MODEL QPF GUIDANCE VERIFICATION BY REGIONS
TEXT UPDATED MAR 19, 2003
NEW REGIONS: TEXT AND REGION MAP UPDATED JAN 13, 2004
GRAPHICS UPDATED WEEKLY AS OF FEB 14, 2005
ADDED SREF MEAN, REMOVED ODDS RATIO, POD, & FAR; TEXT UPDATED MAY 5, 2006
This web site compares the performance of the NAM and GFS for the prediction
of 24-h accumulations of precipitation ending at 12 UT. The continental
US has been broken into 14 regions in consideration of approximate uniformity
of climatology and terrain. The verification was done on a 40-km grid.
Graphical displays are generated for each region,
each of six projection times (24, 36, 48, 60, 72, and 84 hours), and each
of two time ranges (beginning one month and three months prior to the present).
There are a total of 156 graphics to look at. A new set of graphics is
generated every week.
Two measures of performance are displayed on each graph as a function
of accumulation threshold. These measures are computed from the following
traditional contingency table of event counts over a multitude of forecasts:
|
Observed
|
Not observed
|
Forecast
|
a (hit)
|
b (false alarm)
|
Not Forecast
|
c (miss)
|
d (correct rejection)
|
In the following discussion, N=a+b+c+d. The computation of the performances
measures displayed is described below:
-
Equitable Threat Score = the number of correct forecasts minus number expected
by chance divided by the same quantity plus the number of erroneous forecasts
= [a - (a+b)(a+c)/N] / [a + b + c - (a+b)(a+c)/N]. This score ranges
from less than 0 to 1; a perfect score is 1.
-
Bias = the ratio of the total number forecast to the total number observed
= (a+b) / (a+c).
In the graphs, the bias values are plotted along the axis on the right
side of the plot.