On the Interpolation of Three-Dimensional Radar Observations to Flight Levels

In this study, different methods to interpolate radar data to a three-dimensional mosaic at standard flight levels are explored. The first employs a linear interpolation from the three-dimensional Multi-Radar/Multi-Sensor (MRMS) mosaics. This approach yields a consistent overestimate in reflectivity for high-altitude (i.e. greater than 32 kft) flight levels. The second method attempted - a straight mapping from the three-dimensional mosaics - also leads to an overestimate in reflectivity, although it is not as pronounced as with the first method. Ultimately, the high biases are due to beam broadening effects for the highest tilts from each radar. These effects are only slightly mitigated by using a nearest-neighbor approach as opposed to distance-weighted means when producing the mosaics for high-altitude layers. Therefore, a novel approach that makes use of digital-signal processing techniques is attempted. In this method, the vertical profiles of reflectivity from the three-dimensional mosaics are fitted with a curve using a Savitsy-Golay (SG) filter and then interpolated to flight levels. Using this approach, the overestimates in reflectivity are reduced, yielding much closer agreement with independent observations. There are, however, three important caveats that end users should carefully consider before adopting this idea. First, the vertical profiles of reflectivity require different tuning parameters that are dependent on their complexity. Preliminary work shows that one may be able to use the MRMS precipitation-type product as an initial classifier to designate the exact tuning parameters for the SG filter. Second, even with custom tuning, applying a filter to the observations does not always guarantee an improved analysis. There are some occasions, when applying the filter increases the errors in reflectivity. Last, using the SG filter constitutes a considerable computational load. A cost-benefit analysis of using the filter over the more computationally-efficient options discussed above is still pending.