Seasonal forecasts quality assessment report

Agriculture, and in particular winegrowing, is a sector highly dependent on heat, sunlight and water, and therefore very sensitive to climate variability, extremes and impacts of climate change. One mitigation strategy consists on using seasonal predictions, e. g. from one to several months into the future, to modify the vegetative cycle of the grapevines to adapt for the climate anomalies of the upcoming months. This is one of the objectives of this EU funded Research and Innovation VISCA project (Vineyards´ Integrated Smart Climate Application), which seeks to make European wine industries resilient to climate change, minimizing risks through an improvement of the production management. In order to achieve this objective, predictions from different time-scales (from weather to seasonal) will be integrated in a Decision Support System platform that will help the end users adopt better decisions.

In this report, we have assessed the quality of two seasonal forecast systems (European Centre for Middle-range Weather Forecasting (ECMWF) System 4 and SEAS5), two bias correction approaches (calibration and simple bias correction) and a downscaling strategy (calibration). This has been achieved with a common set of deterministic and probabilistic metrics: the correlation of the ensemble-mean (deterministic), the fair Ranked Probability Skill Score (FRPSS, probabilistic), the fair Continuous Ranked Probability Skill Score (FCRPSS, probabilistic) and with reliability diagrams (probabilistic). These metrics are important for the wine partners because they provide information about different aspects of the predictions that contribute to increase the robustness of their decision-making processes at seasonal time-scales.
More specifically, we have verified monthly / 3-month average predictions of temperature at 2 m, maximum temperature, minimum temperature and precipitation from ECMWF System 4 (S4) and ECMWF System 5 (SEAS5) using the Japanese 55-year reanalysis (JRA-55). This validation has also been performed for the three VISCA demo-sites using the observations provided by the end-users: Raïmat (Codorniu), Quinta do Ataide (Symington) and Mirabella-Eclano (Mastroberardino). In addition, we have applied two bias-correction techniques, calibration and simple bias correction, to adjust the forecast statistical properties of the variables to the reference reanalysis. We have also evaluated the effect of these techniques on the skill and reliability of the predictions compared to their raw counterparts. Furthermore, we have selected the calibration approach as a first version of the downscaling to the demo-sites.
The results obtained show that there is some degree of predictability in the three demo-sites in different variables that can provide value beyond the customary use of climatology. Moreover, the verification / bias-correction / downscaling workflow developed in this task provides the basics of all future refinements that we will conduct during the remaining two years of the project, e. g. through the exploration of new seasonal prediction systems and/or downscaling techniques.