Soil Nutrient Levels Prediction
Challenges
- A major agribusiness with large-scale farming operations set out to reduce fertilizer costs and optimize usage.
- It wanted a better soil nutrient balance for higher yields, and to replace manual, vendor-dependent nutrient forecasting.
Approach
An AI model forecasts soil nutrient levels — nitrogen, phosphorus and potassium — in real time to roughly 15% error, combining historical soil samples, weather data and agronomic inputs.
Outlier detection cleans inconsistent measurements, and the outputs are delivered as prescription maps for optimal fertilizer dosage per hectare.
Key features
- Automated cleaning of nutrient sampling data with AI outlier removal
- Real-time forecasts of N, P and K at one-hectare granularity
- Integration with agronomic decision systems to generate prescription maps
- Higher precision than third-party black-box tools
Impact
+1.6–3.1%
EBITDA uplift from precise real-time nutrient forecasting and less manual sampling
~15%
Error of the top-performing model forecasting nitrogen, phosphorus and potassium
100 × 100 m
Area for which nutrient levels are predicted from limited soil samples