B. P. Yadav - National Centre for Medium Range Weather Forecasting

Operational Weather Forecast Verification
at India Meteorological Department
B. P. Yadav, Naresh Kumar, L.S. Rathore
OUTLINE
• Introduction
• Operational Forecast Services
• Performance of Operational Forecasts
• Conclusions
Regional Centres
Delhi, Kolkata, Guwahati,
Nagpur, Mumbai, Chennai
NWFC Delhi/ WF Pune
Six Regions
19 Meteorological Centers
North Region
East Region
North East Region
Central Region
West Region
South Region
Pune Centre for
Research, Training,
Climatology, Agromet and
surface instrumentation
Nowcast, City and Tourism Forecast
Short Range Forecast & Warnings
District /Block level
Quantitative Medium Range Forecast
Other forecast and warnings
 Extended Range Prediction (Week 1, Week 2)
 Long Range Forecast for Southwest &
Northeast Monsoon rainfall and winter
Precipitation
 Cyclone track, intensity & landfall forecast
 Fog Forecast
 Venue Specific Forecasts
 Highway, Expedition, pilgrimage forecast
Basic measures/scores
Correct Non-occurrence
Hit Rate, Probability of Detection
a
H = POD =
(a + c)
Range: 0 to 1,
Perfect score = 1
C − NON =
d
(b + d )
Range: 0 to 1
Perfect score = 1
False Alarm Rate
•Miss Rate, complement score
b
F=
(b + d )
M =1-H= c/(a+c)
Range: 0 to 1
Perfect score = 0
Frequency Bias Index (Bias)
FBI = B =
( a + b)
(a + c)
∞
Range: 0 to
Perfect score = 1
Percentage Correct
(a + d )
PC =
n
Range: 0 to 1
Perfect score = 1
Observed
Forecast
YES
NO
YES
NO
a (Hits)
b (False Alarms)
c (Misses)
d (Correct no events)
Basic measures/scores
Threat Score, Critical Success Index
a
TS = CSI =
(a + b + c)
Range: 0 to 1
Perfect score = 1
No skill level = 0
•poorer scores for rarer events
Heidke Skill Score
2(ad − bc)
HSS =
[(a + c)(c + d ) + (a + b)(b + d )]
∞
Range: to 1
Perfect score = 1
No skill level = 0
Observed
Forecast
YES
NO
YES
NO
a (Hits)
b (False Alarms)
c (Misses)
d (Correct no events)
Error Structure for City/ Tourism forecast
Temperature
+ 10 C
> + 10 C
Correct
Incorrect
 Weather
 Yes/No
Intensity of Rainfall
Error Structure for Verification of District forecast
RAINFALL
Diff ≤ 25% of observed
Correct
25% of observed < Diff ≤ 50% of observed
Usable
Diff > 50% of observed
Unusable
TEMPERATURE
+ 10 C
Correct
+ 20 C
Usable
>+ 20 C
Incorrect
RELATIVE HUMIDITY
+ 10%
Correct
+ 20%
Usable
>+ 20%
Incorrect
Cont…
Wind direction
+ 30 deg
Correct
+ 40 deg
Usable
>+ 40 deg
Incorrect
Wind speed
+ 2 m/s
Correct
+ 4m/s
Usable
> + 4m/s
Incorrect
Cloud cover
+ 2 okta
Correct
+ 3 okta
Usable
>+ 3 octa
Incorrect
CRITERIA for Classification of Forecast
Good- If forecasts are correct and usable
for ≥ 70% of days of the season for
individual district.
Moderate- If forecasts are correct and
usable for 50-70% of days of the season
for individual district.
Poor- If Forecasts are correct and usable
for < 50% of days of the season for
individual district.
Name
FORECAST
VERIFICATION
AT IMD
Theme
Dr. S.K. Roy Bhowmik
MME (Oral)
Y.V. Rama Rao
TIGGE, TC (Oral)
B.P. Yadav, Naresh, L.S. Rathore
SRF , Lead
Dr. Kamaljit Ray
Nowcast (Oral)
V.R. Durai
NWP (Oral)
Neetha Gopal
Aviation (Oral)
M.I. Ansari
Observation (Oral)
Dr. M. Mohapatra
Cyclone (Poster)
Dr. R.K Jenamani
Fog (Poster)
D.S. Pai
LRF (Poster)
B.P. Yadav & Charan Singh
ICC Cup (Poster)
Dr. D.R. Pattanik
ERF (Poster)
Soma Sen Roy
NWP(Poster)
S.D. Kotal
NWP(Poster)
Dr. S.I. Laskar
NWP(Poster)
Ananda Kumar Das
NWP (Poster)
Verification of City rainfall forecast
percentage correctness
95
90
85
2010
80
2013
75
70
WINTER
PRE MON MONSOON POSTMON
Seasons
West Bengal
Verification of Value added Rainfall Forecast 2013
DAY WISE SKILL SCORE VALUE FOR OBSERVED VS VALUE
ADDED R/F
SKILL SCORE IN PERCENTAGE
92
90
88
day1
86
day2
84
day3
82
day4
80
day5
78
76
74
CHB
DJG
MLD
NAD
E.MDP
W.MDP
Multi-model ensemble
(ECMWF, NCEP CFSv2 and JMA)
16 Jan, 2010
17
Extended range Forecast skill for All India
Rainfall during last 4 years (2010-2013)
Correlation Coefficient (CC)
1.0
0.8
0.6
0.4
0.2
0.0
2010 (MME)
2011 (MME)
2012 (MME)
2013 (MME)
-0.2
Week 1 (5-11 Days)
Week 2 (12-18 Days)
Forecast Days
Week 3 (19-25 Days)
Performance of All India Rainfall LRF (1988-2012):







During 7 years error was ≥ 10% with highest during 2002 (20%), 1994 (18%).
Average Abs Error of LPA
Monsoon 2013
1988-2012
= 7.95%
Forecast 98 ± 4 %
1993-2002
= 9.3%
Observed 106%
2003-2012
=6.6%
During 1993-2002, forecast was within ±4% of actual values during 2 years.
During 2003-2012, forecast was within ±4% of actual values during 5 years.
Forecasting date of Monsoon onset over Kerala
(Model error = 4 days)
Year
Actual
Onset Date
Forecast Onset Date
2005
7th June
10th June
2006
26th May
30th May
2007
28th May
24th May
2008
31st May
29th May
2009
23rd May
26th May
2010
31st May
30th May
2011
29th May
31st May
2012
5th Jun
1st Jun
2013
1st Jun
3rd Jun
21-Mar-14
Track forecast error (km)
Average error in km (2009-13)
Lead Error
12 hr – 68.5
36 hr – 163.8
60 hr – 233.8
550
500
450
400
350
300
250
200
150
100
50
0
Lead Error
24 hr- 124.1 km,
48 hr- 202.1 km
72 hr- 268.2 km
Trend (km/year) in improvement in
track forecast (2003-13)
Lead Error
Lead Error
12 hr- -4.8
24 hr- -6.9
36 hr- -30.9 48 hr- -59.0
60 hr- -78.8 72 hr- -78.8
12 hr forecast error(km)
24 hr forecast error (km)
36 hr forecast error (km)
48 hr forecast error (km)
60 hr forecast error (km)
72 hr forecast error (km)
Linear (12 hr forecast error(km))
2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
Track forecast skill (%)
Average Skill (%) - (2009-13)
Lead Error
Lead Error
12 hr- 31.2 24 hr- 35.9
36 hr- 43.9 48 hr- 52.6
60 hr- 58.1 72 hr- 61.8
100
90
80
70
60
50
40
30
20
10
0
-10
-20
-30
-40
-50
-60
-70
Trend (% per year) in track forecast
skill (2003-13)
Lead Error
Lead Error
12 hr- 7.6
24 hr- 3.7
36 hr- 15.3
48 hr- 14.5
60 hr- 14.1 72 hr- 15.0
2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
12 hr track forecast skill (%)
24 hr track forecast skill (%)
36 hr track forecast skill (%)
48 hr track forecast skill (%)
60 hr track forecast skill (%)
72 hr track forecast skill (%)
Mean landfall point forecast error
Average error in km (2009-13)
Lead Error
Lead Error
12 hr - 38.8
24 hr- 75.0 km,
36 hr - 94.5
48 hr- 97.5 km
60 hr – 83.8
72 hr- 123.7 km
600
500
12 hr
48 hr
Linear (12 hr)
Trend (km/year) in improvement in
Landfall time forecast (2003-13)
Lead Error
Lead Error
12 hr- -14.5 24 hr- -31.2
36 hr- -16.6 48 hr- -6.5
60 hr- -04.9 72 hr- -09.1
24 hr
60 hr
Linear (24 hr)
36 hr
72 hr
400
300
200
100
0
-100
2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
Accuracy of Dense Fog Forecast
PC- January 2014
PC- January 2014
PC- January 2014
FOG Day-1
FOG Day-2
FOG Day-3
Region
No. of Dense
Fog Days
Region
<10
11-20
21-30
31-40
41-50
51-60
61-70
71-80
81-90
>90
NA
No. of Dense
Fog Days
East U.P.
26
Bihar
21
West U.P.
28
West M.P.
22
Haryana
30
East M.P.
21
Punjab
21
Jharkhand
05
West Rajasthan
14
SHWB
22
East Rajasthan
18
GWB
05
Skill of Cold Day Forecast
PC - January 2014
PC - January 2014
Cold Day - Day-1
Cold Day - Day-2
<10
11-20
21-30
31-40
41-50
51-60
61-70
71-80
81-90
>90
NA
<10
FAR - January 2014
FAR - January 2014
Cold Day Day-1
Cold Day Day-2
11-20
21-30
31-40
41-50
51-60
61-70
71-80
81-90
>90
NA
Skill of Short Range Forecast and Warnings
36 Meteorological
Sub-Divisions
Content of Bulletin
Terminology
 Threshold limits of all
 Observations, Satellite,
extreme weather warnings
Radar, Synoptic
defined
 Monsoon, Thunderstorm,
Cold, Fog, Heat, Cyclone  The words like “likely”…
done away
Watch
 Probabilistic words like
 Textual as well as
Graphical
“could, may, would, will”
are used with definition:
 Day-wise 3 Days
Category
Probability (%)
Forecast and 3 Days
warnings
Could occur
1-25
 4 days outlook
May occur
26-50
Would occur
51-75
 Updated 4 times a day
Will occur
76-100
Spatial distribution of Met. sub-divisional rainfall
Spatial distribution
Area of the subdivision getting rainfall
Dry
Mainly dry
No rain in the subdivision
< 2.5 mm reported in any station
Isolated
Scattered
Fairly widespread
1- 25% stations report rainfall
26% - 50% stations report rainfall
51% - 75% stations report rainfall
Widespread
76% - 100% stations report rainfall
Verification of spatial distribution of rainfall
Observed
Range
Dry/MD
Isolated
Scattered
Fairly widespread
Wide spread
Total
Forecast range
Dry/MD
Isol
Scat
FWS
WS
Total
a
f
k
p
u
O
b
g
l
q
v
P
c
h
m
r
w
Q
d
i
n
s
x
R
e
j
o
t
y
S
J
K
L
M
N
T
PC = ((a+g+m+s+y)/T)*100
CSI = a/(J+O-a), g/(K+P-g), m/(L+Q-m), s/(M+R-s), y/(N+S-y)
HSS= (a+g+m+s+y-(JO+KP+LQ+MR+NS)/T)/ (T-(JO+KP+LQ+MR+NS)/T)
Spatial Distribution
PC - Monsoon 2013
5 Category
PC - Monsoon 2013
Yes/No Category
PC - Monsoon 2013
3 Category
<10
11-20
21-30
31-40
41-50
51-60
61-70
71-80
Spatial
distribution
rainfall
Dry
No rain
Mainly dry
< 2.5 mm
Isolated
1- 25 %
Scattered
26% - 50%
Fairly
widespread
51% - 75%
Widespread
76% - 100%
Spatial
distribution
rainfall
Dry/ MD
< 2.5 mm
Isol / Scat
1- 50 %
FWS/ WS
51% - 100%
81-90
>90
NA
YES/NO FORECAST
Heidke Skill Score
INTENSITY OF RAINFALL
Intensity
Amount of rainfall
Dry
Very light
Light
Moderate
Rather heavy
Heavy
Very heavy
Extremely heavy
No rain in the subdivision
< 2.5 mm
2.5 – 7.4 mm
7.5 – 24.4 mm
24.5 – 64.4 mm
64.5 mm – 124.4 mm
124.5 mm – 244.4 mm
245 mm or more
Exceptionally heavy
If the rainfall is higher than or nearer to
the earlier record over the station.
However, the rainfall should be 13 cm or
more
Heavy Rainfall during Southwest Monsoon in India
Weather systems
Regions
Off-Shore Trough/
Vortex, LLJ
West Coast
ITCZ & NWly moving
Depressions/ Lows/
Cycirs (Active monsoon)
East, Central, North
Peninsular, NW
India
Mid-tropospheric
circulation
Mainly Gujarat
Interaction of Wly
and Ely Systems
Western Himalayas
& Northwest plains
ITCZ close to
Foothills, N-S trough
( Break monsoon)
Northeastern States
& adjoining East
India
Secondary ITCZ with
Wly moving Cycirs
Southeast
Peninsula
CONTINGENCY TABLE
Observed
Forecast
YES
YES
NO
a (Hits)
b (False Alarms)
c (Misses)
NO
d (Correct no events)
forecast
M
observations
CR
H
F
Heavy rainfall events during the
southwest monsoon season.
HEAVY RF > 10 CM
Very HEAVY RF > 15 CM
Rajeevan et al. 2008, Geophys. Res. Letters
FREQUENCY OF VERY HEAVY RAINFALL EVENTS (R >=15 cm)
OVER CENTRAL INDIA (20-29N, 72-88E)
45
Mean (1901-1975) = 12.3
Mean (1976-2007) = 18.1
40
35
25
20
15
10
5
2001
1997
1993
1989
1985
1981
1977
1973
1969
1965
1961
1957
1953
1949
1945
1941
1937
1933
1929
1925
1921
1917
1913
1909
1905
0
1901
Frequency
30
YEAR
Rajeevan et al. 2008, Geophys. Res. Letters
HEAVY
RAINFALL
TREND OVER
CENTRAL
INDIA
(1°-BY-1°GRIDDED
DATA OF IMD BASED
ON 1803 STATIONS
(90% DATA
AVAILABILITY) FOR
1951–2000)
HEAVY RAINFALL ( ≥64.5 MM) TREND OVER INDIA
(1901-2005)
Increasing trend
Decreasing trend
Significant level 95%
Guhathakurta et al (2011),
J. Earth Syst. Sci.
Increasing trend
Decreasing trend
Significant level 99%
Heavy Rainfall Days
during MONSOON 2013
26
Jammu & Kashmir
Himachal
Pradesh
37 58
Arunachal
Pradesh
Punjab
Uttarakhand
34
25
West Rajasthan
48
52
68
62
74
40 66
59
East
Madhya Pradesh
Chhattisgarh
Vidarbha
20
73 64
Marathawada
Madhya
Maharashtra
61 66
Jharkhand
48
49
87
Orissa
47
41
Gangetic
West Bengal
Coastal
Andhra Pradesh
Rayalaseema
South
Interior
Karnataka
76
19
41
45
Tamilnadu &amp;
Pudducherry
Lakshadweep
14
72
Kerala
Year
% dep.
2010
2
2011
2
2012
-7
2013
6
Nagaland Manipur
Mizoram &amp; Tripura
46
34
85
52
Telangana
North Interior
Karnataka
Coastal
Karnataka
90
Assam &amp;
Meghalaya
Bihar
Gujarat
Region
West
DD&amp;N
Madhya Pradesh
Havelli
Konkan &amp; Goa
79
East Uttar Pradesh
East
Rajasthan
Saurashtra
Kuchh &amp; Diu
Sub-Himalayan
West Benga &amp; Sikkiml
47
Haryana
Chd. &amp;
West
Delhi
Uttar
Pradesh
Andaman
&amp; Nicobar
Islands
Date
24 hrs Chief Rainfall amounts (in Cms)
22
16 - Quant (Gujarat Region)
23
36 - Sagbara ,35 - Umerpada ,30 - Baruch ,20 - Jhagadia, Quant ,19 Jetpur Pavi ,18 - Sankheda, Sankheda Arg, Tilakwada Arg, 15 - Naswadi,
Waghai Arg (All Gujarat Region)
24
43 - Quant ,38 - Rajpipala, Vansda, 35 - Tilakwada Arg , 32 - Nandod 30 Karjan , 28 - Sankheda, Tilakwada, Waghai Arg , 26 - Silvassa ,25 - Sinor
Arg (All Gujarat Region)
25
26
27
37 - Umerpada ,32 - Surat Aws ,26 - Bardoli ,24 - Karjan, Silvassa, Vagra
,23 - Choryasi ,22 - Palsana ,20 - Jalalpor (All Gujarat Region)
39 – Rajkot, 34 - Visavadar , 28 – Babra, 23 - Jasdan ,22 - Chotila , 21 Gadhda, Jamjodhpur , 20 - Kotdasangani (All Saurashtra & Kutch)
25 - Kamrej , 24 – Olpad, 22 - Choryasi, Hansot Arg (All Gujarat Region),
45 - Khambhalia ,35 - Okha ,29 - Jamnagar Iaf , 24 - Dhrol Arg ,23 Kalyanpur , 22 - Kalavad, Naliya, Upleta (All Saurashtra & Kutch)
28
31 – Bhachau, 28 - Anjar ,20- Bachau Aws (Saurashtra & Kutch)
26 - Bhabhar ,23 – Dhanera, 22 - Deesa (Gujarat Region)
20 - Sanchore (west Rajasthan)
29
27 - Mount Abu, 22- Mounntabu Tehsil, 17- Devel, Dungarpur Tehsil,
Kherwara, Pisagan (east Rajasthan) , 17- Merta City (west Rajasthan)
23 – Dantiwada, 17- Deesa (Gujarat Region)
48 hour Forecast
26 Feb
27 Feb
EXTREME
WEATHER
WARNIGS
2014
9 March
11 March
HEAVY RAINFALL WARNINGS (PHAILIN )
12/10/2013
Day 1
Day 2
Day 3
Monsoon 2013 Heavy Rainfall Forecast
Cont…
Cont…
Skill of Heavy Rainfall Warnings
Contd..
Contd..
Contd..
Improvement in Skill of Heavy Rainfall Warning
Improvement in Skill of Heavy Rainfall Warning
(Number of Met. Sub Divisions)
Skill of heavy rainfall warning in recent past
Programme for improving forecast skills
CTCZ PROGRAMME
FDP Fog
Friday, March 21, 2014
SAARC STORM PROJECT
SEVERE WEATHER FORECAST
DEMONSTRATION PROJECT
(SWFDP) – BAY OF BENGAL
FDP Cyclone
Conclusion
In recent times, operational weather forecasts &
warnings of IMD have improved significantly.
However, there is need for further improvement:







Standardize the verification methods
Standardize the threshold limits of scores
Standardize region specific impact of extreme events
Skill beyond day 3
Skill of extreme and rare events
Communicating the uncertainty
Focus on PWS
Critic’s View
 We have done extremely well in improving the
weather forecast and climate prediction
during the past 5 years but have not been that
successful in extracting the full value of the
more accurate forecast for societal benefit
 How should we communicate with the public
about weather Hazards?
PROF. J.SRINIVASAN
DIVECHA CENTRE FOR CLIMATE CHANGE
INDIAN INSTITUTE OF SCIENCE