Soil Organic Carbon (AI model)*

Diamond

tonnes of carbon

10m

Annual

Plot

2018 - 2024

2025-2045

All

Spatialise is a remote-soil carbon monitoring platform that combines satellite imagery, artificial intelligence (AI) and in-situ soil sampling to train and estimate soil carbon. The scientific basis lies in digital soil mapping and pedometrics: using a large collection of ground truth soil samples to train machine-learning models that infer soil property values across space from covariates (such as topography, climate, land cover, remote sensing bands). The rationale is that physical soil variation is governed by factors including parent material, relief, climate, organisms, time and vegetation conditions.

Spatialise’s measurement protocol blends field sampling with remote sensing and model-prediction workflows. They combine “350,000+ soil samples” with “eight satellite systems” and train neural networks to produce predictions at ~10 × 10 m resolution. The protocol therefore implicitly involves: (1) collection of field soil samples across different regions and soil types; (2) geospatial referencing of those samples; (3) assembling remote-sensing covariate layers (terrain, multispectral imagery, time series) for each sample location; (4) training machine-learning models; and (5) applying the trained model to polygons or raster grids for prediction of soil properties for a given year and depth.

Spatialise’s verification framework involves model calibration and validation steps to ensure reliability of the predictions. Their public statements indicate model accuracies of 85%+ across geo-ecological zones. For users adopting Spatialise for outcome-based analysis (e.g., SOC change monitoring or carbon credits) this verification step is critical: ensuring results are defensible for monitoring-reporting-verification (MRV) applications and that predictions align with the field reality and audit requirements.