0

Modelling Evaluation

Deliverable D-B3.3

December 2018

1

0

CONTENTS

Contents ....................................................................................................... 0

1. Executive Summary ................................................................................. 1

2. Introduction .......................................................................................... 1

3. Air Quality Modelling Setup and Configuration ............................................... 1

4. Model Evaluation .................................................................................... 3

5. Modelling Results .................................................................................... 5

5.1. Meteorological modelling Evaluation ............................................................ 5

5.2. Air Quality Modelling Evaluation ................................................................. 7 5.2.1. Particulate Matter ............................................................................ 7 5.2.2. Heavy Metals ................................................................................. 11

6. Conclusions .......................................................................................... 16

7. References ........................................................................................... 17

1

1. EXECUTIVE SUMMARY

The present deliverable presents the modelling evaluation activities in the scope of the

goals of the action B3. The main purpose of this action is to determine the exposure of the

target population (children) to the selected pollutants using a modelling approach. A

dispersion-exposure modelling system was selected and will be adapted to run in

operational mode to allow its use for particulate matter air pollution and exposure control

through its integration in the LIFE Index-Air Management Tool.

The reports describes the modelling setup, the modelling results for the meteorological and

air quality modelling simulations over the Lisbon urban area between the 1st September and

the 31st December of 2017, and the main conclusions of the analysed results.

2. INTRODUCTION

The Action B3 comprises the compilation of emission data, preparation of meteorological

inputs, air quality and exposure modelling setup and application.

Air quality and human exposure numerical modeling are used to determine human exposure:

first a meteorological model was already applied over the different domains, and then

together with emission data, air quality simulations produce the pollutants concentration

patterns.

The air quality modelling system applied in this study includes the Weather Research &

Forecasting (WRF) (Version 3.7.1) model (Skamarock et al., 2008) and Comprehensive Air

Quality Model with Extensions (CAMx) (Version 6.4) (ENVIRON, 2015). The WRF model is a

next generation mesoscale numerical weather prediction system designed to serve both

operational forecasting and atmospheric research needs. CAMx is a three-dimensional (3D)

chemical transport model suited for the simulations of the emission, dispersion, chemical

reactions, and removal of pollutants in the troposphere based on the integration of the

continuity equation for each chemical species on a system of nested 3D grids. Both models

have been extensively tested and shown to produce robust and realistic results (Ferreira et

al., 2012; Nopmongcol et al., 2012; Sá et al., 2016).

The main purpose of this report is to assess the performance of the meteorological

(temperature and wind velocity) and air quality modelling (particulate matter and heavy

metals) simulations over the Lisbon urban area during the air quality experimental

campaign, between 1st September and 31st December of 2017, where relevant data for the

evaluation were specifically acquired.

3. AIR QUALITY MODELLING SETUP AND CONFIGURATION

In this section, the air quality modelling setup and configuration is presented and the

methodology applied to improve the European atmospheric emission inventory over the

study area is described.

The air quality modelling system includes the WRF and CAMx models. The WRF-CAMx

modelling system was applied to the Lisbon region following a nesting technique. Three

online-nested domains with increasing resolution at a downscaling ratio of five were used:

domain 1 (D1) with a spatial resolution of 25 km2, covering the Europe and part of North

Africa; D2 with a spatial resolution of 5 km2, comprising Portugal; and D3 with a spatial

resolution of 1 km resolution, covering the Lisbon urban area. Figure 1 shows the European

2

domain (D1, 25x25 km2 resolution) and nested domains over Portugal (D2, 5x5 km2

resolution) and Lisbon (D3, 1x1 km2 resolution).

Figure 1. Simulation domains used by the WRF-CAMx modelling system: parent grid (D1, 25×25 km2

resolution) and nested domains (D2, 5×5 km2 resolution; D3, 1×1 km2 resolution).

The meteorological model WRF-ARW was initialized with global meteorological fields from

the European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis Interim

(ERA-Interim) model data (http://www.ecmwf.int/en/research/climate-reanalysis/era-

interim) with 1° spatial resolution and temporal resolution of 6 h for surface and pressure

levels. Considering the findings of previously studies performed over Portugal (Aquilina et

al., 2005; Carvalho et al., 2012; Monteiro et al., 2015) the set of parametrizations used in

the model physical configuration included: WRF Single-Moment 6-class Microphysical Scheme

(Hong and Lim, 2006); Dudhia Shortwave radiation scheme (Dudhia, 1989); Rapid Radiative

Transfer Model longwave radiation model (RRTMG) (Mlawer et al., 1997); MM5 similarity

surface layer scheme (Zhang and Anthes, 1982); Noah Land Surface Model (Tewari et al.,

2004) with soil temperature and moisture in four layers, fractional snow cover and frozen

soil physics; Yonsei University Planetary Boundary Layer scheme (Hong and Lim, 2006) and

Grell-Freitas Ensemble Scheme for cumulus parametrization (Grell and Freitas, 2014) (this

last, only for D1 and D2).

For the air quality model, no initial and boundary conditions for the heavy metals were

considered for the coarse domain. For the remaining air pollutants the outputs from the

global chemical model MOZART (NCAR, 2010) at every 6 hours were used. Natural emissions

of sea-salt particles (i.e. sodium, chloride and sulfate) were calculated from the WRF-ARW

outputs using flux equations for open ocean (Ovadnevaite et al., 2014) and breaking waves

in the surf zone (Gong, 2003). Anthropogenic emissions were taken from the most recent

European emission inventory based on Member States submissions for the year 2015. The

European Monitoring and Evaluation Programme (EMEP) inventory, with a horizontal

resolution of 0.1 degrees (approximately 10 km), comprises annual emission totals by

activity sector for gases and particulate species including metals (i.e. cadmium (Cd) and

lead (Pb)). For the Arsenic (As) and Nickel (Ni), total emissions by European country were

3

used and spatially distributed based on the spatial distribution of cadmium/lead/mercury

with higher correlation between emission activity sectors. The large emission sources over

Europe were also considered as point emission sources by the air quality modelling system.

The atmospheric emissions were disaggregated to the Portuguese region (i.e. D2 and D3)

and speciated into the Carbon Bond 6 chemical mechanism (CB6) gaseous species and into

the default particulate species considered by CAMx. The chemical mechanism description

and treatment was adapted to additionally include heavy metals (i.e. Pb, Ni, As and Cd) as

non-reactive fine particles. It is generally considered that these air pollutants are

accumulated on sub-micron fraction of the aerosols and they could be transported in the

atmosphere without modifications in their chemical and aggregate state (González et al.,

2012; Hutzell and Luecken, 2008). In order to improve the spatial resolution of the EMEP

inventory, the emissions were spatially disaggregated to 5 and 1 km2 of horizontal resolution

(i.e. D2 and D3) considering different proxies. The following Selected Nomenclature for Air

Pollution (SNAP) activities were considered: public power stations (S1); commercial and

residential combustion (S2); industrial combustion and production processes (S34);

extraction and distribution of fossil fuels and geothermal energy (S5); solvent and other

product use (S6); road transport (S7); off-road (S81); maritime transport (S82); aviation

(S83); waste treatment and disposal (S9); and agriculture (S10). For the S1, the specific

location of public power stations was used. A “bottom up” approach was applied for the

activity S2, taking into account the wood consumption per district, the type of residential

combustion equipment, and emission factors from the Portuguese Agency for the

Environment (APA) (Silveira et al., 2017). For the S34, the location of industries was used

based on Portuguese land use data (DGT, 2018). The buildings locations, using the

information from the OpenStreeMap (OpenStreetMap contributors, 2017), was considered for

the S5 and S6. The road transport emissions (i.e. S7) were calculated applying the Transport

Emission Model for Line Sources (TREM) (Borrego et al., 2004, 2003, 2000) over Portugal.

The off-road, maritime port and airport activities locations were considered for the S81, S82

and S83, respectively. Finally, the national land use data regarding farming fields was used

to spatially disaggregate emissions of agriculture (i.e. S10) (DGT, 2018).

4. MODEL EVALUATION

The models evaluation consisted on their validation and was performed through the direct

comparison of modelled results against measurements, in the form of quantitative statistical

metrics. Quantitative analysis is considered an accurate and detailed method to assure

quality and demonstrate that a model can provide reliable results for a desirable purpose.

Data from three meteorological stations located in the innermost domain (D3) were used to

evaluate the WRF model performance. The measured data were acquired at the National

Oceanic and Atmospheric Administration (NOAA) (Reynolds et al., 2010) database from the

National Centers for Environmental Information and at the Portuguese meteorological

stations network (IPMA), in an hourly base. The evaluation was performed for two

meteorological variables: temperature and wind velocity. It should be noted that all the

meteorological stations used to this analysis had more than 75% of data collection

efficiency. The location of each meteorological station used in this analysis is presented in

Table 1.

4

Table 1. Location of the meteorological stations considered for WRF model evaluation.

Station Code Name Database Longitude

(°) Latitude

(°)

85790/1200579 Gago Coutinho NOAA/IPMA -9.11 38.77

85350 Geofísica NOAA -9.12 38.73

85360 Aeroporto NOAA -9.13 38.79

Data acquired in air quality monitoring stations operating on a regular basis in the study

region and outdoor measurements of heavy metals from the LIFE INDEX-AIR monitoring

campaigns were used for the validation of the CAMx model application. In the Table 2, the

location and characteristics (if available) of each air quality monitoring station is shown.

Table 2. Location and characteristics of the air quality monitoring stations considered for the evaluation of the CAMx model.

Station Code Name Station type Longitude

(°) Latitude

(°)

ALV Alverca Urban background -9.04 38.90

FID Fidalguinhos Urban background -9.05 38.65

LAR Laranjeiro Urban background -9.16 38.66

LOU Loures-Centro Urban background -9.17 38.83

MEM Mem Martins Urban background -9.35 38.78

OLI Olivais Urban background -9.11 38.77

QMA Quinta do Marquês Urban background -9.32 38.70

REB Reboleira Urban background -9.23 38.75

PPI Paio Pires Suburban industrial -9.08 38.63

ESC Escavadeira Urban industrial -9.07 38.66

LAV Lavradio Urban industrial -9.05 38.67

BEN Santa Cruz de Benfica Urban traffic -9.21 38.75

ODI Odivelas-Ramada Urban traffic -9.18 38.80

AVL Avenida da Liberdade Urban traffic -9.15 38.72

ENT Entrecampos Urban traffic -9.15 38.75

H1 House H1 - -9.09 38.78

H2 House H2 - -9.15 38.75

H3 House H3 - -9.14 38.73

H4 House H4 - -9.17 38.71

H5 House H5 - -9.15 38.74

H6 House H6 - -9.15 38.72

H7 House H7 - -9.12 38.77

H8 House H8 - -9.10 38.78

H9 House H9 - -9.15 38.74

H10 House H10 - -9.12 38.72

I1 House I1 - -9.10 38.78

I2 House I2 - -9.09 38.78

I3 House I3 - -9.09 38.78

The performance of the WRF-CAMx system was assessed by applying the BOOT (i.e.,

bootstrap resampling method) Statistical Model Evaluation Software Package, version 2.0,

by Chang and Hanna (2005). This is widely used in model evaluation exercises (e.g., Mosca

et al., 1998; Nappo and Essa, 2001; Ichikawa and Sada, 2002). Four main statistical

parameters were considered: (i) the correlation coefficient (r) - to provide an indication of

the correspondence of the timing and evolution of observed and simulated values; (ii) the

mean bias error (MBE) - the average difference between simulated and observed values; (iii)

the root mean square error (RMSE) - which gives important information about the skill in

predicting the magnitude of a variable; and (iv) the normalized mean square error (NMSE)

in relation to the multiplication of observed and modelled mean values. Unlike BOOT, the

MBE expresses the arithmetic difference between model predictions and observations

5

(consequently, positive MBE indicates an overestimation). Both RMSE and MBE have the units

of the variables being evaluated and depend on the magnitude of the mean variables. Time-

series were generated to complement the quantitative statistical analysis.

5. MODELLING RESULTS

Given the methodology described in section 4, the meteorological and air quality results are

analyzed following two approaches: i) modelling system evaluation, which include both

quantitative and qualitative analysis; and ii) spatial evaluation, which consists in the

analysis of the spatial distribution of both meteorological variables and air pollutants levels.

5.1. METEOROLOGICAL MODELLING EVALUATION

Table 3 shows the obtained statistical performance indicators for temperature and wind

speed at the three meteorological stations, averaged over the simulation period.

Table 3. Evaluation of the WRF model performance for the period 1st September – 31st December 2017 (2928 hours). The temperature and the wind velocity data were obtained at 2-m and 10-m

above ground, respectively.

Station

Temperature Wind velocity

r (-)

MBE

(°C)

NMSE (-)

r (-)

MBE (m.s-1)

NMSE (-)

“Gago Coutinho” 0.99 -0.04 0.00 0.88 -0.03 0.03

“Geofísica” 0.98 -0.60 0.00 0.60 0.50 0.06

“Aeroporto” 0.99 -0.24 0.00 0.88 -0.45 0.05

The statistical analysis for temperature exhibits a good relation between modelled and

measured data with a correlation coefficient greater than 0.9, for the three analyzed

stations, and with a NMSE that reaches the ideal value (0). There is a consistent

underestimation of the temperature values evidenced by the negative MBE, with values

ranging between -0.6 (at the “Geofísica” station) and -0.04 (at “Gago Coutinho” station).

This slight underestimation could be attributed to the urban heat island developed in the

urban areas, which is not appropriately described by the model (Papanastasiou et al., 2010).

The model presents, for the wind velocity, a correlation coefficient greater than 0.8 for two

of the three meteorological stations; for these stations, a NMSE of 0.03 and 0.05 (close to

the ideal 0) was obtained. At the “Geofísica” station, the correlation between modelled and

measured data is not so high (0.6), having also the higher NMSE (0.06). Regarding the bias,

two distinct behaviors were obtained: the model underestimates the wind velocity at the

location of “Gago Coutinho” and “Aeroporto” stations (-0.03 and -0.45, respectively), while

overestimate this variable at the “Geofísica” station (0.5). This overestimation mainly

occurs during the nighttime (7 p.m. to 6 a.m.), with the land-surface model overestimating

the differences between sea and land temperatures during the night.

Overall, the obtained statistics metrics are in accordance with the acceptance criteria (MBE

within ± 30 % of the mean and NMSE < 1.5) proposed by Chang and Hanna (2005). These

findings guarantee that the most important meteorological variables for the air quality are

well modelled by the WRF model, which strengthens the robustness of the model setup and

gives confidence in the obtained air quality modelling results.

The good model performance is also verified by the analysis of time series, for both

temperature and wind velocity variables. Figure 2 shows these time series for the Gago

Coutinho monitoring station.

6

Figure 2. Modelled and measured daily profiles of temperature and wind velocity for the

simulation period at the Gago Coutinho monitoring station.

The time series shows that WRF is able to reproduce the daily variations of temperature and

wind velocity, and their magnitude, during the entire simulation period. The average

temperature of modelled data ranged between 10.9°C (in December) and 20.8°C (in

October); the measured data ranged between 11.4°C (in December) and 20.8°C (in

October). Maximum values were obtained in September, with values of 26.1°C (modelled

data) and 25.1°C (measured data); minimum values were obtained in December with values

of 7.1°C and 8.0°C for modelled and measured data, respectively. Regarding the wind

velocity, the WRF model provided average values ranging between 2.70 m.s-1 (in October)

and 3.90 m.s-1 (in September); the average measured data varies between 2.90 m.s-1 (in

October) and 4.10 m.s-1 (in September). Maximum wind velocities were obtained in

September for both model and measured data, with values of 6.90 m.s-1 and 6.50 m.s-1,

respectively. The minimum wind velocities were obtained in October.

For a more comprehensive picture of the temperature and wind velocity behavior, Figure 3

shows the spatial distribution of these two variables, as estimated by the model WRF and

averaged for the simulation period. The black arrows represent both wind velocity and wind

direction, being its size proportional to the wind velocity (as higher the wind velocity higher

the arrow size).

7

Figure 3. Spatial distribution of temperature and wind velocity, averaged for the simulation period (1st September to 31st December 2017).

High wind velocities are simulated over the ocean, with a maximum value of 7.5 m.s-1, and

minimum values over the land, with a value of 3.0 m.s-1. Regarding the temperature, the

values vary between 13.4 and 17.2°C, with maximum values over the ocean.

5.2. AIR QUALITY MODELLING EVALUATION

The section 5.2.1 presents the air quality modelling evaluation for PM10 and PM2.5 using the

Portuguese air quality monitoring network. For the heavy metals simulations, the data

obtained from the air quality monitoring campaigns in Lisbon were used and the WRF-CAMx

system performance is shown in the subsection 5.2.2.

5.2.1. PARTICULATE MATTER

Table 4 includes the statistical results for PM10 and PM2.5 using data from the Portuguese

air quality monitoring network over the simulation domain. From the 15 monitoring stations

only 4 were measuring PM2.5. The performance evaluation is based on daily averages.

8

Table 4. Daily model performance for PM10 and PM2.5.

Station Station type

PM10 PM2.5

r (-)

MBE (µg.m-3)

RMSE (µg.m-3)

r (-)

MBE (µg.m-3)

RMSE (µg.m-3)

ALV Urban background 0.42 13.7 19.4 - - -

FID Urban background 0.37 -4.83 14.2 - - -

LAR Urban background 0.40 3.88 14.7 0.32 10.6 16.3

LOU Urban background 0.64 7.72 15.2 - - -

MEM Urban background 0.64 4.00 10.9 0.63 13.3 16.6

OLI Urban background 0.20 6.07 18.6 0.20 14.2 20.5

QMA Urban background 0.53 14.4 18.8 - - -

REB Urban background 0.61 18.9 24.2 - - -

PPI Suburban industrial 0.47 -7.08 17.9 - - -

ESC Urban industrial 0.45 -2.08 14.4 - - -

LAV Urban industrial 0.36 0.78 14.6 - - -

BEN Urban traffic 0.64 -0.98 16.0 - - -

ODI Urban traffic 0.46 5.27 20.1 - - -

AVL Urban traffic 0.32 -1.97 16.6 - - -

ENT Urban traffic 0.27 5.43 17.7 0.24 16.9 21.6

The CAMx model recorded a correlation coefficient between 0.20 and 0.64 and the

simulations tend to overestimate the particulate matter concentrations (positive MBE). The

underestimation of wind velocity by the WRF model could contribute to this behavior,

leading to higher PM10 and PM2.5 over the Lisbon urban area. The magnitude errors ranged

between 10.9 µg.m-3 (MEM) and 24.2 µg.m-3 (REB) for PM10 and between 16.3 µg.m-3 (LAR)

and 21.6 µg.m-3 (ENT) for PM2.5. In general, the obtained statistics metrics are in

accordance with results from previous air quality modelling applications over Portugal.

Monteiro et al. (2013) applied five air quality models (European Air Pollution and

Dispersion—Inverse Model (EURAD-IM), CAMx, CHIMERE, LOTOS-EUROS and The Air Pollution

Model (TAPM)) over Portugal for July of 2006. The authors obtained an underestimation of

PM10 levels (which could be related to the simulated summer month and the occurrence of

PM natural events) and the RMSE ranged between 10 and 50 µg.m-3 (the maximum is twice

higher than observed in this study). Similarly with the results of this work, the correlation

coefficient ranged between 0.2 and 0.6.

Figure 4 provides the time series for the simulation period and the analyzed air quality

monitoring stations over the Lisbon urban area.

PM10

0

50

100

150

1 21 41 61 81 101 121PM10

co

nce

ntr

atio

ns (

µg.

m-3

)

Days

ALV

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM

10

co

nce

ntr

atio

ns

(µg.

m-3

)

Days

FID

Measured WRF-CAMx

9

0

50

100

150

1 21 41 61 81 101 121PM

10

co

nce

ntr

atio

ns

(µg.

m-3

)

Days

LAR

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM

10

co

nce

ntr

atio

ns

(µg.

m-3

)

Days

LOU

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM10

co

nce

ntr

atio

ns (µ

g.m

-3)

Days

MEM

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM10

co

nce

ntr

atio

ns (µ

g.m

-3)

Days

OLI

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM10

co

nce

ntr

atio

ns (µ

g.m

-3)

Days

QMA

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM10

co

nce

ntr

atio

ns (µ

g.m

-3)

Days

REB

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM10

co

nce

ntr

atio

ns (µ

g.m

-3)

Days

PPI

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM10

co

nce

ntr

atio

ns (µ

g.m

-3)

Days

ESC

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM10

co

nce

ntr

atio

ns (µ

g.m

-3)

Days

LAV

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM

10

co

nce

ntr

atio

ns

(µg.

m-3

)

Days

BEN

Measured WRF-CAMx

10

PM2.5

Figure 4. Modelled and measured daily profile of PM10 and PM2.5 for the simulation period.

The air quality modelling system revealed a good performance when reproducing the daily

PM10 and PM2.5 levels during the simulation period. However, some concentrations peaks

were not fully reproduced by the model for locations near industrial areas (i.e. PPI, ESC and

LAV) and where the road traffic is the main source of atmospheric emissions (i.e. BEN, ODI,

AVL and ENT).

Figure 5 displays the spatial distribution of PM10 and PM2.5 averaged concentrations along

the simulation period. The measured averages for the air quality monitoring stations are

also represented by small circles.

0

50

100

150

1 21 41 61 81 101 121PM

10

co

nce

ntr

atio

ns

(µg.

m-3

)

Days

ODI

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM

10

co

nce

ntr

atio

ns

(µg.

m-3

)

Days

AVL

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM10

co

nce

ntr

atio

ns (µ

g.m

-3)

Days

ENT

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM2.

5 co

nce

ntr

atio

ns (µ

g.m

-3)

Days

LAR

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM2.

5 co

nce

ntr

atio

ns (µ

g.m

-3)

Days

MEM

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM2.

5 co

nce

ntr

atio

ns (µ

g.m

-3)

Days

OLI

Measured WRF-CAMx

0

50

100

150

1 21 41 61 81 101 121PM2.

5 co

nce

ntr

atio

ns (µ

g.m

-3)

Days

ENT

Measured WRF-CAMx

11

Figure 5. PM10 and PM2.5 average concentrations for the simulation study period.

Figure 5 shows that the spatial distribution of PM10 and PM2.5 are reasonably well

reproduced. The comparison cannot be direct, because the simulation period is not annual,

but for PM10, some exceedances to the annual limit value were recorded (40 µg.m-3) and for

PM2.5 almost the entire Lisbon urban area did not comply with the annual air quality

standard defined for this air pollutant (25 µg.m-3).

5.2.2. HEAVY METALS

Figure 6 presents the comparison between the WRF-CAMx simulation results and the

measured heavy metals concentrations measured during the air quality monitoring

campaigns, between 1st September and 31st December of 2017 in Lisbon.

Pb

0.00

0.02

0.04

Pb

co

nce

ntr

ati

on

s (µ

g-m

-3)

Year 2017

Measured WRF-CAMx

H1

0.00

0.02

0.04

Pb

co

nce

ntr

ati

on

s (µ

g-m

-3)

Year 2017

Measured WRF-CAMx

H2

0.00

0.02

0.04

Pb

co

nce

ntr

ati

on

s (µ

g-m

-3)

Year 2017

Measured WRF-CAMx

H3

12

0.00

0.02

0.04P

b c

on

cen

tra

tio

ns

(µg-

m-3

)

Year 2017

Measured WRF-CAMx

H4

0.00

0.02

0.04

Pb

co

nce

ntr

ati

on

s (µ

g-m

-3)

Year 2017

Measured WRF-CAMx

H5

0.00

0.02

0.04

Pb

co

nce

ntr

ati

on

s (µ

g-m

-3)

Year 2017

Measured WRF-CAMx

H6

0.00

0.02

0.04

Pb

co

nce

ntr

ati

on

s (µ

g-m

-3)

Year 2017

Measured WRF-CAMx

H7

0.00

0.02

0.04

Pb

co

nce

ntr

ati

on

s (µ

g-m

-3)

Year 2017

Measured WRF-CAMx

H8

0.00

0.02

0.04

Pb

co

nce

ntr

ati

on

s (µ

g-m

-3)

Year 2017

Measured WRF-CAMx

H9

0.00

0.02

0.04

Pb

co

nce

ntr

ati

on

s (µ

g-m

-3)

Year 2017

Measured WRF-CAMx

H10

0.00

0.02

0.04

Pb

co

nce

ntr

ati

on

s (µ

g-m

-3)

Year 2017

Measured WRF-CAMx

I1

0.00

0.02

0.04

Pb

co

nce

ntr

ati

on

s (µ

g-m

-3)

Year 2017

Measured WRF-CAMx

I2

0.00

0.02

0.04

Pb

co

nce

ntr

ati

on

s (µ

g-m

-3)

Year 2017

Measured WRF-CAMx

I3

13

Ni

0

50

100N

i co

nce

ntr

ati

on

s (n

g-m

-3)

Year 2017

Measured WRF-CAMx

H1

0

50

100

Ni

con

cen

tra

tio

ns

(ng-

m-3

)Year 2017

Measured WRF-CAMx

H2

0

50

100

Ni

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H3

0

50

100

Ni

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H4

0

50

100

Ni

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H5

0

50

100

Ni

con

cen

tra

tio

ns

(ng-

m-3

)Year 2017

Measured WRF-CAMx

H6

0

50

100

Ni

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H7

0

50

100

Ni

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H8

0

50

100

Ni

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H9

0

50

100

Ni

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H10

0

50

100

Ni

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

I1

0

50

100

Ni

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

I2

14

As

0

50

100

Ni

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

I3

0.0

0.5

1.0

1.5

2.0

As

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H1

0.0

0.5

1.0

1.5

2.0

As

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H2

0.0

0.5

1.0

1.5

2.0

As

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H3

0.0

0.5

1.0

1.5

2.0

As

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H4

0.0

0.5

1.0

1.5

2.0

As

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H5

0.0

0.5

1.0

1.5

2.0

As

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H6

0.0

0.5

1.0

1.5

2.0

As

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H7

0.0

0.5

1.0

1.5

2.0

As

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H8

0.0

0.5

1.0

1.5

2.0

As

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

H9

15

Figure 6. Modelled and measured concentrations of heavy metals (i.e. Pb, Ni, As) for the simulation period.

There is a strong spatial variation among the different monitored places, but in general the

air quality modelling tends to overestimate the Pb and Ni levels over the study area. For the

As, the air quality measurements are the same for all days and locations because the

measured concentrations were below the detection limit of the methodology used for the

quantification. For this pollutant, the WRF-CAMx system underestimated the air pollution

levels during the simulation period, but the simulated and the measured values are both

very low.

Figure 7 shows the spatial distribution of heavy metals averaged levels between 1st

September and 31st December of 2017.

0.0

0.5

1.0

1.5

2.0A

s co

nce

ntr

ati

on

s (n

g-m

-3)

Year 2017

Measured WRF-CAMx

H10

0.0

0.5

1.0

1.5

2.0

As

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

I1

0.0

0.5

1.0

1.5

2.0

As

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

I2

0.0

0.5

1.0

1.5

2.0

As

con

cen

tra

tio

ns

(ng-

m-3

)

Year 2017

Measured WRF-CAMx

I3

16

Figure 7. Heavy metals average concentrations for the simulation study period.

Based on the modelling results it is possible to assume that no exceedances to the annual

limit value for Pb (0.5 µg.m-3) and for As (6 ng.m-3) were registered. Some areas of Lisbon

did not comply with the annual air quality standards defined for Ni and Cd (Ni = 25 ng.m-3;

Cd = 5 ng.m-3). The highest air pollution levels were mainly registered in industrial areas and

near of Lisbon maritime port.

6. CONCLUSIONS

This study aimed to evaluate the WRF-CAMx system performance to simulate the main

meteorological parameters, particulate matter and heavy metals levels over the Lisbon

urban area. The air quality modelling system revealed a good performance. It tends to

slightly underestimate and overestimate the meteorological variables and air pollution

concentrations, respectively. The quality of the results is in accordance with modelling

acceptance criteria and similar with other applications over Portugal. Thus, the system

setup could be considered robust and the quality of its results high enough to integrate the

LIFE Index-Air Management Tool.

17

7. REFERENCES

Aquilina N., Dudek A.V., Carvalho A., Borrego C., Nordeng T.E. (2005). MM5 high resolution simulations over Lisbon. Geophysical Research Abstracts, 7, 08685.

Borrego, C., Tchepel, O., Barros, N., Miranda, A.I., 2000. Impact of road traffic emissions on air quality of the Lisbon region. Atmos. Environ. 34, 4683–4690. https://doi.org/10.1016/S1352-2310(00)00301-0

Borrego, C., Tchepel, O., Costa, A.M., Amorim, J.H., Miranda, A.I., 2003. Emission and dispersion modelling of Lisbon air quality at local scale. Atmos. Environ. 37, 5197–5205. https://doi.org/10.1016/j.atmosenv.2003.09.004

Borrego, C., Tchepel, O., Salmin, L., Amorim, J.H., Costa, A.M., Janko, J., 2004. Integrated Modeling of Road Traffic Emissions: Application To Lisbon Air Quality Management. Cybern. Syst. 35, 535–548. https://doi.org/10.1080/0196972049051904

Carvalho D., Rocha A., Gómez-Gesteira M., Santos C. (2012). A sensitivity study of the WRF model in wind simulation for an area of high wind energy. Environmental Modelling & Software, 33, 23-34. [doi: 10.1016/j.envsoft.2012.01.019].

Chang J.C., Hanna S.R. (2005) .Technical descriptions and user’s guide for the BOOT statistical model evaluation software package, version 2.0.

DGT, (Direção Geral do Território), 2018. Cartografia de Uso e Ocupação do Solo (COS, CLC e Copernicus) [WWW Document]. URL http://www.dgterritorio.pt (accessed 10.19.18).

Dudhia J. (1989). Numerical study of convection observed during the winter monsoon experiment using a mesoscale two-dimensional model. Journal of the Atmospheric Sciences, 46, 3077–3107.

ENVIRON, 2015. CAMx. User’s Guide. Comprehensive air quality model with extensions. Version 6.2.

Ferreira, J., Rodriguez, A., Monteiro, A., Miranda, A.I., Dios, M., Souto, J.A., Yarwood, G., Nopmongcol, U., Borrego, C., 2012. Air quality simulations for North America - MM5-CAMx modelling performance for main gaseous pollutants. Atmos. Environ. 53, 212–224. https://doi.org/10.1016/j.atmosenv.2011.10.020

Gong, S.L., 2003. A parameterization of sea-salt aerosol source function for sub- and super-micron particles. Global Biogeochem. Cycles 17, n/a-n/a. https://doi.org/10.1029/2003GB002079

González, M.Á., Vivanco, M.G., Palomino, I., Garrido, J.L., Santiago, M., Bessagnet, B., 2012. Modelling some heavy metals air concentration in europe. Water. Air. Soil Pollut. 223, 5227–5242. https://doi.org/10.1007/s11270-012-1274-6.

Grell G.A., Freitas S.R. (2014). A scale and aerosol aware stochastic convective parameterization 346 for weather and air quality modeling. Atmospheric Chemistry and Physics, 13, 845–23.

Hong S.-Y., Lim J.-O. J. (2006). The WRF single-moment 6-class microphysics scheme

18

(WSM6). Asia-Pacific Journal of Atmospheric Sciences, 42, 129–151.

Hutzell, W.T., Luecken, D.J., 2008. Fate and transport of emissions for several trace metals over the United States. Sci. Total Environ. 396, 164–179. https://doi.org/10.1016/j.scitotenv.2008.02.020.

Ichikawa, Y., and K. Sada, 2002: An atmospheric dispersion model for the environmental impact assessment of thermal power plants in Japan – A method for evaluating topographical effects. J. Air & Waste Manage. Asso., 52, 313-323.

Mlawer E.J., Taubman S.J., Brown P.D., Iacono M.J., Clough S.A. (1997). Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-k model for the longwave. Journal of Geophysical Research, 102D, 16663-16682.

Monteiro, A., Ribeiro, I., Tchepel, O., Sá, E., Ferreira, J., Carvalho, A., Martins, V., Strunk, A., Galmarini, S., Elbern, H., Schaap, M., Builtjes, P., Miranda, A.I., Borrego, C., 2013. Bias Correction Techniques to Improve Air Quality Ensemble Predictions: Focus on O3 and PM Over Portugal. Environ. Model. Assess. 18, 533–546. https://doi.org/10.1007/s10666-013-9358-2

Monteiro A., Ferreira J., Ribeiro I., Fernandes A.I., Martins H., Gama C., Miranda A.I. (2015). Air quality over Portugal in 2020. Atmospheric Pollution Research, 6, 788-796.

Mosca, S., G. Graziani, W. Klug, R. Bellasio, and R. Bianconi, 1998: A statistical methodology for the evaluation of long-range dispersion models: an application to the ETEX exercise. Atmos. Environ., 24, 4307-4324.

Nappo, C.J., and K.S.M. Essa, 2001: Modeling dispersion from near-surface tracer releases at Cape Canaveral, Florida. Atmos. Environ., 35, 3999-4010.

NCAR, 2010. Model for Ozone and Related chemical Tracers, version 4 (MOZART-4) [WWW Document]. Natl. Cent. Atmos. Res.

Nopmongcol, U., Koo, B., Tai, E., Jung, J., Piyachaturawat, P., Emery, C., Yarwood, G., Pirovano, G., Mitsakou, C., Kallos, G., 2012. Modeling Europe with CAMx for the Air Quality Model Evaluation International Initiative (AQMEII). Atmos. Environ. 53, 177–185. https://doi.org/10.1016/j.atmosenv.2011.11.023.

OpenStreetMap contributors, 2017. Planet dump [Data file from $date of database dump$]. Retrieved from https://planet.openstreetmap.org." [WWW Document].

Ovadnevaite, J., Manders, A., De Leeuw, G., Ceburnis, D., Monahan, C., Partanen, A.I., Korhonen, H., O’Dowd, C.D., 2014. A sea spray aerosol flux parameterization encapsulating wave state. Atmos. Chem. Phys. 14, 1837–1852. https://doi.org/10.5194/acp-14-1837-2014.

Papanastasiou, D. K., Melas, D., Bartzanas, T., & Kittas, C. (2010). Temperature, comfort and pollution levels during heat waves and the role of sea breeze. International Journal of Biometeorology, 54, 307–317.

Reynolds, R.W., C. L. Gentemann and G. K. Corlett, 2010: Evaluation of AATSR and TMI Satellite SST Data. J. Climate, 23, 152–165.

Sá, E., Martins, H., Ferreira, J., Marta-Almeida, M., Rocha, A., Carvalho, A., Freitas, S., Borrego, C., 2016. Climate change and pollutant emissions impacts on air quality in 2050 over Portugal. Atmos. Environ. 131, 209–224.

19

https://doi.org/10.1016/j.atmosenv.2016.01.040

Silveira, C., Ferreira, J., Monteiro, A., Miranda, A.I., Borrego, C., 2017. Emissions from residential combustion sector: how to build a high spatially resolved inventory. Air Qual. Atmos. Heal. 1–12. https://doi.org/10.1007/s11869-017-0526-4

Skamarock, W.C., Klemp, J.B., Dudhi, J., Gill, D.O., Barker, D.M., Duda, M.G., Huang, X.-Y., Wang, W., Powers, J.G., 2008. A Description of the Advanced Research WRF Version 3, Technical Report. https://doi.org/10.5065/D6DZ069T.

Tewari M., Chen F., Wang W., Dudhia J., LeMone M., Mitchell K., Ek M., Gayno G., Wegiel J., Cuenca R. (2004). Implementation and verification of the unified NOAH land surface model in the WRF model. 20th conference on weather analysis and forecasting/16th conference on numerical weather prediction, 11–15.

Zhang D., Anthes R.A. (1982). A high-resolution model of the planetary boundary layer-sensitivity tests and comparisons with SESAME-79 data. Journal of Applied Meteorology, 21, 1594–1609.