Gro Proves Highly Accurate in Predicting Brazil’s Sugarcane Sucrose Concentration

Talk to our our team about Gro's offering
Talk to our team

With the Brazilian sugarcane harvest wrapping up, Gro Intelligence is excited to announce that its predictive model of cane sugar concentration achieved remarkable accuracy throughout the harvest period. The model allows Gro users to gain insight into the key metric of total recoverable sucrose (TRS) days ahead of the twice-monthly industry reports’ release. It is exclusively available via the Gro API.

Earlier in 2019, Gro Intelligence began forecasting cane sugar concentrations in Brazil’s Sao Paulo state, the largest producer of Brazilian sugarcane. Most sugarcane is harvested from May to October, when rainfall is less frequent and the plant's sugar content is at its highest. In the 2019 season, Gro’s methodology predicted total recoverable sugar with a mean absolute error of 1 kg/metric tonne, less than 1% of the average actual TRS, for each of the last seven semi-monthly periods as reported by UNICA. Anyone trading sugar with our model would have had advance notice of the crop’s progress.

As shown in the table below, Gro’s yield forecast differed from the official report by less than 2% in all seven of the most recent data reporting periods and by less than 1% in four of the seven periods.

Gro’s model is particularly accurate during the July to September period. Backtesting the model over an 11-year period shows that our model can explain 71% of the variation in TRS during this time of year.

Weather conditions in Brazil’s center-south region, dominated by Sao Paulo state, are the primary determinant of Brazilian production, particularly precipitation and temperature during the harvest season. Although crop stress adversely impacts most other crop yields, moderate crop stress is actually beneficial for sugar production during the harvest period. Reduced precipitation limits the growth of the cane’s stem and leaves, but it does not unduly inhibit photosynthesis. This results in the sugarcane plant converting carbon dioxide and water into sucrose, which is stored in the stalk.

Gro’s model uses normalized difference vegetation index (NDVI) as a way to measure crop stress during the harvest. Combined with other geospatial signals in Gro’s data platform—which have been geographically harmonized to the center-south region—NDVI can be a strong predictor of total recoverable sucrose concentrations in Brazilian cane. Combining this yield with mill run rates determines how much sugar is produced.

The chart on the left shows actual total recoverable sugar (TRS) levels reported by UNICA for Brazil’s Sao Paulo state and those forecast by Gro’s model for the past 12 years. The chart on the right compares Gro’s model output with actual UNICA values during those years’ July to September periods.

UNICA provides reports of cane crushing activity, sugar and ethanol production, and total recoverable sugar (TRS) twice a month. During the main harvest period, the reports are an important factor in market forecasts and often impact the price of sugar in world markets. Using UNICA’s historic TRS data for Sao Paulo state, Gro was able to build its predictive model for TRS levels for the semimonthly reporting periods.

The model and the geospatial data series used to construct it are available on a global basis to all of our users exclusively via the Gro API. To construct a TRS yield model specific to your area of interest, or trial the API more generally, please reach out to

Get a demo of Gro
Talk to our enterprise sales team or walk through our platform