Using WRF to downscale rainfall simulations and storm types

Using WRF to downscale rainfall simulations and
storm types
Michaela Bray
Cardiff University
International Symposium on Weather Radar and Hydrology, 18~21 Apr 2011, Exeter
Outline
• Motivation
• Study Area
• Data sets
• WRF Sensitivity to downscaling ratios and storm types
• Effect of Data Assimilation on rainfall prediction with
different storm types
• Conclusions
Motivation
Brue Catchment
Drainage area of 135 square kilometres.
Elevation range between 35 metres to 190 metres above sea level.
Average annual rainfall for the period 1961 to 1990 is 867mm
49 dense rain gauge network
bucket size
gauge aperture
0.2 mm
400cm2
tip time was recorded up to a
time resolution of 10 seconds.
C-band weather radar 30 km to the
south at Wardon Hill
WRF sensitivity to downscaling ratios
and storm types
•
8 different spatial and temporal distributions of rainfall intensity.
Event ID
Storm start time
Storm end time
Accumulated 24 hour rainfall (mm)
a
01/02/2000 04:00
01/02/2000 04:00
23.45
b
02/04/2000 18:00
03/04/2000 18:00
31.36
c
05/11/1999 06:00
06/11/1999 06:00
16.93
d
24/10/1999 00:00
25/10/1999 00:00
29.39
e
07/06/1996 11:00
08/06/1996 11:00
21.26
f
03/08/1994 13:00
04/08/1994 13:00
22.27
g
05/08/1997 10:00
06/08/1997 10:00
30.39
h
06/09/1995 18:00
07/09/1995 18:00
32.41
WRF Domain Configurations
±
Scenario and domain
Scenario1
(S1)
Scenario2
(S2)
Scenario3
(S3)
Scenario4
(S4)
Scenario5
(S5)
Initial lateral and surface boundary conditions: ECMWF 40
Year Re-analysis (ERA-40) resolution of 1° ×1° and updated
every 6 hours.
Vertical levels: 28
Cumulus parameterisation not used in innermost domain,
where convective rainfall generation assumed to be explicitly
resolved.
Domain1
Domain2
Domian3
Domain1
Domain2
Domian3
Domain1
Domain2
Domian3
Domain1
Domain2
Domian3
Domain1
Domain2
Domian3
Domain4
Time step
(hr)
3
1
0.25
3
1
0.25
3
1
0.25
3
1
0.25
3
3
1
0.25
Downscaling
ratio
1:10
1:10
1:7
1:7
1:5
1:5
1:3
1:3
1:3
1:3
1:3
Classification of events
Evenness of rainfall
distribution
Type 1 (Event a, b)
In space
In time
Y
Y
Type 2 (Event c, d)
Y
N
Type 3 (Event e, f)
N*
N*
Type 4 (Event g, h)
N
N
N* means unevenness with highly concentrated distribution in a small area or
a short time period
• POD (probability of detection)
RRi
1 N
 
N i 1 RRi  NRi
• FBI (frequency bias index)
1

N
RRi  RN i

i 1 RRi  NRi
N
• FAR (false alarm ratio)
1

N
RN i

i 1 RRi  RN i
• CSI (critical success index)
1

N
N
N
RRi

i 1 RRi  RN i  NRi
1
M
• RMSE (Root Mean Square Error

• MBE (mean bias error)
1

M
• SD (Standard deviation)

 S
M
j 1
 S
M
j 1
j
 Oj 
2
j
 Oj 
1 M
S j  O j  MBE 2

M  1 j 1
Initial Results:
•
•
•
•
Underestimation of total rainfall amount
Worst downscaling gradient 10:1 (S1)
Best downscaling gradient
1:5 (S3)
Curious result: S4 performed better than S5
Event a
30
(c) domain 3
Scenario 3
Scenario 4
Scenario 5 dom4
10
0
20
Scenario 1
Scenario 2
Scenario 3
Scenario 4
Scenario 5 dom3
10
0
Observed
Cumulative rainfall (mm)
Scenario 2
Observed
Cumulative rainfall (mm)
Cumulative rainfall (mm)
Scenario 1
(a) domain 1
(b) domain 2
Observed
20
30
30
Scenario 1
20
Scenario 2
Scenario 3
Scenario 4
Scenario 5 dom2
10
0
Scenario 5 dom1
1
Observed
Simulated
20
2
10
(b) Event b
30
3
Simulated
20
2
0
0
0
40
40
(c) Event c
1
1
Observed
Simulated
20
30
(d) Event d
20
2
10
0
Best performance of WRF
3
50
Cumulative rainfall (mm)
Rainfall intensity (mm/15min)
50
30
1
Observed
Type 1 :
Even intensity and continuous rainfall.
10
0
Cumulative rainfall (mm)
40
Rainfall intensity (mm/15min)
(a) Event a
0
Observed
Simulated
2
10
3
0
3
Rainfall intensity (mm/15min)
Cumulative rainfall (mm)
40
30
50
0
Rainfall intensity (mm/15min)
Cumulative rainfall (mm)
50
Type2:
Even spatial distribution of rainfall
uneven or discontinuous temporal
distribution.
WRF has difficulty in simulating
rainfall amounts, but spatial
distributions reproduced evenly and
good fit of rainfall occurrence
Observed
Simulated
4
40
6
40
20
20
Observed
2
Simulated
(g) Event g
2
Observed
Simulated
4
40
6
20
3
100
0
80
(h) Event h
60
2
Observed
Simulated
4
40
6
20
8
0
8
Rainfall intensity (mm/15min)
0
Cumulative rainfall (mm)
100
80
Cumulative rainfall (mm)
(d) Event d
0
Rainfall intensity (mm/15min)
8
0
1
30
10
0
60
0
Rainfall intensity (mm/15min)
2
(e) Event e
Cumulative rainfall (mm)
Cumulative rainfall (mm)
80
60
50
0
Rainfall intensity (mm/15min)
100
Type 3
Highly concentrated in small area
can occur over shorter time
periods and rainfall more intense
WRF has the most difficulty in
reproducing these type of events,
Dislocation in time
Type 4
Uneven distributions in time and space,
but rainfall not spatially y concentrated
and more moderate intensities
WRF better performance than type 3
Using Data assimilation to improve WRF
Precipitation
 Flood forecasting in small catchments with short concentration times
depends on rainfall forecasts from high-resolution NWP model to provide
extended lead times;
The accuracy of NWP is negatively affected by the ‘spin-up’ effect and
errors existing in the initial conditions;
Rainfall forecasts can be significantly improved by assimilating
observations, especially the radar data into the NWP model (e.g., Xiao & Sun,
2007; Dixon et al., 2007; Sokol, 2010, etc).
The purpose of this study is to assimilate radar reflectivity and other
surface / upper-air observations into the WRF model, and further compare
the improvements of rainfall forecasts for storm events of different types.
The NWP model and data assimilation scheme
 Advanced Research WRF model version 3.1
Domain configurations:
Time step (hr)
Grid spacing (km)
Domain 1
3
250
Domain 2
3
50
Domain 3
1
10
Parameterisations: the new Kain-Fritsch cumulus parameterization scheme, the WSM3
microphysics scheme, the Dudhia shortwave and the RRTM longwave radiation schemes, etc.
 3D Variational data assimilation system of WRF (WRF-Var)
Observation pre-processing
Boundary condition updating
Background error generation
ECMWF operational forecasting data

ECMWF produces global 10-day forecasts based on 00 and 12 UTC analyses

The spatial resolution of the ECMWF data used here is 2.5 o× 2.5 o

C Band Radar at Wardon Hill + NCAR obs (surface and upper levels of pressure,
temperature, humidity and wind
24hr duration
1995/09/06
09/07
06:00
00:00
06:00
origin2
12:00
18:00
origin1
09/08
12:00
18:00
origin3
run1
run2
run3
run4
run5
run6
run7
run8
00:00
06:00
Four types storm events
Four 24hr storm events are selected from the Brue catchment (135.2 km2)
0
Cumulative rainfall (mm)
40
1
30
20
2
Rainfall intensity (mm/15min)
50
Event C
(2000/04/02/18~04/03/18)
Even in space
Continuous in time
10
100
3
0
80
2
60
4
40
6
20
0
8
Rainfall intensity (mm/15min)
0
Cumulative rainfall (mm)

Event D
(1994/08/03/12~08/04/12)
Very uneven in space
Very discontinuous in time
Four types of storm events
Four 24hr storm events are selected from the Brue catchment (135.2 km2)
0
Cumulative rainfall (mm)
40
1
30
20
2
Rainfall intensity (mm/15min)
50
Event A
(1999/10/24/00~10/25/00)
Even in space
Discontinuous in time
0
3
100
0
80
2
60
4
40
6
20
0
8
Rainfall intensity (mm/15min)
10
Cumulative rainfall (mm)

Event B
(1995/09/06/18~09/07/18)
Uneven in space
Discontinuous in time
Weather radar and the data quality


A C-band radar at Wardon Hill gives a whole coverage of the Brue catchment
Data are taken from the lowest scan (0.5o) on the 2km Cartesian grids with a radius of
76km
Event A
Gauge
29
Radar
10
Event B
31
10
Event C
31
12
Event D
22
10
7
7
6
Gauge
Radar
Event A
6
5
5
4
4
3
3
2
2
1
1
00:00
03:00
06:00
09:00
12:00
15:00
18:00
21:00
00:00
18:00
Gauge
Radar
Event B
21:00
00:00
03:00
06:00
09:00
12:00
15:00
18:00
7
6
Gauge
Radar
Event C
5
21
18
15
4
12
3
9
2
6
1
3
18:00
Gauge
Radar
Event D
21:00
00:00
03:00
06:00
09:00
12:00
15:00
18:00
12:00
15:00
18:00
21:00
00:00
03:00
06:00
09:00
12:00
Data assimilation results of the four storm events

Radar reflectivity is assimilated together with two types of NCAR surface / upper-air
observations (SYNOP and SOUND)
40
35
30
25
20
Gauge
Radar
Run1
Run2
Run3
Run4
Run5
Run6
Run7
Run8
Event A
35
Radar
30
15
10
5
5
45
40
35
30
Gauge
Radar
Run1
Run2
Run3
Run4
Run5
Run6
Run7
Run8
20
10
50
WRF original
25
15
12:00
Rain gauge
Event B
18:00
00:00
Gauge
Radar
Run1
Run2
Run3
Run4
Run5
Run6
Run7
06:00
12:00
18:00
00:00
06:00
12:00
Event C
12:00
25
20
15
18:00
Gauge
Radar
Run1
Run2
Run3
Run4
Run5
Run6
00:00
06:00
12:00
18:00
00:00
24-hr accumulation error:
06:00
Event D
25
20
10
15
10
5
5
12:00
18:00
00:00
06:00
12:00
18:00
00:00
06:00
12:00
18:00
00:00
06:00
12:00
18:00
00:00
Original
run
After
assimilation
A
0.15
(-99%)
25.95
(-12%)
B
17.21
(-46%)
28.18
(-12%)
C
18.68
(-40%)
37.60
(21%)
D
0.06
(-100%)
0.12
(-99%)
Conclusions
The assimilation of radar reflectivity together with other surface /
upper-air observations is able to improve the NWP rainfall forecasts;
Significant improvements are seen with the two types of storm events
having rainfall evenly-distributed across the catchment;
For highly convective storms where rainfall is concentrated in a small
area and happens in a short time period, the improvement is not obvious;
A more frequent assimilation of the radar reflectivity from higher scan
levels and other observations containing information about the cloud
development might help capture the highly convective storms;
Acknowledgements:
• Principal Collaborators: Professor Dawei Han, Dr Jia Liu, Dr
Asnoor Ishak (University of Bristol)
• ARCCA (Advanced Research Computing at Cardiff)
• Wei Wang, Michael Duda (NCAR)
• BADC, NERC-HYREX experiment, ECMWF, NCAR, (Data
providers)
The End
Thanks for your attention!
(a) Observed
Type 1
(b) Simulated
(a) Observed
Type 3