An Evaluation of Quantitative and Probability-of-Precipitation Forecasts during the 1984–85 Warm and Cold Seasons

View More View Less
  • 1 Department of Atmospheric Sciences, University of Illinois, Urbana, IL 61801
© Get Permissions Rent on DeepDyve
Restricted access

Abstract

Objective precipitation guidance has been evaluated for specific regions within the continental United States during the period 1984–85. Cold-season precipitation probability skills for seven locations range from 52% at 12–24 h forecast range to about 21% at 36–48 h range. While these skills show the probability forecasts to be generally useful, an examination of forecasts with an absolute error of greater than 0.5 reveals this smaller sample to contain a disproportionately large number of observed precipitating events. This suggests that large-error-precipitation-probability forecasts have an unexpectedly large number of essentially unforecasted precipitation events, rather than false alarms. Warm-season precipitation probability skills are generally lower and show more variability within a given forecast range, with values ranging from 38 to 6% at 24–36 h range.

Limited-area Fine-Mesh (LFM model, cold-season, quantitative precipitation forecasts (QPFs) for specific cities show no skill beyond a 12-h forecast range. This loss of skill is associated with statistically significant overprediction of precipitation. However, to account for a coding error in the LFM model, we recomputed our statistics by halving all QPFs. The skills of these forecasts rose to respectable overall levels of 18.2, 14.8, 13.1 and 4.0% for the respective forecast ranges of 0–12, 12–24, 24–36 and 36–48 h. These revised forecasts have eliminated all suggestion of precipitation overprediction, and instead show a systematic underprediction of precipitation.

Cold-season, area-averaged QPFs taken directly from the LFM show a loss of skill against the climatological control forecast beyond 24 h. When we halved all forecasts, our area-averaged results showed, generally, more respectable overall skills of 9.3, 20.8, 16.9 and 5.2% for the respective forecast ranges.

Warm-season point and area-averaged QPFs show no skill against the climatological control forecast for any of the four forecast ranges out to 48 h. Statistically significant precipitation underprediction is found for the raw warm season QPFs. When the forecasts are halved, the skills deteriorate to even lower values and systematic underprediction of precipitation is more prevalent.

Abstract

Objective precipitation guidance has been evaluated for specific regions within the continental United States during the period 1984–85. Cold-season precipitation probability skills for seven locations range from 52% at 12–24 h forecast range to about 21% at 36–48 h range. While these skills show the probability forecasts to be generally useful, an examination of forecasts with an absolute error of greater than 0.5 reveals this smaller sample to contain a disproportionately large number of observed precipitating events. This suggests that large-error-precipitation-probability forecasts have an unexpectedly large number of essentially unforecasted precipitation events, rather than false alarms. Warm-season precipitation probability skills are generally lower and show more variability within a given forecast range, with values ranging from 38 to 6% at 24–36 h range.

Limited-area Fine-Mesh (LFM model, cold-season, quantitative precipitation forecasts (QPFs) for specific cities show no skill beyond a 12-h forecast range. This loss of skill is associated with statistically significant overprediction of precipitation. However, to account for a coding error in the LFM model, we recomputed our statistics by halving all QPFs. The skills of these forecasts rose to respectable overall levels of 18.2, 14.8, 13.1 and 4.0% for the respective forecast ranges of 0–12, 12–24, 24–36 and 36–48 h. These revised forecasts have eliminated all suggestion of precipitation overprediction, and instead show a systematic underprediction of precipitation.

Cold-season, area-averaged QPFs taken directly from the LFM show a loss of skill against the climatological control forecast beyond 24 h. When we halved all forecasts, our area-averaged results showed, generally, more respectable overall skills of 9.3, 20.8, 16.9 and 5.2% for the respective forecast ranges.

Warm-season point and area-averaged QPFs show no skill against the climatological control forecast for any of the four forecast ranges out to 48 h. Statistically significant precipitation underprediction is found for the raw warm season QPFs. When the forecasts are halved, the skills deteriorate to even lower values and systematic underprediction of precipitation is more prevalent.

Save