P56  Points and Pixels: Assessing the Impact of Observation Choices on Verification Results

Tollerud, Edward, National Oceanic and Atmospheric Administration/ESRL, Tara Jensen, NCAR/RAL, Tressa Fowler, NCAR/RAL, John Halley Gotway, NCAR/RAL, Seth Gutman, NOAA/ESRL, Kirk Holub, NOAA/ESRL, Paul Oldenburg, NCAR/RAL, and Barb Brown, NCAR/RAL

Selection of observations for verification of numerical forecasts presents several important issues relating to verification uncertainty. Representativeness remains the most significant, particularly when using point values versus datasets representing analyzed areal estimates. Rain gauge and radar quantitative precipitation estimates (QPE) are good examples of this kind of error. Significant differences can result from the nature of the analysis method. Understanding the impact of the analysis scheme on the magnitude of verification score differences is a critical step toward developing verification comparisons that are fair and useful.

During recent Hydrometeorolgy Testbed winter exercises, a set of WRF-based regional ensemble focused on heavy precipitation forecasts for the American River basin.  Since improvement in quantitative precipitation forecasts (QPF) was the principal objective of the exercises, comparison of QPE from the StageIV product with individual gauges was performed. Different accumulation periods (24h and 6h) were also compared. In both cases, gauge-based verification scores were superior to those from the analyses. This past season, evaluation was extended to integrated water vapor (IWV) forecasts verified with point GPS measurements and LAPS analysis of IWV observations; these results showed the opposite tendency of better scores for the analyses. Reasons for these two disparate results will be suggested.