Dynamically downscaled climate projections for the California Current Upwelling System

Most future projections for upwelling ecosystems are based on earth system models (ESMs) with resolutions too coarse to properly resolve coastal winds and upwelling dynamics. Here we use a high-resolution (0.1˚ in horizontal resolution) regional ocean circulation model coupled with a biogeochemical model to dynamically downscale ESMs and produce climate projections for the California Current System under the rcp8.5 scenario. To capture the spread of projections, we select three earth system models (ESMs) that span the CMIP5 range for future changes in both the mean and variance of physical and biogeochemical CCS properties: GFDL-ESM2M, HadGEM2-ES, and IPSL-CM5A-MR. To debias the model forcing obtained from the ESMs (i.e., correct for their systematic offsets with observed climate), we apply a “time-varying delta” method in which the regional model’s surface and lateral boundary conditions are constructed by adding the transient (1980-2100) ESM anomalies to the observed historical (1980-2010) climatology. Relative to a “fixed delta” method that compares a historical period to a future one, the time-varying delta method has the advantages of (1) capturing projected changes in interannual variability, and (2) resolving intermediate time frames (e.g., 2010-2070), when climate changes will have considerable impacts on the CCS ecosystem. Changes in surface and subsurface biogeochemical variables relevant to ecosystem processes (e.g., chlorophyll biomass, dissolved oxygen concentration, and pH) and the physical mechanisms that drive them are further investigated in the downscaled projections and in their ESMs counterparts. This analysis sheds light on the uncertainty that results from insufficient resolution in ESMs relative to the uncertainty due to spread among the ESMs themselves. Until large ensembles of eddy-resolving global or regional models are computationally feasible, we suggest that a fruitful approach is to combine coarser resolution large ensembles with dynamical downscaling of select runs informed by analyses like the one presented here.