Expanding the biophysical ensemble: hybrid dynamical-statistical downscaling methods based on spatial/temporal scale

While coupled global models of the atmosphere and ocean have demonstrated significant skill in predicting broad-scale SST patterns on seasonal timescales, they typically lack the fine resolution necessary to capture biophysical details which strongly impact fish recruitment (for example, regional prey biomass and regional subsurface temperatures). Dynamical downscaling - the use of coarse-scale global predictions to force fine-scale regional dynamical models - can be used to bridge this gap, and to generate useful predictions for biogeochemical variables not included in the global simulations. However, the computational expense of dynamical downscaling has strongly limited its wider use. Indeed, at present, readily available ensembles of coarse-scale global output are far larger than what can be affordably downscaled. Given this computational barrier, statistical relationships derived from a limited set of downscaling output can be used as a proxy to generate a much larger ensemble of regional predictions from the global forcing. Here we review several related hybrid statistical-dynamical methods to expand dynamically-generated regional ensembles on both seasonal and multi-decadal timescales. These methods are based on dominant spatial/temporal correlations between the global forcing and the regional response, and can include correlations across different biophysical variables. Depending on the spatial and temporal scale of the target predictand, time-lagged correlations may be a crucial element in these hybrid methods. Examples include present and planned use of hybrid methods for projecting the seasonal and multi-decadal futures of the Bering Sea cold pool (a feature with strong connections to major fisheries of the Bering Sea), the coastal Pacific Northwest, and the Gulf of Alaska.