← Back

ATLAS-SOC Core v1.0

Our ATLAS Soil Organic Carbon (SOC) Measurements provide remote estimations of SOC between 10m and 500m resolution. Gain high-quality insights without physical soil sampling for efficient and scalable land analysis.

To access high-resolution data for your area(s) of interest, contact our team.

High-Level Description

ATLAS-SOC Core v1.0-demo represents a significant advancement in Perennial Climate Inc.'s digital soil mapping-based soil organic carbon (SOC) quantification framework. Building upon the foundations of earlier versions (see Atlas-SOC v0.1), v1.0-demo introduces enhanced capabilities for comprehensive SOC assessment, uncertainty quantification, and broader geographic scalability. Designed to predict both SOC % by mass and bulk density, the model’s outputs can be directly combined to yield SOC stock (t C/ha), delivering greater precision and versatility for soil carbon quantification across diverse agricultural landscapes.

This version pioneers biogeochemically-informed digital soil mapping, integrating deeper ecosystem dynamics into SOC predictions. It leverages advanced machine learning techniques, high-resolution remote sensing data, and an expansive soil sample archive to provide accurate, scalable, and dynamic SOC assessments. Enhanced temporal stability and confidence in change detection enable more reliable monitoring of carbon stock changes over time, addressing critical needs in carbon markets and sustainable land management.

The Core class of models is designed for general-purpose SOC quantification without requiring new soil sample data, making it ideal for large-scale analyses, such as policy development, identification of degraded lands, carbon accounting at national or regional scales, and baseline assessments. With over 350,000 soil samples integrated, the Core class of models provides reliable SOC stock estimates while balancing spatial coverage and predictive performance.

Unlike ATLAS-SOC Fine-Tuned models—optimized for specific farms or carbon projects using localized soil data—Core v1.0 provides an "out-of-the-box" solution with wide applicability but without site-specific calibration.

Technical Specifications

• Release Year: 2024
• Native Resolution: 10-30m
• Spatial Extent: Expanded global applicability, including Continental US (CONUS), LATAM, Australia, and Europe

• Domain: Agricultural croplands (grasslands and rangelands available in a different model version)

• Depth of Prediction: 0–30 cm (primary) and up to 1 m (extended capabilities)

• Units

  • SOC % by mass
  • Bulk density (g/cm³)
  • SOC stock (t C/ha)
  • SOC stock standard deviation (t C/ha)
  • SOC stock change (t C/ha/yr)

• Inputs:

  • Remote Sensing-Based Vegetation and Tillage Indices
  • Vegetation, soil and residues seasonal dynamics
  • Vegetation, soil, and residues inter-annual dynamics
  • Weather and climate variables, long-term and short-term
  • Advanced biogeochemical predictors (e.g., NPP, microbial respiration, carbon use efficiency)

• Temporal Range: 2014- Present

• Temporal Frequency: Annual

Algorithm Theoretical Basis

ATLAS-SOC Core v1.0-demo integrates a hybrid machine learning framework that combines ensemble-based methods with multi-temporal data integration techniques to provide robust and dynamic SOC predictions. The core model leverages XGBoost for predictive modeling, supplemented with Quantile Regression Forests (QRF) methods for comprehensive uncertainty quantification.

1. Multi-Temporal Digital Soil Mapping (DSM):

The model capitalizes on the temporal dimension of environmental variables to capture seasonal and interannual variability in SOC dynamics. Multi-temporal DSM techniques allow the integration of time-series data from diverse remote sensing platforms (e.g., Landsat, MODIS, Sentinel) to model the biophysical processes that drive SOC changes. This includes the use of various forms of regression to model periodic vegetation dynamics and anomalies in land surface temperature, enabling better detection of SOC stock fluctuations over time.

2. Ensemble Machine Learning Framework:

• XGBoost: Serves as the primary predictive model due to its robustness, ability to handle non-linear interactions, and efficiency with large, complex datasets. It is particularly effective in managing the high-dimensional, multi-temporal feature space generated by remote sensing data.

• Quantile Regression Forests (QRF): Provides distributional estimates of SOC predictions, enabling robust uncertainty quantification by modeling prediction intervals. This method captures the variability in SOC estimates, especially important in heterogeneous landscapes.

3. Temporal Prediction and Change Detection:

The model employs time-series analysis techniques to create predictors indicative of SOC stock changes. By integrating short- and long-term ecosystem dynamics, ATLAS-SOC v1.0-demo achieves increased stability in the magnitude and direction of SOC changes. This allows for the detection of gradual carbon accumulation or loss, as well as abrupt changes due to land-use shifts or management practices.

4. Biogeochemically-Informed Modeling:

The inclusion of biogeochemical indicators—such as net primary productivity (NPP), microbial respiration rates, and water availability—enables the model to account for the processes that drive carbon cycling in soils. This approach integrates ecosystem function with soil science, providing a deeper understanding of the factors influencing SOC stocks and enhancing the model’s predictive power.

5. Uncertainty Aggregation and Spatial Autocorrelation:

Advanced geostatistical methods are employed to aggregate pixel-level uncertainties to larger spatial units (e.g., fields, projects) while accounting for spatial autocorrelation. Techniques such as variogram analysis and spatial bootstrapping are used to adjust uncertainty estimates, ensuring that aggregated SOC stock and change estimates maintain statistical rigor across varying spatial scales.

Model Assumptions & Constraints

1. Uncertainty Quantification: Fully implemented using Quantile Regression Forests, enabling reliable aggregation from pixel to project scale.

2. Bulk Density Prediction: Allows for direct SOC stock calculation without external bulk density datasets.

3. Temporal Sensitivity: Improved modeling of SOC change over time, validated with longitudinal data from diverse geographies.

4. Geographic Scalability: Optimized for global application, including the Americas, Australia, and Europe.

5. Remote Sensing Dependency: Accuracy varies with data quality and temporal resolution; high revisit rates improve change detection.

Known Issues

• Limited Legacy Data Integration: In regions without extensive proprietary sample archives, prediction confidence may be lower.

• Computational Intensity: Full-scale uncertainty quantification requires significant computational resources, especially for large areas.

• Out-of-sample Uncertainties: For areas in which there was limited training data, estimated uncertainties tend to be smaller than expected after aggregation.

What’s New in v1.0?

• Uncertainty Quantification and Aggregation: Advanced statistical methods for precise SOC stock and change assessments.

• Temporal Prediction Enhancements: Improved stability and confidence in SOC change magnitude and direction.

• Bulk Density Modeling: Direct prediction of bulk density for SOC stock calculation.

• Expanded Geographic Coverage: Now applicable to LATAM, Australia, and Europe.

• Biogeochemically-Informed DSM: Integration of short- and long-term ecosystem dynamics into SOC prediction.

Data Acknowledgements & Partners

• PRISM – Short-term and long-term weather data

  • The PRISM Climate Group at Oregon State University develops high-resolution spatial climate datasets for the United States, incorporating data from various monitoring networks.

• WorldClim – High-resolution global climate data

  • WorldClim offers high-resolution global climate data, including historical and future temperature and precipitation layers, widely used in ecological modeling and climate impact studies.

• ECMWF – Global weather and climate data

  • The European Centre for Medium-Range Weather Forecasts (ECMWF) provides global weather forecasts, reanalysis datasets (like ERA5), and climate data used in research and operational meteorology. Some datasets are openly accessible, while others require licensing.

• SoilGrids v2.0/ SoilGrids+ – Global soil texture, pH, and bulk density datasets

  • SoilGrids is a global digital soil mapping system developed by ISRIC – World Soil Information. It utilizes machine learning to produce maps of soil properties, including texture, pH, and bulk density, at a 250 m resolution.

• ISRIC - Soil samples (WOSIS)

  • The World Soil Information Service (WoSIS) by ISRIC, compiles standardized soil profile data globally, supporting various soil mapping and modeling applications.

• LUCAS - Soil samples

  • The LUCAS topsoils dataset, made available by the European Soil Data Centre, is a continent-wide effort to monitor soil properties over time, supporting various environmental and modeling applications.

• RaCA – Soil samples

  • The Rapid Carbon Assessment (RaCA), led by the USDA NRCS, provides nationwide soil organic carbon (SOC) data across diverse U.S. land uses, supporting carbon stock assessments and land management. The dataset is publicly available and widely used in soil science and carbon markets.

Key References:

• Fu et al., 2024. Accurate Quantification of 0–30 cm Soil Organic Carbon in Croplands over the Continental United States Using Machine Learning. Remote Sensing, 16, 2217.

• Kellner et al., 2025. Digital soil mapping in support of voluntary carbon market programs in agricultural land. PLOS ONE. In review.

• Wadoux & Heuvelink, 2023. Uncertainty of spatial averages and totals of natural resource maps. Geoderma

Accessing Data

Contact our team to request high-resolution data for your area(s) of interest.