E'Minguzzi, G'Bonafe, M'Deserti, S'Jongen, M'Stortini, ARPASIM

1
1st Chimere workshop USE OF LOKAL MODELL FOR
THE METEOROLOGICAL INPUT OF CHIMERE Palaiseau,
France March 21-22, 2005 Enrico Minguzzi,
Giovanni Bonafè, Marco Deserti, Suzanne Jongen,
Michele Stortini HydroMeteorological Service of
Emilia Romagna Region (SIM), Bologna, Italy
2
Overview

• Objectives
• Use Lokal Modell (our operational meteorological
  model) as input for Chimere
• Analysis (and, wherever possible, verification)
  of LM outputs relevant for this application
• Tuning Chimere implementation for simulations
  over Northern Italy
• Summary (work in progress)
• LM and how we plan to use it
• LM verification (how LM errors will impact on
  Chimere performance?)
• PM underestimation testing erosion/resuspension
  scheme (analysis of Qsoil and U)
• Choice of operational time-step
• New scheme for nucleation routine
• Future work/Open questions
• Looking for advices

3
Lokal Modell

• The model
• Non-hydrostatic, limited area model (same class
  as MM5, Aladin,)
• First designed by the German Weather Service,
  presently developed by the COSMO consortium
  (weather services of Germany, Switzerland, Italy,
  Greece, Poland)
• Used for operational forecasts and research
  programs (see www.cosmo-model.org)

• Implementation at ARPA-SIM
• 7 km horizontal resolution
• 35 vertical levels (first levels 35, 110, 200 m)
• Two daily forecasts, lasting 72 hours
• Initial and boundary conditions by GME (German
  GCM)
• Data assimilation 12 hours nudging of GTS data

Operational domain of LM _at_ ARPA-SIM
4
Lokal Modell re-analysis
A re-analysis dataset is being built by storing
LM fields during assimilation cycle.

• Applications
• Long-term and scenario simulations
• Simulations whit simple dispersion models
• Meteorological characterisation of areas where no
  measurements are available
• Features
• Available from April 2003
• 10 parameters on model levels 26 surface fields
  included hourly resolution
• to prevent model drift, some surface fields are
  updated from GCM every 12 hours (so this is no
  exactly a continuous assimilation)
• presently only GTS data are included in
  assimilation cycle (i.e. relatively low
  resolution)

5
LM Chimere interface (1)

• It does exactly the same job as interf-mm5
• Input GRIB archive. Output input for diagmet.f
• All questionable calculations are left to
  diagmet
• Interface steps
• Extraction form archive
• Horizontal interpolation (rotated coordinates)
• Temporal interpolation and de-cumulation
• Calculation of pressure and mixing ratio, units
  conversions

• This is NOT a general GRIB-to-Chimere interface!!
  (sorry for that)
• GRIB format is very general and not as standard
  as it pretends to be. It is very difficult (and
  probably not worth) to handle all possibilities
• a lot of different options for validation times,
  geographic projections, vertical levels
• different models store different parameters (ex.
  Humidity, pressure)
• We have substantially modified the structure of
  Chimere calling scripts, to make it possible to
  prepare input files prior to model integration

6
LM Chimere interface (2)

• Chimere implementation at ARPA-SIM
• Input preparation split from model run
• All user modifications are in a single command
  file (keywords)
• All programs making calculations unchanged
• Prevent duplication of data
• When testing different model configurations,
  input files can be prepared only once
• Easier analysis of inputs

7
LM verification

• Background
• Meteorological fields are a very critical input,
  especially In Po valley
• LM has hardly ever been used to drive a chemical
  model
• Objectives
• Verification of operational forecasts focused on
  environmental applications (routinely LM
  verifications concentrated on precipitation)
• Find the best way to produce the input (select
  the most reliable parameters)
• How will LM errors affect Chimere performance?
  Are they common to most LAMs?
• Some systematic errors found in LM output
• Temperature profiles
• 10 meters wind

8
LM verification winter temperature
Examples of winter Temperature profiles at
S.P.Capofiume (rural site) Observations (black)
and short term LM forecasts at different
resolutions (colours)

• PBL looks always too cold in LM
• During night, LM strongly underestimates the
  strength of surface inversion (a 6 to 8 degrees
  inversion is frequent in Po valley)
• Possible causes surface fluxes (sensible vs
  latent heat?), turbulent diffusion in PBL
• Effects on Chimere wrong vertical mixing, high
  level emissions (stacks) not being considered
  above inversion

9
LM verification summer temperature
Examples of summer Temperature profiles at
S.P.Capofiume Observations (black) and short term
LM forecasts at different resolutions (colours)

• Temperature in the PBL is underestimated also in
  summer (known problem of LM), both in the diurnal
  mixed layer and in the nocturnal residual layer.
• No night-time inversion in LM (which often occurs
  in Po valley)
• Possible cause LHF overestimated, SHF
  underestimated (errors in soil moisture, soil
  type)
• Effects on Chimere ??

10
LM verification 10 m wind

• Verification dataset
• 74 stations in Po valley (46 plain, 10 hills, 18
  mount.)
• Hourly values, 1 year (apr 2003 mar 2004)
• Wind speed
• Overestimated on plain and hills, esp. during
  night
• MAE similar ?in plains/hills errors are more
  systematic
• Errors do not grow with validation time
• Wind direction
• Plains slightly better than mountains (MAE 60 vs
  75)

Wind direction, MAE
11
PM10 underestimation (1)

• The most pressing problem is PM10 underestimation
  
• ? activate erosion/resuspension scheme
• Forced by u and Soil Humidity (Qsoil)
• Qsoil from LM
• u (and usalt) estimated by Chimere (diagmet.f)
  starting form LM values of wind, q, (Note
  thermal mixing is taken into account through a
  term proportional to w)
• Note in the following, LM re-analysis were used

• Erosion emissions (negligible in this case)
• Increase with usalt
• Decrease with Qsoil
• Switched off over sea and if Qsoil gt 0.3 m3/m3
• Resuspension emissions
• Proportional to u1.43
• Decrease with Qsoil if Qsoil gt 0.15 m3/m3
• Switched off over sea and if Qsoil gt 0.3 m3/m3

12
PM10 underestimation (1)

• The most pressing problem is PM10 underestimation
  
• ? activate erosion/resuspension scheme
• Forced by u and Soil Humidity (Qsoil)
• Qsoil from LM
• u (and usalt) estimated by Chimere (diagmet.f)
  starting form LM values of wind, q, (Note
  thermal mixing is taken into account through a
  term proportional to w)
• Note in the following, LM re-analysis were used

• Erosion emissions (negligible in this case)
• Increase with usalt
• Decrease with Qsoil
• Switched off over sea and if Qsoil gt 0.3 m3/m3
• Resuspension emissions
• Proportional to u1.43
• Decrease with Qsoil if Qsoil gt 0.15 m3/m3
• Switched off over sea and if Qsoil gt 0.3 m3/m3

13
PM10 underestimation (2)

• Resulting additional emissions are not exactly
  what we expected
• biogenic PM emissions are comparable to
  anthropogenic in mountain areas
• but much smaller (at least 2 orders of magnitude)
  in Po valley

? Analysis of Soil Moisture and Friction
Velocity
14
Soil moisture analysis

• Horizontal distribution dominated by soil type
  (low resolution!)
• Little time variability (except annual cycle and
  precipitation events)
• May be sistematically overestimated
• No measurements available (at this time)
• Erosion/resusp often switched off, especially in
  winter days with no precipitation (where PM
  concentrations are higher!)

15
Friction velocity analysis
Chimere with LM input (wind, q, qv)
LM direct output (momentum flux)

• Pattern is similar
• LM values are almost double (0.2 vs 0.1)
• Chimere much lower during night
• Chimere diurnal cycle much stronger
• Note u affects also dry deposition, Kz, Zi

16
Friction Velocity validation (1)

• Metodology (preliminary)
• U measurements (sonic anemometer) available from
  a campaign held in winter 2002 at S.P.Capofiume
  (rural site in eastern Po Valley)
• U estimated by meteorological pre-processor
  Calmet (forced by surface observations and
  radiosoundings Holtslag and Van Ulden 1983) is
  available for both 2002 and 2004
• Calmet output for 2002 is in good agreement with
  observations we suppose that it is a good
  estimate also for 2004 data.
• Note
• In winter 2004 surface wind speed is
  significantly different from 2002, especially in
  afternoon hours
• Routine measurements of soil humidity and
  turbulence parameters at S.P.Capofiume will begin
  in the next months

17
Friction velocity validation (2)
18
Friction velocity validation (3)

• During night
• Chimere underestimates very low in specific days
  (0.01)
• LM overestimates (by a factor of 2)
• During day
• Chimere (probably) underestimates very strong
  diurnal cycle because of W term
• (this will be even stronger in summer)
• LM looks good
• Further work required

19
PM10 underestimation

• Possible solutions
• Retuning the scheme in order to get higher
  additional emissions (soil type, salt. u ...)
• if erosion/resuspension is really not important,
  try something else (ex. multiplying SOA)
• Take into account urban areas
• Approximately 10 of Po valley is urbanized (see
  pictures)
• PM underestimation may not so large in real
  rural stations
• A parameterisation for urban erosion/resuspension
  could be useful

Urbanized areas in Northern Italy (according to
Corine 1990)
Nocturnal illumination in Northern Italy
(satellite view)
20
Time step

• We have a problem with computer time
• looking for the longest possible time-step
• Chimere suggestion
• 60 (step1) for resolution gt 0.25
• 15 (step4) for resolution 5-10 km
• If we could use 20 (step3)
• CPU time reduced from 1h15 to 55 per day
• 1 hour saved in a 3 days forecast
• Test with 20 and comparison with 10 (control)
• Model did not explode
• Errors are usually negligible
• Errors do not accumulate during the simulation
• Some differences in secondary species (PM10,
  PM25), where high concentrations predicted
• Local differences in primary pollutants (NH3,
  H2SO4, NO) close to strong emitting sources
• Promising results test with strong wind required

21
Nucleation scheme
Surface PM10 concentration, ?g/m3, 18/02/2004 h
22Z. Old (left) and new (right) nucleation scheme

• A new nucleation scheme is being tested
• Different formulation (Kulmala et. al 2002
  instead of 1998)
• Allows description of very dry conditions
  (RHlt10)
• First test there are some differences, but
  rather small
• Further investigations required

Difference new-old
22
Recap.

• LM-Chimere interface has been built
• LM output looks promising, but it shows some
  systematic errors
• Wind speed overestimation
• Surface inversions
• The erosion/resuspension scheme needs to be
  adapted to Northern Italy
• Either tune the scheme
• Or improve inputs (soil water)
• Or change approach (urban)
• Friction velocity deserves further investigations
  (it also affects dry deposition, Kz, PBL height,
  ...)

23
Future work

• Near future work
• Test on a summer episode
• Operational simulations over Northern Italy
• Long-term verification of our regional forecasts
  (GEMS project)
• Extend LM verification (surface inversion,
  micromet. station at S.P.Capofiume...)
• Test direct use of other optional meteorological
  parameters (Zi, surf. fluxes, cloud water)
• Analysis of wet/dry deposition (we have a
  monitoring network for wet dep.)
• Improve soil type dataset
• Far future work (looking for advices,
  cooperation, common interest)
• Treatment of point sources (stacks)
• PM verification with satellite data
• Urban parameterisation for erosion/resuspension
• Measuring campaign for PM speciation
• Data assimilation of air quality monitoring data
  to initialize Chimere runs

24
References

• Vehkamäki, H. Kulmala, M. Napari, I.
  Lehtinen, K. E. J. Timmreck, C. Noppel, M.
  Laaksonen, A, 2002. An improved parameterization
  for sulfuric acid-water nucleation rates for
  tropospheric and stratospheric conditions
  Journal of Geophysical Research (Atmospheres),
  Volume 107, Issue D22, pp. AAC 3-1.
• Kulmala, Markku Laaksonen, Ari Pirjola, Liisa,
  1998 Parameterizations for sulfuric acid/water
  nucleation rates Journal of Geophysical
  Research, Volume 103, Issue D7, 8301-8308.
• Holtslag, Van Ulden, 1983 A simple scheme for
  daytime estimates of the surface fluxes from
  routine weather data Journal of Climate and
  Applied Meteorology, Volume 22, 517-529

25
Extra
26
LM verification 2m Temperature

• LM operational forecasts, 1year (apr 2003 mar
  2004), 284 stations in Northern Italy
• Plains (blue lines)
• diurnal cycle underestimated (positive bias in
  min, negative in max)
• annual variability overestimated (positive bias
  in summer max, negative in winter)
• RMS ? 2-3 C, better than in mountains
• Mountains (green lines)
• Night and winter are too cold
• A lot of possible sources of errors (altitude
  difference, extrapolation form 1st model level

BIAS
RMSE
27
LM verification 3D temperature evolution
LM forecast
Observations (twice daily radiosoundings)
Examples of time evolution of Temperature
profile, winter (left) and summer (right)

• Although the surface daily temperature excursion
  is underestimated, in the 200-1500 m layer this
  could be correct or even overestimated
• Further analysis required

28
Time step (2)

• Sensitivity to time step doubling NH3 surface
  concentrations, in an area of large emissions.
• This is one of the largest differences observed
  between 10 and 20 time-step simulations