1. Introduction
Hurricanes develop and are maintained by heat energy they receive from the sea surface. The warmer the sea surface temperature (SST) is below the hurricane, the more energy is available to the hurricane (e.g. Emanuel 1986; 1999). Wind-induced mixing of the upper ocean by a hurricane can cool the sea surface via entrainment of cooler water into the oceanic mixed layer (OML) from below (e.g. Shay et al. 1992; Ginis 2002). Therefore, the future intensity of a given hurricane depends not only on the initial temperature of the sea surface below the hurricane, but also on the magnitude of the wind-induced sea surface cooling in the region providing heat energy to the hurricane (Bender and Ginis 2000; Shay et al. 2000; Cione and Uhlhorn 2003). The magnitude of the wind-induced cooling depends on the magnitude of the surface wind stress, the depth of the OML, and the temperature gradient at the base of the OML (e.g. Price 1981).
In order for a hurricane prediction model to capture the effect of wind-induced sea-surface cooling, it must be coupled to an ocean prediction model. Since an uncoupled hurricane model is restricted by a static SST valid only at the initialization time, the resulting hurricane forecast is necessarily unphysical except in rare cases where the wind-induced sea-surface cooling is negligible (Bender and Ginis 2000). Even if an uncoupled hurricane model forecast is accurate, the danger exists that the forecast is right for the wrong reason. For example, model air-sea fluxes for a given storm case may be tuned to optimize that particular storm’s forecast intensity, but if the flux parameterization is not changed, a model forecast for another storm that either (a) goes over a different part of the ocean or (b) goes over the same place at a different time when the SST is similar but the OML depth is different would likely be less accurate. So, even though proper parameterization of air-sea fluxes is extremely important and much work still needs to be done in this area, tuning of flux parameters should be done in tandem with accurate ocean initialization of a coupled model.
The Geophysical Fluid Dynamics Laboratory/University of Rhode Island coupled hurricane prediction system (hereafter GFDL model) has been run operationally at the NOAA’s National Centers for Environmental Prediction (NCEP) to forecast hurricane track and intensity since 2001 (Bender et al. 2007; Falkovich et al. 2005). From 2001 to 2006, yearly upgrades were made to the atmospheric and/or ocean components of the GFDL model, and one of the major improvements has been the ocean initialization procedure. One major challenge for proper ocean initialization in a coupled hurricane-ocean model is accurate representation of mesoscale oceanic features that do not follow an annual (or even a regular) cycle, such as the penetration of the Loop Current (LC) into the Gulf of Mexico (GoM) and the shedding (and perhaps reattachment) of Loop Current eddies (LCEs) (e.g. Oey et al. 2005). For these features, neither a monthly climatology nor a set of historical observations is sufficient, and the potential improvement afforded by near real-time observations is well-documented (e.g. Goni and Trinanes 2003; Goni et al. 2003). There are at least four ways to initialize these features more accurately: (1) Adjust the climatological LC position and structure and directly insert LCEs using a feature-based modeling procedure with real-time in situ and/or remotely-sensed observations, (2) continuously run an ocean model that is capable of predicting, with reasonable accuracy, the future position of the LC and LCEs, (3) nudge a model such as the one described in (2) towards the observations described in (1), or (4) some combination of (1)-(3). In the GFDL model initialization procedure, method (1) is employed.
Scientists at the NOAA’s Hurricane Research Division (HRD) occasionally measure ocean temperature profiles in the pre-hurricane environment via airborne expendable bathythermograph (AXBT) instruments dropped from aircraft (Cione and Uhlhorn 2003). As the thermistor on the AXBT descends from the sea surface, it provides accurate, high-resolution measurements (every 1.5 m) of the ocean temperature to a depth of ~350-450 m. In section 3, it will be shown how AXBT profiles can be used for (1) direct assimilation into the ocean model initialization procedure, (2) adjusting the position of oceanic features such as the LC or LCEs, and (3) validation of model temperature profiles.
The goals of this paper are (1) to describe the most recent version of the GFDL model ocean initialization, including a data assimilation strategy in the Gulf of Mexico, (2) to evaluate the accuracy of this initialization using in-situ data, and (3) to compare the results to an alternative ocean data assimilation technique currently available.