Adjoint data assimilation of idealized microburst observations

A technique is examined whereby a numerical model is adjusted to match several radar-observed radial wind fields of a developing microburst. Unobserved model variables are reconstructed to produce a first order representation of the two-dimensional horizontal wind and mass fields of a microburst and the thunderstorm downdraft which produces it.; Adjoint variational assimilation is employed. A dynamically simple numerical model is used, the nonlinear shallow water equations. A strong constraint is placed on the model which requires the evolving model winds to match the radar-observed winds at evenly spaced times throughout the assimilation period. A cost function is constructed which measures the error in this match. Observed winds are simulated from a previous integration of the model.; The system successfully reconstructs the initial conditions of the model such that the evolving modelled velocity field closely approximates the observations. Unobserved fields are reproduced, including the thunderstorm downdraft. Six spatially complete observed radial velocity fields are required to produce a high quality match. It is shown that typical random measurement errors in the radar velocity observations are removed from the solution. Data void regions typical of radar observations do not substantially degrade the solution. Sparse observations in space or time degrade the solution slightly. The effect of an inaccurate model is examined by varying the acceleration of the modelled flow compared with the observed wave. With mismatched speeds, the agreement between incorrectly modelled and observed states is poor, leading to a compromised solution.; Internal choices in the assimilation system are important. The first guess of the initial model state will determine the required adjustments. The model-observation is fed to the adjoint model. This information propagates to other variables to direct the adjustment of each variable. Although they reduce the cost function, these adjustments are often suboptimal and need to by undone later. Bounding feasible and infeasible values of initial model state variables increases the rate of convergence toward a good model-observation match.; Short radar-microburst distances produce better matches between modelled and observed winds. This method could be included in a short term forecast of microburst wind hazards near airports.