Tested by the University of Córdoba.
An EU-backed independent validation of SoilSense soil moisture sensors against gravimetric reference, on contrasting clay and sandy soils. Córdoba, Spain. March 2026.
Within spec. On heavy clay.
2.77%
RMSE — Vertisol global model, after outlier removal
Below the manufacturer's ±3% nominal accuracy. R² 0.91 against gravimetric reference, six weeks of evaporation.
Best individual sensor:
- 1.69%
- MAE
- 1.70%
- RMSE
- 0.99
- R²
Eight pots. Two soils. The data behind the numbers.
AgrifoodTEF is one of the European Testing & Experimentation Facilities established under the EU's Digital Europe Programme — set up specifically to validate digital and AI technologies in real agricultural conditions. Its Spanish node is hosted by the University of Córdoba, a Mediterranean agronomy institution with deep standing in soil physics and irrigation.
At UCO's Rabanales Experimental Farm in Córdoba, eight pots were prepared. Four held a Calcic Vertisol — heavy clay (66% clay, 18% sand), the kind of soil that cracks as it dries and seals again as it wets. Four held a Stagnic Luvisol — coarse-textured surface (24% clay, 66% sand). Each pot was brought to saturation and weighed gravimetrically as it dried, while a SoilSense sensor at 12 cm depth recorded volumetric water content in parallel. Error was measured with MAE, RMSE, and linear regression; outliers identified through robust standardized residuals on the global dataset.
Vertisol — heavy clay
Three of four sensors within or close to the ±3% spec.
| Sensor | MAE (%) | RMSE (%) | R² |
|---|---|---|---|
| 1 Verle | 3.64 | 3.72 | 0.86 |
| 2 Verle | 2.64 | 2.88 | 0.83 |
| 1 Jasmin | 5.15 | 5.46 | 0.18 |
| 2 Jasmin | 1.69 | 1.70 | 0.99 |
Luvisol — sandy
Higher variability on coarse-textured soil.
| Sensor | MAE (%) | RMSE (%) | R² |
|---|---|---|---|
| 1 Verle | 5.61 | 5.72 | 0.75 |
| 2 Verle | 3.91 | 4.06 | 0.53 |
| 1 Jasmin | 4.33 | 4.40 | 0.75 |
| 2 Jasmin | 2.75 | 2.78 | 0.65 |
On coarse soils, more variability — and the honest reason why.
The Luvisol global model came in at RMSE 3.95% after outlier removal, slightly above the ±3% nominal spec. Sandier soils retain water unevenly and the small pot volume amplifies that. The system tracks trend faithfully; if your operation relies on calibrated absolute-value thresholds in coarse soils, plan for a touch more variance than on clay.
Cite this work
Universidad de Córdoba / AgrifoodTEF (2026). Validation Report: SoilSense — UCO. Service ID S00383. Córdoba, Spain.
Want SoilSense in your soil?
Book a free 30-minute call. We'll size the system to your crops and the soils you actually grow on — and walk through how the dashboard helps you decide when to irrigate.