Predicting Crop Yields

Assistance The Security Of The National Food Supply Chain By Predicting Crop Yield Based On Satellite Images & Weather By Using Alternative Intelligence/Machine Learning

The exact yield of crops in the US is measured by the amount of goods brought to market.  It is hard to determine the impact of crop yield on the supply chain until it may be too late to make other adjustments.  We need to predict crop yield.  Existing approaches rely on survey data and other variables related to crop growth (such as weather and soil properties) to model crop yield.

Unfortunately, this is not an easy problem.  In the US, the size of the planting area is massive.  Weather and temperature conditions and patterns differ by region. Soil and topographical elements can be unique to each acre.  Farmers plant their crops at different times of the year, and change what they plant each season for crop rotation and based on market conditions. Optimal planting time depends on a complex variety of factors, including growing season, temperature, water, soil condition, and even availability of shared farm equipment.

To help predict the agriculture supply chain in relation to the COVID-19 pandemic and for decades to come, we propose to the Joint Artificial Intelligence Command a scalable and accurate method to predict crop yield before harvest in a given season based on satellite images and weather data.

Our deep learning approach has proven to predict crop yield with county-level (perhaps acre level) resolution several months before harvest, using only globally available factors.  The model will be made more accurate with additional extensive data with which the DoD and Intelligence Community has.

We believe our solution can potentially help inform planting decisions, set appropriate food reserve levels, identify low-yield regions, and improve risk management of crop-related derivatives, and help increase the security of the nations' food supply chain.

Academic Paper