About Gro Forecasts

Data providers customarily offer only trade estimates, or market polls, that cover a small portion of the global supply and demand balance. These reports are impossible to confirm because the industry analysts do not provide their individual methodologies.

Gro has developed a range of methods for deriving crop forecasts that are entirely reproducible using either the Gro web application or the Gro API. These methods enabled Gro to fill the void left when the US government shutdown prevented the release of the USDA’s WASDE report in January 2019.

Gro Intelligence was able to quickly mobilize resources from Gro to provide a comprehensive outlook for global agriculture. Satellite-generated knowledge offered insights into crop conditions, and various non-US data sets available in Gro were used to better inform cumulative year-over-year comparisons. Using linear regressions from Gro’s vast amounts of data allowed us to generate quantitative forecasts for thousands of crop-country pairs.

Source data might be sparse or delayed for reasons other than a government shutdown, and Gro has developed ways of accurately predicting future yield with minimal data input. Using just NDVI—a measure of plant biomass and therefore crop health—and historical yield values, Gro API users can run a Python notebook to generate forecasts.

The notebook provides a very basic model that uses crop-production-weighted NDVI to forecast current season yield. The code computes the "weight" of each district, based on a full history of the area’s production quantity, and then applies the weights to the NDVI values for the districts.