The ability to assimilate microwave radiance observations of the Earth's atmosphere affected by precipitation is investigated with an emphasis on channels that are sensitive to frozen hydrometeors. Mesoscale numerical weather prediction and radiative transfer models, as well as their corresponding adjoint models are utilized in sensitivity and data assimilation experiments. Special Sensor Microwave / Imager observations of Hurricane Bonnie (1998) are compared with model results that are transferred to radiance space with the radiative transfer model. Sensitivity results indicate that the model error in radiance space in areas of precipitation at the initial time is most dependent on the initial hydrometeor fields. At later forecast times, the model error is more sensitive to initial conventional model variables such as water vapor and temperature. When radiance data is assimilated, the model fields have better agreement with the observations compared to a control experiment for all observed channels at the initial time. However, at the next observation time 12 h later, the quantitative error measurements for the control and post assimilation forecasts are approximately the same value. Although this study demonstrates the ability to assimilate observations sensitive to atmospheric ice (as well as liquid) concentrations in a variational framework, important aspects such a background error correlation and bias have been ignored for simplification. More observations in a more complex data assimilation system will be needed in order to fully maximize the forecast impact of these observations.
An adjoint limited area numerical weather prediction model with multiple nests has been developed. The adjoint modeling system has the capability to pass gradient information from the finer spaced nest to the coarser spaced domain. Therefore, gradients of scalar functions calculated from small scale features can be computed with respect to the large scale model state. Sensitivity experiments were performed to show that the nested adjoint model produces physically meaningful gradients. Results from data assimilation experiments will be presented at the conference. An adjoint model such as the one presented here, could be an important tool for variational assimilation schemes that intend to make use of high resolution remotely sensed data.
Important issues involving the assimilation of rain-affected observations using an adjoint mesoscale modeling
system are addressed in this study. The adjoint model of the explicit moist physics parameterization is included
in the modeling system, which allows for the calculation of gradients with respect to the initial hydrometeor
concentrations (cloud water/ice, rain, snow, and graupel). Cloud-scale idealized four dimensional variational data
assimilation experiments demonstrate the benefit of assimilating precipitation information and the ability of the
adjoint model to produce useful gradients with respect to the hydrometeor fields. The agreement between model
fields and observations is greater (especially for the early forecast hydrometeor fields) when rainy observations
are incorporated into the assimilation process versus only assimilating conventional model data (windspeeds,
temperature, pressure). Additional data assimilation experiments are conducted with microwave radiances.
These data improve the initial precipitation structure of a tropical cyclone. These experiments are promising
steps for the incorporation of rain-affected observations in operational data assimilation systems.