Institute of Heavy Rain, Wuhan, CMA

advertisement
Institute of Heavy Rain, Wuhan, CMA
Assimilation of Radar Observations
for Heavy Rain Numerical Prediction
Wang Yehong Cui Chunguang
Zhao Yuchun Li Hongli
Institute of Heavy Rain, Wuhan, CMA
1.
2.
3.
4.
5.
Introduction
1DVAR system and experiments for heavy rain
GRAPES_3DVAR system and experiments
4DVAR system and experiments
Summary and future work
Introduction
1998 China Flood
Introduction
Institute of Heavy Rain, Wuhan, CMA
observation
forecast
24h accumulate rainfall from 20BST,20July1998
to 20BST,21July 1998
Introduction
Institute of Heavy Rain, Wuhan, CMA
Radiosonde station
shade: 1h rainfall from 20BST to 21BST 20July 1998
Introduction
Institute of Heavy Rain, Wuhan, CMA
Only by conventional observation net and
traditional initialization procedure, it is very
difficult to get the accurate initial field needed in
the meso-scale model, especially the meso-scale
systems vital to torrential rain generation. One
way improving the initial field of the meso-scale
model is assimilating non-traditional data into
meso-scale model with variational method. In
meso- or micro-scale studies, the assimilation of
Doppler radar data, which have relatively high
spatiotemporal resolution and contain adequate
meso-scale cloud-water, precipitation and wind
information, is significant to the improvements of
meso-scale initial field.
1DVAR system
Institute of Heavy Rain, Wuhan, CMA
One-dimensional Variational Assimilation System of
Radar Derived Rainfall
• Developed at WHIHR for the research of radar derived
rainfall data assimilation
• Focused on initialization and forecasting of heavy rain
using a regional numerical model
• The physics includes horizontal diffusion,large-scale
condensation and evaporation and Betts convective
adjustment parameterization scheme.
• Spatial distributions of the variables are on a E-grid with
37km horizontal resolution and 16 levels in the vertical.
The vertical coordinate isη
1DVAR system
Institute of Heavy Rain, Wuhan, CMA
1DVAR method
The 1DVAR seeks optimum values of numerical
model prognostic variables by minimizing the following
one-dimensional cost function :
1
1  R  X   R0 
b T
1
b

J  X   X  X  B X  X   
2
2
0

Background term
Xb
2
Observation term
Where X and
denote the optimum values and
background values of the model prognostic variables;
R(X) and Ro denote the observation operator logarithm
of precipitation and the radar-derived rainfall respectively.
B is the error covariance matrix of the background. is
the standard deviation
 o of the observation error.
1DVAR system
Institute of Heavy Rain, Wuhan, CMA
1DVAR System Flow Chart
Eta-model
Data analysis
radiosonde at
20LST
background
rainfall
One-dimensional variational assimilation system
Initial humidity
at 20LST
Forecast rainfall at 21LST
Radar-derived rainfall
at 21LST
Eta-model
1hr forecast
24hr forecast
analysis
rainfall
observation
rainfall
1hr forecast
24hr forecast
Application to rain forecast
Institute of Heavy Rain, Wuhan, CMA
One-Dimensional Variational Assimilation of
Radar-Derived Precipitation Data for “98·7”
Torrential Rain
• limited area meso-scale numerical model underη–coordinate
with 37km resolution
• One dimensional variational assimilation system
• data: radiosonde data and reflectivity observations
• Retrieval of 1 hour precipitation using reflectivity
observations
• Application: 0~24h rain forecast
Application to rain forecast
(a)
(c)
(b)
The accumulated rainfall
distribution of observation
Ro (a), background Rb (b)
and analysis Ra (c) in
Hubei from 20:00 to 21:00
on the 20th of July 1998.
Application to rain forecast
Fig. the relative humidity
profiles distribution at 20LST
20 July 1998
Without assimilation
With assimilation
Application to rain forecast
The relative humidity distribution of the differences
between with and without assimilation at 700hPa at
20:00 on the 20th of July 1998.
Application to rain forecast
observation
Without assimilation
With assimilation
(b)
(c)
Institute of Heavy Rain, CMA, Wuhan http://www.whihr.com.cn
(a)
(b)
The 12 hours rainfall distribution of model forecast
differences between with and without assimilation .
The former 12 hours is from 20:00 of 20th to 08:00
of 21st(a), and the latter 12 hours is from 08:00 of
21st to 20:00 of 21st(b).
Grapes-3dvar system
Institute of Heavy Rain, Wuhan, CMA
GRAPeS_3DVAR system





Global-Regional Assimilation and Prediction System
It is a new global and regional assimilation and
prediction system being developed by CMA
It is analyzed in the horizontal grid and vertical isobaric
surface level, and the analysis variables include potential
height, wind and humidity.
The horizontal background correlation is realized by
spacial recursion filter(finite regional version)
The minimization of control variables is carried out by
LBFGS
Application to rain forecast
Institute of Heavy Rain, Wuhan, CMA
On the Three Dimensional Variational
Assimilation of Radar Wind Data related to
2003-7-8 Catastrophic Torrential Rain
Regional -coordinate
Model version 2.1
 Advanced
 GRAPES_3dvar
 Retrieval
system
wind data from Wuhan and
Yichang Doppler radar
Application to rain forecast
Institute of Heavy Rain, Wuhan, CMA
700hPa
宜昌
武汉
retrieval wind from Wuhan and Yichang Doppler radar (vector vane)
wind detected by radiosonde (barb)
Institute of Heavy Rain, Wuhan, CMA
3DVAR experiments for “7·8” heavy rain
Observed
fields
radiosonde
Retrieval wind
from Wuhan
Doppler radar
Retrieval wind
from Yichang
Doppler radar
Control
experiment
√
Experiment
1
√
√
√
Experiment
2
√
√
Experiment
3
√
√
initial fields improvement by
assimilation of retrieval wind
without assimilation of
retrieval wind
with assimilation of
retrieval wind
without assimilation of
retrieval wind
with assimilation of
retrieval wind
Vertical-latitude cross
section of the
differences of specific
humidity along 1110E
The differences of
specific humidity at
850hPa
observation
大庸379mm
大
悟
石门182mm
大悟154mm
大
庸
Without assimilation
石门
With assimilation
evolution of 1h rainfall
observation
Test 1
Control test
observation
Test 1
Control test
Institute of Heavy Rain, Wuhan, CMA
evolution of accumulate rainfall
Institute of Heavy Rain, Wuhan, CMA

The prescribed results show that in the simulation
of an extremely heavy rain on July 8 in middle
Yangtze river, the assimilation of retrieved wind
from Wuhan and Yichang Doppler radar by
3DVAR greatly improve the initial field which is
consistent with model and conducive for strong
precipitation generation both in thermodynamics
and dynamics with the results that the model
reproduces 24h accumulated and hourly rainfall
close to observation in the middle Yangtze River.
It proves that effective usage of retrieved wind
data from Doppler radar can make positive
contributions to numerical simulation and
forecasting.
observation
大
悟
大
庸
石门
Grapes-3dvar system
Institute of Heavy Rain, Wuhan, CMA
4DVAR system


MM5V1 and Adjoint-model Assimilation System. The
main physical processes include nonhydrostatic
equilibrium scheme, Grell cumulus parameterization
shceme, Blackadar high-resolution planetary boundary
scheme, simple ice explicit moisture scheme, simple
cooling atmosphere radiation scheme and time-dependent
flowing-in/out lateral boundary condition.
NCEP data, regular observations and simple-Doppler radar
data. The initial time is 00h 23th June,2004. The total
integration time is 24h. The center of model is located at
(113ºE,29ºN). The number of horizontal grid is 61×61.
The grid length is 15km.
Institute of Heavy Rain, Wuhan, CMA
A study on 4-dimensional variational assimilation
of single-Doppler radar wind data
---a heavy rainfall in middle reach of the Yangtze River
Scheme I: Sigh the initial field made by the model when using
NCEP data and regular observations as A field.
Scheme II: First, the model analysis field of the initial time serves
as the model first-guess field. Then the regular observations of 06h
is input into assimilation model. And finally by adjusting the firstguess field through restrict conditions we have the optimal initial
field as B field.
Scheme III: We use radar data retrieved with the variational method
to replace the value of A field so as to obtain C field that includes
radar data,
Scheme IV: It is similar with scheme II. The regular observations
and retrieved radar data of 06h is input into assimilation model.
And finally by adjusting the first-guess field through restrict
conditions we have the optimal initial field as D field.
radiosonde
Institute of Heavy Rain, Wuhan, CMA
4DVAR:radiosonde
observed rainfall field
radiosonde + retrieval wind
from Wuhan Doppler radar
4DVAR:radiosonde +
retrieval wind from Wuhan
Doppler radar
Institute of Heavy Rain, Wuhan, CMA
Summary
The established 1DVAR assimilation system can
effectively assimilate radar-derived rainfall data and the
heavy rain forecast can be improved by adjusting
humidity profiles of the initial field.
 After retrieved wind from Doppler radar has been
assimilated by Grapes_3dvar system, more meso-scale
information is presented in the initial wind field,
humidity field and potential height field in such a
manner that rainfall prediction can be greatly improved.
 Effectively using radar wind field information in
4DVAR assimilation system can improve rain belt
simulation.

Institute of Heavy Rain, Wuhan, CMA
Future work
Three-dimensional variational assimilation of radar
retrieved wind field data in heavy rain forecasts should be
studied on the basis of Doppler radar wind field data
collected as much as possible.
 Real-time forecasting experiments using retrieved wind
data from Doppler radar should be conducted step by step
in the light of meso-scale operational forecasting model
AREM and Grapes_3dvar system.
 Direct variational assimilation of radial velocity of
Doppler radar should be studied.
 On the basis of LAPS developed by FSL, several kinds of
local observation data such as radar data, satellite data,
wind profiler data and etc, should be combined into mesoscale numerical models (AREM, GRAPES, MM5) in order
to improve the initial field and better rainfall forecast.

中国气象局武汉暴雨研究所 http://www.whihr.com.cn
THE END
THANKS!
Download