TRANSLATE: from climate data to climate services for Ireland

The TRANSLATE project was initiated by Met Éireann, the Irish meteorological service, in Jan. 2021, with the objectives of developing standardised future climate projections for Ireland, and hence to develop a range of sector-specific climate services.  

The project has already produced an initial set of standardised projections out to the end of the century, based on a selection of CMIP5 global model projections using 3 different forcing scenarios (RCP 2.6, 4.5 and 8.5).  For each scenario, a 6-member ensemble of CMIP5 simulations were dynamically downscaled to high-resolution (4 km) over Ireland using the COSMO and WRF regional models, while a larger ensemble (up to 30 members, depending on scenario) were downscaled to 12 km by the EURO-CORDEX project.  The future of the 21^st century was divided into three 30-year periods (2021-2050, 2041-2070, and 2071-2100), and for each of these the downscaled simulations were detrended and bias-corrected (using quantile-delta mapping).  Ultimately, most fields were also statistically downscaled to the 1.5km observational grid.  The ensemble of downscaled simulations for each scenario and each time-period was further decomposed into low, medium, and high-sensitivity members (depending on the mean temperature change over Ireland projected by each member), as a representative way to portray future uncertainty.

In practise (as will be explained), the post-processing of the few 4 km-resolution simulations was necessarily different to that of the many 12 km-resolution CORDEX simulations.  Even so, the final climate charts generated by each set of simulations are climatologically indistinguishable from each other.  Moreover, while the distribution of absolute projected values over Ireland can be complex (as determined mainly by local geography), the difference between future projections and historical fields is relatively simple and bland, with temperature changes showing just a gradual increase from west to east across the country.  The emergence of such simple, clean, and consistent climate change signals after all the numerical complexity involved in global simulations, regional downscaling, and statistical post-processing provides quite convincing evidence (to us at least) that those signals are real.

As “input” for the development of climate services, each future climate projection consists of a detrended, bias-corrected ensemble of 30-year-long daily values for each variable of interest (initially daily mean, min and max temperatures and daily precipitation) at each grid-point.  Each ensemble is used to generate a standard set of statistics (means, variances, percentiles, and frequency distributions), and may also be queried to produce standard climate indices (e.g., heat wave occurrences) as well as more customised indices (e.g., length of growing or grazing seasons). 

The model selection, downscaling, detrending and bias-correction processes will be discussed, and a representative selection of results will be shown.  Further work based on CMIP6 global simulations is already underway.