Application of statistical models in the detection of grapevine phenology changes

In recent decades, there have been significant changes in temperature and precipitation, as well as in the occurrence of phenological stages of the grapevine in most wine-growing regions around the world. These changes are not the same for each variety, nor in all locations. Due to the vulnerability of the viticulture sector, including the possible losses in production in the current winegrowing areas, as well as the planting of vineyards in new areas, it is of great importance to examine the trends in the occurrence of individual stages and to link them as successfully as possible with changes in meteorological parameters. The simplest approach to this is using agrometeorological indices (e.g., Growing degree day, GDD) which can determine the possibility of growing a certain variety. There is also the possibility of developing and testing simple statistical phenological models that serve to predict the occurrence of phenological stages. Four such models were tested for the prediction of four phenological stages (budburst, flowering, veraison, and harvest) for four grape varieties ('Gras ˇevina', 'Chardonnay', 'Merlot', and 'Plavac mali') in Croatia. The first two models are commonly used GDD models with a temperature base of 10 degrees C or 5 degrees C, and thresholds necessary for phenological stage to start depending on variety or variety and location. The other two models are based on the determination of the best multi-linear regression using as predictors monthly and multi-month averages of minimum temperature, maximum temperature, mean temperature, and total precipitation. The increase in temperature index values from the 1990s to today is particularly significant. Statistical phenological models also proved to be a good indicator of the occurrence of individual phenological stages. GDD models proved to be somewhat better in prediction, GDD models that use a temperature of 5 degrees C as a base proved to be better for predicting budburst, those that use a base of 10 degrees C proved to be better for the other stages and particularly for flowering (with agreement index d up to 0.8 and root mean square error of prediction RMSE from 5 to 10 days). Linear regression that uses temperature as a predictor and the same equation regardless of location proved to be very good in predicting the harvest of autochthonous varieties ('Gras ˇevina' and 'Plavac mali') with low RMSE (up to 10 days). The presented results indicate that these models could be applied to future scenarios and with that help to make decisions in the wine sector in Croatia and worldwide.