QPF verification at the model grid versus at the stations

Scores commonly used to verify model predicted precipitation, such as the Bias Score, require that forecasted and observed precipitation be at the same location. Therefore, either the observed values must be brought to the model's grid, or the forecasted precipitation interpolated to the station positions. Although these analysis procedures have a large impact on the final scores obtained for verification, they are seldom discussed in the literature of forecast verification.

Verification scores for high-resolution (8 km grid-spacing) precipitation forecasts produced by a local model for the 1996 Atlanta Olympic Games will be presented. A comparison will be made between verification at the stations and verification at the model grid.

Results show that the Bias Score computed at the stations is larger than the one computed at the grid for the 0.25 mm threshold, but is lower than the one computed at the grid higher thresholds. Differences can be as large as a factor of 2.

Interpretation of the results, to be discussed in more depth at the Conference, indicate that the discrepancy is caused by differences in the density of stations compared to the density of model grid points, heterogeneity of station distribution, and to the radius of influence of each rain event in the analysis algorithm.