Assessment of Operational Airborne-Radar Quality Control Methods for NOAA Hurricane Reconnaissance

Airborne Tail-Doppler Radar (TDR) data collected during NOAA hurricane reconnaissance missions consist of both meteorological data (MD) and non-meteorological data (NMD). In order for TDR data to be useful in NOAA operations and research, the NMD must be filtered out beforehand. Prior to 2005, filtering NMD required manual editing, taking an expert radar meteorologist several weeks to complete just one flight leg. In 2005, the current automated rules-based quality control (QC) method, NOAA-QC, was developed. Having eliminated the need for manual editing, the TDR data are now available to be used operationally by the National Hurricane Center (NHC) and assimilated into the Environmental Modeling Center’s (EMC’s) hurricane models. Although experience proves most NMD from the current TDR system are removed, NOAA-QC has never been quantitatively assessed on its performance. The main goal of this study is to quantitatively evaluate NOAA-QC’s performance in comparison to manually-edited TDR datasets. Overall, we found that NOAA-QC is accurate in terms of identifying NMD, removing over 99%, yet frequently misidentifies MD as NMD, removing 25-30% depending on storm structure. The most MD are removed from the boundary layer and upper troposphere, areas that provide crucial information on tropical cyclone structure and dynamics. These findings support a shift towards a new machine-learning QC (MLQC) method that would retain significantly more MD than the current rules-based method. We briefly describe initial steps in training MLQC to identify NMD unique to the current airborne TDR system that have proven most challenging to removal by both NOAA-QC and manual QC.