Inter-comparison of numerical models of fog
and low clouds : a proposal
Météo-France / CNRM – T. Bergot
Inter-comparison : Why?
Paris-CDG fog field experiment
2) The methodology of the inter-comparison
3) Phase 1 : cases study
4) Phase 2 : forecast quality over a long period
To be discussed precisely at the end of phase 1
inter-comparison : why?
1) The goal : link with COST722 objectives
NOT to create a competition between the different participants!
Learn about the value of different existing physical
Improve our understanding of the sensitivity to different physical
Hope : lead to some improvement in parameterisations
Investigate the potential (limitation?) of the different types of
2) The data
Fog field experiment at Paris CDG
Performed by Météo-France/CNRM
Available following a convention between Météo-France/CNRM
and the participants
Paris CDG fog field experiment
Ground measurements :
T / W inside the soil (between
ground and –50cm)
short- and long-wave radiations
Since December 2002
Airport terminal 1:
T / H%
Meteorological tower of 30m :
T / Hu%
measurements / 6min
2 winter seasons
frequency of dense fogs (visi < 600m) / hours
The distribution is characteristic to events dominated by
Dense fog :
1) Phase 1 : cases study
inter-comparison : the methodology
Goal : exhibit model deficiencies or weaknesses due to imperfect
representation of physical processes
Focus on specific cases well defined and observed : radiation fog
and low clouds (formation, evolution and dissipation)
Lead to improvements in the physical parameterisations
2) Phase 2 : evaluation of the forecast quality
Goal : investigate the potential and limitation of forecast
performed by numerical models over a long period in order to get
representative results in statistical sense
Phase 1 : methodology
Exhibit deficiencies due to imperfect representation of physical
processes involved in the formation and evolution of fog and low
The study of simulated boundary layer at local scale using highquality
observational data + effect on fog/low clouds simulations
Focus on vertical processes
Same initial conditions
No meso-scale flow (no advection + no vertical
Phase 1 : numerical models used
France : 1D COBEL-ISBA :
COBEL : http://www.rap.ucar.edu.staff/tardif/COBEL
ISBA : http://www.cnrm.meteo.fr/mc2/index.htm
Spain : 1D HIRLAM
Germany : 1D PAFOG
Switzerland : 1D COBEL-OSU
Denmark : ?
U.K. : ?
First level : 0.5m
20 levels below 200m
High resolution radiation scheme (232 spectral intervals)
Turbulence scheme : turbulent kinetic energy (TKE)
Fine mesh vertical grid
(Bergot 1993 ; Bergot and
Guedalia 1994 ;Guedalia
and Bergot, 1994)
Phase 1 : input data
All participants will used the same initial conditions given by
Météo-France/CNRM issued from the Paris CDG fog field
experiment (send on CD to participants)
Initial vertical atmospheric profiles
T, q, ql, U, V, other?
Between 0 and 5500m ? Step?
Initial soil profiles
T, Wl, Wi, other?
Profile between ground and 2m in depth?
Spatial heterogeneities : every 3h?
Cloud cover (low, medium, high)
Assimilation at local scale
1) Initialisation of dry atmosphere
Methodology : variational assimilation 1D-Var
Data : local observations, operational 3D NWP forecast
2) Initialisation of fog / low clouds
Define the depth of the cloudy area (minimization of errors on
the radiation fluxes divergence)
Correction of the atmospheric profiles below and inside the
cloudy area (dry / moist mixed area)
3) Initialisation of soil parameters
Soil temperature and moisture have a strong influence on the
surface cooling (energy budget at the surface : spin-up problem!)
Offline version of the ISBA model, driven by observed
Phase 1 : output data
Frequency : 30min? – duration : up to 12h?
Microphysics : visibility at 2m, ceiling, height of cloud/fog
Vertical profiles : T, Q, Ql – levels?
Radiation : short- and long-wave at 2m and 45m, other?
Turbulent exchanges : TKE? Turbulent fluxes? Mixing length?
Soil – vegetation – atmosphere exchanges : H, LE, other?
Send on CD to other participants
Phase 1 : evaluation
Comparison for a given validity (e.g. 06UTC) and a given lead
time (e.g. +6h)
Comparison between the output of participants + comparison
More efficient if performed centrally, but all participants should be
associated in the evaluation processes
Evaluation of microphysical parameters : formation, evolution and
dissipation (fog, low cloud and LVP conditions -visibility < 600m
and/or ceiling < 200ft)
Evaluation of boundary layer processes : profiles? evolution of
Evaluation of radiation processes : short-wave and long-wave?
Turbulent exchanges : TKE? turbulent fluxes? profiles? evolutions?
Soil – vegetation – atmosphere exchanges : H, LE? evolution?
Phase 1 : schedule
Participants : before October 2004
Description of the numerical model : before October 2004
Input data (convention between participants and Météo-
France/CNRM + distribution on CD) : before the end of 2004
Collection of output on CD : before April 2005
First analysis of results : mid 2005
Inter-comparison : Phase 2
Goal : learn about the quality of the different numerical models (1D,
3D, …) used for fog and low clouds forecasting in a statistical sense.
To be discussed precisely at the end of phase 1
Input : observations from CDG fog field experiment
Output : visibility, ceiling, LVP, T-Hu at 2m, wind at 10m, shortwave
and long-wave radiation at ground
Evaluation : Statistical verification – ROC curves, scatter-plots,
statistical scores (bias, RMS, other)
should be completed in collaboration with WG3 - task 1
“Determine how to evaluate the potential of existing methods”
!! END !!
Mesoscale terms :
Exchanges between soil,
vegetation and atmosphere
Radiative processes (IR+vis)
Turbulent processes (stable
Initialization / forcing
requirements / forecast
Local fog forecasting
visibility visibility / vertical thickness
Temperature at 1m (CI Cobel-Isba)
1D-Var : T / q surface boundary layer
Guess = previous COBEL-ISBA forecast (3h)
Altitude « observations » = 3D NWP Aladin forecast
Surface observations = local data (30m tower, 2m obs.)
Temperature at 1m
Temperature at 1m (observation)
Bias = 0.0°C
Std. Dev. = 0.3°C
Low clouds from Aladin
Initialisation of low clouds
cloud = mixed layer (moist variables)
Assimilation of radiation fluxes at 2m and 45m
IR fluxes when low
clouds are detected
low clouds from local assimilation
3D operational NWP models are not able to give realistic
forecasts (occurrence) of low clouds!