Post on 09-Feb-2021
transcript
© ECMWF October 12, 2015
Update on ECMWF forecast performance
Thomas Haiden, Martin Janousek
October 29, 2014
Contents
2
• Upper-air performance
• Surface parameters
• Model upgrade 2016 (cycle 41r2)
• High-impact weather
• Scale-dependence of skill
October 29, 2014
Upper-air forecast skill
3
October 29, 2014
Upper-air forecast skill
4
ERA-Interim
HRES
October 29, 2014
Upper-air forecast skill
5
HRES - ERA-Interim
October 29, 2014
Model intercomparison
6
October 29, 2014
Forecast skill: vertical correlation
7
Geer (2015)
← 500 hPa height
850 hPa temperature →
October 29, 2014
Upper-air ensemble forecast skill
8
𝐶𝑅𝑃𝑆𝑆 = 1 −𝐶𝑅𝑃𝑆𝐸𝑁𝑆𝐶𝑅𝑃𝑆𝐶𝐿𝐼𝑀
60%
40%
25%
October 29, 2014
Upper-air ensemble forecast skill
9
CRPS
D+4
D+6
D+8
October 29, 2014
Wave forecast: verification against buoys
10
Wave height Wind speed
October 29, 2014
Wave height – model intercomparison
11
DJF 2015-16
October 29, 2014
10 m wind speed
12
SDEV
BIAS
D+3
October 29, 2014
2 m temperature
13
SDEV
BIAS
D+3
October 29, 2014
2 m temperature bias at 00 UTC (DJF)
14
Winter 2014-15 Winter 2015-16
October 29, 2014
Precipitation
15
Simbach, Germany, 2 June 2016
October 29, 2014
Performance during individual events
16
Large-scale events that actually happened
October 29, 2014
Headline score for HRES precipitation: SEEPS
17
Skill in discriminating
between
• Dry
• Light
• Moderate-to-
heavy
Adapted to local
climatology
HRES
October 29, 2014
HRES precipitation
18
Skill in discriminating
between
• Dry
• Light
• Moderate-to-
heavy
Adapted to local
climatology
ERA-Interim
October 29, 2014
HRES precipitation
19
Skill in discriminating
between
• Dry
• Light
• Moderate-to-
heavy
Adapted to local
climatology
DIFFERENCE
October 29, 2014
HRES precipitation – other centres
20
Skill (ETS)
Frequency
bias
D+5
Extra-tropics
Thr=10mm/24h
October 29, 2014
ENS precipitation
21
CRPSS=10%
October 29, 2014
ENS precipitation – other centres
22
October 29, 2014
Model upgrade in March 2016 (Cycle 41r2)
23
Nhem Shem Tropics Europe N-Atl N-Amer N-Pac E-Asia Aust/NZ Arctic Antarctic
Upper-air: improvements 2-3%
Surface: up to 5% in the medium range
October 29, 2014
Improvements in surface parameters
24
uniform resolution
HRES 10 m wind speed ENS 2 m temperature
October 29, 2014
High-impact weather
25
Monitoring of forecast skill for extremesTropical cyclones
2m temperature 10m wind speed
24-h precipitation
Extreme forecast index (EFI), D+4
D+3
D+5
Position
Intensity
Speed
Position
spread/error
Heavy precipitation
ECMWF Newsletter 144 (Summer 2015)
Largest gains in skill at days 4-7
October 29, 2014
EFI for severe convection (CAPE, CAPESHEAR)
28
Verification against severe weather reports
(tornadoes, large hail, severe wind gusts)
Summer 2015, United States
65%
35%
October 29, 2014
29
Forecast skill horizon (schematic)
ECMWF Newsletter No. 145 (2015)
October 29, 2014
30
Forecast skill horizon: 500 hPa geopotential
Buizza and Leutbecher (2015)
Temporal averaging →
Spatial
averaging
↓
T120 T30
T7
October 29, 2014
Precipitation verification against gridded data (NEXRAD)
31
Frequency bias (Oct 2014 – April 2015) Fractions skill score FSS (April - Aug 2014)
16km16km
…1 2 n
i
j
Forecast Fn=5/25
Observed On=3/25
MSEn = (Fn-On)2
FSS concept
October 29, 2014
32
Total cloud cover forecast skill
ECMWF Newsletter No. 143 (2015)
October 29, 2014
Summary
33
• Upper-air performance continues to increase
• ECMWF maintains overall lead among global centres
• Challenge of 2 m temperature biases
• Significant improvements from Cycle 41r2
• Increased focus on verification of high-impact weather
• Forecast skill horizon: scale-dependent predictability
October 29, 2014
34
Deadline: 31 October 2016
Price: Invited keynote and paid attendance at the
7th International Verification Methods Workshop in
May 2017 (Berlin, Germany)
October 29, 2014
Model intercomparison DJF
35
October 29, 2014
Potential economic value of extreme wind forecasts
36
moderately
strong
extreme
extreme and
large-scale
see also ECMWF Newsletter No. 139 (2014)
October 29, 2014
EFI for severe convection (CAPE, CAPESHEAR)
37
Verification against severe weather reports
(tornadoes, large hail, severe wind gusts)
Summer 2015, United States
65%
35%
I. Tsonevsky (2016)