(English) 2016 – Forecasting rice yield at rice district scale from WARM


ERMES WARM regional modeling solution already tested in 2015 was applied in real time to predict the 2016 productivity in the Italian, Greek and Spanish rice districts in two moments of the season (i.e., around flowering and around maturity).

The forecasting system was trained on 2003-2014 data by relating some selected indicators of the modeling solution with the series of official rice yields, provided by the ERMES Italian, Greek and Spanish partners. The indicators used in this statistical post-processing, based on multiple linear regression analysis, could be grouped in five categories:

  1. state variables of the WARM model (e.g., aboveground biomass, LAI, ..) simulated in potential conditions;
  2. state variables of the WARM model (e.g., aboveground biomass, LAI, number of potential infections, ..) simulated in blast-limited conditions;
  3. state variables obtained by forcing the model with remote sensing values of LAI;
  4. state variables obtained by the recalibration of the model parameters using remote sensing values of LAI;
  5. agro-meteorological indicators.

First, an agro-meteorological analysis was performed to study if weather conditions were favorable to rice growth (e.g., occurrence of relevant cold and heat waves, distribution of precipitation,..); moreover blast potential infections were simulated by the modeling solution.

On the basis of this analysis the forecasting system was applied to forecast in real time the 2016 yields. Results obtained in the three study area are shown below. They evidence a generally “normal” 2016 rice season, which forecasted yield changes of maximum plus/minus 5 % with respect to the 2010-2015 average.


Forecasted rice yield fort the Italian, Greek and Spanish rice districts obtained at the end of the cultivation season.