Bellat et al. (2025) Soil information and soil property maps for the Kurdistan region, Dohuk governorate (Iraq)

Identification

• Journal: Earth system science data
• Year: 2025
• Date: 2025-09-15
• Authors: Mathias Bellat, Mjahid Zebari, Benjamin Glissman, Tobias Rentschler, Paola Sconzo, Nafiseh Kakhani, Ruhollah Taghizadeh‐Mehrjardi, Pegah Kohsravani, Bekas Brifany, Peter Pfälzner, Thomas Scholten
• DOI: 10.5194/essd-18-2507-2026

Research Groups

• CRC 1070 “ResourceCultures”, University of Tübingen, Tübingen, Germany
• Department of Geosciences, Working group of Soil Science and Geomorphology, University of Tübingen, Tübingen, Germany
• Ludwig-Maximilians-Universität München, München, Germany
• Nawroz University, Duhok, Iraq
• Institute for Ancient Near Eastern Studies (IANES), University of Tübingen, Tübingen, Germany
• Digital Humanities Center, University of Tübingen, Tübingen, Germany
• Department of Culture and Society, University of Palermo, Palermo, Italy
• Faculty of Agriculture and Natural resources, Ardakan University, Ardakan, Iran
• College of Agriculture, Shiraz University, Shiraz, Iran
• Dohuk Directorate of Antiquities and Heritages, Dohuk, Iraq
• DFG Cluster of Excellence “Machine Learning: New Perspectives for Science”, University of Tübingen, Tübingen, Germany

Short Summary

This study provides the first detailed, high-resolution (30 m) soil property and depth maps for the Dohuk governorate, Kurdistan Region of Iraq, outperforming global models and offering crucial data for local land management in a data-poor, arid/semi-arid environment.

Objective

• To develop the first detailed, high-resolution digital soil property and soil depth maps for the Dohuk governorate, Kurdistan Region of Iraq, across five depth increments.
• To provide an updated soil classification based on WRB taxonomy for the region.
• To compare the performance of the developed regional models with global soil products like SoilGrids 2.0.

Study Configuration

• Spatial Scale: Dohuk governorate, Kurdistan region of Iraq (Simele and Zakho districts), covering a total area of 2280 km². Maps produced at 30 m pixel resolution.
• Temporal Scale: Soil sampling campaigns conducted between 2022 and 2023. Legacy samples from 2017–2018 were used for MIR calibration.

Methodology and Data

• Models used:
  • Cubist model (regression-based machine learning) for predicting soil properties from mid-infrared (MIR) spectra.
  • Quantile Random Forest (QRF) for digital soil mapping of soil properties and soil depth.
  • Revised Universal Soil Loss Equation (RUSLE) for erosion risk modeling (used as a covariate).
  • Boruta package for feature selection.
• Data sources:
  • Soil Samples: 532 soil samples from 122 sites collected at five depth increments (0–10, 10–30, 30–50, 50–70, and 70–100 cm). A subset of 108 samples underwent laboratory analysis.
  • Spectroscopy Data: Mid-infrared (MIR) spectra measured for 561 soil samples.
  • Laboratory Measurements: pH, calcium carbonate (CaCO₃), total nitrogen (Nₜ), total carbon (Cₜ), organic carbon (OC), electrical conductivity (EC), and texture (sand, silt, clay) for 108 samples. Mean weight diameter (MWD) was also estimated.
  • Remote Sensing Data: Landsat 8 SWIR, EVI, SAVI; Sentinel 2 SWIR; MODIS Vegetation Index (MOD13Q1), MODIS/Terra Land Surface Temperature/Emissivity (MOD11A2); Copernicus DEM (GLO-30).
  • Climate Data: ERA5-Land monthly averaged data (potential evapotranspiration, wind speed, solar radiation), WorldClim 2 (precipitation, temperature).
  • Geological Data: Compiled geological maps of the region.
  • Land Use/Cover: Copernicus (2019) for RUSLE C factor.
  • Existing Soil Data: SoilGrids 2.0 for comparative analysis.

Main Results

• Soil Property Predictions (MIR-based): Calcium carbonate, total carbon, sand, and clay showed strong predictive performance (high R², low RMSE). pH, total nitrogen, organic carbon, and mean weight diameter showed moderate performance. Electrical conductivity and silt predictions showed weak performance.
• Digital Soil Mapping (DSM): Fifty prediction maps for ten soil properties across five depth increments, and a soil depth map, were produced at 30 m resolution. Key predictors included Landsat 8 SWIR bands, EVI, SAVI, Sentinel-2 SWIR bands, potential evapotranspiration, and solar radiation.
• Spatial Patterns: Soil properties exhibited clear contrasts between the flat Selevani and Zakho plains (characterized by sedimentation) and the mountainous Bekhair anticline and Little Khabur Valley (dominated by erosional processes). Finer soil fractions (clay and silt) prevailed in the plains, while coarser textures (sand) were found in more eroded areas.
• Soil Depth Mapping: The soil depth model explained 39% of the observed variability with a root mean square error (RMSE) of 30.76 cm. Shallow soils were prevalent in mountainous regions, badlands, and along active wadis, while deeper soils were mapped in the Zakho Plain and Little Khabur Valley plateaus.
• Comparison with SoilGrids 2.0: Our models outperformed SoilGrids 2.0 across all evaluation metrics for pH, total nitrogen, organic carbon, sand, silt, and clay at all depth intervals, offering higher resolution and accuracy. Our models predicted significantly higher organic carbon (~1000%), total nitrogen (~300%), and sand (~25%) values, and slightly lower clay (~15%) compared to SoilGrids 2.0.

Contributions

• Generated the first detailed, high-resolution (30 m) digital soil property and soil depth maps for the Dohuk governorate, Kurdistan Region of Iraq, addressing a significant data gap in a data-poor region.
• Developed a comprehensive and reproducible workflow for regional digital soil mapping, integrating conditioned Latin hypercube sampling, mid-infrared spectroscopy, and machine learning models (Cubist, Quantile Random Forest).
• Provided an updated soil classification map aligned with modern World Reference Base (WRB) standards, offering improved spatial detail over previous exploratory maps.
• Demonstrated superior performance of regional models compared to global products (SoilGrids 2.0) in terms of resolution and accuracy, emphasizing the importance of local calibration for operational land management.
• Created a FAIR-compliant (Findable, Accessible, Interoperable, Reusable) soil dataset, contributing to reducing geographical imbalances in soil science data and promoting transparency and reproducibility.
• Offers substantial insights for soil knowledge in aridic and semi-arid areas, with a transferable workflow applicable to regions facing similar environmental and data constraints.

Funding

• Deutsche Forshungemeischaft (DFG) Collaborative Research Center (CRC) 1070 “ResourceCultures” (grant no. 215859406).
• Open Access Publication Fund of the University of Tübingen.

Citation

@article{Bellat2025Soil,
  author = {Bellat, Mathias and Zebari, Mjahid and Glissman, Benjamin and Rentschler, Tobias and Sconzo, Paola and Kakhani, Nafiseh and Taghizadeh‐Mehrjardi, Ruhollah and Kohsravani, Pegah and Brifany, Bekas and Pfälzner, Peter and Scholten, Thomas},
  title = {Soil information and soil property maps for the Kurdistan region, Dohuk governorate (Iraq)},
  journal = {Earth system science data},
  year = {2025},
  doi = {10.5194/essd-18-2507-2026},
  url = {https://doi.org/10.5194/essd-18-2507-2026}
}