I don’t think there is a simple answer to this question. If you google it, you will realise that there are several papers covering this. As a very simple (and thus often incorrect) rule of thumb: statistical downscaling techniques tend to be less expensive to run but may cover only a sub-set of the correlation (spatial, temporal and cros-variable) experienced in the real world. On the end of the spectrum, dynamical downscaling technique are often more computational expensive to run but often provide a better 4D corss-variable coherency than statistical tools.
Overview
Community Forums
Content Tools
