Extended Abstract (148K)7A.5
High resolution radar precipitation evaluation
Dennis A. Miller, NOAA/NWS, Silver Spring, MD, MD; and S. Wu, D. Kitzmiller, and F. Ding
Since the summer of 2008, the Weather Surveillance Radar-1988 Doppler (WSR-88D) has the capability to process base reflectivity data at 8 times higher spatial resolution than previously (nominally 250 m x 0.5º vs. 1000 m x 1.0 º). While this “super-resolution” is only being applied to some base radar products during the initial phase of implementation, we wished to determine what benefits may be achieved from its application to rainfall accumulation products such as those of the WSR-88D Precipitation Processing System (PPS). For example, higher radar spatial resolution may be expected to result in better detection of small-scale, heavy rain patterns and, in turn, better distribution of that rain to topographic features such as small-scale stream basins. However, these potential benefits might be offset by several factors that are known to cause discrepancies between QPE and the amounts and distribution of rainfall realized at the ground, including sub-beam advection, evaporation, and hydrometeor interactions. Indeed, the impact of these influences may be exacerbated when QPEs are analyzed at finer spatial scales.
We therefore carried out a study to perform rain gauge-radar (G-R) statistical comparisons over a span of discrete radar resolutions, ranging from that of the legacy WSR 88D to approximately that of the new "super-resolution". Utilizing a methodology analogous to the WSR-88D PPS, we generated one-hourly radar accumulation estimates and compared them to complementary measures from rain gauges in closely-spaced networks. Our primary dataset was from an experimental system deployed in central Florida during the summer of 1998, with the radar data supplied by NCAR's S-band, dual polarization, Doppler radar known as “S Pol”, and rain gauge data from NASA's Tropical Rainfall Measurement Mission-Ground Validation (TRMM-GV) network. Another dataset was obtained from the NOAA/National Severe Storms Laboratory's research-prototype version of the WSR-88D radar, located in Norman, Oklahoma, and coincident gauge reports from the Oklahoma Climatological Survey Mesonet, obtained during the warm seasons of 2004 & 2005. The S-Pol/TRMM Florida dataset was chosen as primary due to the tighter clustering of its rain gauges and the greater number of cases of heavy rainfall available. For the purposes of this study in assessing only the consequences of spatial variability, only rainfall estimates determined from the horizontally-polarized, base-radar reflectivity field via the traditional “Z-R” relationship were evaluated, and only over update-time scales consistent with those of the current volume scan patterns of the operational WSR-88D (typically ~5 minutes).
From the primary (Florida) dataset, after numerous quality control procedures were applied, a set of correlation and error measures were determined, with over 8,000 G-R pairs incorporated during 96 data hours. This was done across a span of (six) discrete radar-spatial resolutions, with the finest corresponding to the highest available in the S-Pol system (i.e. 150m x 1.0º) and the coarsest being close to that of the legacy WSR-88D (i.e. 900m x 1.0º). The intermediate resolutions (i.e. 300m, 450m, 600m, 750m), as well 900m, were determined via successive aggregation of the rain-rates determined at the 150m base resolution. Similar processing was applied to Z R rainfall estimates from the Oklahoma radar, with QPEs being estimated at 250-, 500-, 750-, and 1000-m x 1º spatial aggregation levels. One-hour QPEs were then correlated with collocated rain gauge estimates, with nearly 2,500 G-R pairs being incorporated into the dataset.
The analysis methodology undertaken included evaluation of all G-R pairs (point values) with non-zero, hourly accumulation under the entire radar umbrella; successive application of higher criteria for minimal rainfall amount; and application of a threshold for rainfall gradient. Furthermore, for some of these datasets, we performed, stratification of the G-R pairs into annular rings by distance from the radar; evaluation of densely packed sub-clusters of gauges (as closely spaced as 1-km) that could be considered representative of networks of multiple, small basins; and evaluation of gauge reports from those same sub-clusters now averaged together to produce mean areal precipitation amounts.
Examination of the results indicates that, overall, we found relatively small differences (of little or no statistical significance) in our statistical measures of QPE accuracy across the range of fine-to-coarse spatial resolution. We did, though, find somewhat better results at the finer resolutions in situations of: 1) large precipitation amounts; 2) large precipitation gradients; and 3) at mid–range distances from the radar, where neither low-altitude contamination (such as from residual clutter, AP, radar side lobes or biota) nor sub-beam effects (such as from wind advection) are likely to play a significant role.
Session 7A, Quantitative Precipitation Estimation
Tuesday, 6 October 2009, 4:00 PM-6:00 PM, Auditorium
Previous paper Next paper
Recorded presentation