Thursday, 7 August 2003: 11:45 AM
Error statistics of VPR corrections in stratiform precipitation
Errors in surface rainfall estimates caused by ignoring the vertical profile of reflectivity, (VPR), have been assessed by simulating how fine resolution 3-D reflectivity measurements at close ranges are sampled by the radar at various ranges and heights. This approach has the advantage of isolating errors due solely to the VPR while ignoring all other sources of errors that would be present when comparing radar measurements with gauges. Over 200 hours of stratiform precipitation distributed among 21 events and with a distinct melting layer have been used to derive uncorrected and corrected 1-hour accumulations using various procedures. The latter include (a): "local" VPR obtained by a space-time averaging procedure over 30 minutes, (b): intensity dependent "climatological" VPR derived from the entire data set and (c) "event" VPR. The rainfall estimates, 0.2 km apart in height up to 5km and 40 km apart in range up to 210 km, are compared with the "ground truth" of the original near-range data at the lowest height of 1 km. The RMS error structure has thus been derived as a function of height and range and for verification areas ranging from (2km x 2km) to (20km x 20km). However, it is the errors at 1.5 km in height up to 90 km and along the height of the lowest elevation angle afterwards that are most relevant since this corresponds to the height of rainfall estimates of the Canadian radars. The stratification of the results in terms of the height of the bright band is essential in order to understand the influence of the bright band with range. The largest errors (> 100% at near ranges without correction) are encountered with lower bright bands that are both stronger and occurring at heights potentially used for surface estimates. The "local" VPR correction reduces these errors to 30 or 40% at 2 km resolution and by an additional 10% at 10 km resolution. This result is considered as a "best-case" scenario achievable only under conditions of homogeneity of the bright band at all ranges. Surprisingly, the "climatological" and "event" based VPR performed equally well, with errors about 10% higher than those from the "local" VPR. Errors for accumulations ranging from 5 minutes two 2 hours are also obtained.