Downscaling Method

In climate change analysis, problems occur when dealing with the scale differences between global (or regional) climate models and local hydrologic models. The resolution of available GCM output is typically between 200 and 1000 km, for many impacts models, however, information is required on regional scale (i.e. 100km) and/or local scale (i.e. 10km). One potential solution to this problem is to downscale the output from the GCM to a local scale.
It is noted that downscaling should prove to be a useful tool for assessments involving multiple air issues, including climate change, because of the physical links between these issues and large-scale atmospheric flow. Downscaling schemes can be grouped into two broad categories; dynamical and statistical downscaling schemes. Dynamical downscaling (DD) employs a regional climate model (RCM). These models are driven by boundary conditions derived from observations and from the output of larger-scale global atmospheric models over a limited spatial domain. Wilby et al. (2002) found that DD methods are computationally costly and still need to downscale the output in order to use it in impact models.
The second method involves the development of statistical relationships between observed large scale variables of observed climate and local variables such as site-specific temperature and/or precipitation. These relationships are assumed to remain constant under a changed climate, and are applied to predict future local climate form the future large scale conditions simulated by a GCM (Wilby and Wigley, 1997). Statistical downscaling (SD) techniques are commonly based on methods from linear or non linear regression analysis, stochastic processes, artificial neural networks, etc. Each methodology has its unique strengths for reproducing specific statistical features of the fields. Notably, SD methods are computationally
inexpensive and do not required detailed information about physical processes.