WP3: Testing the quality of different operational meteorological forecasting systems for ...... data, like a meteorological pre-processor (e.g. CALMET in ARPA-SMI), as a further source ...... workshop proceedings, Risø National Lab, 55-58.
FUMAPEX Integrated Systems for Forecasting Urban Meteorology, Air Pollution and Population Exposure
Evaluation and inter-comparison of operational mesoscale models for FUMAPEX target cities Barbara Fay Lina Neunhäuserer
Deutscher Wetterdienst Research and Development Air Pollution Modelling 63067 Offenbach Germany José Luis Palau, Gorka Pérez-Landa, José Jaime Dieguez (CEAM) Viel Ødegaard (DNMI) Giovanni Bonafé, Suzanne Jongen (ARPA-SIM) Alix Rasmussen, Bjarne Amstrup, Alexander Baklanov (DMI) Ulrich Damrath (DWD) Contract number: EVK4-CT-2002-00097 WP3: Testing the quality of different operational meteorological forecasting systems for urban areas Deliverable D3.4, extended and final version, June 2005
1
1
METEOROLOGICAL CLASSIFICATION OF EPISODES.................................................................. 3 1.1 1.2
2
METEOROLOGICAL DESCRIPTION OF AIR POLLUTION EPISODES .................................................................. 4 MEASUREMENTS AND STATIONS ................................................................................................................ 6
ORGANISATION OF MODEL COMPARISON..................................................................................... 7 2.1 HARMONISED SET-UP OF MODEL SIMULATIONS FOR ALL PARTNERS ........................................................... 7 2.2 MODEL EVALUATION ................................................................................................................................. 8 2.2.1 Evaluation methodology .................................................................................................................. 8 2.2.2 Restrictions of evaluation and comparability .................................................................................. 9 2.3 TARGET PARAMETERS FOR COMPARISON AND EVALUATION .................................................................... 10 2.4 MODEL INTER-COMPARISON .................................................................................................................... 12
3
INFLUENCE OF HIGHER HORIZONTAL AND VERTICAL RESOLUTION............................... 14 3.1 HELSINKI ................................................................................................................................................. 14 3.1.1 Summary ........................................................................................................................................ 15 3.2 OSLO ........................................................................................................................................................ 15 3.2.1 Winter episodes November 2001 and January 2003 ..................................................................... 17 3.2.2 Summary ........................................................................................................................................ 23 3.3 VALENCIA ................................................................................................................................................ 24 3.3.1 Summer episode September 1999 .................................................................................................. 25 3.3.2 Summary ........................................................................................................................................ 29 3.4 BOLOGNA................................................................................................................................................. 29 3.4.1 Winter episode January 2002 ........................................................................................................ 30 3.4.2 Summer episode June 2002 ........................................................................................................... 35 3.4.3 Summary ........................................................................................................................................ 40
4
MODEL PERFORMANCE AND COMPARISON FOR SINGLE CITIES ........................................ 40 4.1 HELSINKI ................................................................................................................................................. 40 4.1.1 Single model characteristics, including effects of increased resolution ........................................ 41 4.1.2 Inter-comparison of model evaluation results according to single parameters ............................ 42 4.2 OSLO ........................................................................................................................................................ 44 4.2.1 Winter episodes November 2001 and January 2003 ..................................................................... 44 4.2.2 Summary ........................................................................................................................................ 51 4.3 VALENCIA ................................................................................................................................................ 52 4.3.1 Iberian Peninsula scale ................................................................................................................. 52 4.3.2 Highest model resolution............................................................................................................... 55 4.3.3 Summary ........................................................................................................................................ 63 4.4 BOLOGNA................................................................................................................................................. 64 4.4.1 Winter episode January 2002 ........................................................................................................ 64 4.4.2 Summer episode June 2002 ........................................................................................................... 66 4.4.3 Summary ........................................................................................................................................ 70
5
MODEL EVALUATION STATISTICS .................................................................................................. 71 5.1 EPISODES ................................................................................................................................................. 71 5.2 LONGER-TERM MODEL EVALUATION AND COMPARISON WITH EPISODE PERFORMANCE ........................... 72 5.2.1 DNMI ............................................................................................................................................. 74 5.2.2 Comparison DNMI HIRLAM/MM5 and LM/LAMI for Oslo-Blindern.......................................... 78 5.2.3 Lokalmodell LM (DWD), LAMI (ARPA) ....................................................................................... 80 5.2.4 FMI ................................................................................................................................................ 89 5.2.5 DMI................................................................................................................................................ 90 5.3 OVERVIEW OVER MODELS ........................................................................................................................ 92 5.4 CONCLUSIONS .......................................................................................................................................... 95
6
STRENGTHS AND WEAKNESSES OF MODELS FOR TASK CITY EPISODES.......................... 96 6.1 6.2 6.3 6.4
IMPACT OF INCREASED MODEL GRID RESOLUTION AT STATION LOCATIONS ............................................. 96 SINGLE MODEL CHARACTERISTICS FOR THE EPISODE SIMULATIONS ......................................................... 97 INTER-COMPARISON OF MODEL EPISODE EVALUATION RESULTS ACCORDING TO SINGLE PARAMETERS . 101 REASONS FOR MODEL DEFICIENCIES APPARENT IN THE EPISODE SIMULATIONS ...................................... 103
7 SUMMARY AND CONCLUSIONS FOR URBAN MESOSCALE MODELLING FOLLOWING FROM THE FUMPEX WP3 EVALUATION AND INTER-COMPARISON ............................................ 105 8
ACKNOWLEDGEMENTS..................................................................................................................... 107
9
LITERATURE ......................................................................................................................................... 107
2
Introduction An important aim of FUMAPEX is the provision of improved meteorological input data for the urban air pollution (UAP) models used in urban air quality information systems (UAQIFSs). Modelling and campaign experience show this to be an important prerequisite to the improvement of urban episode forecasting (and dispersion modelling in general). Another main objective lies in evaluating and inter-comparing operational NWP and mesoscale models for the same target cities and pollution episodes covering representative European regions of varying climates, topographies, and characteristic episode types. Several partners participate in FUMAPEX using different numerical weather prediction (NWP) or research mesoscale models for providing these meteorological input data. As a starting point and comparison standard for further improvements in the following WPs, therefore, the initial modelling task (WP3) is the analysis, evaluation, and inter-comparison of these operational models for urban episodes to investigate their performance in the urban environment. The emphasis lies on operational models and models settings, using just some higher pre-operational resolution, but not improved, urbanised or scale-adapted parameterisations. It must be stressed that this procedure highlights the capabilities and deficiencies of the models in their usual environment but makes a really systematic and objective model inter-comparison difficult. These results provide the basis for judging the improvements in physiographic data and physical parameterisations for the models performed in WP4, for improved interface development in WP5, and for the sensitivity studies with improved models in WP6. A detailed model overview was provided in deliverable D3.1 (M3.1) (Fay, 2003a). Purpose, set-up, and realisation of a successful comparison and evaluation exercise were outlined in the design of the model comparison study (M3.2, D3.2, Fay, 2003b) The parallel runs and preliminary analysis of operational meteorological models for one target city comprised phase 1 of the NWP modelling and comparison (M3.3, D3.3, Fay et al, 2004). Phase 2 then consist of the performance and final analysis of the comparison study for all major target cities (M3.4, D3.4).
1 Meteorological classification of episodes The agreed FUMAPEX target cities for air pollution episode analysis and UAQIFS implementation in FUMAPEX are Helsinki, Oslo, Turin, Bologna, and Castellon/ Valencia (plus Copenhagen for radioactivity emergency systems). During the search for one common city for phase 1, Helsinki was chosen because the largest number of partners agreed to provide model calculations, its orography is less complex than for the other target cities, and because of the availability of additional mast data. The Helsinki episodes were simulated by partners from all regions and their models while for the remaining target cities Scandinavian and Mediterranean partners will mainly simulate their regional target cities with a smaller choice of partners and models per city. Tab. 1 shows the models and partners participating in the episode simulations. P11 (UH) may later also simulate Helsinki episodes with their version of the nested MM5 making an interesting comparison with the integrated ECMWF/DNMI HIRLAM/MM5 simulations of P6 (DNMI).
3
Partner
Cities
P1 (DMI, Denmark)
Helsinki, Oslo, Bologna P2 (DWD, Germany) Helsinki, Oslo, (P4, ARPA-SMR, Valencia, Italy) Bologna, Turin P4 (CEAM, Spain)
Model
Resolution Model type
DMI HIRLAM LM (LAMI)
15, 5, 1.4 Operational NWP model km at weather service 7, 2.8, 1.1 Operational NWP model km at national weather service DWD (and regional service ARPA-SMR) RAMS 40, 13, 4.5, Often-used mesoscale 1.5 km research model (DNMI + 10(9), 3, 1 Operational NWP FMI km models at national HIRLAM) weather service +MM5
P9 (DNMI), Norway
Helsinki, Valencia Helsinki, Oslo
P6 (FMI, Finland)
Helsinki
FMI HIRLAM
P11 (UH, England)
Helsinki (London)
MM5
22 (33km) Operational NWP model at national weather service 81,27,9,3,1 Often-used mesoscale km research model
Tab. 1: WP3 and additional partners for the model comparison study and their operational mesoscale/NWP models. Thus, all NWP models in FUMAPEX will participate in modelling the Helsinki episodes: DMI HIRLAM, DNMI HIRLAM and MM5, FMI HIRLAM, LM/LAMI, RAMS, MM5. A detailed description of the NWP models and information on interfaces to UAP models are compiled in the Model Overview (D3.1) (Fay, 2003a).
1.1 Meteorological description of air pollution episodes The following description is based on the episode classification in D1.2 'Identification and classification of air pollution episodes in terms of pollutants, concentration levels and meteorological conditions' (Valkama, 2003) and on information from the FUMAPEX data website at FMI (http://fumapex.fmi.fi) provided by WP1. An episode is defined as the period when air pollutant concentrations exceed a set threshold value. This threshold is given for Helsinki measurements as the currently applicable EU limiting values (NO2: 200µg/m³ hourly with 18 allowed exceedances/year, O3: 180µg/m³ hourly, CO: 10µg/m³ gliding 8h-average, PM10: 50µg/m³ for 24h with 35 yearly exceedances). The EU directives require practical measures to be taken if air quality limits are exceeded. Pollution episodes typically vary in different climatic regions of Europe. However, low wind speeds and stable atmospheric stratification seem to be a key causing factor. In northern Europe, ground-based inversions, stable atmospheric stratification, low wind speed, and topography are the key meteorological factors (Sokhi et al., 2003, Kukkonen et al., 2005). Wintertime episodes prevail which involve particle formation or re4
suspension. The best meteorological predictors for elevated concentrations were found to be the temporal evolution of temperature inversions, atmospheric stability, and wind speed. In the Dec 1995 episode, the ground-based inversions were intensified by intensive cooling of the snow-covered surface due to long-wave radiation and were exceptionally strong for Southern Finland and for Europe also. An elevated atmospheric pressure is probably a necessary, but not a sufficient condition for the occurrence of an episode. In southern Europe (like Valencia), photochemical summertime episodes are typical and may persist for many weeks, interrupted by short destruction phases only. In central Europe, like Germany, France, but also the Po valley, both summer and winter episodes occur. The following episodes and target cities are involved: Helsinki: Episode 1: 27 to 29 Dec 1995 Local-emission wintertime inversion-induced period, high NO2, CO and PM10 (extremely strong inversion, NO2 and PM produced by local traffic). cold, dry air from North Atlantic, anti-cyclonic high-pressure belt above Finland, with warm advection aloft low/moderate westerly wind, little cloud, very strong ground inversion (-16°C in lowest 90m, -18°C in 120m), persistent stable to very stable stratification. sea ice belt only ca. 11km broad along coast, temperatures –25 to –2°C,. slight snowfall on 28 Dec warm front on 29 Dec with larger wind speeds ends episode simulated days: 27 to 30 December 1995 Episode 2: 22 to 31 March 1998 Local-emission springtime re-suspended particle episode (dry weather, melting snow, suspended dust), 2 periods 22-24 and 27-31 March anti-cyclonic very high pressure system very low, south(westerly) wind, ground inversion, moderately to extremely stable (nighttime) stratification with strong diurnal variation. Temperatures –10 to +4°C simulated days: 23 to 26 March 1998 Episode 3: 8-13 April 2002 Local-emission springtime re-suspended particle episode (inversion, prolonged high PM from re-suspension from sanded streets) anti-cyclonic pressure very low wind speed, sunny, dry (20% of monthly average precipitation.), cold cloudless nights, ground inversion, no snow or ice. simulated days: 9 to 12 April 2002 Oslo: Episode 3: 12 -22 Nov 2001 Local-emission re-suspended particle episode (high PM10) high pressure, dry, cold intermittent ground inversion and stronger northwesterly foehn winds simulated days: 17 to 19 Nov 2001 Episode 5: 1 - 11 Jan 2003 Local-emission inversion induced episode caused by warm advection above (highest NO2, high PM2.5) 5
high pressure, strong inversion, dry and cold near ground snow cover simulated days: 6 to 8 Jan 2003 Bologna: Episode 1: 17 Jan – 23 Jan 2002 regional episode, high PM10 exceedances strong thermal inversion over lower part of troposphere high pressure, weak winds from varying directions decrease of PM10 values due to widespread precipitation on 24 Jan 2002 simulated days: 21 to 23 Jan 2002 Episode 2: 13 – 24 June 2002 regional episode, high O3 the only episode of photochemical smog in a particularly rainy summer high pressure, low winds, clear sky nocturnal inversions fast moving thunderstorm outbreak ends episode on 24 June 2002 simulated days: 10 to 14 June 2002 Valencia: Episode 3: 29 Aug - 1 Oct 1999 O3 episode, no long-range transport high pressure stable atmospheric conditions, breeze circulation cold front passage from NW on 30 Sep 1999 simulated days: 26 to 30 Sep 1999
1.2 Measurements and stations The measurements of 3 to 7 meteorological stations were used for the evaluation of the episode simulations or for the longer-term evaluation: WMO (nation al) no.
Name
Lat.
Long.
02963
Jokioinen
60.82
23.50
02974 02978 02988
Helsinki-Vantaa Airport Helsinki-Kaisaniemi Isosaari
60.32 60.17 60.10
24.97 24.94 25.07
(0334) 014920 1384 (01992) (01991) 06180 06181
Kivenlahti mast OSLO-BLINDERN OSLO/GARDERMOEN TRYVANN HOVIN KOEBENHAVN/KASTRUP KOEBEN./JAEGERSBORG
60.18 59.93 60.20 59.98 59.92 55.62 55.75
24.65 10.721 11.08 10.67 10.80 12.65 12.52
6
Barom. Type height (station height) 103 Sounding, synoptic 56 synoptic 4 synoptic 5 Autom. Wea. 44 Sounding 96 204 515 95 5 40
Distance from city centre Rural
90km
Suburban 10km Urban centre Rural/island 8km Rural
15km
16140 16144 (20018) (20005) (13513) 16059 16061 16080
08284 08285 08286
03772 03779 03882 06601 07650 12556
BOLOGNA/BORGO PANIGALE S. PIETRO CAPO FIUME SALA BOLOGNESE SASSO MARCONI P.ZZA VIII AGOSTO TORINO/CASELLE TORINO/BRIC DELLA CROCE MILANO-LINATE AIRPORT TORINO-CSELT TORINO-ALENIA TORINO-CONSOLATA DRUENTO ORBASSANO VALENCIA/AEROPUERTO VALENCIA CASTELLON CORACHAR VILLAFRANCA SAN JORGE PENETA ONDA GRAO PORTO SAGUNTO LONDON/HEATHROW AIRPORT LONDON WEATHER CENTRE NORTH KENSINGTON BLOOMSBURY HERSTMONCEUX (SAMOS) BASEL-BINNINGEN MARSEILLE MARIGNANE AIR KRAKOW/BALICE
44.53 44.65 44.59 44.44 44.50 45.22 45.03 45.43 45.11 45.08 45.08 45.14 45.01 39.50 39.48 39.95 40.69 40.43 40.55 40.01 39.96 39.98 39.67 51.48 51.52 51.521 51.522 50.90 47.55 43.45 50.08
11.30 11.62 11.25 11.24 11.35 7.65 7.73 9.28 7.77 7.62 7.68 7.58 7.55 -0.47 -0.38 -0.07 0.08 -0.25 0.37 -0.06 -0.25 0.01 -0.02 -0.45
49 11 25 275 90 287 710 103 246 246 246 337 268 62 11 35 1231 1137 168 93 167 10 10 (24)
-0.12
(20)
-0.2142 -0.1247 0.32 7.58 5.23 19.78
(3m) (3m) (52) 317 36 (241)
Tab. 2: Meteorological stations used for the episode and longer-term evaluations, in blue: non-WMO stations. Further meteorological stations exist in some target cities, but are not used for various reasons, like in the Helsinki area as they were not measuring during all episodes and/or are not managed by the FUMAPEX partner. Observation frequency varies from station to station even inside one country: At the Kivenlahti mast, e.g., observations are taken at 10 min intervals while at all other stations they are 3-hourly. For the comparison with the simulated NWP results the mast data for the last 10 min before the full hour were used. Quality assurance of the measured data is performed by the relevant FUMAPEX partners or their associates and described on the FUMAPEX web http:\\fumapex.fmi.fi set up by WP1 (lead by FMI).
2 Organisation of model comparison 2.1 Harmonised set-up of model simulations for all partners As the end users of UAQIFSs attempt pollution forecasts, mainly NWP forecast data are compared and evaluated in WP3. The forecast error introduces an additional element of uncertainty into the comparison and evaluation but model analyses are not free of error, either. Depending on the model initialisation used, they suffer from initial imbalances for the parameters not initialised (many boundary layer parameters relevant to urban modelling). 7
Due to some episodes lying many years back and the sometimes necessary non-operational nesting of the single models, a use of forecasts/analyses diverging from operational procedures and/or between the single models could not altogether be avoided (details in chapter 2.4). Depending on the type of episode, the critical daily periods to be forecasted in UAQIFSs may be early in the morning (especially for winter/spring inversion episodes like in Helsinki) or in the afternoon (summer O3 episodes) or whole days for daily averages (i.e. PM10). The information is normally provided in the morning with the aim of preparing a forecast for the next day (e.g. for tomorrow's morning inversion or afternoon O3 maximum) but also as a short-term forecast for the same day to confirm and improve yesterday's forecast. For covering these different circumstances • 48h forecasts, starting at 00UTC for each day of the episode were performed. In the course of a 48h forecast, the forecasts for the smallest inner nest areas of episode calculation will increasingly be dominated by the boundary values of the outer nest even in the calm or light wind conditions which are frequently connected with pollution episodes. Partners chose inner nest areas between about (80km)² and (230km)². Still, 48h forecasts were evaluated in accordance with the varying requirements of UAQIFSs forecasting in the target cities. An overview of simulation specifications is provided in Tab. 3. Period Episode
Type of Forecast Period simulated calculation length Forecast 48h from Episodes (4 days) 00UTC
Comparison with FUMAPEX data sets
Comparison of
Model results + statistical scores Longer-term Forecast >=48h from Monthly + 1 year Mainly WMO Only statistical (analyses 00UTC statistics: SYNOP scores optional) 2002/2003/2004 stations (before Sep 04) Episode calculation, Ca. 150 x 150 km² to 250 x 250 km² for the inner target city proposed size of inner domains modelling nest (city area): Tab. 3: Calculations for model comparison.
2.2 Model evaluation 2.2.1 Evaluation methodology Systematic model evaluation is a complex task and may only be performed in a concentrated way with the limited resources available in FUMAPEX. Evaluation terminology, strategy, and methodology are defined in D3.2 'Model study design' (Fay, 2003b). The evaluation strategy will be integrative-diagnostic ( using only space/time distributions of modelled and measured parameters) and, wherever possible, process-oriented (additionally investigating the valid modelling of separate components or processes on which the modelled parameters depend) (Schluenzen, 1997). Also the model inter-comparison is regarded as an important step in the evaluation of numerical models (Thunis et al., 2003). The detailed conditions of model validity and other specifications (like physical parameterisations) are listed in the model documentation (D3.1) (Fay, 2003a), which is a necessary prerequisite for a successful model evaluation (Schluenzen, 1997).
8
All the mesocale models in FUMAPEX are in operational or frequent use and are thus well evaluated for their operational domains and resolutions: All NWP models are validated daily with real weather observations and operational statistics for their operational domains and resolutions of 5 to 15km, although not specifically for urban measurement stations. Both RAMS and MM5 are widely used and investigated by a large research community. Most models, however, are not operationally employed for NE Europe (Helsinki). For resolutions below 5km the NWP models are under development and many of the current physical parameterisations for the boundary layer, physiographic parameters etc. are not valid or unsuitable, especially for urban conditions which are not considered in the usual NWP application. Therefore, the models are not strictly 'valid' for urban applications in their operational state or when used with increased resolution only. The WP3 model evaluation is thus planned to quantify the (expected) deficiencies for the urban application using suitable evaluation techniques and to set a basis for judging the improvements and evaluations in the following workpackages. The tasks performed in D3.3. and D3.4 are: • episode validation only for meteorological parameters • single model validation against measurements • inter-model comparison of direct model output and statistical scores. • longer-term evaluation of model where available. The meteorological parameters are evaluated using inspection by eyeball and • horizontal field plots • vertical profiles at station locations • time series at station locations • time series of vertical profiles at station locations • vertical cross-sections • standard statistical scores (all agreed in D3.2 (Fay, 2003b) and at Valencia workshop). For harmonised graphics, GrADS graphics (free web software, identical visualisation and interpolation of partners’ grid point results to the location of the measurement stations) is to be used. A comprehensive GrADS tool for the performance of the above evaluations and interface modules to MMAS (see below) were provided by DWD (Neunhäuserer, 2004) to the partners. M. Sofiev (FMI) adapted the statistics tool Model and Measurement Analysis Software (MMAS) (Zlatev et al., 2001) originally designed for evaluating longer air quality simulations against measurements for use for standard meteorological parameters and scores. Both systems and their application in WP3 are specified in ‘Guidelines to the handling of FUMAPEX city episode data and modelling results’ (Neunhäuserer, 2004). 2.2.2 Restrictions of evaluation and comparability The quality and representativeness of observations and model results have to be considered. Partners describe their quality assurance measures in the FUMAPEX data sets; the used FMI (WMO) SYNOP station and mast data passed the standard QA checks. All operational NWP model results contain forecast errors but are quality-assured due to daily and long-term evaluation procedures. Model grid point simulations are regarded as representative for the model grid box which varies with the different resolutions of the partners’ simulations. The representativeness of observations depends on the qualified choice of station location and set-up and on the variability of the topography. The station set-up was not investigated, measurements are 9
assumed to represent 2m temperature and 10m winds and to comply with WMO station standards. It also has to be kept in mind that the modelling is still performed on the mesoscale and that microscale aspects of urban structure (like street canyons) are neither resolved nor yet parameterised in the models investigated. Simulation results are normally provided hourly whereas at most SYNOP stations observations are only reported every 3h. Some meteorological parameters are modelled as instantaneous values while the measurements are reported as time averages (wind speed and direction may be reported as 10min or hourly averages), in some cases observations are reported in different units (cloudiness in octas or decas of the sky dome etc.). Additionally, representativeness is influenced by the largely diverging volumes/scales for observations and model grids which also differ between the various models and model resolutions even for the same model. For avoiding large height deviations between the location of the observation station and the various locations of the nearest grid point of each different model and/or model resolution especially in mountainous areas, an identical interpolation of partners’ grid point results to the location of the measurement station was performed with the interpolation module of the GrADS visualisation program. No height correction is performed for any parameter between station and grid point levels. NWP simulations have always been validated against station measurements, but with increasing detail of the simulations the probability of larger deviations due to reduced averaging also increases at the station location. These deviations cast a lot of uncertainty on model validation against measurements and on model inter-comparison. Therefore, deterioration of model evaluation at station locations with increasing model evaluation must not be taken to signify that the model did not improve. It is important to also look at the behaviour of the meteorological fields in space and time for judging model improvements. The aim of the evaluation is to broaden the comparison of modelled against observed data in space and time (Skouloudis, 2000). Evaluation is performed for meteorological parameters as horizontal or vertical fields and for single observation stations. In the modelling context, it is important to show the spatial consistency and plausibility of parameters in the whole modelling region of interest. At measurement stations, the simulations are directly compared to observations providing useful information to pinpoint model problems for the evaluation in the following WPs also. Statistical scores for just single stations are hampered by the restricted amount of comparable results, non-identical averaging volumes for observations and model grid points which also differ between the various models and model resolutions even for the same model. With the common use of the GrADS at least the interpolation of the model grid results to the stations is performed with the same algorithm. Temporal consistency is investigated by comparing time series of simulations and observations at station locations and for whole parameter fields.
2.3 Target parameters for comparison and evaluation In WP3 for the evaluation of NWP models, only meteorological parameters are investigated. For the choice of meteorological target parameters, the following criteria were considered: • Standard meteorological parameters in operational NWP model validation (reference: operational validation at the DWD (Damrath, 2002)) • Meteorological parameters available in the FUMAPEX data sets • Meteorological parameters important for describing and modelling physical processes in cities. Tab. 4 lists the slightly modified list of D3.2 (Fay, 2003b) agreed upon by the partners at the FUMAPEX WP3-workshop in Valencia: 10
Parameter
FUMAPEX Calculated: measurements Met preproc / ABL models Blue: parameters to be (in evaluated by all partners FUMAPEX data sets) T2M (air temp. 2m above X ground) DD (wind direction 10m X above ground) FF (wind speed 10m X above ground) RH (relative humidity) X (rather than:) (TD) (dew point temp. (x) 2m above ground) snow cover, Tsnow, sea ice Total precipitation (x) Total cloudiness (x) Low cloud Medium cloud High cloud Ps (surface pressure) (x) Gusts (different classes) (x) Tmin (2m above ground) Tmax (2m above ground) Vertical profiles: (radio(x) P, T, TD, RH, DD, FF sounding) Radiation (short + long (x) (x) wave) Vis (visibility) (x) T25M (air temp. 25m (x) above ground) DeltaT (temp. Difference (x) between 25m and 2m) Mixing height (x)
Important Propose for city d for: ABL Episode
Proposed for:
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
Roughness length zo (as table for stations) Land-sea partitioning (as table for stations) Convective velocity scale w* Friction velocity u* and theta* or Stability (e.g. Pasquill stability class, Monin-Obukhov length L, vert. turb/Ri-no. /theta profiles)
X (x) (x) (x) (x) (x) X X X
X X
X X X
X
X
(x) X
(x) (x)
X
(x)
X
(X)
(x)
X
X
(x)
X
X
(x)
X
X
(x)
X
X
11
LongerTerm
X
X X (x) (x) (x) X (x) X X X
Parameter
FUMAPEX Calculated: measurements Met preproc / ABL models Blue: parameters to be (in evaluated by all partners FUMAPEX data sets) Sensible,latent heat (x) flux,total flux (=sen+lat+rad), upward flux = positive sign! Soil temperature and soil water External parameters (physiographic data) Initial soil conditions
Important Propose for city d for: ABL Episode X
X
X
Check !
X
Check !
X
Check !
Proposed for: LongerTerm
Check !
Tab. 4: Overview of measured / proposed meteorological parameters for model comparison and evaluation in FUMAPEX cities. In brackets: only available in some data sets / only optional for proposed evaluation. In blue: main parameters to be validated. Many of the standard validated parameters are measured in the FUMAPEX data sets or can be calculated from other measured parameters (like RH (measurements) from simulated T and TD). The main parameters for episode and longer-term evaluation are given in blue in Tab. 5. These will be evaluated in a more thorough way (e.g. using standard statistics). Additional parameters and processes of special importance in urban meteorology are the modified energy budget, surface heat and moisture fluxes (including the extra urban heating leading to the heat island effect (Oke,1987; Oke and Grimmond, 2000)) and altered stability conditions, the developing internal urban boundary layer causing modified 'mixing heights' during day and night - with modified shallow night-time boundary layers crucial to many episodes (Fisher et al., 2002). Most of these effects are not included in the basically ‘rural’ NWP models and contribute to the observed deficiencies of the simulated results.
2.4 Model inter-comparison When the operational NWP model chains are used by the partners, very heterogeneous models are compared. Unlike in some large European (dispersion) model exercises (ATMES, ETEX (ENSEMBLE, 2001), MESOCOM (Thunis et al., 2003)), the partners do not use standardised meteorological input data for their nested model chains, like ECWMF analyses, but they employ their own, if possible operational combination of global data, external parameter sets, and physical parameterisations as detailed in the Model Overview (D3.1) (Fay, 2003a). Thus, the real operational modelling situation may be portrayed but the search for causes of the differing model results is made more complicated. The operational NWP model versions may not be applied for specific target cities due to - city situated outside the operational model domain - standard analyses and forecasts not being available for older episodes. For Helsinki, only FMI could run their model with all resolutions in the original model domain. The other partners had to move some of their model domains for higher resolutions to Helsinki (DMI, DNMI) while others had to re-define all mesoscale model domains (CEAM, DWD). This also included procuring sets of more highly resolved external parameters for the new model domains, and in some cases of analyses, too (CEAM), and may 12
also imply less model skill due to less experienced tuning or data assimilation (no snow cover assimilation used in CEAM). Due to restricted availability of forecasts and analyses as initial and boundary values for the city simulations (archiving or re-configuration problems) for 'older' episodes, non-standard analyses or forecasts were used to start and/or run the model forecasts. At CEAM, RAMS is operated not with the usual NCEP data but with ERA-40 re-analyses of ECMWF as a standardised, high-quality, and well-documented data set, with 6hourly analyses providing the initial and boundary values for the RAMS forecasts. At DNMI, e.g. Helsinki episode 3 is run with the nesting of DNMI HIRLAM of 50 and 11km resolution nested into ECMWF forecasts (as operational before Mar 2003), with 2 inner MM5 nests of 3 and 1km, while the 2 older episodes are started from FMI HIRLAM 22km forecasts with 3 inner MM5 nests. The standard DWD global forecasts are not readily available before Dec 1999, and thus a single 1hour analysis of the ERA-40 re-analysis data set of ECMWF was used to start the Globalmodell GME 60km(operational until Sep 2004) forecasts for the Helsinki episodes 1 and 2 and the Valencia 1999 episode while at the historic time of the episodes before 2000 the much coarser (200km) Globalmodell GM was operational. 3 LM nests of 7km (operational) and 2.8 and 1.1km (experimental) are driven with these 1hourly GME forecast files, with full operational data assimilation only for GME and the 7km LM nest. Helsinki episodes were variously simulated at FMI with FMI HIRLAM versions 4, 5, and 6 using 6hourly ECMWF analyses for the episode calculations, with HIRLAM 5 and 6 using 3D variational data assimilation instead of the upper air optimum interpolation technique in HIRLAM 4 (Rantamäki et al., 2004). The UH (University of Hertfordshire) will perform their MM5 simulations of Helsinki episodes with several nests of their MM5 constellation driven with ECMWF data. Therefore, some models are driven with global analyses, others with global forecasts which adds an additional error to the simulations. An example is given for Helsinki episode 3 in Tab. 5. All episode calculations should at least be performed with the identical NWP model version per partner. For the longer-term calculations, operational changes in the NWP model code have to be accepted. Episode 3 (April 2002) Part.
Model version
DMI
ECMWF (analyses)
Hydrostat. Operational DNMI ECMWF Forecasts Hydrostat. Operational
Model version Forecasts DMI HIRLAM G 45km, 40 lay. 1-way nesting hydrostat. Operational DNMI HIRLAM50 50km, 17 lay. 1-way nesting hydrostat. Operational
Model version Forecasts DMI HIRLAM E 15km, 40 lay. 1-way nesting hydrostat. Operational DNMI HIRLAM10 11km, 17 lay. 1-way nesting hydrostat. Operational
13
Model version Forecasts DMI HIRLAM D 5km, 40 lay. 1-way nesting hydrostat. Experimental MM5 3.4 3km, 17 lay. 2-way nesting
Model version Forecasts DMI HIRLAM
hydrostat. Operational
hydrostat. Operational
1.4km, 40 lay. 1-way nesting hydrostat. Experimental MM5 3.4 1km, 17 lay. 2-way nesting
DWD
FMI
Globalmodell GME forecasts, 60km, 31 layers 1hourly data, hydrostat. Operational ECMWF, 6hourly data (analyses)
Hydrostat.
LM 3.5 7km, 35 lay., hourly 1-way nesting non-hydrostat. Operational FMI HIRLAM 4 44km, 31 layers, 1-way nest,hydro was operational FMI HIRLAM 6 22km,40 layers, 1-way nest,hydro operational
LM 3.5 2.8km, 45 lay., hourly 1-way nesting non-hydrostat. Experimental FMI HIRLAM 4 22km,31 layers, 1-way nest,hydro was operational
LM 3.5 1.1km, 45 lay., hourly 1-way nesting non-hydrostat. Experimental
Tab. 5: Details of the partners' model configuration for the example of Helsinki episode 3 (April 2002) simulations.
3 Influence of higher horizontal and vertical resolution The evaluation for episodes and cities was performed in the following steps: Chapter 3: Evaluation of each of the NWP models (also against measurements) • inspection of single meteorological parameters − covering the influence of increased resolution (if available) • statistical evaluation (if available) using the statistics package MMAS − for single meteorological parameters. Chapter 4: Inter-comparison of NWP model results of highest resolution (also against measurements) • inspection of single meteorological parameters in comparison • statistical evaluation (if available) using the statistics package MMAS − for single meteorological parameters. The main parameters evaluated are temperature and inversion characteristics, horizontal wind speed and direction, humidity, sensible and latent heat fluxes, radiation, stability parameters, and simulated PBL/ inversion height. They are chosen for their relevance to the description and dispersion modelling of European urban pollution episodes, and to the FUMAPEX work in other WPs. Still, only a very limited choice of results and their interpretation for all models can be presented in this report. More figures showing inter-comparison of partners’ results not mentioned in the main text and an additional DMI Helsinki evaluation are listed in two separate D3.3 appendices. The partners’ specific model versions, data assimilation, numerics, nesting set-up, physical parameterisation etc. are described in the ‘Model Overview’ D3.1 (Fay, 2003) and in the Helsinki comparison study D3.3 (Fay et al., 2004a).
3.1 Helsinki Helsinki is the city modelled by most partners and models in FUMAPEX. The results of the single models and the inter-comparison are presented in D3.3. More results and evaluation were provided by P1 for DMI HIRLAM and are presented in Baklanov et al., 2005, a new
14
appendix to D3.3 (FUMAPEX_D3.3_appendix_DMI-HIRLAM_Helsinki) about the Helsinki simulations. 3.1.1 Summary The December 1995 episode (extreme inversion and even stable daytime stratification) constitutes by far the most extreme FUMAPEX episode considering meteorological parameters. It also turned out to be the most difficult to simulate satisfactorily with operational NWP models resulting in large errors in temperature and inversion. This could be expected as most models perform badly in exactly those very stable inversion situations. The results are described in full in D3.3 and are summarised according to models as follows: Model DNMI MM5
Sim. resolution Results of increasing resolution 9 / 3 / 1km More detailed topography, no change in land-sea mask at station locations. Small change of v10m and vert. temp gradient near surface, some larger changes in T2m explained by cloud cover changes. Both improvement and deterioration compared with point measurements. DMI 15 / 5 / 1.4 km More detailed topography and increased roughness, changes in HIRLAM surface land cover (land-sea mask) and soil type at station locations. Kaisaniemi strongly influenced by sea grid point characteristics for all three resolutions, largest impact of grid refinement at Vantaa and Jokioinen probably due to changes in soil type / land use. FMI 22km ; Changes in ground surface properties and physiographic HIRLAM 33km (better parameters at station locations with substantial influence on humidity) results (Rantamäki et al., 2003). Some larger changes in T2m. LM 7 / 2.8 / 1.1km More detailed topography, influential changes in surface land cover (land-sea mask) and soil type at coastal station location of Kaisaniemi. Generally, small changes in T2m and v10m and in higher levels, larger in vert. wind and TKE. Partially large impact on T2m, v10m, surface fluxes, RH2m due to changed land-sea mask and thus soil type. Improvement mainly due to better land/sea mask and ext. Parameters and some deterioration compared with point measurements depending on station and parameter (also in MMAS statistics scores). CEAM 27/9/4.5/1.5km More detailed topography, no comparison due to overwriting of RAMS low-resolution results.
3.2 Oslo Oslo is situated in a mountain basin location between the shores of the narrow inland Oslo fjord and neighbouring hills reaching 600m thus experiencing land-sea and mountain-valley effects. Topography now has a large influence on meteorological fields. It induces wind channelling, orographic circulations, enhanced vertical motions as well as increased stability in dynamically cut-off valley bottoms. In mountain basins, the strongest inversions are observed in the valleys due to radiation cooling, cold air drainage from the mountains and decreased wind speeds due to sheltering. In the last event, stagnant pools form for many days and lead to very high pollutant concentrations as observed in Oslo (D1.2, Valkama and 15
Kukkonen, 2004). Changes of orography with increasing model resolution are therefore expected to influence simulations substantially. Both episodes also show a decisive influence of dynamical features. In November, strong surface inversions break down due to intermittent wind peaks related to northwesterly foehn, e.g. in the early hours of 19 Nov 2001. Warm advection aloft enhances the strong groundbased inversions in Jan 2003. The Oslo episodes of November 2001 and January 2003 were simulated with DMI HIRLAM, DNMI HIRLAM/MM5 and DWD LM. The Oslo fjord region is shown exemplarily for the inner LM-nest topography of 1.1km resolution (Fig. 1) which as a medium orography still underestimates the height differences (compare Tab. 6 and Tab. 7). Nevertheless, the increase in horizontal grid resolution leads to a considerable improvement of the model topography (Tab. 7).
Fig. 1: Topography of Oslo fjord region from LM with 1.1km resolution. Three Oslo meteorological measurement stations in lower right: Blindern (A), Valle Hovin (B), Tryvann (C). Two more locations for additional model evaluation in the Hallingdal area (D: mountain top, E: valley). Two vertical cross sections for model evaluation (red lines, cross section E = Hallingdal upper left, cross section B = Grorud valley lower right). Name
ID
h[m]
Type
Blindern
A
92
Urban (temp obs in 25m + 8m height)
Valle Hovin
B
89
Urban
Tryvann
C
515
Rural hill
Tab. 6: Description of Oslo meteorological measurement stations. Name
ID
Blindern
Model height H[m] MM5 1.0km
LM 7.0km
LM 2.8km
LM 1.1km
A
100
118.4
39.7
42.8
Valle Hovin
B
88
149.2
121.1
90.6
Tryvann
C
363
249.1
285.3
442.5
Mountain top
D
-
1039.9
1183.0
1245.1
Valley
E
-
740.7
456.7
261.9
Tab. 7: Model height at measurement stations and additional model evaluation locations. 16
3.2.1 Winter episodes November 2001 and January 2003 The impact of the refined topography on the model results is striking. The vertical temperature distribution, here given along the Hallingdal cross section E (Fig. 2), exhibits deeper and warmer valleys and higher and colder mountain tops. Depending on the overall situation, vertical velocities increase both in up- and downward direction in the presence of the more structured orography (Fig. 3), and the temperature layers become accordingly more disrupted.
Fig. 2: Oslo simulations with LM of 7, 2.8, 1.1 km resolution. Vertical cross section E along Hallingdal with temperature (shaded contours) and horizontal wind field (arrows), 17 Nov 2001, 00UTC +18h.
Fig. 3: Oslo simulations with LM of 7, 2.8, 1.1 km resolution. Vertical cross section along Hallingdal with vertical velocity (shaded contours) and horizontal wind field (arrows), 17 Nov 2001, 00UTC +18h. A similar topography influence on the temperature but less distinct due to the smoother terrain and the smaller differences in height is observed for the time series at the meteorological stations in the Oslo city area. Generally, LM time series of 2m temperature for the city stations show only little influence of the grid resolution for the November 2001 episode. In Fig. 4, 2m temperature is slightly decreasing with resolution, performing better for Blindern (Fig. 4, left) but underpredicting the Tryvann values (Fig. 4, middle). The impact of grid refinement is much larger for the January 2003 episode, leading to some improvements but for the urban stations to some deterioration as well (Fig. 5, left, middle). Reasons for the large overestimation of T2m are discussed in Chapter 4.2. In Hallingdal (location "E"), 2m temperatures rise as expected with increasing grid resolution (Fig. 4, right, Fig. 5, right).
17
Fig. 4: 2m temperature for LM 7.0km, 2.8km, 1.1km and observations, 18 Nov 2001. Left: Blindern. Middle: Tryvann. Right: Valley station Hallingdal (location "E").
Fig. 5: 2m temperature for LM 7.0km, 2.8km, 1.1km and observations, 06 Jan 2003. Left: Blindern. Middle: Tryvann. Right: Valley station Hallingdal (location "E"). Fig. 6 exemplarily shows LM time series of 10m wind speed for Blindern, Tryvann and the Hallingdal location. For the two episodes investigated here, the 10m wind speed decreases with increasing resolution for the valley stations of Blindern, Hovin and in Hallingdal, and it increases for the mountain stations Tryvann and location "D". The 1.1km (and 2.8km) LM simulations describe well even the persistently low wind speeds (below 2m/s especially for the Jan episode) that are a key factor in the valleys for both episodes. Higher resolution thus improves the forecasts for Blindern and Hovin compared to measurements, but leads to increased overprediction for Tryvann hill. In November, strong surface inversions break down due to intermittent wind peaks of 5-7m/s related to northwesterly foehn (D1.2, Valkama and Kukkonen, 2004), e.g. around midnight of 18 Nov in Blindern and midnight of 19 Nov 2001 in Hovin. Fig. 6 (left) shows LM time series of 10m wind speed for Blindern where the episode-relevant low wind speed, but also the observed occurrence and phase of the narrow wind peak are very well represented with the 1.1km LM (though with underpredicted wind maximum). The very sharp Hovin peak, however, that follows the 19 Nov with the maximum NOx and particle concentrations at all Oslo AQ stations for the whole episode, is not forecasted at all with any LM version (nor any other model, see Chapter 4.2.1).
Fig. 6: 10m wind speed for LM 7.0km, 2.8km, 1.1km and observations, forecast of 18 Nov 2001. Left: Blindern. Middle: Tryvann. Right: Valley station Hallingdal (location "E").
18
In Fig. 7, forecasts of the temperature inversion and observations for Blindern for 6 Jan 2003 are presented. The predictions above the inversions agree remarkably well with the soundings. Inversions are predicted with all model resolutions but underpredicted in strength and often elevated instead of ground-based. Near the surface, temperature overprediction may reach up to 11°C (Fig. 7, right). Above the lowest model layer, the model inversion strength actually decreases with increasing resolution and the forecast thus deteriorates (Fig. 7, right). This is contrary to the Helsinki episodes where increased model resolution always led to at least minor improvement in the inversions. The reasons are discussed in the section on wind fields below.
Fig. 7: LM simulations with 7, 2.8 and 1.1km resolution for Blindern on 06 Jan 2003. Left: time series of vertical temperature distribution, 48h forecast starting 00UTC+00h. Right: 00UTC+24h, LM and Blindern radiosonde measurements. Caution: Left fig. does not show the surface temperature, only T of lowest model level which is usually distinctly larger than T2m and Tsurface in winter inversions in LM, see right fig. The temperature inversion was further investigated using temperature gradients available from observations. At Hovin, temperature is measured at 8m and and 25m height above ground. Consequently, the temperature gradient has been defined as gradTHov (obs ) =
T 25m − T 8m 17 m
for observations and gradTHov (mod) =
Tk − T 2m ∆z
with Tk being the temperature value of the lowest model level at Hovin and ∆z the difference between the height of the lowest model level above ground at Hovin and 2m. For the LM, Tk is roughly 34m for 7.0km grid resolution and 20m for 2.8km and 1.1km grid resolution. To assess the model behaviour over a larger vertical range, another temperature gradient was introduced using the 2m temperatures measured at Tryvann and Blindern: gradTTry =
T 2m(Tryvann) − T 2m( Blindern) ∆z
with ∆z being the height difference, in reality or in the model respectively, between the stations of Tryvann and Blindern. In all cases, inversions are indicated by values larger zero. 19
Fig. 8 exemplarily shows time series of the temperature gradient for Hovin (Fig. 8, left, middle) and for Tryvann/Blindern (Fig. 8, right). As a general feature, the Hovin gradient is smaller for LM 7.0km grid resolution than for 2.8km and 1.1km. Similar computed gradients for Blindern, Tryvann and the Hallingdal locations reveal that this behaviour also is true for Blindern and the valley location "E" while Tryvann and the mountain top location "D" only show small differences due to grid refinement. Some influence of the horizontal resolution on the temperature gradient may be attributed to the fact that vertical resolution decreases (LM lowest model level height 7.0km: ca. 34m, 2.8km/1.1km: ca. 20m) and that for strong inversions, 2m temperature may also be overestimated by specific features of the stabilitydependent interpolation between surface and the lowest prognostic level in the LM (see below). Compared to the measurements at Valle Hovin, the gradient improves with increasing resolution for the November 2001 episode but totally misses the January 2003 observations which hardly show an inversion for the lowest 25m (Fig. 8, left, middle). The inversion between Tryvann and Blindern is underpredicted in most cases but generally improves with increasing horizontal resolution (Fig. 8, right). Obviously, the structure especially of strong inversions like those during the Jan 2003 episode or the Helsinki Dec 1995 episode is not correctly captured with the LM. For very stable winter episodes, the overestimation of 2m temperatures and 10m winds may be due to suspected insufficient stability influence on the transfer coefficients and the introduction of minimum diffusion coefficients for stable situations causing overestimation of vertical exchange that reduces the overall inversion strength (Raschendorfer et al., 2003).
Fig. 8: Temperature gradient [°C/m] for LM 7.0km, 2.8km, 1.1km and observations. Left: Valle Hovin, 18 Nov 2001. Middle: Valle Hovin, 06 Jan 2003. Right: Tryvann/Blindern, 06 Jan 2003. Topography has a large influence on temperature and wind fields in the complex Oslo location. Winter inversion episodes in Oslo are induced and strengthened by stagnating air in the mountain basin. A characteristic pattern emerges when looking at time series of vertical temperature profiles compared to vertical soundings (which are only available for Blindern). Warm advection mainly between 300 and 1500m starts on 6 Jan but seems to have been strong enough in layers between about 300 and 900m only in the 7km LM forecast when compared to the radiosounding profile for 7 Jan, 00UTC, in Fig. 7. For the higher resolutions, therefore, the upper part of the inversion is not enhanced enough by warm advection and remains too cold while the temperatures below about 300m are too high. Warm advection and thence the upper level forecast of the inversion are well predicted for some forecast hours, like for 6 Jan, 00UTC+48h and 7 Jan, 00UTC +00h and +24h. In the lowest model layer below 20m (34m in LM 7km), temperatures drop rapidly with the use of the profile functions leading to the sharp, non-observed near-surface inversion gradients shown in Fig. 7 and Fig. 8.
20
Deficient inversion forecasts may be accompanied, caused or intensified by deviations in the forecasts of the wind fields. Time series of LM vertical profiles of wind speed show increased values aloft of 300m above Blindern, Tryvann and Hovin already by 6 Jan, 18UTC, with overestimation of wind speed in the valley basin for LM 7km. This rise occurs at the same time for the valley stations Blindern and Valle Hovin as at Tryvann hill but is 12-15h premature at the valley stations (Fig. 5). Time series of vertical wind speed and direction profiles show a boundary layer wind regime below 600-900(1200)m distinct from the N-erly tropospheric wind above, with largest low-level shear of horizontal wind speed and direction in the 1.1km version from noon of 6 Jan until 8 Jan and sometimes beyond. According to vertical profiles of potential temperature, the stratification always remains stable but often with reduced stability in the lowest 200m for the higher resolutions. Available radiosoundings for Blindern for 7 Jan, 00UTC, describe varying over- and underprediction of modelled southerly to northwesterly winds in the lowest 600m for all LM versions compared to observed easterly winds (Fig. 9). Below 400m, air from S-erly directions is advected in the 2.8 and 1.1km LM versions possibly caused by the higher and steeper model topography leading to diversion of streamlines around the hills and a mountain blocking effect. With the diverted SW-erly winds in the higher resolutions, additionally warm air lingering for some hours above the unfrozen Oslo fjord (always 3° to 12°C warmer than the surroundings in LM) may be advected instead of colder inland air from the observed low level easterly winds (possibly drainage winds from the hills towards the also in reality unfrozen, comparatively warm fjord). This fits the overpredicted temperatures in the lower boundary levels in Fig. 7 together with the reduced stability in the high resolution versions. The same observation applies to Valle Hovin station, but here with a westerly model wind bias due to the station position east of the fjord. The better inversion quality for LM 7km above 400m is explained with the modelled and observed NW-erly warm advection. Nevertheless, due to the overpredicted boundary layer wind speed, vertical exchange is strong and lower-level temperatures too high (Fay and Neunhäuserer, 2005) The dependencies of meteorological parameters in this episode are highly non-linear with positive feedback mechanisms of possible topography, thermal and dynamic differences between the LM versions themselves and between simulations and observations. Predictability is even reduced in areas of large temperature gradients which are prone to erroneous temperature advection. The model skill for forecasting the inversion thus depends on the combined skill of soil, surface and surface layer parameterisations as well as model dynamics throughout the lower atmosphere, in this case of vertical wind shear and temperature advection, in the lowest levels and aloft. Nevertheless, winds below 200m remain low especially for LM 1.1km for the main episode days 7 to 9 Jan and show that the highly resolved LM is capable of preserving this cold pool characteristic for the Oslo valley stations.
Fig. 9. DWD LM vertical profiles of wind forecasts with 7, 2.8 and 1.1km, and observations, for Blindern on 6 Jan 2003, 00UTC+24h. Left: wind speed, right: wind direction.
21
Further potential reasons for the decrease of model inversion forecasting skill with increasing resolution may again be based on topography: The more detailed orography induces larger vertical velocities and influences the temperature layering not only where the differences in height are large (Fig. 2, Fig. 3) but also in the smoother Oslo area (Fig. 10, Fig. 11).With increasing height and steepness of mountains around Oslo, errors may partly stem from the terrain-following vertical coordinate system that was shown to deteriorate valley inversions close to mountains in comparison with simulations with a step orography model (Steppeler, 2002). Information to Oslo mixing heights with the LM gradient Richardson number scheme are provided in D5.4.
Fig. 10: Oslo simulations with LM of 7, 2.8, 1.1 km resolution. Vertical cross section B through Hovin along Grorud valley with temperature (shaded contours) and horizontal wind field (arrows), 07 Jan 2003, 00UTC+06h.Caution: Models do not show the surface temperature, only T of lowest model level which is usually distinctly larger than T2m in winter.
Fig. 11: Oslo simulations with LM of 7, 2.8, 1.1 km resolution. Vertical cross section B through Hovin along Grorud valley with vertical velocity (shaded contours) and horizontal wind field (arrows), 07 Jan 2003, 00UTC+06h. The DMI HIRLAM results improve with increased resolution compared to measurements in most of the cases for the two Oslo episodes discussed (Fig. 12, Fig. 13). As for the LM, differences in temperature due to grid refinement are largest for Blindern and Hovin during the January 2003 episode (Fig. 12, left) and smaller for Tryvann (Fig. 12, middle) and for the November 2001 episode (Fig. 12, right). The Tryvann values for the November 2001 episode are underpredicted and do not improve with resolution (Fig. 4, middle, and Fig. 12, right). Low 10m wind speeds are generally well represented with the highest DMI HIRLAM resolution of 1.1km not only for Blindern and Hovin (Fig. 13, left, middle) but also for Tryvann (Fig. 13, right, compare to Fig. 6, middle). The short-lived peaks in wind speed (e.g. around midnight of 18 Jan in Blindern and 19 Jan in Hovin) that interrupt the episode and influence concentrations values (see above) are hardly or not all captured with any resolution. There are no vertical DMI HIRLAM profiles available for the evaluation of inversions and potential mountain sheltering effects. The planetary boundary layer height of DMI HIRLAM increases with increasing horizontal grid resolution for both the November 2001 and the January 2003 episode (Fig. 14). The general pattern fits the episode with higher mixing heights/reduced concentration levels on 18 Jan and maximum concentrations/low mixing 22
heights on 19 Jan. As there are no radiosoundings for the November 2001 episode an absolute evaluation is not possible, but the PBL heights with the coarsest resolution generally follow expected values best. This is confirmed for the January 2003 episode with mixing heights
Fig. 12: 2m temperature for DMI HIRLAM 15.0km, 5.0km, 1.4km and observations. Left: Blindern, 06 Jan 2003. Middle: Tryvann, 06 Jan 2003. Right: Tryvann, 18 Nov 2001.
Fig. 13: 10m wind speed for DMI HIRLAM 15.0km, 5.0km, 1.4km and observations, 18 Nov 2001. Left: Blindern. Middle: Hovin. Right: Tryvann. from radiosoundings for Blindern between 350 and 600m compared to DMI HIRLAM 15km ones between 300 and 800m and mostly far larger ones for the higher resolutions. There is also the problem of using traditional mixing height schemes (even Richardson number schemes) for determining PBL heights in stable (nocturnal) conditions (D3.3, Fay et al., 2004a, Baklanov et al., 2002).
Fig. 14: Planetary boundary layer height for DMI HIRLAM 15.0km, 5.0km and 1.4km, 18 Nov 2001. Left: Blindern. Middle: Hovin. Right: Tryvann. DNMI MM5 model results for the highest resolution with MM5 1km are discussed and intercompared in Chapter 4.2. 3.2.2 Summary Contrary to Helsinki, the Oslo area is a mountain basin location where the impact of increased horizontal resolution on the model topography and consequently on the model results is large. 23
The 2m temperature generally improves slightly with increased resolution for both DMI HIRLAM and DWD LM for the November 2001 episode and shows larger differences with improvements and for the January 2003 episode. Both models underestimate the Tryvann November 2001 temperature observations increasingly with increasing resolution. 10m horizontal wind speeds generally perform quite well and improve with higher resolution except for the DWD LM forecast for Tryvann where model results increase due to the steeper topography and overpredict the measured values. The short-lived wind maxima of 5-7m/s that interrupt the episode and influence concentrations are better predicted with DWD LM but may also fail for DMI HIRLAM, DNMI MM5 (see chapter 4.2) and DWD LM. The DMI HIRLAM planetary boundary layer height increases (and deteriorates) with resolution for both episodes. The influence of horizontal resolution on inversions and vertical wind profiles could only be evaluated for the DWD LM. Inversions deteriorate with horizontal grid refinement especially in the lower levels for the strong inversion episode in January 2003. Additionally, temperature gradients between the lowest model level and 2m and between Tryvann and Blindern indicate that the structure mainly of strong inversions is not correctly described with the DWD LM. This may be due to a still too coarse vertical resolution close to the ground and/or to the constant flux assumptions for vertical profiles which is not valid in all cases. Detailed investigation and evaluation against radiosoundings for the January 2003 episode show that the quality of the inversion forecast depends on the skill of the combined forecast of vertical wind speed and direction, i.e. of vertical wind shear and temperature advection near the surface and aloft. The DWD LM performance in forecasting inversions varies from good to poor during the episode and the forecast, the low level temperatures are always overpredicted. The correct forecasting of the model dynamics also determines the existence and persistence of stagnant cold pools forming in the Oslo mountain basin and leading to the strongest inversion episodes. Further explanations may be the more detailed orography inducing increased vertical exchange and the use of terrain-following coordinates which both reduce inversions.
3.3 Valencia The modelling domain for the Valencia episode is centred on the Castellón conurbation (Fig. 15). This Mediterranean coastal site is surrounded by a variety of mixed industries including ceramic factories, nitrogen fertiliser plants and a power station. The atmospheric dynamics in the area is governed by the sea and the close mountains. When anticyclonic conditions prevail in the Iberian Peninsula, mesoscale circulations develop, and the numerous river valleys transversal to the coast channel the coastal air mass loaded with emissions towards inland areas (Millán et al., 1992; Millán et al., 1997; Millán et al., 2000). Moreover, repeated experimental results evidence that non-local (mesoscale) effects strongly determine PBL properties at urban scales. When modelling, the interaction between the different scales has to be reproduced properly.
24
Fig. 15: Valencia modelling area. Left: Iberian Peninsula scale. Right: Highest resolution domain including the positions of the stations used for model evaluation (COR: Coratxar, GRA: Grao de Castellón, JOR: Jordi, OND: Onda, POR: Port Sagunto, VIL: Vilafranca, ZOR: Zorita) and the position of a cross section (red line).
3.3.1 Summer episode September 1999 The Valencia September 1999 episode was simulated with CEAM RAMS and DWD LM. For model evaluation, the observations at seven stations are used, comprising four coastal stations and three mountain stations (Fig. 15, right). For both models, increasing the horizontal resolution results in a more detailed topography with higher mountains and deeper valleys like for the other FUMAPEX target cities. This directly affects the complexity of the wind field and the direction of the near-surface wind. Fig. 16 and Fig. 17 exemplarily show the streamlines of the 10m wind solution of the model suites of CEAM RAMS and DWD LM, respectively, for the forecast time 28 Sep 1999, 00UTC+24h. Comparing the results of the Peninsula scale (resolution: RAMS: 13.5km, LM: 7.0km) and the high-resolution scale (resolution: RAMS 1.5km, LM 1.1km) demonstrates the overall influence of the mesoscale conditions on the finer scale, but also the clearly increased channelling effect (in this case drainage flow along the valleys at night) due to the highly resolved orography. Influences of the improved topography can also be found for the 2m-temperature, mainly in the more mountainous areas. In Fig. 18 and Fig. 19, the 2m temperature results of CEAM RAMS and DWD LM are given for the forecast time 27 Sep 1999, 00UTC+13h. The 2m temperature distribution exhibits a much more detailed structure with higher resolution, as can be expected, with the mountain tops experiencing lower and the valley floors higher temperatures than at the Peninsula scale. This is also confirmed by comparing the 2m temperature time series of the different resolutions with observations at the station locations (Fig. 20).
25
Fig. 16: Streamlines of 10m wind velocity and orography, CEAM-RAMS, 28 Sep 1999, 00UTC+24h. Left: Iberian Peninsula scale. Right: High-resolution scale, black dots indicate station locations.
Fig. 17: Streamlines of 10m wind velocity and orography, DWD LM, 28 Sep 1999, 00UTC+24h. Left: Iberian Peninsula scale. Right: High-resolution scale, black dots indicate station locations.
Fig. 18: 2m temperature, CEAM-RAMS, 27 Sep 1999, 00UTC+13h. Left: Iberian Peninsula scale. Right: High-resolution scale, black dots indicate station locations.
26
Fig. 19: 2m temperature, DWD LM, 27 Sep 1999, 00UTC+13h. Left: Iberian Peninsula scale. Right: High-resolution scale, black dots indicate station locations. For the DWD LM, grid refinement leads to changes in the physiographic parameters at the station locations like for other target cities, as can be seen in Tab. 8. All parameters but the soil type are interpolated directly to the station location, while for the oil type the value of the closest grid point is given. The surface roughness shows a clear tendency to decrease with decreasing resolution at mountain stations (Coratxar, Vilafranca, Zorita) while it increases with decreasing resolution at coastal stations (Jordi, Onda, Port Sagunto). Name
Model height H[m] LM LM LM 7.0 2.8 1.1 Coratxar 1012.1 1151.1 1222.1 Grao 31.0 2.3 4.2 Jordi 167.1 162.7 165.5 Onda 220.4 178.0 148.0 Port Sagunto 14.8 13.2 9.2 Vilafranca 1104.0 1129.3 1087.3 Zorita 802.2 698.8 660.3
Surface land cover LM LM LM 7.0 2.8 1.1 1 1 1 0.60 0.59 0.94 1 1 1 0.97 1 1 0.61 0.94 1 1 1 1 1 1 1
soil type LM LM 7.0 2.8 5 5 6 6 5 5 6 5 6 6 5 5 5 5
LM 1.1 5 6 5 5 6 5 5
surface roughness LM LM LM 7.0 2.8 1.1 0.83 0.58 0.40 0.48 0.26 0.62 0.47 0.88 0.97 0.74 0.76 0.89 0.46 0.79 0.82 0.60 0.34 0.26 0.80 0.62 0.62
Tab. 8: LM physiographic parameters at the observation stations. Soil type: 5=loam, 6=clayey loam. Fig. 20 exemplarily shows time series of 2m temperature results of the three LM grid resolutions, compared to station observations. As already anticipated from the interpretation of the horizontal fields (Fig. 18 and Fig. 19), the 2m temperature exhibits nearly no influence of grid refinement at the coastal station of Onda (Fig. 20, left). For the mountain stations of Coratxar and Zorita, model results improve with increasing grid resolution, compared to measurements. In both cases, the model height approaches the real station height with increasing grid resolution, and the modelled 2m temperature decreases (Coratxar, Fig. 20 middle) or increases (Zorita, Fig. 20 left) accordingly.
27
Fig. 20: 2m temperature for LM 7.0km, 2.8km, 1.1km and observations, 28 Sep 1999. Left: Onda. Middle: Coratxar. Right: Zorita. 10m wind speed improves slightly with increased grid resolution, both at coastal and mountain stations, as is demonstrated in Fig. 21. Nevertheless, as stated before in the context of this project (Fay et al., 2004a; Neunhäuserer and Fay, 2004), the DWD LM tends to overestimate very low wind speeds. The diurnal cycle of daily sea breeze and drainage flow at night typical for this kind of episode as well as the front passing on 30 Sep 1999 is generally captured by the model (Fig. 22). The 10m wind direction results often improve with increasing grid resolution (Fig. 22, left) but sometimes also deteriorate (Fig. 22, middle).
Fig. 21: 10m wind speed for LM 7.0km, 2.8km, 1.1km and observations. Left: Port Sagunto, 26 Sep 1999. Middle: Onda, 26 Sep 1999. Right: Coratxar, 30 Sep 1999.
Fig. 22: 10m wind direction for LM 7.0km, 2.8km, 1.1km and observations. Left: Port Sagunto, 26 Sep 1999. Middle: Onda, 26 Sep 1999. Right: Coratxar, 30 Sep 1999. Sensible and latent heat fluxes generally show little influence of grid refinement (e.g. Fig. 23, left, and Fig. 24, left). Exceptions are the stations of Port Sagunto and Grao. In Port Sagunto, the latent heat fluxes decrease with the increase of land fraction at that station (compare Tab. 8), while the sensible heat fluxes increase at the same time. At Grao de Castellón, the latent heat fluxes increase with decreasing surface roughness from 7.0km to 2.8km resolution, and they decrease again with increasing surface roughness and land fraction from 2.8km to 1.1km. The behaviour of the sensible heat fluxes at Grao is vice versa.
28
Fig. 23: Latent heat fluxes for LM 7.0km, 2.8km, 1.1km, 26 Sep 1999. Left: Coratxar. Middle: Port Sagunto. Right: Grao.
Fig. 24: Sensible heat fluxes for LM 7.0km, 2.8km, 1.1km, 26 Sep 1999. Left: Coratxar. Middle: Port Sagunto. Right: Grao.
3.3.2 Summary Like Oslo, the Valencia/Castellón area shows a distinctly more detailed topography with increasing horizontal resolution. This affects the complexity of the wind field and the direction of the near surface winds. In the model results, increased channelling effects in the valleys and more clearly defined convergence/divergence lines are observed. The 2m temperatures generally improve slightly with increasing resolution for both CEAM RAMS and DWD LM, depending on the changes in the topographic height of the investigated stations, i.e. improvements are larger at mountain than at coastal stations. The 10m horizontal wind speed improves slightly at all stations, but low winds are sometimes overpredicted. The DWD LM description of the diurnal breeze circulation also improves in several, but not in all cases with increasing resolution. As for Helsinki, the physiographic parameters at the station locations may change significantly with increasing resolution. DWD LM latent and sensible heat generally show only little influence of horizontal resolution except for the coastal stations where a growing land fraction and changes in the surface roughness have a large impact on the model results.
3.4 Bologna Bologna is situated in the Northern Italian Po valley at the foot of the hills close to the Apennine mountains and is thus prone to local and meso-scale mountain-valley circulations transporting pollutants around the area and long-lasting stagnant valley basin episodes with capping inversions. Similar to the northern cities of Helsinki and Oslo, temperatures, inversions and wind fields play a key role for the development of winter air pollution episodes. As for Valencia, the main meteorological factors for summer ozone episodes are marked daily cycles and gradually increasing maximum daily values of temperature and wind
29
fields including local breeze circulations that determine the transport of ozone precursors and ozone concentrations. The inner modelling domain for the Bologna episodes comprises Bologna with the Po valley north and the Apennine Mountains south. It is shown exemplarily for the inner LAMI (ARPA version of LM) nest (Fig. 25).
Fig. 25: Bologna inner modelling area including the positions of the stations used for model evaluation (BOP: Borgo Panigale, P8A: Bologna Piazza VIII Agosto, SBA: Sala Bolognese, SMA: Sasso Marconi, SPC: San Pietro Capofiume). Left: topography 7.0km grid resolution (LAMI). Right: topography 1.1km grid resolution (LAMI). The Bologna January 2002 and June 2002 episodes were simulated with DMI HIRLAM and ARPA LAMI. Contrary to the LM simulations for the target cities, ARPA LAMI retains the original 35 layers for all resolutions. For model evaluation, the observations at five stations are used, comprising two urban-suburban stations, two rural stations in the Po valley and one rural station on the hills close to the Apennine Mountains (Fig. 25, Tab. 9). Name
ID
Altitude [m]
Type
Bologna Piazza VIII Agosto
P8A
90
Urban
Borgo Panigale
BOP
49
Suburban (airport)
Sala Bolognese
SBA
25
rural / industrial
San Pietro Campofiume
SPC
11
rural (more central Po valley)
Sasso Marconi
SMA
275
rural hill
Tab. 9: Information on observation stations for the Bologna episode evaluation.
3.4.1 Winter episode January 2002 Fig. 26 exemplarily shows an ARPA LAMI 2m temperature distribution for 7.0km and 1.1km horizontal resolution. The much more refined 1.1km topography leads to a finer structure of the temperature as expected. Generally, a slight decrease in the 2m temperature can be observed with increasing resolution for some regions in the modelling domain while in some cases a temperature increase is found (in Fig. 26, right, in Bologna and to the west). This behaviour is confirmed in the time series of 2m temperature (Fig. 27) with the values mainly 30
decreasing with increasing horizontal resolution and only increasing for the 22 Jan 2002 around noon and the urban (Fig. 27, left) and Po valley (Fig. 27, middle) stations. The time series in Fig. 27 also show that the ARPA LAMI generally overestimates the 2m temperatures for this episode, to some degree in the urban area (Fig. 27, left), considerably in the Po valley (Fig. 27, middle) and less at the mountain station (Fig. 27, right). A slight improvement is found due to increased horizontal resolution which may be attributed to differences in the cloudiness and accordingly in the incoming short wave radiation / available net radiation (Fig. 28).
Fig. 26: 2m temperature, ARPA LAMI, 22 Jan 2002, 00UTC+12h. Left: 7.0km grid resolution. Right: 1.1km grid resolution.
Fig. 27: Time series of 2m temperature for ARPA LAMI 7.0km, 2.8km, 1.1km and observations, 22 Jan 2002. Left: Bologna Piazza VIII Agosto. Middle: San Pietro Capofiume. Right: Sasso Marconi.
Fig. 28: Time series of surface net total radiation [Wm-2]for ARPA LAMI 7.0km, 2.8km, 1.1km and observations, 22 Jan 2002. Left: Bologna Piazza VIII Agosto. Middle: San Pietro Capofiume. Right: Sasso Marconi. The vertical temperature distribution does not show larger differences due to grid refinement as can be observed from the time series of vertical profiles and the single vertical profiles at station locations given in Fig. 29 and Fig. 30. As in the 2m temperature time series, slight 31
differences can be found in the daytime maximum value distribution (Fig. 29, left). At San Pietro Capofiume, observations from radiosoundings indicate a continuous temperature inversion (Fig. 29, right, bottom) which is not reproduced with the ARPA LAMI (Fig. 29, right, 1st to 3rd line). The corresponding single vertical profiles show that the inversion are elevated during the day and are situated below 400m-600m. The model overpredicts the temperature values in the inversion layer and hardly captures an inversion at all, but forecasts temperature quite well above (Fig. 30).
Fig. 29: Temperature time series of vertical profiles, 22 Jan 2002. Left: Bologna Piazza VIII Agosto, ARPA LAMI 7.0km (top), 2.8km (middle) and 1.1km (bottom). Right: San Pietro Capofiume, ARPA LAMI 7.0km (top), 2.8km (2nd line), 1.1km (3rd line) and observations, available every 12h only (bottom). Caution: Models do not show the surface temperature, only T of lowest model level which is usually larger than T2m and Tsurface in surface inversions.
Fig. 30: Temperature vertical profiles for ARPA LAMI 7.0km, 2.8km, 1.1km and observations, San Pietro Campofiume, 22 Jan 2002. Left: 00UTC+12h. Middle: 00UTC+24h. Right: 00UTC+36h. Caution: Models do not show the surface temperature, only T of lowest model level which is usually larger than T2m and Tsurface in surface inversions. The January 2002 episode exhibits very low wind speeds until 24 Jan around noon when widespread precipitation occurs. The values are generally captured quite well or slightly overpredicted with the ARPA LAMI except for the 22 Jan around noon when the model results are too large mainly for the two urban stations close to the Apennine (Fig. 31, left) and the hill station Sasso Marconi. The increase in wind speed towards the end of the episode is predicted with the model but started too early (Fig. 31, right). Differences due to grid refinement are small. This is mostly true for wind direction as well.
32
Fig. 31: Time series of 10m wind speed for ARPA LAMI 7.0km, 2.8km, 1.1km and observations. Left: Bologna Piazza VIII Agosto, 22 Jan 2002. Middle: San Pietro Capofiume, 22 Jan 2002. Right: Bologna Piazza VIII Agosto, 23 Jan 2002. Fig. 32 shows time series of vertical profiles for the horizontal wind direction. Above 300m600m (corresponding to the observed inversion), the model simulates synoptic scale wind directions changing from Northwest on 21 Jan 2002 over West on 22 Jan to Southwest on 23 Jan and than back to North-northwest on the afternoon of 24 Jan. In the surface layer, model winds often come from easterly directions during the day while they adjust to the southwesterly wind direction of the above layers during the night. Compared to measurements (attention: only available every 12h, therefore interpolated in time!), the simulated wind directions are close to the observations in the higher atmospheric levels and even capture strong vertical wind shear of horizontal winds as around 6UTC on 22 Jan even for low simulated wind speeds (
