With the Brazilian soybean crop now fully planted, the market has turned its attention to expected yields from the world’s largest soybean producer. So today, Gro Intelligence is excited to announce the launch of our in-season, district-level Brazilian soybean yield forecast model. The model’s yield predictions can be combined with Gro’s soybean area harvested forecasts or official CONAB area data to estimate Brazil’s final soybean production numbers.
The Brazilian soybean model adds an important new element to Gro’s portfolio of yield forecasting models. Together with our U.S. and Argentine soybean models, Gro is now able to forecast more than 80% of worldwide production in a given year, thereby providing users with a highly accurate, real-time window into global soybean supply. As of Jan. 9, our Brazilian model is forecasting yield of 3.20 tonnes per hectare, implying total Brazilian soybean production of 117.75 million tonnes, based on area estimates from CONAB.
For ground-truthing and training, Gro’s model used 17 years of historical yield data from IBGE, Brazil’s main government provider of data and information about the country. Over the last 10 years, Gro’s yield forecasts at the completion of the Brazilian harvest in March have diverged from the final IBGE ground-truthed value released in December by an average of 3.41%.
As with all of Gro’s models, our machine-learning-based Brazilian soybean model forecasts yields at the district level using a broad range of data sources on the Gro data platform. Features used in the model include NDVI, evapotranspiration anomalies, land surface temperature, soil moisture, potential evapotranspiration, soil properties, and latitude to name a few.
Daily updates from the Brazilian soybean model are exclusively available through the Gro API and the Gro web application. The output is accessible at the national, provincial, and district levels. We will be publishing a separate blog post on how we developed our Brazilian harvested area forecasts at a later date.