Questions related to downscaling techniques

What is downscaling?

To increase resolution in climate model output dynamical or statistical downscaling can be applied to include information representative of finer scales than those represented by the model. This is in contrast to simple interpolation that, despite providing more high-resolution data, does not add such additional fine-scale information.
Global climate models typically provide output at resolution of about 100-200 km. This may not be adequate for various impact studies at local scales. Consequently, there may be a need for further refining this information to local scales. Simple interpolation can provide information at any resolution but does not add additional information on finer scales. As an example, a small-scale (with respect to the climate model) mountain, will in the real climate system show different climate characteristics compared to its low-lying surroundings. In a coarse-scale model where this mountain is not present, interpolation between two surrounding low-level grid points will not give information that is adequate for the high-altitude mountain. In statistical and dynamical downscaling, such information can be added to provide more realistic information also at the local site.

Why is there a need for downscaling?

The coarse resolution of global climate models implies that they have a relatively poor representation of local features such as land-sea distribution and height of mountains. Also, relevant processes of the climate system, such as low-pressure systems and associated areas of precipitation implies that such models are not always fit for purpose for representing regional or local features. This is particularly problematic for some high-impact events such as high-intensity precipitation and wind storms.
GCMs provide good information about climate characteristics on large regional scales. This includes differences between continents and oceans and between high and low latitudes in general. But, also, they can provide information about climate details in parts of a continent like southern and northern Europe etc. However, at a typical GCM-resolution of 100-200 km grid spacing, complex features of the European geography such as the Mediterranean and Baltic Sea, are not well resolved. Also, mountain chains are far too low implying that their role as barriers for precipitation are often underestimated. Moreover, high-intensity events on daily and sub-daily time scales are poorly represented by such coarse models.
Techniques for downscaling global model information, either dynamically through regional limited area models, or through empirical statistical methods, have been developed for this purpose.

What is dynamical downscaling? What is required to run a regional climate model?

Dynamical downscaling with limited-area models is principally very similar to the global climate models as they are numerical models of the climate system. The key difference is that they are applied at higher horizontal resolution for a limited area taking lateral boundary conditions from the global model as input.
In addition to lateral boundary conditions for the atmosphere from the global model, regional climate models mostly need sea-surface temperatures and sea-ice conditions. Exceptions are if they are applied in continental areas or if the regional models also include modules for treating the ocean and sea ice. Such regional earth system models have been set up for the Mediterranean Sea and for the Baltic Sea and the North Sea albeit the actual number of existing simulations is still low.
The EURO-CORDEX projections provided by C3S RCMs have not been run in coupled mode, implying that sea-surface conditions are taken directly from the underlying global model. This also applies for most other CORDEX domains. For the Med-CORDEX domain, however, there exists also simulations in which the Mediterranean Sea has been explicitly simulated in a regional ocean model coupled to the atmosphere.
Depending on what the data should be used for the relevance of sea surface temperature and sea-ice conditions should be carefully considered.

What is empirical or statistical downscaling?

Empirical downscaling implies that large-scale climate information, for instance from a climate model, is downscaled with help from observations. Statistical relationships between the local scale observation and the large-scale model fields are identified for the historical time period. For future climate change the same empirical relationships are assumed to be valid which may not always be the case.
As empirical downscaling relies on observations it is required that such observations exist and that they are representative for the scales that should be addressed. A complication with the empirical downscaling methods is that downscaling is most often done one variable at the time. An implication is that time series of several variables may not be completely consistent.
It is also noted that the assumption that the empirical relationships are the same also in the future climate may not always be true. Examples when this assumption may be violated include areas where non-linear changes are seen. This could for instance involve areas with retreating snow cover where temperature distributions changes differently in their different parts (e.g. Kjellström, 2004). Another example would be areas where strong drying is seen that would exacerbate warming and thereby further increase the response.
It is recommended that users of empirically downscaled information carefully consider if such potential problems may have any implication on their applications.

What are advantages/disadvantages with dynamical/statistical downscaling?

A major advantage with empirical downscaling over dynamical downscaling it can be relatively easily used for downscaling large ensembles of climate model data requiring only limited computational capacity. The key drawback is that the empirical methods assumes that future relationship between local and large scales remain the same as in the historical climate. Dynamical downscaling, on the other hand, allows for such changes over time. Furthermore, regional models provide internally consistent climate states implying that several variables from the model can be used simultaneously.

References

Kjellström, E., 2004. Recent and future signatures of climate change in Europe. Ambio, 33(4-5), 193-198.