Forecasts from fits of frontal fluctuations 1 Arthur

A primitive equation ocean model is fit with strong constraints to non-synoptic hydrographic surveys in an unstable frontal current region, the Iceland–Faeroe Front. The model is first Ž . initialized from a time-independent objective analysis of non-synoptic data spanning 2 to 6 days . A truncated set of eddy-scale basis functions is used to represent the initial error in temperature, salinity, and velocity. A series of model integrations, each perturbed with one basis function for one dependent variable in one layer, is used to determine the sensitivity to the objective-analysis initial state of the match to the non-synoptic hydrographic data. A new initial condition is then determined from a generalized inverse of the sensitivity matrix and the process is repeated to account for non-linearity. The method is first tested in ‘identical twin’ experiments to demonstrate the adequacy of the basis functions in representing initial condition error and the convergence of the method to the true solution. The approach is then applied to observations gathered in August 1993 in the Iceland–Faeroe Front. Model fits are successful in improving the match to the true Ž data, leading to dynamically consistent evolution scenarios. However, the forecast skill here . defined as the variance of the model–data differences of the model runs from the optimized initial condition is not superior to less sophisticated methods of initialization, probably due to inadequate initialization data. The limited verification data in the presence of strong frontal slopes may not be sufficient to establish forecast skill, so that it must be judged subjectively or evaluated by other quantitative measures. q 1999 Elsevier Science B.V. All rights reserved.