Improve the accuracy and timing of yield forecasting

Commodity trading, Agriculture

VanderSat was approached by a leading international company in the animal feed industry to improve the accuracy and timing of their yield forecasting. After all, precise water availability data for fields across the globe can tell you all you need to know about crop yields.

Finding a competitive edge in today’s yield forecasting market seems like an almost impossible task. A whole range of companies and publicly funded organizations now have specialised products for this exact purpose, most providing an acceptable degree of accuracy.

However, there was still room for improvement in speeding up accurate yield forecasts. Beating the USDA by a couple of weeks is one thing, but outpacing the specialised yield forecasters at private companies by the same margin is quite a challenge. To achieve this, you need an extraordinary and globally available data source with a long time series and higher predictive value than traditional datasets.

VegetationOpticalDepth(VOD)

The historical Vegetation Optical Depth (VOD) dataset was used to train the yield forecast models for regions across the globe. VOD is a solid proxy for crop biomass. VanderSat improved the timing of these forecasts by up to a month while ensuring that accuracy remained on par with the competition.

Inanutshell

OPPORTUNITY: To obtain a crop yield forecast earlier in the season to optimize resource allocation across the globe.

CHALLENGE: The client had experience of commodity trading data terminals based on standard datasets, but needs a data source with more predictive value.

SOLUTION: VanderSat’s VOD data set provides very strong predictive value up to two months ahead of harvest.

KEY BENEFITS: The client now has a competitive edge of one month over traditional yield forecast methods.

Yield forecasting is a subject that has always attracted data scientists, statisticians and agronomists. Though each model is different, results often vary only slightly. One reason for this is the predictive power of the datasets being fed into the model. A model will always be limited by the quality and usefulness of the dataset it draws on.

With its VOD satellite product, VanderSat has been able to introduce a previously unavailable dataset into the yield forecasts and has drastically improved the timing of the model’s outcome: the one area in the forecasting business where there was still significant room for improvement.