 0.3 m, 0.6 m, 1.0 m).
• Temporal Scale: NABODAT soil data collected over several decades; environmental covariates from various periods (e.g., LANDSAT NDVI 2005-2020, CHELSA climate 1991-2020, surface reflectance 2018-2022).

Methodology and Data

• Models used: Quantile Random Forest (QRF) algorithm, implemented using the ranger package in R.
• Data sources:
  • Soil Data: National Swiss Soil Information System (NABODAT) version 6 dataset (3,089 samples from 773 locations).
  • Environmental Covariates:
    • Vegetation: Normalized Difference Vegetation Index (NDVI) from LANDSAT (2005-2020).
    • Climate: CHELSA dataset (1991-2020) resampled to 30 m.
    • Topography: swissALTI3D Digital Elevation Model (2018) resampled to 30 m (metrics include elevation, slope, flow accumulation, terrain ruggedness, curvature).
    • Surface Reflectance: Averages and standard deviations from 2018-2022.
    • Parent Material: Lithological map and soil/sediment deposit thickness data.
    • Geographic Coordinates: Latitude and longitude.
    • Soil Depth: Included as a covariate.
  • External Validation Data: Independent soil samples from Mayerhofer et al. (2021) (233 samples at 0-0.2 m depth).

Main Results

• Model Performance (Internal Validation): The Quantile Random Forest model achieved a coefficient of determination (R²) of 0.42, a concordance correlation coefficient (CCC) of 0.57, a bias of -1.5 kg/m³, and a root mean square error (RMSE) of 280 kg/m³. The 90% prediction interval coverage probability (PICP) was 91.3%.
• Model Performance (External Validation): Using an independent dataset, the model yielded an R² of 0.36, a CCC of 0.39, a bias of 170 kg/m³, and an RMSE of 290 kg/m³. The PICP was 87.3%.
• Spatial Patterns of Bulk Density:
  • Land Use: Croplands exhibited the highest mean bulk density (1250 kg/m³), followed by grasslands (1050 kg/m³) and forestlands (970 kg/m³).
  • Regions: The Central Plateau (average 1160 kg/m³) and Jura Mountains (average 1000 kg/m³) showed higher bulk density compared to the Alpine regions (average 730-810 kg/m³).
  • Depth: Bulk density consistently increased with soil depth.
• Key Predictors: Soil depth was the most influential covariate, followed by elevation, temperature, slope, and latitude. Surface reflectance showed low importance.
• Uncertainty: Higher prediction uncertainty was observed in the Central Plateau, Jura Mountains, and complex alpine terrains.

Contributions

• Generated the first high-resolution (30 m) national soil bulk density maps for Switzerland at multiple depths (0 m, 0.3 m, 0.6 m, 1.0 m), addressing a significant data gap (the "blank spot") and improving upon existing coarser or land-use-limited maps.
• Provided comprehensive coverage across all major land cover types in Switzerland.
• Rigorously validated the maps using both internal cross-validation and an independent external dataset.
• Offered valuable insights for national-scale applications such as soil carbon stock estimation and compaction assessment.

Funding

• European Union Horizon Europe research and innovation program (Grant 101086179 - AI4SoilHealth)
• Swiss State Secretariat for Education, Research and Innovation (SERI)

Citation

@article{Gupta2026Mapping,
  author = {Gupta, Surya and Scheper, Simon and Alewell, Christine},
  title = {Mapping Swiss soil bulk density at 30 m Resolution: Insights from Machine Learning, environmental Covariates, and national data},
  journal = {International Journal of Applied Earth Observation and Geoinformation},
  year = {2026},
  doi = {10.1016/j.jag.2026.105112},
  url = {https://doi.org/10.1016/j.jag.2026.105112}
}