Source Data

The analysis of future climate conditions is based on simulations from the most recent generation of global climate models (GCMs). These models have evolved rapidly over the past decades, primarily in terms of the number of interconnected processes and feedbacks within the climate system that they are able to represent. The spatial resolution of GCMs has also improved; however, for a detailed overview of local climate conditions, it is still necessary to apply methods that refine the model output, a process known as downscaling. This can be done either by incorporating more detailed regional climate models (RCMs) or by using statistical methods applied directly to GCM output, for example the incremental method.

Because climate model outputs are associated with systematic errors due to the necessary simplification of complex real-world processes, they must be corrected in order to obtain meaningful results regarding the simulated properties of the climate system. In general, when working with expected values of meteorological elements, such as seasonal and annual averages, the changes projected by the models can be used as they are, without modification. Problems arise when analysing daily data and extreme values, such as maximum and minimum temperatures or precipitation totals above specified thresholds, because of the low spatial resolution of GCMs.

The authors’ experience with the presented methodology has so far led to a preference for refining GCM outputs for local-level applications using statistical methods rather than dynamic downscaling with RCMs. This is mainly due to the systematic errors of RCMs, which tend to produce a more humid and colder climate during the historical control run, and also to the fact that, while GCM simulations are available from the latest generation of climate models, CMIP6, used in the latest IPCC report (AR6), the available Euro-CORDEX RCM simulations for Europe are driven by the older generation of global climate models, CMIP5, used in the previous IPCC report (AR5). As a result, the available RCM simulations do not reflect the latest scientific findings.

Model Selection

Many of the latest CMIP6 GCM simulations include models with varying levels of spatial detail. Most simulations of 21st-century climate evolution have a horizontal spatial resolution of approximately 100 to 250 km. There is also a small subset of GCMs with resolutions around 50 km, but their simulations end in the mid-21st century. Individual GCMs also differ in the complexity with which they describe processes in the climate system, the way smaller-scale phenomena are parameterised, and the formulation and numerical solution of the underlying physical equations. It is inevitable that the simulated climate diverges to some extent from reality and that this divergence varies across space, time, and physical variables. Therefore, GCMs that best represent the climate of Central Europe were preferred for simulations of the region’s future climate. At the same time, it was necessary to ensure that the preferred GCMs, which form only a subset of all available GCMs, affect future climate evolution in the same way, and with the same degree of uncertainty, as the full set of available GCMs. In other words, the selected subset of GCMs should not consist of models that, under the same conditions, project, for example, higher temperature increases, or different changes in precipitation, wind, sunshine duration, and so on, than models outside the selection. For this purpose, a selection methodology for narrowing down the model ensemble was proposed and is described here.

In accordance with this methodology, models that were unable to reliably simulate the climate of Central Europe in the recent past were excluded from the set of approximately twenty CMIP6 GCMs on the basis of validation. From the remaining models, seven GCMs were selected so that the shortlist remained statistically representative of the original model set while allowing work with fewer simulations. One reason for narrowing down the GCM ensemble is that individual GCMs are often used as sources of input meteorological data for hydrological models or for models simulating the impacts of climate change on landscapes and their management. In such cases, integrating the entire original GCM set places enormous demands on computing power and is often practically infeasible. The selection of GCMs was made with regard to all the basic meteorological elements that are further analysed and used for the calculation of reference evapotranspiration and soil moisture by the SoilClim model, which was calibrated and validated for the conditions of the Czech Republic by Hlavinka et al. (2011). The selected models, together with the available climate change scenarios, are presented in the following table. GCMs with finer spatial resolution, 100 km or less rather than 250 km, were preferred. We also included the Czech model ALADIN-Climate/CZ from the CP-RCM family.

Climate Scenario Preparation

GCM outputs cannot be used directly unless we are concerned only with the relative change in meteorological elements. They are affected by systematic errors, for example an underestimation of temperature by 1 °C or an overestimation of precipitation by 25% in Central Europe, which must first be corrected. Alternatively, one can work with the climate change signal resulting from climate model simulations and relate it directly to observed data. This latter approach is referred to as the incremental method, or direct modification, and is traditionally used in the Czech Republic for modelling the impacts of climate change on, for example, the hydrological balance, because it is more robust than using climate model simulations with bias correction. To apply the incremental method on a daily time step, it is appropriate to use transformations that account not only for changes in averages but also for changes in variability. This is made possible, for example, by the Advanced Delta Change (ADC) method. The ADC method makes it possible to include changes in variability in the transformation. In practical terms, this means that extremes can change differently from the mean, which correctly reflects the situation observed in the real world. When deriving changes in precipitation from a climate model, the ADC method also takes into account systematic errors in the simulation, which may be non-linear. Further details can be found in van Pelt et al. (2012).

When applying the ADC method, daily values of meteorological elements are processed on a weekly basis in order to preserve their annual cycles, and the transformation parameters are smoothed. Air temperature is transformed linearly, unlike precipitation. Other meteorological variables, namely global radiation, relative humidity, and wind speed, are modified by multiplying them by the ratio of the averages for the GCM control run period and the GCM scenario simulation period. These transformation parameters are then smoothed as well.

In the case of the Czech Republic, station measurements are used in the form of technical data series. These are measurement time series that have undergone thorough quality control, error correction, homogenisation, and gap filling (tpánek et al., 2011, 2013). Data from stations operated by the Czech Hydrometeorological Institute (CHMI) are interpolated to a fine spatial grid with a resolution of 500 m before the ADC method is applied. The same approach is used for climate model data on a daily time step. The resulting information on future climate change is therefore presented in the form of maps with a spatial resolution of 500 m and a temporal resolution of one day.

For processing Central Europe, the E-OBS dataset (version 27.0e) of gridded station observations was used instead of CHMI station measurements. This dataset is based on station observations collected in the ECA&D database.

With respect to the interpretation of the results, it is important to note that, in addition to the 1981-2010 reference period, 30-year time frames are used for future climate conditions: 2015-2044, referred to as “2030”; 2035-2064, referred to as “2050”; 2055-2084, referred to as “2070”; and 2070-2099, referred to as “2085”. These periods overlap. Within these time frames, statistical characteristics, including outliers, can be evaluated over the whole period. As with climate model simulations, it does not make sense to analyse and present individual days or years; only statistics for the whole period are meaningful. Long-term trends can then be evaluated by comparing individual moving periods in the future climate.