|
Each 'storm-total' snowfall contest must have at least seven entries to be a valid. The scoring system for the
'regular' season is designed to determine objectively
which forecast had the best overall synoptic quality.
An accurate ... high-quality ...
synoptic-scale snowfall forecast is defined as one capturing the magnitude
and distribution of the storm/s total snowfall.
Forecast quality is evaluated by calculating its 'goodness of fit'
between all forecast
and observed snowfall amounts.
The scatter plot/s black diamonds
show the forecast snowfall (x-axis) plotted
against the observed snowfall (y-axis). The dashed line
represents the
forecast/s 'goodness of fit.' The magenta points represent a perfect forecast ...
where each station forecast matched each station
observation.
The 0.82 RČ value indicates the
forecaster accounted for 82% of the variability
in the observed snowfall amounts. A 'perfect' forecast
would have an RČ value of 1 ...
meaning 100% of the observed snowfall/s
variability was accounted for by the forecast.
The forecast error statistic is the Sum of Squared
Errors (SUMSQ).
The arithmetic difference between
the Forecast and Observed snowfall at each
station is squared and the squared errors for all stations are
summed. This skill
measure rewards (penalizes) station forecasts with small (large)
errors.
The smaller the SUMSQ ... the
better
the synoptic-scale forecast.
If the absolute difference between your
Forecast snowfall and the Observed snowfall
is:
one (1) inch = one (1) error point
two (2) inches = four (4) error points
three (3) inches = nine (9) error points
Each forecaster/s SUMSQ Error score is normalized to a
Z-Score
for use in
calculating Interim and End-of-Season standings.
Forecasts with normalized Z-scores greater than or equal to 2.5 standard
deviations above the mean are statistical outliers and may be excluded
from the final
Z-Score calculation.
If Snow is Observed and Snow is Not Forecast...
then the Forecaster/s Snowfall Amount is set to
'0' for that station.
If Snow is Forecast and Snow is Not Observed...
then the Observed Snowfall Amount is set to '0'
for that station.
If Snow is not Forecast and Snow is Not
Observed...
then the Forecaster is not scored for that
station.
Other reported forecast statistics ... such as average
absolute error (AAE) ...
total absolute error (TAE) ... total snowfall error (STP) ... RČ
(coefficient of determination)
... and 'batting averages' are 'measures of forecaster skill' but do not
factor into
forecast rankings. |
