Tropical observability and predictability

Abstract

Many studies have investigated tropical data assimilation in the context of global models or specifically for tropical cyclones, but relatively few have focused on the mesoscale predictability and observability of the general tropical environment. This work constructs an ensemble data assimilation system for the tropics using a state of the science mesoscale prediction model, and tests the effect of a sparse observational network of wind and moisture in constraining the estimate of the state. A perfect model framework is used as a necessary first step to ease interpretation of results. Ensemble assimilation allows for state-dependent error covariances, foregoing prederived balances and correlations and allowing for the use of the full nonlinear model. Boundary conditions are necessary for limited-area models, and the perturbed lateral boundaries and initial conditions are taken from a global ensemble using a non-perturbed sea surface temperature analysis. In the mesoscale model, this uniform surface had a profound effect on moisture levels in the lower levels, rapidly bringing the spread of vapor mixing ratio to near zero. Comparing the mesoscale forecast with a downscaled global model forecast showed that the interior solution was not completely dependent on the boundary conditions. Observing system experiments that assimilated synthetic moisture and wind component observations in the boundary layer and in the free atmosphere had a small effect on the state estimate when compared with an unconstrained control case. The largest improvement was in the upper troposphere obtained by observing upper-level moisture, but several analyses were degraded by the data, due in part to the sparse network and small localization radius.