Methodologies for converting
NWP outputs into QPF and PQPF
PQPF
Krzysztofowicz approach
MODEL OUTPUT STATISTICAL
METHODS THAT CAN BE USED TO DEVELOP PROBABILITIES
Brier skill scores for six
statistical methods for 3 precipitation thresholds
Non linear techniques like
logistic regression improve forecasts for the higher probabilities (figures
from Applequist et al.)
VERY USEFUL FOR LIGHTER
THRESHOLDS BUT TREND TOWARDS CLIMATOLOGY AT LONGER RANGES
STATISTICS METHODS
Ensemble forecasting
Bias (Forecast
precipitation/Observed precipitation) for various models during September 2000
for northeast
Perfect Prog approach,
Assume members are unbiased. Can show
where at least 60% of the members exceeded a threshold.
Man-machine mix
A better way of using
enemble forecasts to generate probability forecasts
The brier scores for higher
thresholds were higher for forecasts applying various statistical methods to
ensemble forecasts than those from the NGM MOS
If ensemble forecasts are
used, which more cost effective
Slide 16
Accuracy of model forecasts
decreases rapidly as threshold increases and as the scale of the event
decreases
Because higher amounts or
thresholds are relatively rare, it is difficult getting a big enough sample to
calibrate forecasts using traditional statistical techniques. Verification of
24 hour QPF for various thresholds
For extreme events is there
a better way?
Or is it? How Good Are the Forecasts?
What about assessing
probability of precipitation occurring anywhere within a circle or hexagon? The
probabilities would be influenced by the density of observations, but your probabilities might be high enough
to allow emergency managers to assess the risk within their area.
Then how do we approach
forecasting QPF , Probabilistically?
For high end, rare events