Climate predictions and downscaling to extreme weather

Downscaling and bias correction of present climate and decadal predictions (12 km)

Dynamical downscaling in BINGO was performed with the COSMO-CLM regional climate model. Using ERA-Interim reanalysis as boundary forcing, a regionalization of the past and present climate (1979-2015) over Europe was produced at 12 km resolution, with a temporal resolution of 1- or 3-hours depending on the variable ; the years 1989-2008 had already been simulated as part of the EURO-CORDEX project. At the same resolution, a ten-member ensemble of MiKlip decadal predictions (2015-2024) was produced over sub-regions of Europe encompassing the research sites.

The 12 km simulations were bias corrected using the CDF-transform method and the best available reference data (deliverablesD2.1 and D2.2) . Users then had the option to receive data via DECO either with or without bias correction.

The long record for the past and a decade ahead produced by BINGO served to determine average conditions as a baseline to define floods and droughts and to provide data for successive use in the integrated analysis phase.

High-resolution downscaling O(1 km), with an emphasis on floods

The largest added value in very-high-resolution O(1 km) climate data is found in the representation of precipitation extremes. Such data are, however, computationally expensive to produce via dynamical downscaling. BINGO thus adopted a two-pronged approach to producing O(1 km) climate data at affordable computational expense.

In the first approach, large-scale weather patterns with an elevated risk of extreme precipitation were identified from observations and reanalysis via a clustering algorithm, for each research site. Local-scale meteorological predictors of intense precipitation are also identified for each site. These factors were then combined in a classification algorithm (Meredith et al., 2018), which tests each day in the 12 km simulations for an enhanced risk of intense precipitation. Identified “potential extreme days” are then dynamically downscaled to 2.2 km resolution at hourly frequency. The resulting catalogue of O(1 km) extremes at each site is ideal for stress-testing of hydraulic infrastructure, design situations and process-orientated case studies.

In the second approach, a stochastic weather generator was produced. This uses the 12 km simulations as boundary conditions to inexpensively produce large ensembles of BINGO-relevant hydrological variables at 1 km and hourly resolutions, which are consistent with the large-scale situation in the forcing data. Such large ensembles of continuous data would not be computationally feasible via dynamical downscaling.

Additional Products

While BINGO’s main focus is decadal predictions, the EURO-CORDEX archive was also exploited to study climate projections up to 2100 under different climate scenarios (RCP2.6, RCP4.5 and RCP8.5). For inter-comparability, 12 km simulations which used the same global and regional models as in the decadal predictions were added to the DECO platform at daily resolution. Bias correction of these data was also provided.

Spatial maps of return-levels for extreme precipitation events for the research sites Badalona, Bergen, Tagus and Wupper – for various return-periods and event durations – were also computed. The return-levels were obtained based on a duration-dependent Generalized Extreme Value distribution with spatial covariates.

What was achieved?

• BINGO has produced a data exchange platform "DECO" for disseminating climate data in formats specific to individual hydrological models.
• We dynamically downscaled climate data between 1979 and 2015 (evaluation period), and from 2015-2024 (decadal predictions). O(1 km) resolution simulations were also produced for extremal episodes via the development of a classification algorithm (Meredith et al., 2018), greatly reducing computational expense.
• Bias correction via the CDF-transform was applied to all 12 km data.
• Using these data and observations, we developed (i) precipitation and weather generators, and (ii) spatial maps of rainfall return levels and intensity-duration-frequency curves for all research sites. A key result for end users is spatial maps of return periods.
• Guidelines for Dynamical downscaling to 1 km scale – method, rainstorms

MUST-READ publications:

• Meredith, E. P., Rust, H. W., and Ulbrich, U.: A classification algorithm for selective dynamical downscaling of precipitation extremes, Hydrol. Earth Syst. Sci., 22, 4183-4200 , https://doi.org/10.5194/hess-22-4183-2018, 2018.
• Guidelines for Dynamical downscaling to 1 km scale – method, rainstorms
• Dynamical downscaling of European reanaysis to 12km and daily values for the period 1979/2014*
• Data downscaled to 12km/daily, Europe, for the period 2015-2024
• Definition of extremal circulation patterns, present climate
• Present climate extremal episodes downscaled to 2km/1h / maps of return levels
• Ensembles for decadal prediction extremal episodes downscaled to 3-1km/ 1h); Spatial stochastic precipitation generator for catchments
• Ensembles for decadal predictions extremal episodes downscaled to 1km1h for Cyprus research site
• Stochastic weather generator producing large ensembles for decadal predictions
• Maps of IDF curves
• Technical Factsheet for Dynamical downscaling to 1 km scale – method, rainstorms

* This project deliverable (in particular, values for periods 2009-2015 and 1979-1989) can be made available upon request. If you’re interested in having access to the data, get in contact with us .