Severe Weather Forecasts

1
Severe Weather Forecasts

• Ervin Zsoter

2
Outline

• Extreme weather forecast products, EFI and other
  forecast indices, extreme weather risk maps
• New EFI climate
• Verification of extreme weather forecasts
• Case studies

3
Extreme Weather Forecast Indices
Extreme Forecast Index (EFI) Scaled integral
distance between the forecast and climate
probability distributions
Ff(p) p
Shift of Tails (SPS) Proportional distance
between the forecast and climate distributions in
the meteorological variable space at the
p-percentile
SPS (p)
p (Fc(p))
Qc(p)- Qc(1)
SOT(p)
Qf(p)- Qc(1)
Shift in Probability Space (SPS) Distance
between the forecast and climate distributions in
the probability space at the p-percentile
SPS (p)
4
2m temperature EFI-s
12 UTC EFI
Daily max EFI
Daily min EFI
5
Examples with the Extreme Weather Indices
SOT an SPS may highlight areas relatively far
from the EFI maximum
2m temperature
120-hour total precipitation
6
Multi-parameter Extreme Weather Risk map, based
on the EFI
http//w3ec2.ecmwf.int/metops/efitest/
7
Extreme Weather Risk map
8
Distribution diagrams attached to the Extreme
Weather Risk map

• Evolution of EPS forecast distributions,
  relative to the climate
• EFI is indicated as a measure of the level of
  extremety
• Currently prepared for the location of maximum
  EFI in a 1010 degrees gridbox
• Different diagram for different verifying lead
  times (D1 D5)
• For 2m temperature, 10m wind speed and 24h total
  precipitation

9
Recent progress Severe Weather Risk map for the
Monthly Forecast System
10
New Medium range Model Climate

• Running an additional EPS control re-forecast
  suite in dynamical adaptation mode (48h forecast)
• Initial conditions from ERA40 (1971-2000)
• Use latest resolution/ physics
• 12 UTC daily runs (for 30 years each day)
• Same post-processing as for EPS (surface a few
  pressure and PV levels)
• Cost is only like adding six 10-days Control runs
  per day
• Will allow an immediate adaptation of the EFI to
  any EPS model upgrade, and also an extension of
  the EFI (and other calibrated probabilistic
  products) to other parameters than rain, T2m and
  wind
• In MARS Ensemble Forecast Atmospheric Hindcast
  efhc and Ensemble Forecast Wave Hindcast
  ewhc.
• Operational since 1 February 2006 (high
  resolution upgrade)

11
New EPS Control Climate for the EFIExample for
Reading, diff. distribution sampling window from
1 to 15 days
12
EFI with old and new model climatology
3 calendar month of archived EPS forecasts from
previous years
EFI with old climate
EFI with new climate
EPS control model reruns for years 1971-2000, 31
days from each year
13
EFI with old and new model climatology
Difference between the old and new EFI climate
(measured by the EFI)
Observed anomalies July 2003-2005 referenced to
ERA-40 (1971-2000)
14
Verification of Extreme Weather Forecasts

• Follow up of the study made by Francois
  Lalaurette Federico Grazzini in 2005 (for 24h
  total precipitation EFI, 2003 Oct 2005 May)
• Verification of the EFI (based on the old
  pseudo-climate and the new EPS control
  re-forecast model climate) and also the EPS
  probability for extreme events
• EFI is regarded directly as probability and
  truncated at 0 (no transformation is applied)
• For parameters of 2m maximum temperature and
  24-hour total precipitation
• D1, D3, D5, D7 and D9
• For period of 2005 July 2006 May
• Observational climate is based on the Climate
  Atlas of Europe (Météo-France) and the new EPS
  control model climate

15
Definition of the extreme events

• The atlas contains monthly means, upper and lower
  quintiles, daily extreme values and different
  frequency indices for temperature, precipitation,
  wind gust and sunshine duration for 700
  selected European stations. The climate period is
  1971-2000.
• Monthly extreme events are defined as at least
  as extreme as the 99.5 percentile of the daily
  climate distribution based on sampling over the
  month
• Average return period of 6-7 years. Apprx. the
  5th most extreme case (99.5 percentile -gt 1/200)
  in the monthly sample (size 900,
  30days30years).
• It is assumed that the model climate behaves
  similarly to the observed climate (similar shape
  characteristics of the distributions).

16
Definition of the extreme events

• Monthly thresholds for extreme events (DayObs995)
  
• MonObsQ80 (DayEpsQ995 / MonEpsQ80)
• MonObsQ80 MonEpsQ80 Upper quintile of
  distribution of monthly means in the observation
  EPS model climate (per station, per month)
• MonEpsQ80 Upper quintile of distribution of
  monthly means in the EPS control re-forecast
  climate (per station, per month)
• Daily extreme thresholds are created by
  interpolating between adjacent months
• Using GTS synop 1700 (tp24) 2500 (tmax24)
  extreme events were found during the period July
  2005 May 2006

17
Examples of extreme thresholds
24-hour total precipitation May 16th
2m daily maximum temperature July 16th
18
EFI based on new versus old climateROC, 24-hour
total precipitation 2005 July 2006 May
F/(RF)
F/(HF)
19
EFI based on new versus old climateROC, 24-hour
total precipitation 2005 July 2006 May
F/(RF)
F/(HF)
20
EFI based on new versus old climateROC, 24-hour
total precipitation 2005 July 2006 May
F/(RF)
F/(HF)
21
EFI based on new versus old climateROC, 24-hour
total precipitation 2005 July 2006 May
F/(RF)
F/(HF)
22
EFI based on new versus old climateROC, 24-hour
total precipitation 2005 July 2006 May
F/(RF)
F/(HF)
23
EFI new versus EPS probabilitiesROC, 24-h total
precipitation, 2005 July 2006 May
F/(RF)
F/(HF)
24
EFI new versus EPS probabilitiesROC, 24-h total
precipitation, 2005 July 2006 May
F/(RF)
F/(HF)
25
EFI new versus EPS probabilitiesROC, 24-h total
precipitation, 2005 July 2006 May
F/(RF)
F/(HF)
26
EFI new versus EPS probabilitiesROC, 24-h total
precipitation, 2005 July 2006 May
F/(RF)
F/(HF)
27
EFI new versus EPS probabilitiesROC, 24-h total
precipitation, 2005 July 2006 May
F/(RF)
F/(HF)
28
EFI new versus EPS probabilitiesROC, 2m maximum
temperature 2005 July 2006 May
F/(RF)
F/(HF)
29
EFI new versus EPS probabilitiesROC, 2m maximum
temperature 2005 July 2006 May
F/(RF)
F/(HF)
30
EFI new versus EPS probabilitiesROC, 2m maximum
temperature 2005 July 2006 May
F/(RF)
F/(HF)
31
EFI new versus EPS probabilitiesROC, 2m maximum
temperature 2005 July 2006 May
F/(RF)
F/(HF)
32
EFI new versus EPS probabilitiesROC, 2m maximum
temperature 2005 July 2006 May
F/(RF)
F/(HF)
33
EFI new versus EFI old EPS probReliability
diagram, 24-hour total precipitation 2005 July
2006 May
F/(RF)
F/(HF)
34
EFI new versus EFI old EPS probReliability
diagram, 24-hour total precipitation 2005 July
2006 May
F/(RF)
F/(HF)
35
EFI new versus EFI old EPS probReliability
diagram, 24-hour total precipitation 2005 July
2006 May
F/(RF)
F/(HF)
36
EFI new versus EFI old EPS probReliability
diagram, 24-hour total precipitation 2005 July
2006 May
F/(RF)
F/(HF)
37
EFI new versus EFI old EPS probReliability
diagram, 24-hour total precipitation 2005 July
2006 May
F/(RF)
F/(HF)
38
Extreme event case study - I Swiss flood -
heavy precip in August 2005
Observed 24-hour total precipitation 22 August
2005
Extreme events
39
Swiss flood - heavy precip in August 2005
D9
40
Swiss flood - heavy precip in August 2005
D7
41
Swiss flood - heavy precip in August 2005
D5
42
Swiss flood - heavy precip in August 2005
D3
43
Swiss flood - heavy precip in August 2005
D1
44
Extreme event case study - II Heat wave in West
Europe - in October 2005
Observed 24-hour maximum temperature, 30 October
2005
Extreme events
45
Heat wave in West Europe - in October 2005
D9
46
Heat wave in West Europe - in October 2005
D7
47
Heat wave in West Europe - in October 2005
D5
48
Heat wave in West Europe - in October 2005
D3
49
Heat wave in West Europe - in October 2005
D1
50
Summary - I

• After careful consideration and survey for
  demands, the range of extreme forecast products
  (forecast indices, maps) can be extended
• New extreme indices (SOT, SPS) ?!
• New parameters (Tmax, Tmin, Waves, etc) ?!
• Extended forecast range (D6 to monthly range
  VAREPS) ?!
• The new ERA40 based EPS control model
  re-forecasts provide more reliable base for any
  extreme forecast product at present or in the
  future
• Further developments (possible use of EPS
  re-forecasts made for calibration purposes) ?!
• The first set of verification results are
  encouraging
• Need for detailed, quality controlled,
  representative observation climate

51
Summary - II

• The first set of verification results are
  encouraging
• The better climate sampling with the 30-year
  model climate seems to be reflected in the
  verification result by slightly better ROC for
  the new EFI
• In terms of resolution the EFI also seems to over
  perform little bit the raw EPS probabilities for
  extreme events
• However statistical calibration is heavily needed
  in order to decrease the strong over forecasting
  tendency of the EFI (taken directly as
  probability)
• Lot of more work is needed to further investigate
  the characteristics and value of these extreme
  forecast products