A statistical-dynamical scheme for reconstructing ocean forcing in the Atlantic. Part II: Methodology, validation and application to high-resolution ocean models

A novel statistical-dynamical scheme has been developed to reconstruct the sea surface atmospheric variables necessary to force an ocean model. Multiple linear regressions are first built over a so-called learning period and over the entire Atlantic basin from the observed relationship between the surface wind conditions, or predictands, and the anomalous large scale atmospheric circulations, or predictors. The latter are estimated in the extratropics by 500 hPa geopotential height weather regimes and in the tropics by low-level wind classes. The transfer function further combined to an analog step is then used to reconstruct all the surface variables fields over 1958-2002. We show that the proposed hybrid scheme is very skillful in reproducing the mean state, the seasonal cycle and the temporal evolution of all the surface ocean variables at interannual timescale. Deficiencies are found in the level of variance especially in the tropics. It is underestimated for 2-m temperature and humidity as well as for surface radiative fluxes in the interannual frequency band while it is slightly overestimated at higher frequency. Decomposition in empirical orthogonal function (EOF) shows that the spatial and temporal coherence of the forcing fields is however very well captured by the reconstruction method. For dynamical downscaling purposes, reconstructed fields are then interpolated and used to carry out a high-resolution oceanic simulation using the NATL4 (1/4°) model integrated over 1979-2001. This simulation is compared to a reference experiment where the original observed forcing fields are prescribed instead. Mean states between the two experiments are virtually undistinguishable both in terms of surface fluxes and ocean dynamics estimated by the barotropic and the meridional overturning streamfunctions. The 3-dimensional variance of the simulated ocean is well preserved at interannual timescale both for temperature and salinity except in the tropics where it is underestimated. The main modes of interannual variability assessed through EOF are correctly reproduced for sea surface temperature, barotropic streamfunction and mixed layer depth both in terms of spatial structure and temporal evolution. Collectively, our results provide evidence that the statistical-dynamical scheme presented in this two-part study is an efficient and promising tool to infer oceanic changes (in particular those related to the wind-driven circulation) due to modifications in the large-scale atmospheric circulation. As a prerequisite, we have here validated the method for present-day climate; we encourage its use for climate change studies with some adaptations though.