Danish Meteorological Institute

1
Danish Meteorological Institute

• EPS Forecast of Weather Scenarios and Probability
  Presented at www.dmi.dk
• by
• Michael Steffensen
• Acknowledgments
• Gorm Dybkjær Danish Meteorological Institute, who
  made much of the work in building the product.

2
Danish ByVejr (CityWeather) consists of three
different forecast approaches

• First 24 hours (day 1) are based on the Danish
  version of HIRLAM updated 4 times a day
• Day 2 to 5 based on ECMWF T511 updated once a day
• Day 7 to 9 based on ECMWF EPS updated once a day

3
ByVejr is located at the www.dmi.dk frontpage
4
Each red dot is a city with a local forecast.
5
The upper graph is HILAM and lower graph is T511
HIRLAM
T511
6
Presentation of EPS forecast should

• Look like the previous forecast, days in columns,
  weather symbol, temperature and wind speed
• Show the probability for the occurrence of the
  combination of weather parameters shown in the
  column, called a weather scenario
• Show the three most likely alternative weather
  scenarios
• The result is shown in the next three slides

7
probability
day temperature
night temperature
wind speed
8
(No Transcript)
9
(No Transcript)
10
Probability of the combination of parameters
shown in a scenario

• Is determined for temperature and precipitation
  together, since these are the most important for
  the public
• Three regimes for precipitation are used.
  lt1mm/24hr, between 1 and 5mm/24hr and gt5 mm/24hr
  meaning no rain, some rain and more rain
• Temperature is determined in a 5 degree interval.
  
• The most probable 2D square is choosen as first
  scenario
• The intervals for night temperature and wind
  speed are determined as the spread of the EPS
  members making up the respective 2D square

11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
Probability of the combination of parameters
shown in a scenario

• The second most probable 2D square not covering
  the first square is chosen as the second scenario
• The third scenario is chosen similarly

16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
Verification

• Next slide show the ranked histogram with the raw
  data in blue an calibrated data in red.
• The ranked histogram for the raw data is very
  asymmetric.
• Bin 52 for raw data is very high indicating that
  the ensemble generally is to cold.
• This is corrected by a linear translation of all
  members, determined by the difference between the
  Kalman filtered ensemble mean and the ensemble
  meanshift KalltEPSgt - ltEPSgt
• Precipitation is also asymmetric, but not
  corrected.

20
(No Transcript)
21
ltEPSgt
KalltEPSgt
22
(No Transcript)
23
Verification, Reliability

• The most popular probabilities are around40, 50
  ,60 and 70 percent, because of the combination of
  parameters
• Though still overestimating probabilities, the
  translation of the ensemble gives some
  improvements in reliability
• The overestimation of probabilities are probably
  due to too small spread of the ensemble

24
translated ensemble
raw ensemble