WMOD00579

• Title: An Exploratory Forecast Experiment to Predict Supercooled Liquid Water in the Sierra Nevada
• Date: 11/1/1988
• Weather Modification - Doc Type: Report

3 . 0 DATA k'lD STATISTICAL 1.mTHODS

In this article we will objectively evaluate the skill in
forecasting the magnitude (i.e., concentration), onset, and
duration of SLW, the frontal type expected during the 12-hour
forecast period, and the contributions of a selected group of
forecast inputs to the final forecast.

Post season frontal analysis was used to verify the
predicted frontal type. These analyses largely relied on
satellite data and rawinsonde time cross-sections which have
been documented by Heggli and Rauber (1988). Verification of the
SLW was based on quality controlled radiometer data averaged by
2-hour time blocks. The data base used was the 2-minute average
radiometer data recorded to tape at KGV. If data were missing,
then the icing rate data from Squaw Peak were used by
quantitatively translating icing rate into SLW categories
(Henderson and Solak, 1983). In the 1985-86 season the icing
rate data were used 35% of the time for determining the liquid
water concentration but only 9% in 1986-87. (It is understood
. . ,~
that the lClng rate dataionly are a crude back-up for the
radiometer data.)

The verification analyses utilize the contingency table
approach and its summarizations by graphical and statistical
measures (e.g., bivariate frequency diagrams, conditional bias
plots, conditional probability of detection (CPOD), and
association measures; see Flueck, 1988 for more details). A
number 0= measures of association are available (e.g., Conover,
1971) but the True Skill Statistic (TSS) , which focuses on the
residuals from the expected counts due to chance, seemed most
appropriate (?lueck, 1987, 1988). This statistic produces a
value of +1.0 when all residuals (i.e., observed minus predicted
counts) reside on the left to right diagonal of a k x k table, -
1.0 whe~ all non-zero residuals reside on the opposite diagonal,

7