Back to opportunities | Category Index
These are the intended use-case/justification for one or multiple variable groups. Opportunities are linked to relevant experiment groups. Identifying opportunities helps to provide a structure to map variables against requirements. Each opportunity description will convey why this combination of variables and experiments is important and how they contribute to impact.
| Attribute | Value |
|---|---|
| description | This opportunity covers empirical-statistical downscaling (ESD) activities in CORDEX, including hybrid downscaling (emulators). These tools are trained using large-scale fields from reanalysis (or RCM driving GCMs, in the case of emulators) and can then be applied to compatible fields in the GCMs (e.g. CMIP7 output). This requires a common set of standard levels sampling the atmospheric column. Machine learning-based statistical approaches are now used to emulate the behavior of high-resolution climate models such as RCMs or CPRCMs. When trained, those tools have the capacity to replace the dynamical downscaling approaches for a variety of applications such as creating multi-member ensembles or emulating a SSP not run with the emulated RCMs. Emulators need GCM inputs to work. Both standard ESD techniques and emulators require a similar set of variables including daily temperature, wind, moisture and geopotential height across the atmosphere. The sub-daily temporal resolution is an opportunity for downscaling at a sub-daily scale, which is increasingly demanded. Ideally, GCMs should use a common extrapolation algorithm below ground, or let those areas as NaN. Surface pressure is key to deal with these areas. An additional set of variables which could save much processing is monthly aggregated precipitation (monthly wet-day frequency and monthly wet-day mean precipitation, both with a 1 mm/day threshold). While these statistics can be estimated from daily data, such a job is computationally demanding and makes such statistics inaccessible. The wet-day frequency and mean precipitation are two key parameters that can be used as both predictor and predictand for downscaling 24-hr precipitation statistics (e.g. DOI:10.5194/egusphere-2024-1463). ESD can be applied to larger ensembles and it is feasible to downscale decadal predictions (DCPP experiment) if these pressure-level fields are saved. This CORDEX request is similar to CMIP6, but with a temporal extension in the past to cover the GWL+0 and in the future, as 2100 is too soon now. Extension to 1850 is more important than extension to 2150. If the full period data are too much to store, the period 1901-1949 could be skipped (i.e. 1850-1900 for GWL+0 reference, and 1950-2150 for historical+scenario). |
| expected_impacts | As dynamical downscaling takes long to be available, ESD is the method of choice in many climate services (e.g. in Spain or Germany) to provide timely downscaled information after major CMIP cycles. Also, it is the only method that can reach point information, using station data as predictands. Therefore, this opportunity provides input for many local studies, where historical records are available and need to be projected into the future. |
| justification_of_resources | Daily 3D data on an standardized set of levels covering the troposphere allow the use of predictor fields ESD and emulators compatible with the training set. Cost-effective ESD and emulators have the potential to fill gaps in the costly dynamical downscaling simulations. In this sense, despite requesting a single member in the ensemble size, the more members provide inputs for ESD, the better these models can be used to sample the internal variability uncertainty, which is underexplored in the CORDEX dynamical downscaling activity due to computational constraints. |
| lead_theme | Impacts & Adaptation |
| minimum_ensemble_size | 1 |
| name | Empirical Statistical Downscaling and Emulators |
| opportunity_id | 80 |
| data_request_themes | Impacts & Adaptation, Atmosphere |
| experiment_groups | dcpp, fast-track, scenarios, historical |
| mips | DCPP |
| time_subsets | 80ac3156-a698-11ef-914a-613c0433d878 |
| variable_groups | global_emulators, statistical_downscaling_core_6hourly, statistical_downscaling_core_daily, statistical_downscaling_extended |