The Climate Services Team at the Barcelona Supercomputing Center has deployed a climate service for vineyard management in the context of the vitiGEOSS project. This dataset results from post-processing, i.e. by downscaling, calibrating and assessing, the seasonal climate prediction system SEAS5 (ECMWF).

Probabilistic predictions have as output several solutions (ensemble members) to account for forecast uncertainty. The forecast information is conveyed as probabilities, in this case as the probabilities of occurrence of three categories or terciles (below normal, normal and above normal).  The categories are defined based on the terciles of the model climatology distribution over a period in the past.  Additional information regarding the probability of occurrence of extremes is also provided, considered as the probability of not reaching the 10th percentile or surpassing the 90th percentile of the model climatology distribution. The skill scores provide information on the forecast quality (fair Ranked Probability Skill Score for the tercile categories and fair Brier Skill Score for the probabilities of extremes). A positive skill score indicates that the prediction is good (better than using average past conditions) in the long term. In contrast, a negative skill score indicates a prediction is not beating the climatological forecast.

• Prediction system: European Center for Medium-Range Weather Forecasts  (ECMWF) SEAS5 and post-processed by BSC.

• Issue frequency: Monthly (~15th of each month)

• Lead times: months 1 to 3 (e.g. For a forecast initialised in June, forecasts will be monthly averages for July, August and September). The initialization date is indicated in the name of each file (e.g. 20210701). 

• Variables: mean, minimum and maximum 2 m temperature, accumulated precipitation, incoming solar radiation

• Ensemble size: 51 members

• Postprocessing: Downscaling from original (1°x 1°) resolution to 0.1°x 0.1° for three domains and monthly calibration with variance inflation.  

• Spatial coverage of the domains: 

• Catalonia region is indicated by ‘cat’ and covers latitudes [10 N, 44 N], and longitudes [1 W, 4 E]. The latitude indices range [1:41], and the longitude indices range [1:51].

• Douro region is indicated by ’douro’ and covers latitudes [40 N, 43N ] and longitudes [9 W, 6 W]. The latitude indices range [1:31], and the longitude indices range [1:31].

• Campana region is indicated by ‘campania’ and covers latitudes [39 N, 43 N] and longitudes [13 E,17.3 E]. The latitude indices range [1:41], and the longitude indices range [1:44]. 

For each prediction, there are several files containing the seasonal variables values, probabilities, definition of the categories and skill scores.

• Forecast probabilities

E.g t2_campania_prob_20210701.ncml

The file name contains the name of the variable, domain, the label ‘prob’ and the initialization date of the forecasts (1st of the month).

It contains the forecast probabilities in (%) of each tercile category below normal (prob_bn), normal (prob_n) and above normal (prob_an) and the probability of lower extreme (prob_bp10) and the probability of upper extreme (prob_ap90). The latitude, longitude, and lead time (months 1 to 3) can be selected.

• Forecast ensemble members

  E.g. t2_campania_20210701.ncml

The file name contains the name of the variable, domain and initialization date of the forecasts (1st of the month).

It contains the 51 absolute values of the forecast variables in their corresponding units (see Table 2).  The latitude, longitude, and lead time (months 1 to 3) can be selected.

• Category limits

E.g. t2_campania-percentiles_month07.ncml

The file name contains the name of the variable, domain, the label ‘percentiles’ and the month for which the category limits apply. 

It contains the limits of the predicted categories ( below normal, normal and above normal). These categories are defined with respect to a period in the past. The 33rd, 66th percentiles (p33 and p66) divide the model climatological distribution into 3 equiprobable categories. The 33th percentile is the boundary between below normal and normal, and the 66th percentile is the boundary between the normal and above normal categories. The 10th and 90th percentiles, which define the threshold for the lower and upper extreme conditions, are also provided (p10 and p90). It should be noted that the definition of the categories is specific to each location (latitude and longitude), initialization month and lead time (valid month).

• Skill scores

E.g t2_campania-skill_month07.ncml

The file name contains the name of the variable, domain, the label ‘skill’ and the month for which the skill scores apply. 

It contains the measures of forecast quality, the fair Ranked probability score for terciles (rpss) and the fair Brier Skill Score for lower and upper extremes (bsp10 and bsp90). It should be noted that the skill level is specific to each location (latitude and longitude), initialization month and lead time (valid month).