Lower and middle-tropospheric observations are disproportionately sparse, both temporally and geographically, when compared to surface observations. The limited density of observations is likely one of the largest constraints in numerical weather prediction. Atmospheric measurements performed by the Tropospheric Airborne Meteorological Data Reporting (TAMDAR) sensor of humidity, pressure, temperature, winds aloft, icing, and turbulence, along with the corresponding location, time, and altitude from built-in GPS are relayed via satellite in real-time to a ground-based network operations center.

Since December 2004, the TAMDAR sensors have been operating on a fleet of 63 Saab 340s operated by Mesaba Airlines in the Great Lakes region as a part of the NASA-sponsored Great Lakes Fleet Experiment (GLFE). Equipage of sensors on additional aircraft (>112), including regional jets, across the continental US and Alaska is well underway. Currently, Horizon, Republic, Chautauqua, Shuttle America, PenAir, and Mesaba Airlines, as well as a few research aircraft are hosting the TAMDAR sensor and transmitting data. Upon completion of the 2008 installation, more than 5000 daily sounding will be produced in North America.

A 3-way study is conducted where various cases are analyzed using 3DVAR, 4DVAR, and RT-FDDA assimilation methods, and where the three experimental (control) runs include (withhold) TAMDAR data. Forecast skills are verified against other observing platforms (e.g., RAOBs and ASOS), as well as radar, Stage-IV precipitation analysis, and North America Regional Reanalysis (NARR).

The objective of this study is to ascertain the optimal relationship between assimilation method, model domain configuration, model physics and various parameterization choices that best utilize the TAMDAR data.

Preliminary results suggest that the addition of TAMDAR data improves the experimental simulations for certain output parameters. Results also suggest that more accurate lower and mid-tropospheric data assimilation and model initializations feed back to the surface and spread in the full 3D domain volume, thus improving surface parameters, as well as quantitative precipitation forecasts.