Statistical downscaled sub-seasonal forecast (Sub-SEA)

The Highlander project aims to improve the subsets used for predicting and analyzing climate data. See here for more details.

This includes the statistical downscaling of sub-seasonal forecast (Sub-SEA), thanks to the application of the ECMWF's ecPoint technique. CINECA uses Galileo100, Cineca's HPC resource, to produce these sub-SEA forecasts twice a week. This post-processing delivers probabilities for point measurements of rainfall and temperature at 2 meters above ground (within a forecast model gridbox). ecPoint is based on the concept of conditional verification, and it consists of identifying (through calibration) and correcting (through post-processing) errors in model rainfall or temperature forecasts according to a diagnosed grid box weather-type. Its main purpose is to activate local-weather-dependent adjustments for gridscale bias and for sub-grid variability, to predict, probabilistically, values as measured by weather stations.

The ecPoint Sub-SEA products are based on global rainfall accumulated over 24 hours, and daily minimum, average, and maximum temperatures from day 16 up to day 30 of the forecast. These data are very important for animal or human wellbeing.

For the Highlander project, the post-processing is applied to the ECMWF's 51-member forecast ensemble (ENS), compounded by a control member and 50 perturbing members (currently with 36 km of horizontal resolution) and the final outputs are 99 percentiles.
What are the 51 members? In other words, they are the equally likely predictions related to the question of, for example. "What will happen in 15 days in the area" X "?" that I obtained by making very small changes in the initial conditions. The changes are used to account for the uncertainties in the measurement of the data in the initial data assimilation process and the uncertainties in the forecast prediction at each lead time. Using all 51 ensemble members, we calculate the probabilities of the occurrence of specific thresholds (e.g. what are the probabilities of 30 mm of rain accumulating in 24 hours) but these probabilities are based on the average values of each 36 km grid box, which ignores any local variations in the grid box itself (sub-grid variability). As a result, ecPoint post-processing will provide much more accurate forecasts at point-scale (how many possible local scenarios of rain or 2m temperature do I have within a grid-box?).

These new Sub-SEA ecPoint products are now available in Italy and the surrounding areas.