Development of a prototype to forecast rice yields and grain quality in the Northern Italian district

The consideration of pre-harvest grain quality in yield forecasting systems represents a crucial challenge to preserve the competitiveness and sustainability of rice sector in many regions worldwide. This is even more urgent in developed countries, where the key objective of farmers is the achievement of superior grain quality standards, given the strict technical policies imposed by national and supranational supervisory bodies (Tesio et al., 2014). This contribution deals with the improvement of available models for the reproduction of the seasonal variability of rice grain quality, aiming at developing a quali-quantitative rice yield forecasting system for Northern Italy. The quality variable considered in this first prototype is head rice yield (HRY - i.e., the relative weight presence of entire kernels after milling), which represents the main determinant of rice market price at global level. The study was conducted using the WARM model, which accounts for quantitative (Confalonieri et al., 2009) and qualitative (Cappelli et al., 2014) aspects of rice production. The model for HRY was improved using field data referred to Loto and Gladio rice varieties collected in six experiments performed by the Ente Nazionale Risi in Castello d’Agogna (Pavia province) during the seasons 2006-2013. After model improvement, simulations were carried out in three sites of the province in the period 1998-2013, considering actual sowing dates and meteorological data. The model outputs were then used as independent variables to build multi-regression models to explain the variability of historical yields and HRY statistics (dependent variables). At field level, the evaluation metrics computed to evaluate model performances denote a noticeable agreement between observed and simulated HRY and biomass values for both varieties. Modelling efficiency (EF) and determination coefficient (r2) were always higher than 0.9 for aboveground biomass, whereas they ranged between 0.72 (Loto) and 0.94 (Gladio) for HRY values. At province level, the performances of the forecasting prototype were overall satisfactory, in light of the absence of a technological-driven trend influencing rice quali-quantitative yields.. Best results were always obtained when the prototype was applied to Loto variety, as the regression model was able to explain 63% and 72% of the yield variability and HRY respectively and it also revealed a good accuracy while reproducing official statistics (0.36 < EF < 0.72). The present study allowed to realize a pilot system to forecast rice yields and grain quality in Northern Italy, capable to give timely in-season indications about rice cropping systems performance to be used by agricultural stakeholders at different levels, from farmers to regional politicians; furthermore it represents a first prototype to be extended for the consideration of other qualitative variables influencing rice prices and/or a methodological example that may be exported to other rice-growing districts or applied to other crops.