Downscaling is the process of translating broad-scale projections from GCMs to finer spatial scales, making them more relevant for local or regional impact assessments.
Downscaling is crucial for assessing the impacts of climate change on local environments, as GCMs typically operate at scales too coarse for detailed regional analysis.

Generally speaking, three methods are available for downscaling GCM output to higher spatial resolutions:
• Dynamical Downscaling: uses regional climate models to simulate climate processes at a finer scale within a specific region.
• Statistical Downscaling: uses statistical relationships between large-scale atmospheric variables and local climate variables to derive local-scale projections.
• Calibration of current climate surfaces with GCM output

Regional/dynamical downscaling
Regional downscaling involves using high-resolution physicially-based regional climate models (RCMs) to simulate climate processes over a specific area. These models are nested within GCMs, incorporate the same physical principles as GCMs but focus on a specific region.

The dynamical downscaling using a RCM yields the best results, even in areas with complex topography, and directly generates climate surfaces. RCMs capture detailed topography, land use, and regional weather patterns, which can lead to more accurate simulations of local climate phenomena such as precipitation, temperature and extreme weather. Moreover, RCMs typically have resolutions ranging from 10 to 50 kilometers, compared to GCMs, which often have resolutions of 100 kilometers or more.

The superior performance of RCMs is counterbalanced by several disadvantages:
• Running RCMs requires significant computational resources because they solve complex equations at a higher resolution than GCMs.
• RCMs depend on the quality and accuracy of the driving GCM data. Different methods of dynamical downscaling are linked to specific GCMs, thus transferring inherent flaws in particular models from a lower to a higher resolution.

Statistical downscaling
Statistical downscaling uses statistical methods to relate large-scale climate variables from GCMs to local climate conditions. This approach relies on historical climate data to establish statistical relationships between large-scale and local-scale climate features.

Statistical downscaling yields good results, in terms of reproducing current climates from GCMs. They are computationally less demanding than RCMs and can be applied to output of different GCMs.

On the down side, statistical relationships have to be established individually for each station and GCM, requiring high-quality historical data. Surfaces have to be created from point data, a problem in data scarce regions. Moreover, the approach assumes that statistical relationships between large-scale and local-scale variables remain constant over time, which may or may not hold under changing climate conditions.

Calibration method
In situations where no downscaled data are available from a RCM, it is still possible to generate these by using the calibration method of GCM downscaling.
This involves essentially the superposition of a low-resolution future climate change field on top of a high-resolution current climate surface, for instance of precipitation or temperature.

NOTES
1. a step-by-step explanation of downscaling using the calibration approach is provided in Pub.105 pp. 11-15
2. In the CLIMAFRICA project, in which I participated (Job 40), both dynamical and statistical downscaling methods were used, resulting into downscaled data to 25 km resolution.