Crop yield prediction

Modern remote sensing techniques together with random sampling provide enough data for a pre-trained computer model to predict crop yield at global as well as at regional scale.

Crop monitoring and yield prediction is an essential task for agricultural economics. Crop production is very complex process, influenced by a large number of parameters. This makes yield prediction an ideal object for machine-learning techniques. Computer solutions show impressive results in finding complex dependencies between attributes of different types, combining geospatial images, weather, soil, sampling data in one model. BitRefine group works on a number of approaches that depend on available input data.

For each crop species our prediction model uses 10 years’ worth of historical data together with currently monitored attributes:

• Geospatial images
• Solar radiation
• Rainfall or irrigation volumes
• Temperature monitoring
• Soil profile
• Sampling data

Our main approach to accurately yielding prediction at regional scale includes four base steps. First, we build and train a machine-learning model based on historical data. Then we include observed attributes, which allow the model to calculate relative yield indicators. Next with precise sampling information added, the model is able to accurately predict according to the sample data collection. Finally, the model combines relative yield indicators across a whole field or a number of fields together with precise calculation from sampling points and provides the farmer with accurate numbers across all areas of interest.