It is shown that four-dimensional variational analysis (4DVAR) is a data assimilation technique that can be modified to calculate optimal perturbations for weather modification. Experiments demonstrate the ability of 4DVAR to calculate perturbations to influence the evolution of a simulated tropical cyclone. In "target" experiments, the Penn State/NCAR Mesoscale Model 5 4DVAR system determines the optimal atmospheric state trajectory which simultaneously minimizes the size of the initial perturbation and the difference (using a quadratic norm) between the new model state and a target in which the simulated cyclone has been repositioned. In experiments for Hurricane Iniki, the simulated hurricane was successfully repositioned after the 6 hour 4DVAR period, and then continued on a track parallel to its original track missing the Island of Kauai. Perturbations to the initial conditions are small relative to the hurricane.
In "damage cost function" experiments, 4DVAR simultaneously minimizes the size of the initial perturbation and an estimate of property loss that depends on wind speed. We had anticipated that the damage cost function would also have the effect of steering the hurricane away from built up areas, but this did not occur. Instead the hurricane surface winds decrease over the built up area at landfall. In damage cost function experiments the optimal perturbations usually include quasi-symmetric concentric features centered on the hurricane. It appears that the perturbation evolves as a wave disturbance that propagates to a focus at the hurricane center, and converts the kinetic energy of the hurricane into thermal potential energy at the appropriate time. The hurricane surface winds regenerate soon thereafter, so the effect is that the high intensity winds surrounding the hurricane eye "blink" off for a short period.