Uncertainty of Verification Data

Ed Tollerud

Forecast Systems Laboratory

Abstract presented for FAA-sponsored verification Workshop

July 30-August 1, 2002

When precipitation observations are used to verify model-simulated precipitation fields, uncertainty in these verification data has direct effects on the interpretation and even the credibility of scores. An assessment of these effects is necessary before subtle changes in scores can be confidently interpreted as real changes in model performance. Using hourly and daily gage-measured precipitation observations from real-time and retrospective climate networks, we describe several sources of observation uncertainty, including natural variability (which is high for precipitation), systematic differences in measurement practices at different observing sites, and data quality problems. We then quantitatively estimate the resulting scoring uncertainty by sub sampling techniques and by direct comparison of scores computed with independent sets of gage measurements.