The use of radar in evaluating precipitation in LMK and ARPS: two precipitation cases over Belgium

1
The use of radar in evaluating precipitation in
LMK and ARPS two precipitation cases over
Belgium
Kwinten Van Weverberg and Ingo Meirold-Mautner

• 6 March 2007 International PhD-studens and
  Post-docs meeting on QPF

2

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Forecasting of precipitation is still one of the
challinging tasks in Numerical Weather Prediction

• Discontinuous distribution of water in space and
  time in the atmosphere in all its three phases.
• Verification of model predicted precipitation
  variables is not straightforward.
• We want to learn more about the strengths and
  weaknesses of both the LMK and ARPS model in
  simulating precipitation processes
• A correct representation of precipitation in
  numerical models is indispensable for e.g.
  studying the sensitivity of precipitation
  processes to temperature changes

3

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Until recently rain gauge measurements were the
main input for evaluation of precipitation in
atmospheric models

• But rain gauges always have a too low spatial and
  temporal coverage and only provide us with ground
  precipitation data.

4

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

During the last two decades, new methods of
remote measurement gained importance as
alternative high quality data for hydrometeor
model evaluation
5

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

During the last two decades, new methods of
remote measurement gained importance as
alternative high quality data for hydrometeor
model evaluation
6

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

The C-band weather radar of the RMI in Wideumont

• Radar sends electromagnetic pulse and receives
  the reflected pulse
• The waiting time between sending and receiving is
  a measure for the distance of the target, the
  power of the returned beam is a measure for the
  size of the object
• Radar scans in one direction on a turning
  platform (360) and at different elevation angles
  (0.5 to 17.5) to provide a full volume scan

7

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

The C-band weather radar of the RMI in Wideumont

• C-band Doppler radar (3.7 4.2 GHz)
• Positioned at a height of nearly 600 m in the
  south of Belgium
• Radar beam scans each 5 min at 5 and each 15 min
  at 10 different elevation angles
• Horizontal resolution is 250 m in range and 1
  degree in azimuth

8

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Advanced Regional prediction System (CAPS)
Lokal Modell Kürzesfrist (DWD)

• Mesoscale nonhydrostatic model
• Subgrid scale turbulence 1.5 order Turbulent
  Kinetic Energy closure
• Kain and Fritsch convection parameterization in 9
  km runs, no parameterization in 3 km runs.
• Kessler warm rain microphysics scheme was used in
  the 9 km run, Lin-Tao 3-category ice scheme was
  used in the 3 km run
• Initial and boundary conditations derived from
  ECMWF operational analysis
• Double one-way nesting procedure
• 9 and 3 km horizontal resolution Vertically
  stretched grid
• 240 km x 240 km domain, covering Belgium
• No data assimilation

• Mesoscale nonhydrostatic model
• Subgrid scale turbulence 1 eq. Turbulent Kinetic
  Energy closure.
• Moist convection following Tiedtke (1989) for
  shallow convection, no paramterization for deep
  convection
• Grid scale clouds saturation adjustment
• Precipitation formation bulk microphysics
  parameterization including water vapour, cloud
  water, rain and snow.
• Initial and boundary conditations derived from
  ECMWF operational analysis
• Double one-way nesting procedure
• 7 and 2.8 km horizontal resolution
• Vertically stretched grid
• 500 km x 500 km domain, covering Belgium
• No data assimilation

9

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Advanced Regional prediction System (CAPS)
Lokal Modell Kürzesfrist (DWD)
10

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Two different cases were selected with each
different precipitation characteristics...
Frontal stratiform case
convective supercell case
23/10/2006
01/10/2006
11

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Two different cases were selected with each
different precipitation characteristics...
Frontal stratiform case
convective supercell case
23/10/2006
01/10/2006
12

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

• An extensive model evaluation is necessary before
  using the model in experiments in order to gain
  insight in the models strengths and weaknesses
  in simulating the variables of interest
• Using radar as a tool for atmospheric model
  evaluation has great advantages over the use of
  rain gauges due to the very high spatial and
  temporal coverage
• We can also gain insight in the vertical
  distribution of hydrometeors and compare them to
  the modeled distribution

13

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Model evaluation was done using the Wideumont
radar

• But... radar does not measure atmospheric
  constituents represented by the model, but
  measures only the reflectivities
• Two approaches exist observation to model
• Precipitation intensities are derived from radar
  reflectivities and compared to model
  precipitation intensities ? based on empirical
  relations Z 200 x R1.6 (Marshall and Palmer)
• model to observation
• Radar reflectivity is derived from model
  variables and compared to observed radar
  reflectivities ? less uncertainty because model
  variables can be described much more accurately.

14

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Model evaluation was done using a model to
observation approach
15

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Model evaluation was done using a model to
observation approach
Observation to model approach (preliminary
results) comparing radar derived (Marshall and
Palmer) and model precipitation fields
24h-Accumulated precipitation on 1 October 2006
LMK 2.8 km
ARPS 9 km
Radar 1 km
16

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Model evaluation was done using a model to
observation approach
Observation to model approach (preliminary
results) comparing radar derived (Marshall and
Palmer) and model precipitation fields
24h-Accumulated precipitation on 23 October 2006
LMK 2.8 km
ARPS 9 km
Radar 1 km
17

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Model evaluation was done using a model to
observation approach
But large errors in the radar derived
precipitation rates due to the Marshall Palmer
relation, which is not constant in time.... Radar
is prone to errors varying in time attenuation,
overshooting beam broadening. Further, the ZR
relation depends on the hydrometeor type, which
is not known
18

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Model evaluation was done using a model to
observation approach
Model to observation approach (preliminary
results) comparing radar reflectivities with
simulated reflectivities based on model output.
We do know the hydrometeor type in the model
Simple forward operator (Keil et al, 2003), based
on modeled formulas of Fovell and Ogura (1988)
and the assumption of a Marshall-Palmer size
distribution for the hydrometeors.
Simple forward operator (Smedsmo et al, 2005)
19

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Case 1 convective supercell (01/10/2006)
Radar reflectivities at 2 km above the surface on
1 October 2006 at 15 UTC following Smedsmo (2005)
LMK 2.8 km
ARPS 9 km
Radar 1 km
20

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Case 1 convective supercell (01/10/2006)
Mixing ratios of rain at 2 km above the surface
on 1 October 2006 at 15 UTC
LMK 2.8 km
ARPS 9 km
Radar 1 km
21

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Case 1 convective supercell (01/10/2006)
Vertical cross section through radar
reflectivities (W-E) at 1 October 2006 at 15
UTC following Smedsmo et al (2005)
LMK 2.8 km
ARPS 9 km
Radar 1 km
22

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Case 1 convective supercell (01/10/2006)
Vertical cross section through rain mixing ratio
(W-E) at 1 October 2006 at 15 UTC
LMK 2.8 km
ARPS 9 km
Radar 1 km
23

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Case 1 convective supercell (01/10/2006)
Spatially averaged vertical Profiles of
Reflectivity at 1 October 2006 at 15 UTC
ARPS 9 kmSmedsmo
Radar 1 km
24

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Case 2 stratiform case (23/10/2006)
Radar reflectivities at 2 km above the surface on
23 October 2006 at 19 UTC following Smedsmo (2005)
LMK 2.8 km
ARPS 9 km
Radar 1 km
25

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Case 2 stratiform case (23/10/2006)
Mixing ratios of rain at 2 km above the surface
on 23 October 2006 at 19 UTC
LMK 2.8 km
ARPS 9 km
Radar 1 km
26

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Case 2 stratiform case (23/10/2006)
Vertical cross section through radar
reflectivities (W-E) at 23 October 2006 at 19
UTC following Smedsmo (2005)
LMK 2.8 km
ARPS 9 km
Radar 1 km
27

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Case 2 stratiform case (23/10/2006)
Vertical cross section through rain mixing ratio
(W-E) at 23 October 2006 at 19 UTC
LMK 2.8 km
ARPS 9 km
Radar 1 km
28

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Case 2 stratiform case (23/10/2006)
Spatially averaged vertical Profiles of
Reflectivity at 23 October 2006 at 19 UTC
ARPS 9 km Smedsmo
Radar 1 km
29

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Model evaluation was done using a model to
observation approach
Model to observation approach (preliminary
results) comparing radar reflectivities with
simulated reflectivities based on model
output. But a simple forward operator does not
take the errors in the radar observations into
account (atmospheric refraction and attenuation)
Advanced forward operator (Haase and Crewell,
2000), involving two steps 1. simulation of the
radar beam propagation including the effects of
the Earths curvature and atmospheric
refraction 2. determination of radar
reflectivity and attenuation
30

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Model evaluation was done using a model to
observation approach

• Once models are both having a satisfying set up,
  more advanced and quantitative evaluation
  techniques will be applied
• Traditional verification Scores False Alarm
  Ratio, Hit Rate, frequency bias, RMSE, Equitable
  Threat Score
• Evolution Histograms
• Categorical verification socres, discriminating
  between different sources of error minimisation
  of RMSE (Hoffman et al. 1995 and Du et al. 2000),
  isolating individual precipiation events and
  minimising MSE (Ebert and McBride, 2000),
  allowing a distinction between errors due to
  displacement, volume and pattern error.

31

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Preliminary conclusions

• ARPS is clearly having a problem in simulating
  the convective storms. The amount of
  precipitation at the ground is more or less ok,
  but there are no reflectivities from the
  convective storms at all due to very low rain
  mixing ratios. ARPS is able to simulate the
  ground precipitation more or less, but the
  precipitating area is too large in the horizontal
  and extends to high into the atmosphere
• LM captures the precipitation patterns for the
  convective case quite well, but tends to
  underestimate the ground precipitation amounts.
  The simulated reflectivities on the other hand
  are too high
• The model set up needs to be much more improved
  for both models in order to start a more
  extensive evaluation, using the Radar Simulation
  Model and applying a forward operator (Radiative
  Transfer Model), e.g. The Cloudy RTTOV-6
  (Chevalier et al, 2001) to investigate the
  models ability to simulate clouds.

32

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Preliminary conclusions

• ARPS is clearly having a problem in simulating
  the convective storms. The amount of
  precipitation at the ground is more or less ok,
  but there are no reflectivities from the
  convective storms at all due to very low rain
  mixing ratios. ARPS is able to simulate the
  ground precipitation more or less, but the
  precipitating area is too large in the horizontal
  and extends to high into the atmosphere
• LM captures the precipitation patterns for the
  convective case quite well, but tends to
  underestimate the ground precipitation amounts.
  The simulated reflectivities on the other hand
  are too high
• The model set up needs to be much more improved
  for both models in order to start a more
  extensive evaluation, using the Radar Simulation
  Model and applying a forward operator (Radiative
  Transfer Model), e.g. The Cloudy RTTOV-6
  (Chevalier et al, 2001) to investigate the
  models ability to simulate clouds.

33

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

Preliminary conclusions

• ARPS is clearly having a problem in simulating
  the convective storms. The amount of
  precipitation at the ground is more or less ok,
  but there are no reflectivities from the
  convective storms at all due to very low rain
  mixing ratios. ARPS is able to simulate the
  ground precipitation more or less, but the
  precipitating area is too large in the horizontal
  and extends to high into the atmosphere
• LM captures the precipitation patterns for the
  convective case quite well, but tends to
  underestimate the ground precipitation amounts.
  The simulated reflectivities on the other hand
  are too high
• The model set up needs to be much more improved
  for both models in order to start a more
  extensive evaluation, using the Radar Simulation
  Model and applying a forward operator (Radiative
  Transfer Model), e.g. The Cloudy RTTOV-6
  (Chevalier et al, 2001) to investigate the
  models ability to simulate clouds.

34

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

prospectives

• Both models will be improved testing the
  microphysics schemes, advection schemes,
  convection schemes and damping parameters.
• ARPS will be run on a 3 km resolution, similar to
  the current LMK horizontal resolution
• A more advanced forward operator will be applied
  (the Radar Simulation Model (Haase 2004)), a
  training at the SMHI is planned for the last week
  of March 2007
• Once both models seem to simulate both cases well
  enough, an extensive and much more quantitative
  model evaluation will be performed, also looking
  at the models ability to reproduce clouds
• Advanced techniques will be applied for the
  precipitation verification, discriminating
  between the different sources of forecast error
  (Hoffman et al (1995), Du et al (2000), Nehrkorn
  et al (2003), Ebert and McBride (2000).
• The most appropriate model with the most
  convenient model set up will be used in the
  furhter research to study the sensitivity of the
  precipitation characteristics to temperature
  increases in Belgium.

35

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

prospectives

• Both models will be improved testing the
  microphysics schemes, advection schemes,
  convection schemes and damping parameters.
• ARPS will be run on a 3 km resolution, similar to
  the current LMK horizontal resolution
• A more advanced forward operator will be applied
  (the Radar Simulation Model (Haase 2004)), a
  training at the SMHI is planned for the last week
  of March 2007
• Once both models seem to simulate both cases well
  enough, an extensive and much more quantitative
  model evaluation will be performed, also looking
  at the models ability to reproduce clouds
• Advanced techniques will be applied for the
  precipitation verification, discriminating
  between the different sources of forecast error
  (Hoffman et al (1995), Du et al (2000), Nehrkorn
  et al (2003), Ebert and McBride (2000).
• The most appropriate model with the most
  convenient model set up will be used in the
  furhter research to study the sensitivity of the
  precipitation characteristics to temperature
  increases in Belgium.

36

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

prospectives

• Both models will be improved testing the
  microphysics schemes, advection schemes,
  convection schemes and damping parameters.
• ARPS will be run on a 3 km resolution, similar to
  the current LMK horizontal resolution
• A more advanced forward operator will be applied
  (the Radar Simulation Model (Haase 2004)), a
  training at the SMHI is planned for the last week
  of March 2007
• Once both models seem to simulate both cases well
  enough, an extensive and much more quantitative
  model evaluation will be performed, also looking
  at the models ability to reproduce clouds
• Advanced techniques will be applied for the
  precipitation verification, discriminating
  between the different sources of forecast error
  (Hoffman et al (1995), Du et al (2000), Nehrkorn
  et al (2003), Ebert and McBride (2000).
• The most appropriate model with the most
  convenient model set up will be used in the
  furhter research to study the sensitivity of the
  precipitation characteristics to temperature
  increases in Belgium.

37

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

prospectives

• Both models will be improved testing the
  microphysics schemes, advection schemes,
  convection schemes and damping parameters.
• ARPS will be run on a 3 km resolution, similar to
  the current LMK horizontal resolution
• A more advanced forward operator will be applied
  (the Radar Simulation Model (Haase 2004)), a
  training at the SMHI is planned for the last week
  of March 2007
• Once both models seem to simulate both cases well
  enough, an extensive and much more quantitative
  model evaluation will be performed, also looking
  at the models ability to reproduce clouds
• Advanced techniques will be applied for the
  precipitation verification, discriminating
  between the different sources of forecast error
  (Hoffman et al (1995), Du et al (2000), Nehrkorn
  et al (2003), Ebert and McBride (2000).
• The most appropriate model with the most
  convenient model set up will be used in the
  furhter research to study the sensitivity of the
  precipitation characteristics to temperature
  increases in Belgium.

38

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

prospectives

• Both models will be improved testing the
  microphysics schemes, advection schemes,
  convection schemes and damping parameters.
• ARPS will be run on a 3 km resolution, similar to
  the current LMK horizontal resolution
• A more advanced forward operator will be applied
  (the Radar Simulation Model (Haase 2004)), a
  training at the SMHI is planned for the last week
  of March 2007
• Once both models seem to simulate both cases well
  enough, an extensive and much more quantitative
  model evaluation will be performed, also looking
  at the models ability to reproduce clouds
• Advanced techniques will be applied for the
  precipitation verification, discriminating
  between the different sources of forecast error
  (Hoffman et al (1995), Du et al (2000), Nehrkorn
  et al (2003), Ebert and McBride (2000).
• The most appropriate model with the most
  convenient model set up will be used in the
  furhter research to study the sensitivity of the
  precipitation characteristics to temperature
  increases in Belgium.

39

• Introduction Radar Model set
  up evaluation experiment
  preliminary conclusions

prospectives

• Both models will be improved testing the
  microphysics schemes, advection schemes,
  convection schemes and damping parameters.
• ARPS will be run on a 3 km resolution, similar to
  the current LMK horizontal resolution
• A more advanced forward operator will be applied
  (the Radar Simulation Model (Haase 2004)), a
  training at the SMHI is planned for the last week
  of March 2007
• Once both models seem to simulate both cases well
  enough, an extensive and much more quantitative
  model evaluation will be performed, also looking
  at the models ability to reproduce clouds
• Advanced techniques will be applied for the
  precipitation verification, discriminating
  between the different sources of forecast error
  (Hoffman et al (1995), Du et al (2000), Nehrkorn
  et al (2003), Ebert and McBride (2000).
• The most appropriate model with the most
  convenient model set up will be used in the
  further research to study the sensitivity of the
  precipitation characteristics to temperature
  increases in Belgium.