Brazil Sao Paulo Sugarcane Crop Model

Summary & Output

This model enables users to project total recoverable sugar (TRS) levels in the state of Sao Paulo during the crushing season. By drawing on key environmental variables this model is capable of predicting TRS, known in Brazil as ATR, in advance of UNICA’s semimonthly reports. Users can therefore get earlier insights on ethanol and sugar availability in the market.


  • See how K-means are used and apply to other agricultural challenges.
  • Input your own assumptions and improve the model with your own proprietary data.
  • Use environmental conditions in the model to identify acreage in other countries that will impact production.
Accessibility: The code for Gro’s framework is functional exclusively with the Gro API Client
NDVI gives an indication of crop stress. It is therefore a good predictor of TRS since moderate crop stress is beneficial for sucrose concentration during the harvest season. 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. Combining this yield with mill run rates can determine how much sugar is produced.


This framework uses several environmental variables including precipitation, temperature, potential evapotranspiration and normalized difference vegetation index (NDVI). The model also uses UNICA’s semimonthly TRS values. Each input source is profiled below.


Model Specific Data: 8-day averages for districts of Brazil’s soybean-producing areas


Model Specific Data: 8-day averages for districts of Brazil’s soybean-producing areas


Model Specific Data: Daily potential evapotranspiration in Sao Paulo


Model Specific Data: Daily precipitation quantity in Sao Paulo


Model Specific Data: Daily land surface temperature in Sao Paulo


Gro: Brazil Sugarcane Model Methodology
  • Compare TRS to environmental indicators. Select indicators with greatest correlation to TRS.
  • Cluster data points together using K-means to better organize data and discover patterns.
  • After clustering, the impact of environmental indicators on TRS proves to be seasonal. Dataset is split into three distinct seasonal periods.
  • Run a linear regression of TRS on the predictive variables (temperature, precipitation, evapotranspiration, and NDVI) that is unique to each clustered period.


Contact sales