Model-based decision support for farmers

Soil conditions and environmental context are a crucial factor determining farming success. Remote monitoring kits like TerraPrima’s Ladybird can read parameters like soil moisture and temperature alongside external factors like air temperature and relative humidity, light, weather,etc., and provide them over the internet. This empowers the farmer to make informed farming decisions, e.g. on the timing of irrigation.
However, mere monitoring of those values still puts the burden on the farmer to interpret the readings correctly and schedule actions. The readings need to be put in context to be interpreted correctly. For example, critical factors like moisture depletion rate and nutrient retention depend on the crop and specific variety, the type of soil used (organic, mineral, coconut husk, etc), among other factors.
An experienced farmer knows how to interpret these values for the specific conditions they are familiar with. However, when experimenting with e.g. a new type of greenhouse, soil, or irrigation system, or even a new type of crop, they may lack the experience required to make the best decisions. The high level of experience required also complicates delegation of these decisions to farm employees.
Our goal in this project is to support the farmer’s decision making by leveraging artificial intelligence and machine learning. We will develop a system that learns, for a specific farm and crop, to predict the timing of future farming events, e.g. irrigation, depending on conditions like current temperature, humidity, soil moisture, and the weather. Initially we will focus on irrigation, but the modeling technique is applicable also to other events like fertilisation timing or even specific crop treatments. The system will therefore be able to provide model-based support for farming decisions.
Moreover, models learnt on one site could be applied to another one with similar conditions. Provided with the trained model, a farmer could experiment with new soil types that have successfully been used elsewhere, while minimising the risk associated with the lack of experience, since the system will be able to suggest safe decisions that have worked previously on other farms. Ultimately a decision support system may even help a farmer automate parts or even all of the operation.
This project combines the disciplines of agriculture and machine learning to deliver an innovative step that no discipline could achieve in itself.