USING 4D-VAR TO MINIMIZE PROPERTY DAMAGE IN AN MM5-SIMULATED HURRICANE R. N. Hoffman, J. M. Henderson, and S. M. Leidner Atmospheric and Environmental Research, Inc. 1 INTRODUCTION Hoffman (2002) has discussed the possibility of controlling the global weather by introducing a se- ries of small, precisely calculated perturbations. In the preliminary work reported here, we take a small component of the global weather control system of Hoffman (2002) and put it into practice, in an admit- tedly crude manner. We demonstrate the ability of a currently available data assimilation technique, four- dimensional variational analysis (4d-VAR), to esti- mate the perturbations needed to locally “control” the weather. The motivation to modify the weather is especially strong in the case of tropical cyclones. The AMS policy statement “Hurricane Research and Fore- casting” (AMS 2000) summarizes the hazards of tropical cyclones over land: loss of life and nearly $5 billion (in 1998 dollars) annually in damage due to the storm surge, high winds, and flooding. The economic cost continues to rise due to growing pop- ulation and wealth in coastal regions. According to the Hurricane Andrew Reanalysis Project (HRD 2002), Hurricane Andrew (1992) is only the third Category 5 (Simpson 1974) hurricane to make landfall in the US since records began. Damage of approximately $26 billion was inflicted on southern Florida and Lousiana, and 23 people were killed. The storm first made landfall on the US mainland near Homestead at 0905 UTC 24 August with a central pressure of 922 hPa. Maximum sus- tained winds at landfall were estimated at 75 ms . Andrew is a fine example of a storm that would have had less impact, in terms of wind damage, on the US coastline if the track had been displaced farther south by as little as 150 km. To this end, we apply the Penn State/NCAR Mesoscale Model 5 (MM5) 4d-VAR-system with the goal of repositioning a simulation of Hurricane An- drew farther to the south. MM5 produces very de- tailed and accurate simulations of tropical cyclones when high resolution and advanced physical pa- rameterizations are used (e.g., Liu et al. 1999). Corresponding author: R. N. Hoffman, Atmospheric and En- vironmental Research, Inc., 131 Hartwell Avenue, Lexington, MA 02421. e-mail: rhoffman@aer.com However, in the current experiments, coarse res- olution is used for computational efficiency. For the purpose of our demonstration, the unperturbed MM5 simulation is taken to be reality. 2 MESOSCALE MODEL AND DATASETS The MM5 is described in detail by Grell et al. (1994). In our experiments, the dimensions of the MM5 computational grid are 200 200. The horizontal resolution is 20 km with ten sigma layers in the ver- tical from the surface to 50 hPa. Only basic physi- cal parameterizations are currently available in the MM5 4d-VAR system: MRF PBL scheme, Kuo cu- mulus convection, stable explicit moisture, and sim- ple radiative transfer. The first simulation we present is a 24-h integra- tion with initial and boundary conditions provided by the 6-hourly NCEP-NCAR reanalysis fields (Kalnay et al. 1996); the inital time of this “unperturbed” run is 1800 UTC 23 August 1992. A vortex of inten- sity equal to the observed was bogussed into the initial conditions using the NCAR/AFWA MM5 tropi- cal cyclone bogussing system (Davis and Low-Nam 2001). At the initial time (Fig. 1), Andrew is a Cat- egory 4 storm. The second simulation (hereafter, “controlled”), with an initial time of 0000 UTC 24 August 1992, is an 18-h integration using initial con- ditions modified by 4d-VAR. The 6-h forecast fields from the unperturbed simulation provided input to 4d-VAR. The simulations cover Andrew’s westward translation towards south Florida. 3 CALCULATION OF PERTURBATIONS The MM5 implementation of 4d-VAR is described by Zou et al. (1997). 4d-VAR can be used to find the smallest global perturbation, as measured by the a priori, or background, error covariances, at the start of each data assimilation period so that the so- lution best fits all the available data. 4d-VAR solves this complex nonlinear minimization problem itera- tively (we permit 10 iterations), making use of the