Scholarly article on topic 'A coupled road dust and surface moisture model to predict non-exhaust road traffic induced particle emissions (NORTRIP). Part 2: Surface moisture and salt impact modelling'

A coupled road dust and surface moisture model to predict non-exhaust road traffic induced particle emissions (NORTRIP). Part 2: Surface moisture and salt impact modelling Academic research paper on "Earth and related environmental sciences"

CC BY-NC-ND
0
0
Share paper
Academic journal
Atmospheric Environment
Keywords
{"Air quality" / "Non-exhaust emissions" / "Road dust" / Suspension / "Road surface moisture"}

Abstract of research paper on Earth and related environmental sciences, author of scientific article — B.R. Denby, I. Sundvor, C. Johansson, L. Pirjola, M. Ketzel, et al.

Abstract Non-exhaust traffic induced emissions are a major source of airborne particulate matter in most European countries. This is particularly important in Nordic and Alpine countries where winter time road traction maintenance occurs, e.g. salting and sanding, and where studded tyres are used. Though the total mass generated by wear sources is a key factor in non-exhaust emissions, these emissions are also strongly controlled by surface moisture conditions. In this paper, Part 2, the road surface moisture sub-model of a coupled road dust and surface moisture model (NORTRIP) is described. We present a description of the road surface moisture part of the model and apply the coupled model to seven sites in Stockholm, Oslo, Helsinki and Copenhagen over 18 separate periods, ranging from 3.5 to 24 months. At two sites surface moisture measurements are available and the moisture sub-model is compared directly to these observations. The model predicts the frequency of wet roads well at both sites, with an average fractional bias of −2.6%. The model is found to correctly predict the hourly surface state, wet or dry, 85% of the time. From the 18 periods modelled using the coupled model an average absolute fractional bias of 15% for PM10 concentrations was found. Similarly the model predicts the 90'th daily mean percentiles of PM10 with an average absolute bias of 19% and an average correlation (R 2) of 0.49. When surface moisture is not included in the modelling then this average correlation is reduced to 0.16, demonstrating the importance of the surface moisture conditions. Tests have been carried out to assess the sensitivity of the model to model parameters and input data. The model provides a useful tool for air quality management and for improving our understanding of non-exhaust traffic emissions.

Academic research paper on topic "A coupled road dust and surface moisture model to predict non-exhaust road traffic induced particle emissions (NORTRIP). Part 2: Surface moisture and salt impact modelling"

Contents lists available at ScienceDirect

Atmospheric Environment

journal homepage: www.elsevier.com/locate/atmosenv

A coupled road dust and surface moisture model to predict non-exhaust road traffic induced particle emissions (NORTRIP). Part 2: Surface moisture and salt impact modelling^

CrossMark

B.R. Denbya*, I. Sundvora, C. Johanssonb,e, L. Pirjolac, M. Ketzeld, M. Norman K. Kupiainenf, M. Gustafsson g, G. Blomqvistg, M. Kauhaniemih, G. Omstedti

a The Norwegian Institute for Air Research (NILU), PO Box 100, 2027 Kjeller, Norway b Department of Applied Environmental Science (ITM), Stockholm University, Sweden c Helsinki Metropolia University of Applied Sciences, Finland d Department of Environmental Science, Aarhus University, Roskilde, Denmark e Environment and Health Protection Administration of the City of Stockholm, Sweden f Nordic Envicon Oy, Helsinki, Finland

g Swedish National Road and Transport Research Institute (VTI), Sweden h Finish Meteorological Institute (FMI), Helsinki, Finland ' Swedish Meteorological and Hydrological Institute (SMHI), Norrkoping, Sweden

HIGHLIGHTS

> A coupled road dust and surface moisture non-exhaust emission model is applied to seven sites over 18 different periods.

> Surface moisture is shown to be the dominant cause of variability of the road dust emissions.

> The model explains half the variability seen in roadside PM10 measurements.

Salt is found to have an impact on the surface moisture and on the variability of emissions.

ARTICLE INFO

ABSTRACT

Article history: Received 20 June 2013 Accepted 2 September 2013

Keywords: Air quality

Non-exhaust emissions Road dust Suspension Road surface moisture

Non-exhaust traffic induced emissions are a major source of airborne particulate matter in most European countries. This is particularly important in Nordic and Alpine countries where winter time road traction maintenance occurs, e.g. salting and sanding, and where studded tyres are used. Though the total mass generated by wear sources is a key factor in non-exhaust emissions, these emissions are also strongly controlled by surface moisture conditions. In this paper, Part 2, the road surface moisture submodel of a coupled road dust and surface moisture model (NORTRIP) is described. We present a description of the road surface moisture part of the model and apply the coupled model to seven sites in Stockholm, Oslo, Helsinki and Copenhagen over 18 separate periods, ranging from 3.5 to 24 months. At two sites surface moisture measurements are available and the moisture sub-model is compared directly to these observations. The model predicts the frequency of wet roads well at both sites, with an average fractional bias of -2.6%. The model is found to correctly predict the hourly surface state, wet or dry, 85% of the time. From the 18 periods modelled using the coupled model an average absolute fractional bias of 15% for PM10 concentrations was found. Similarly the model predicts the 90'th daily mean percentiles of PM10 with an average absolute bias of 19% and an average correlation (R2) of 0.49. When surface moisture is not included in the modelling then this average correlation is reduced to 0.16, demonstrating the importance of the surface moisture conditions. Tests have been carried out to assess the sensitivity of the model to model parameters and input data. The model provides a useful tool for air quality management and for improving our understanding of non-exhaust traffic emissions.

© 2013 The Authors. Published by Elsevier Ltd. All rights reserved.

q This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial-No Derivative Works License, which permits noncommercial use, distribution, and reproduction in any medium, provided the original author and source are credited. * Corresponding author. Tel.: +47 63898164; fax: +47 63898050. E-mail addresses: bde@nilu.no, bruce.denby@nilu.no (B.R. Denby).

1352-2310/$ — see front matter © 2013 The Authors. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.atmosenv.2013.09.003

1. Introduction

Non-exhaust emissions of particulate matter are a dominant component of traffic emissions in many countries in Europe. This is particularly true in Nordic and Alpine regions where studded tyres and winter maintenance activities, such as salting and sanding, are used (e.g. Gustafsson et al., 2008; Norman and Johansson, 2006). Even in regions where these practices are not carried out, non-exhaust emissions from suspended road dust and other vehicle wear sources, such as brake and tyre wear, have been shown to contribute significantly to the total vehicle emissions (Pant and Harrison, 2013; Bukowiecki et al., 2010). Currently non-exhaust emissions are not directly regulated but controls on these emissions are required if concentrations are to be reduced. There is thus a need for improved modelling of non-exhaust emissions that can be further applied to assess abatement strategies and improve air quality management.

This paper presents the second part of a study describing a coupled road dust and road surface moisture model (NORTRIP) for calculating non-exhaust traffic induced emissions. The first part presented in Denby et al. (2013), and further referred to as 'Part 1', provided a description of the road dust sub-model. In that case the road dust sub-model was applied at two sites, over seven years, where observed surface moisture measurements were available in order to assess the road dust and suspension part of the model. In this second part the surface moisture sub-model is described and the model is applied to seven different sites over 18 periods, ranging from 3.5 months to two years.

Modelling the surface moisture and the impact it has on the surface particle retention is essential if the suspension of the surface particles is to be modelled correctly. Often highly para-meterised forms are used to simulate this retention, e.g. based on time since last rainfall and/or humidity (e.g. Berger and Denby, 2011; US EPA, 2006; Pay et al., 2011). In the road dust suspension model from Omstedt et al. (2005) a surface mass balance approach was used where precipitation was balanced by drainage and evaporation, based on a simplified energy balance scheme. However, none of these schemes take into account processes such as condensation, snow, freezing processes, vehicle spray or impact of salting on surface vapour pressure. To achieve this, a more comprehensive surface moisture model is required that reflects the physical processes in a more generalised way. Such a model is similar to road weather models (e.g. Sass, 1997; Karlsson, 2001; Möller, 2006) used to predict surface temperature for traffic maintenance and safety applications.

Within the road surface moisture sub-model the following parameterised processes are included and described in Section 2.

1. Mass balance for surface moisture (water and ice/snow)

2. Production through precipitation (rain and snow)

3. Production through road maintenance wetting activities

4. Removal through drainage

5. Removal through vehicle spray

6. Removal of snow through snow ploughing activities

7. Evaporation and condensation using energy balance modelling

8. Melting and freezing

9. Impact of salt solution on vapour pressure and freezing temperatures

The model coupling, between the road dust and surface moisture sub-models, is through three main mechanisms. Firstly by use of the surface moisture to determine the retention and suspension of surface particles, secondly through the surface moisture removal processes (drainage, spray, snow ploughing) which will also remove dust and salt from the surface and thirdly

through the impact of salt on the vapour pressure of the surface moisture. This allows not just the description of salting for de-icing applications (NaCl) but also provides a mechanism to describe the impact of dust binders (e.g. MgCl2 and CMA) on the surface moisture. In this study only the application of NaCl is considered.

In Section 2 the surface moisture sub-model is described along with the coupling mechanisms and in Section 3 the seven different sites where the model is applied are briefly described. A comparison is made with observed surface temperature and surface moisture at two of these sites, where such observations are available. The coupled model is then further applied to all seven sites for 18 separate periods and a comparison is made with observed concentrations (Section 4). A model sensitivity analysis is provided at one of the sites. Not all details concerning the model are presented in this paper and the reader is referred to the earlier model documentation provided in Denby and Sundvor (2012) and to Part 1 of this study, Denby et al. (2013).

2. Surface moisture sub-model

The major aim of the surface sub-model is to predict the surface retention of particles and to describe removal processes related to the dust and salt mass. In Part 1, Equations (18) and (19), the surface retention factor fq was defined as a function of surface moisture. A road surface is considered to be completely non-retentive (dry) below a minimum surface wetness (gretention-mm = 0.04 mm, fq = 1) and completely retentive (wet) above a threshold surface wetness (gretention-thresh = 0.1 mm, fq = 0), see Part 1 Section 3.7. The following section describes how the surface moisture is calculated.

2.1. Mass balance for road surface moisture

As with the dust and salt loading (Part 1) we establish a mass balance equation for water and ice (moisture) and determine the production and sink terms for the road moisture balance. The road moisture is separated into water (groad) and snow/ice (sroad), given in mm w.e. (water equivalent for snow). The surface moisture mass balance is specified by

dgroad dt

Vsroad

Pg ~ Sg

The production of road water (Pg) is determined by the processes of rain, snow/ice melt, wetting (during cleaning or salting) and condensation

Pg — Pg, rain + Pg, snowmelt + Pg,wetting + Pg, condens

The sink terms (Sg) for the road surface water include drainage, spray, evaporation and freezing (converting water to ice).

Sg = Sg, drain + Sg, spray + Sg, evap + Sg,freeze (4)

Note that evaporation/condensation are the same process but in reverse directions.

The production of road snow/ice (Ps) is determined by the processes of snow fall, freezing and deposition (condensation of ice)

Ps = Ps, snow + Ps, freeze + Ps, condens (5)

The sink terms for the road surface snow/ice (Ss) include snow melt, snow ploughing and sublimation (evaporation of ice)

s — Ss,snowmelt + Ss, ploughing + Ss,evap

The surface of the road is considered to be impermeable to moisture and no consideration is given in the model to subsurface water content. The moisture sub-model in its current form is thus only suitable for paved roads.

2.2. Precipitation

Precipitation in the form of rain or snow is added to the road surface. The rate of production by precipitation (mm h_1) is simply written as

and P.

s,snow

Where the total rain/snow for the period Dt is given in mm (water equivalent).

2.3. Wetting

Wetting reflects the addition of water in salt solutions to the surface or if water is used for cleaning the surface and is given by

g,wetting

groad-wetting (^wetting)

where groad-Wettmg is the amount of water used in the wetting (mm or litre m-2) at the time twetting. If salt is provided in solution then the amount of water applied will depend on the salt solution concentration. This can be specified directly as input to the model or may be determined using the salting rule model (Section 2.10).

2.4. Drainage

Drainage is treated in the model as an instantaneous process, since the time scale for drainage is much less than the model time step of 1 h. The amount of water drained from the road in the period Dt is specified by

groad,drainable = max(groad — groad,drainable-min> (9)

and the water sink rate is specified by

Sg, drainage = groad,drainable/Dt (10)

The parameter groad,drainable-min indicates the minimum moisture level below which drainage does not occur. Typically drainage will depend on pavement characteristics such as surface macro texture but also large scale features such as road slope. Values for groad,drainable-min are not clearly defined, Omstedt et al. (2005) applied a value of 1 mm. Based on sensitivity tests and comparison to observed moisture, Section 4.2, we use a value groad,drainable-min = 0.5 mm. Drainage is used, in combination with efficiency factors (Table C.2), to remove dust and salt mass from the road surface. The efficiency factors reflect how well the mass is mixed with water, and since salt is highly soluble it is given an efficiency of unity. The dust efficiency is unknown and is set to a low value of 0.001. This part of the model is also described in Part 1, Section 3.6.

2.5. Vehicle spray

Spray is the mechanism by which water is emitted from the road surface through contact of the tyre with the road, removing water

from the wheel tracks and redistributing it on the road surface or removing it completely from the road. We consider spray as a road moisture sink term that can be described using a rate equation and that occurs down to a threshold surface moisture level (groad,-sprayable-min)

Sg,spray — Rg,spray 'groad for groad > groad,sprayable-— 0 for groad < groad,sprayable-

■min min

The rate equation is dependent on traffic volume (Nv), a spray rate factor (/spray) and the vehicle speed ( Vveh). The index v refers to vehicle type, either light (li) or heavy (he) duty, and the spray removal rate is summed over these two vehicle types.

vehicle £

t\g,spray - ^ j spray

v = he,li "lanes

/spray (Vveh)

Here the spray factor fspray (veh 1) is given by a quadratic dependence on vehicle speed.

( Vv . \

fv lyv \ _ fv veh i

J spray veh J — J 0,spray I Vref spra I

The basic spray factor /ospray for the two vehicle types (v) is defined by the user. Möller (2006) derived a value for this spray

factor of f0

0 , spray

= 5 x 10~3 veh

for light duty and

f0hepray = 6/0'spray for heavy duty vehicles at a reference speed of Vref spray = 70 km/hr. However that study concentrated on the wheel tracks and included all other processes as well, e.g. evaporation. Since the model describes the average surface moisture of the road surface, not just the wheel tracks, we choose a value significantly lower than this of /0'spray = 1 x 10~4 veh-1 and a value for groad,sprayable-min of 0.1 mm. The sensitivity of the model to this parameter is shown in Section 4.5.

Spray will also remove mass from the road surface and will be an additional sink for the road dust and salt loading Mrmoad (Part 1, Section 3.1). The spray rate Rg,spray provides the basis for the mass removal which is modified by the mixing efficiency for spray hm>ray-eff for different mass types m. The same mixing efficiencies are used for spray as for drainage. We write the spray sink term for both dust and salt as:

Sspray — Mroad ' Rg; spray ' hspray-eff

2.6. Snow ploughing

Ploughing of the road will remove snow and if no information concerning its timing is available then a snow ploughing rule is applied (Section 2.10). The efficiency of the ploughing, i.e. the fraction of snow removed by a ploughing event, is given by the factor hpnoughmg-eff. This should be fairly high and a value of 0.8 is used. Similarly an efficiency factor for the removal of dust and salt (hmoughing-eff) during ploughing can be applied. This is expected to be very low for dust but is set to unity, as in drainage and spray, for salt.

2.7. Evaporation, condensation and energy balance modelling

Both evaporation and condensation processes are described by the water vapour flux to, or from, the surface. This flux is the direct result of the energy balance of the road surface, which also

determines the surface temperature and surface humidity. In order to describe these processes an energy balance model is applied.

The surface energy balance, i.e. the net energy passing through the top of the road surface, is given by the following:

+ Hs + Ls + Ht

traffic

RL-in,s = £eff sTK,a

where s is the Stefan—Boltzmann constant ( ), 7K,a is the

atmospheric temperature in Kelvin and the effective emissivity (eeff) is a function of cloud cover (nc) and water vapour pressure (ea). We use a version from Konzelmann et al. (1994) given as

where Hs and Ls are the sensible and latent turbulent heat fluxes, Rnet,s is the net radiation flux and Htraffic is an additional traffic induced heat flux. We use the convention that heat fluxes are positive into the surface. If the surface energy flux (Gs) is positive then this means that the surface is being warmed.

2.7.1. Surface heat flux

The surface heat flux Gs is used to warm a surface layer slab of depth Dzs as follows

vis dt

1 Gs G

where Gsub is the flux out of the slab into the under laying subsurface. The sub-surface flux is specified using a relaxation term

Gsub = m{Ts - Tsub)

where Tsub is the subsurface temperature for the period considered, in this case the mean of the last three days. The parameter m is specified so that the model provides the correct surface temperature for a sinusoidal varying surface flux with a period of one day (U), similar to the force restore method described in Garratt (1994) where

fisCs2û)

m = QpsCsDzs

Given typical road parameters of density ps = 2400 kg m

Ceff = £cs (1 - n2) + ecl n2

and where the clear sky emissivity (ecs) is further parameterised as:

£cs = 0.23 + 0.443-( ^

A constant value for cloudy sky emissivity (eci = 0.97) is used as suggested by Konzelmann et al. (1994). In addition the long wave radiation from the surrounding street canyon walls, assuming them to be at the atmospheric temperature, is also taken into account.

The outgoing long wave radiation will depend on the surface temperature following Boltzmanns law,

RL-out,s = -£ssTK,s

We linearise this equation for the surface temperature in Kelvin (TK,s) around the near surface atmospheric temperature (TK,a) in order to implicitly solve the surface temperature (Equation (16)). Equation (24) can thus be rewritten as:

out,a '

1 - 4M +

^KaJ TK,a

es sTK,a

RL-out,a

The surface emissivity (es) is taken to be unity.

2.7.3. Sensible and latent heat fluxes

We use a bulk atmospheric surface layer formulation to describe the sensible and latent heat fluxes as:

specific heat cs = 800 J kg 1 K and thermal conductivity Hs = Pa'Cp'{Ta — Ts)/tt ks = 2.0 W m—1 K—1 then we find that the appropriate choice of

Dzs = 0.08 m and that m = 11.8 W m-2 K-1.

2.7.2. Radiation

The net radiation flux at the surface (Rnet,s) is given by

Rnet,s = RS-in,s(1 - aroad) + RL-in,s + RL-out,s

Ls = Pa "As '{qa - qs)/rq

where RS-jn,s is the incoming short wave global radiation at the surface, aroad is the road surface albedo (0.15 for road, 0.4 for snow) and RL-in,s and RL-out,s are the incoming and outgoing long wave radiation respectively.

The incoming global radiation (RS-in,s) is required input for the model and is available from meteorological models or from measurements. This is adjusted to account for shading of the road within the street canyon. In addition the global radiation is used, in conjunction with a clear sky radiation model based on Konzelmann et al. (1994) and Iqbal (1983), to estimate the cloud cover if this is not available, see Denby and Sundvor (2012) for details concerning these parameterisations.

The incoming long wave radiation is based on the Boltzmann equation for blackbody radiation written as

where rTand rq are the aerodynamic resistance for temperature and water vapour respectively, ra is the density of air and Cp is the heat capacity of dry air. Ts and Ta are the surface and atmospheric temperature and qs and qa are the surface and atmospheric specific humidity respectively.

The aerodynamic resistance factors rq and rT for atmospheric turbulence are described using classic similarity theory under neutral conditions. i.e.

FF{z)'k2

■wind

log{Z/Z0)'log {Z/Zqj)

Where k is the von Karman constant (0.4) and z0, zqj are the respective roughness lengths for momentum, water vapour and temperature. According to surface renewal theory (Denby and Snellen, 2002) zq,T z z0/10 under typical conditions. The resistance factors may be reduced due to traffic induced turbulence. This will be dependent on the vehicle speed (Weh) and the traffic

volume per hour (Nv) for the given vehicle type (v). Heavy duty vehicles will induce more turbulence than light duty. We relate these parameters as simply as possible to each other in the following way

rtraffic

3600'3.6

y^. atraffic '

' = li,he

nlanes

where the constants convert the traffic speed (km h 1) and volume

and veh s 1 respectively. The coefficient and represents the aerodynamics of the

(veh hr—1) to units of m s— atraffic has units of s veh—

vehicles. We suggest values of around 1 x 10—3 and 1 x 10—2 s veh— for light duty and heavy duty vehicles respectively. Use of these values will half the resistance factors on busy roads, e.g. ADT of 40,000 and vehicle speeds of 70 km h—1, compared to typical atmospheric induced turbulence. The total resistance is calculated using

rtraffic

it must be below the lower limit of the surface retention parameters (Part 1, Equations (18) and (19)) to ensure that roads can be 'dry' even in relatively humid conditions. We use a value of 0.02 mm, half of the minimum retention threshold, and investigate the models sensitivity to this parameter in Section 4.5.

2.7.6. Solving the energy balance to determine the evaporation and condensation

The surface energy balance is solved by prognosis of the surface temperature, using Equations (15) and (16), and diagnosis of the related heat fluxes. From this the evaporation sink and condensation production is calculated from the latent heat flux as

'g,evap

maxf 0, -j-

g,condens

max J-

Where the coefficient of latent heat (1s) depends on whether the surface is snow or water. Similar equations apply for the surface snow cover.

2.7.4. Vehicle induced heat flux

Heat fluxes are produced by vehicles due to both motor warmth and friction of tyres with the surface. This is parameterised in the model using the following form

traffic —

v = li,he

veh nlanes Vveh

Here the traffic induced heat flux (Htraffic in W m—2) is determined by the individual heat flux from a single vehicle (Hveh, W m—2 veh—1), the number of vehicles in the vehicle category v (Nv) and the time the vehicles spend over any part of the road, determined by the vehicle speed (Weh) and the vehicle length (f^). Heat flux per light vehicle is given as H|ieh = 10 W m—2 veh—1. Heavy vehicles are considered to be three times as long and to give off three times as much heat. These parameter values are uncertain and require further assessment, but the model is not highly sensitive to the choice of these.

2.7.5. Surface humidity

The surface water vapour pressure is expected to decrease below the saturated value once the surface moisture starts to fall below a threshold value (groad,evap—thresh). This mimics the patchi-ness of the drying surface and moisture contained within the pores of the road surface. We write this in terms of the surface relative humidity (RHs) which is described as a discontinuous linear function

RHs gr°ad + road—100 for groad + sroad < groad,evap-thresh

groad,evap-thresh 100

for groad + sroad > groad,evap-thresh

The surface specific humidity (qs) in Equation (28) is then specified by

qs = mq*

2.8. Melting and freezing

Snow can melt once the snow temperature, i.e. surface temperature, reaches the freezing or melting point (Tmelt). For pure water this is 0 °C but when salt is present this will be lower (Section 2.9). The amount of melt depends on the surface energy flux and is given as a sink term for the surface snow and as a production term for the surface water

's,snowmelt = j

G for Gs > 0 and Ts > rmdt

g,snowmelt = Ss,snowmelt

where 1m is the latent heat of fusion of ice and Gs is the surface energy flux (Equation (16)).

Similar to snow melt, surface water may freeze when the surface temperature is at the melting/freezing temperature (Ts = Tmelt) and the surface heat flux is negative (Gs < 0). The amount of freezing depends on the surface energy flux and is given as a sink term for the surface water and as a production term for the surface snow/ice

'g, freeze

for Gs < 0 and Ts < Tm

s,freeze = Sg,freeze

2.9. Vapour pressure and freezing point temperature dependence on salt concentration

The addition of salt changes the vapour pressure of the surface moisture which may impact on the evaporation and condensation. The vapour pressure of a salt solution can be described as depending on the salt content, salt type and temperature. Vapour pressure for saturated salt solutions may be determined by fitting Antoine's function to experimental data (Morillon et al., 1999). Antoine's function is described using three parameters

(34) log10 es*alt = Asalt -

Csalt + T

where q* is the saturated specific humidity, determined using Equations B.2 and B.3. The value of groad,evap—thresh is not known but

where the saturated vapour pressure of the salt solution e*alt (mm Hg) is dependent on the temperature (T) in °C and the

Table 1

Antoine coefficients and other parameters for saturated salt solutions in the temperature range 10 °C to -25 °C, Morillon et al. (1999). Small corrections are made to the salt vapour pressure curve to ensure an intersection at the correct eutectic point

(ecorrection).

Variable Units Water/ice NaCl MgCl2

Atomk weight Mom^t) g mol-1 18.0 58.4 95.2

Saturated freezing °C 0 -21 -33

temperature

(Tmelt,salt-saturated)

Saturated solution by % 23 22

mass fraction

Saturated solution by % 0.085 0.05

molar fraction

(Saturatedsalt)

Saturated relative % 100 75 33

humidity at 0 °C

(RHs,salt-saturated)

Asalt 10.3 7.4 7.2

Bsalt 2600 1566 1581

Csalt 270 228 225

ecorrection mm Hg +0.013 +0.118

experimentally fitted coefficients Asalt, Bsalt and Csalt. Values for these parameters are listed in Table 1 for water/ice, NaCl and MgCl2. The intersection of these Antoine functions for saturated salt and pure ice indicate the eutectic temperature at which the salt solution freezes, see Table 1.

When the solution is above saturation then salt will crystallise out and the impact on the vapour pressure will diminish. For NaCl this results in a return to normal vapour pressures for over-saturated solutions with mass solutions >26%. This is prescribed in the model by a linear increase in vapour pressure from the saturated vapour pressure fraction (0.75 at 0 ° C) to the oversaturated vapour pressure fraction of 1. For MgCl2 this increase in vapour pressure for oversaturated solutions also occurs but it never completely returns to the water vapour pressure (Vaa and Meland, 2005).

From these data the temperature dependence of freezing temperature and surface relative humidity on the salt solution is specified and these are shown in Fig. 1, see Denby and Sundvor (2012) for more details. The final surface relative humidity (RHs,salt) is specified using

RH5salt = ef#fRHs (40)

eice('s)

Freezing temperature of salt solution

-5 -10 -15 -20

-25 -30 -35

V > \\

\ x \ \ / / /

.... s \ N / / / ...../...........

-----MgCI2 \ \ \ \ / / t

\ V / t \ /

10 15 20 25 Salt solution (% of mass)

¡2 0.9

1 0.8 o

.1 0.7

t 0.6 =J

Vapour pressure fraction (salt/ice) of salt solution at 0 °C

\ \...............

\ V \ / / /

-----MgCI2 \ V / /

\ \ y \ /

2.10. Road salting and snow ploughing rules

Information concerning road salting activities is generally not available and so a salting model, based on a set of rules, is implemented in the model. The salting rules are described using the following logic: A window of time is established (trule-window) that can be used both backwards and forwards in time, around the current time (t0) in which a number of meteorological parameters are searched for. If these parameters are found within specified bounds then salting can occur at predefined times of the day (trule-hour). A minimum 'delay' time between salting events is prescribed (trule-delay). The following rules apply based on temperature (Ta), Humidity (RHa) and precipitation (Prec).

10 15 20 25 Salt solution (% of mass)

Fig. 1. Freezing temperature (top) and vapour pressure ratio (salt/ice) (bottom) used in the model for NaCl and MgCl2 salt solutions.

Fig. 2. Road configuration parameters used in defining the street site data. In this example there are 4 lanes.

Table 2

Input data requirements for the NORTRIP emission model.

Site and road data

Traffic (hourly)

Meteorology (hourly)

Road maintenance activity (hourly)

Number of lanes Width of lane (m)

Road width (m)

Street canyon width (m)

Street canyon height north (m)

Street canyon height south (m)

Street orientation (degrees

from north) Latitude (decimal degrees)

Longitude (decimal degrees)

Elevation (m a.s.l.)

Height of observed wind (m) Height of observed temperature

and RH (m) Surface albedo (0—1) Time difference with UTC (hr) Surface pressure (mbar) Driving cycle (index) Pavement type (index)

Time and date

Total traffic volume (veh/hr)

Total heavy duty vehicle traffic volume (veh/hr)

Total light duty vehicle traffic volume (veh/hr)

Studded tyre heavy duty vehicle traffic volume (veh/hr)

Studded tyre light duty vehicle traffic volume (veh/hr)

Winter friction tyre heavy duty vehicle traffic volume (veh/hr)

Winter friction tyre light duty vehicle traffic volume (veh/hr)

Summer tyre heavy duty vehicle traffic volume (veh/hr)

Summer tyre light duty vehicle traffic volume

Heavy duty vehicle speed (km hr-1) Light duty vehicle speed (km hr-1)

Time and date Observed atmospheric temperature (°C) Wind speed (m s-1)

Relative humidity (%)

Rain fall (mm hr-1)

Snow fall (mm hr-1)

Global radiation (W m-2)

Cloud cover (fraction)a

Road surface wetness (mV or mm)b Road surface temperature (°C)b

Time and date Sanding mass per hour (gm-2)

Salting mass NaCl per hour (gm-2)

Salting mass MgCl2 per hour (gm-2)

Road wetting per hour (mm) Road cleaning occurrence Snow ploughing occurrence

a Not obligatory, cloud cover can be estimated from the global radiation.

b Not required directly for the emission modelling but can be used for comparison with observed concentrations.

If to = trule-hour and to > tlast-application + trule-delay then (salting is allowed this hour):

If Trule-min < Ta (t) < Trule-max for t = to to to + trule-window then Tallowed is true

If RHa (t) >RHrule-min for t = to to to + trale-window then RHal-lowed is true

If Prec(t) >Precmle-min for t = to - trule-window to to + trule-window then Precallowed is true If Tallowed and (RHallowed or Precallowed) then tapplication = to (salt is applied)

An additional rule concerning the wetting of the salt, at a predefined solution (e.g. 2o% salt), is included. Salt is applied in solution if the road surface moisture is below a threshold value

(gmle-mm). Values of the various parameters are listed in Table C.3. For snow ploughing a much simpler condition is applied. When snow depth on the road surface exceeds a limit value (sroad > 3 mm w.e., which for fresh snow is around 3 cm in depth) then snow ploughing occurs and snow is removed with a specified effiriency (hpnoughing -eff >

3. Model data requirements and application datasets

3.1. Input data

The NORTRIP model requires a number of model parameters (Table B.2 and Table A.1 in Part 1), static information about the site

Table 3

Summary information concerning the sites used in this study.

Site name City Modelling period(s) Road configuration Average daily traffic (vehicle) Average vehicle speed (km hr-1) Winter time max studded tyre fraction (%) Pavement type factor (hpave)

Hornsgatan Stockholm 2ooo, 2oo6—2o11: 1 calendar year and 5 non-summer periods of 9 months 4 lane street canyon 23,000 44 70-33 0.83a

Essingeleden Stockholm 2oo8—2oo9: 18 months 9 lane motorway 134,000 71 70 0.83a

Riksvei 4 (RV4) Oslo 2oo4—2oo6: 3 4—6 month winter periods 5 lane highway 42,000 64 28-18 1.32a

Nordby Sletta Oslo 2oo2: 3.5 month winter period 4 lane highway 35,000 82 32 1

Mannerheimintie Helsinki 2oo7—2oo8: 24 months 4 lane open street 19,000mod 37mod 80 1

canyon 21,000mod 48mod

Runeberginkatu Helsinki 2oo4: 4 month winter period 4 lane street canyon 80 1

H. C. Andersen Copenhagen 2oo6—2o11: 5 periods of 6—12 6 lane open street 58,000mod 43mod 0 4—2a

Boulevard (HCAB) months canyon

mod principally derived from traffic models. a Derived from road pavements characteristics using the Swedish road wear model.

and road characteristics, see Fig. 2, and temporal information (hourly) concerning traffic, meteorology and road maintenance activities to calculate emissions. In addition information concerning air quality measurements and exhaust emissions can be included for comparison and, in the case of NOx, for converting emissions to concentrations (Part 1, Section 3.9). These data are listed in Table 2.

3.2. Model parameters

The model parameters for the road dust sub-model are listed in Part 1, Table A.2 and for the surface moisture model these are listed in Table C.2. In all cases the same model parameters are used, though the pavement wear scaling factor may vary depending on information available about the pavement, Table 2. For pavements where no information is available on pavement type then this factor is given the default of 1.

Some model parameters are slightly different to those applied in Part 1 to Hornsgatan and HCAB. The road dust suspension factor has been halved to 2.5 x 10-6 veh-1 and the fraction of PM10 in the suspendable dust (/PM,ref) has been increased from 0.18 to 0.21. This choice better reflects the entire range of data presented here and is within uncertainties of these parameters. New information concerning the pavement type at HCAB after repaving in summer 2008 indicates pavement characteristics with lower wear than with the previous pavement. A pavement wear factor of hpave = 2 is applied after 2008, compared to 4 in the previous years. The efficiency factors for salt mixing for drainage, spray and snow ploughing have been set to unity and the efficiency factors for dust mixing in drainage and spray are given as 0.1%. These new efficiency factors are listed in Table C.2.

3.3. Application datasets

In this paper seven different sites covering 18 separate periods are used for assessing the NORTRIP model. Table 3 provides an overview of the sites and the datasets used, more detailed information on the sites is available in Denby and Sundvor (2012). Two of these sites, Hornsgatan and HCAB, have observed surface temperature and moisture available for some of the periods and these are used for a direct comparison with the surface moisture submodel. At Hornsgatan conductivity measurements at three positions on the road determine surface moisture and at HCAB the thickness of the surface water film is measured. All sites measure PM10 and NOx, some also measure PM2.5, at both a traffic site and at a nearby urban background site. At all sites the net PM10 concentrations, traffic minus urban background, are used for comparison with the model. Meteorological data is either measured at the sites, at roof top or at nearby meteorological stations. At all sites the studded tyre season lasts from around November to April, with the exception of HCAB in Copenhagen where no studs are used.

3.4. Salting data and salting rules

Information is only available at three sites concerning salting application on the particular road, Mannerheimintie, Hornsgatan 2010-2011 and HCAB. For the other sites the salting rule model is applied. Comparison of the salting rule model with the Hornsgatan 2010-2011 salting information gives a very similar number of salting events, 87 and 86 respectively, but the timing of the events is not exactly the same. On HCAB the salting rule model under predicts the number of salting events by a factor of two and on Mannerheimintie the salting rule model over

predicts the salting events by a factor of two, but the salting rule model is never applied for these two roads. Salting activities vary from city to city and one single set of rules are not expected to be valid at all sites.

4. Model application to the datasets

4.1. Methodology

The model is applied to all the available datasets given in Table 3 and a comparison with measurements (temperature, surface moisture and concentrations) is carried out for the coupled and uncoupled (no surface retention or suspension) models. As in

Fig. 3. Scatter plots showing the hourly observed and modelled surface temperature difference ATS = Ts - Ta for HCAB 2006-2007 (a) and Hornsgatan 2010-2011 (b). Also indicated is the orthogonal linear regression fit to the data with coefficients a0 (intercept) and a1 (slope).

Part 1 (Section 3.9) emissions are converted to concentrations, for comparison with measurements, using NOx emissions and observed NOx concentrations. Sanding is not included in the current modelling though this may contribute to emissions at some sites.

The following comparisons will be presented to assess the surface moisture sub-model at the Hornsgatan and HCAB sites:

1. Comparison of hourly modelled and measured surface temperature

2. Comparison of hourly modelled and measured surface moisture

3. Comparison of hourly modelled and measured concentrations, using both modelled and measured surface moisture

The following comparisons are made to assess the model:

4. Comparison of modelled and measured concentrations for the coupled and uncoupled model at all sites

5. Sensitivity to model inputs and parameters at one site

4.2. Modelled and measured surface temperature

For five years from Hornsgatan and two years from HCAB, surface temperature measurements are available. Though this parameter is not a prerequisite for surface moisture modelling, predicting the surface temperature successfully is a necessary step towards this. A comparison is made between the modelled and

Fig. 4. Top: daily mean surface moisture layer thickness, modelled and observed, for HCAB 2oo7—2oo8. Middle: modelled surface snow/ice depth. Bottom: Modelled and observed surface retention factor fq (o is wet and 1 is dry).

observed hourly surface temperature difference, i.e. DTs = Ts - Ta, and presented in the form of scatter plots (Fig. 3).

Modelled hourly surface temperature differences are well correlated with those observed, ranging from R2 = 0.52-0.63 for Hornsgatan and R2 = 0.72 and 0.75 for HCAB. There is a small negative bias of the model at both sites (Intercept a0 = 0.140.60 °C) with the regression slope ranging from a1 = 0.84-0.88 for Hornsgatan and a1 = 1.23 and 1.24 for HCAB.

4.3. Modelled and measured road surface moisture and its sensitivity to salting

We compare modelled and measured road surface moisture for the two sites Hornsgatan and HCAB. Comparison is made for two types of model calculations, one 'with' and one 'without' the impact of salt on the surface vapour pressure, both calculations include the application of salt. To indicate the temporal variability in surface moisture we show, in Figs. 4 and 5, two examples of daily mean

surface moisture modelling for HCAB and Hornsgatan with salt impact included in the modelling.

The following indicators are used to further compare the modelled and observed moisture at these two sites:

1. comparison of modelled and observed wet surface frequency (percentage of total hours when surface is wet)

2. the percentage of hours with correctly modelled surface moisture (number of 'hits')

3. correlation of the modelled and observed hourly PM10 concentrations

The results are shown in Fig. 6 for all seven years at these two sites. Both cases, with and without salt impact, are shown. The model tends to slightly under predict wet road frequency (Fig. 6 top) for Hornsgatan but slightly over predicts for HCAB. The average fraction bias for all years is -10.7% without salt impact and -2.6% with salt impact. From the model hit rate

Fig. 5. As in Fig. 4 but for Hornsgatan 2010-2011. Only conductivity measurements are available at this site to indicate surface moisture and this is indicated by the observed retention factor fq (0 is wet and 1 is dry).

Impact of salting on road surface moisture: frequency of wet road

□ No salt impact

■ With salt impact

■ Observed moisture 53

100 95 90

I» 80

1 75 a 70

S 65 | 60 55 50

Hornsgatan Hornsgatan Hornsgatan Hornsgatan Hornsgatan HCAB HCAB

2006-2007 2007-2008 2008-2009 2009-2010 2010-2011 2006-2007 2007-2008

Impact of salting on road surface moisture: correctly modelled surface moisture

85.0 85.4

□ No salt impact 92.2 ■ With salt impact

83.5 83.0 83.6 83.0

Hornsgatan Hornsgatan Hornsgatan Hornsgatan Hornsgatan HCAB HCAB

2006-2007 2007-2008 2008-2009 2009-2010 2010-2011 2006-2007 2007-2008

Impact of salting on road surface moisture: correlation of hourly mean net PM

l 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

Hornsgatan 2006-2007

□ No salt Impact

■ With salt Impact

■ Observed moisture

Hornsgatan 2007-2008

Hornsgatan 2008-2009

Hornsgatan 2009-2010

HCAB 2007-2008

Fig. 6. Top: Frequency of modelled and observed wet road surface (hourly). Middle: Frequency of correctly modelled surface moisture, or 'hits'. Bottom: Hourly mean PMio correlation of modelled concentrations with observations. Shown are the model calculations both with and without salt impact on the surface vapour pressure as well as the model calculations using the observed surface moisture to determine retention.

(Fig. 6 middle) it can be seen that 76-90% of road moisture conditions are correctly predicted without the inclusion of salt impacts. This increases to 80-92% with the inclusion of salt impacts, with slight reductions for HCAB. The hourly mean net PMio concentration correlation (Fig. 6 bottom) is seen to improve slightly with the inclusion of salt impacts for five of the seven years. In general the moisture sub-model performs fairly well with average moisture hit rates of 85% for all years. The improvement with salt impact is only seen in the Hornsgatan data.

4.4. Comparison of modelled and observed concentrations

The coupled model, including salt impact, is now applied to the 18 separate periods of varying duration, from 3.5 months to 2 years (Table 3), at the 7 sites. The daily mean PM10

concentrations for two cases (Mannerheimintie, 2oo7—2oo8 and RV4 2oo5) are shown in Figs. 7 and 8. Plots for all 18 periods for the 7 sites are provided as supplementary material, see Appendix A. The reader is also referred to Part 1 (Section 4.2) for similar plots for Hornsgatan and HCAB using observed surface moisture. As can be seen much, but not all, of the variation in the emission concentrations and the emission factors are captured by the model.

The results of each of the 18 separate model applications are summarized in Figs. 9—11 where the net PM1o mean concentrations, 9o'th percentile daily mean concentrations and daily mean correlation (R2) are presented. Two cases are shown: the coupled model, with surface retention and salting, and the uncoupled model, with no retention and no salting. This last simulates an emission factor based calculation which is not affected by surface conditions.

Fig. 7. Results of the model application at Mannerheimintie for the period January 2007 to December 2008. Top: Daily mean modelled and observed concentrations of PMi0 including salt contribution. Middle: Surface mass loading of suspendable dust and salt.

The fully coupled model (with retention and salting) provides mean concentrations within ±35% of observations for all of the datasets (Fig. 9). The largest differences are seen for NB 2002 and RV4 2006. There is no clear reduction of bias as a result of the application of the coupled model compared to the uncoupled model, 9 of the 18 datasets have less bias for the coupled model. The uncoupled model has a minor impact on the mean concentrations since removal of dust through drainage and spray is very limited. The addition of salt slightly increases the means but the added surface retention will also reduce these, particularly for the shorter datasets where mass remains on the surface at the end of the modelling period. The 90'th percentiles are well simulated for most datasets with only NB 2002 and RV4 2006 lying outside the ±30% range (Fig. 10). Surface retention has a more obvious impact on the percentiles. 13 of the 18 datasets show reduced bias in the 90'th percentile with the application of the coupled model.

Surface retention has a major impact on the timing, and hence, correlation of the daily mean concentrations (Fig. 11). For many data sets the correlation is very low for the uncoupled model, when surface retention is not included. Only for one

dataset (HCAB, 2006-2007) is the correlation higher for the uncoupled model. The correlation presented in Fig. 11 can be directly compared to the results for Hornsgatan and HCAB presented in Part 1 (Fig. 8). In that case observed surface moisture was used and lead to daily mean correlations (R2) of between 0.77 and 0.91 for Hornsgatan and for HCAB of 0.51 and 0.39. Using the moisture sub-model leads to correlations for Horns-gatan of between 0.43 and 0.68 and for HCAB of 0.30 and 0.31. As we have seen in Section 4.3 the surface moisture is not modelled correctly for 15% of the hours and this leads to the reduction in modelled correlation.

Two of these sites and periods have previously been modelled using the model from Omstedt et al. (2005). For Hornsgatan 2000 (Omstedt et al., 2005) predicts a correlation R2 = 0.59 and a fractional bias of -3% and -10% for the net mean concentrations and 90'th percentile respectively. The NORTRIP model provides similar correlation (R2 = 0.60) but under predicts the observed concentrations with fractional biases of -30% and -14% for the two indicators. In the case of Runeberginkatu 2004 (Kauhaniemi et al., 2011) the modelled correlation for the total concentrations, including background, were R2 = 0.64 with a fractional bias of

Fig. 8. As in Fig. 7 but for RV4 in the period November 2005 to April 2006.

FB = +3%. For NORTRIP these two indicators are R2 = 0.77 and FB = -4% respectively.

4.5. Sensitivity assessment

There are a large number of model parameters and input data so it is interesting and necessary to assess the sensitivity of the model to these. We do this by perturbing a number of parameters and input data, or removing various processes, and assessing the sensitivity of the model to these changes for the one dataset Hornsgatan 2010-2011. It should be noted that the sensitivity to model parameters is also dependent on the dataset being modelled. For datasets such as Hornsgatan, where there are long cold wet periods, then small changes to the input data and the model parameters may lead to significant changes in the surface conditions on a day to day basis.

The results are shown in Figs. 12-14 where the relative surface wetness (modelled/observed), the daily mean correlation and the mean concentrations are shown as a result of the perturbations. We divide the results into the categories of 'Salting', 'Moisture processes', 'Dust processes' and 'Meteorological data'.

The modelled surface wetness frequency shows a range of approximately ±10% for the selected parameter perturbations. These changes can lead, for this particular dataset, to significant changes in correlation. For example a 10% reduction in surface wetness can reduce the correlation by a factor of two, as is the case when street canyon shadowing is removed from the model. In general we see that the surface wetness and correlation is sensitive to salting, some of the moisture process parameters and importantly to meteorological input data. The mean concentrations are generally less sensitive to these changes and most sensitive to uncertainties in the wear and, for this period and site, suspension rates. The sensitivity assessment provided here indicates both uncertainties in the model and real dependencies of the model to input data.

5. Discussion and conclusions

In this paper we have applied a coupled road dust and surface moisture model to predict traffic induced non-exhaust emissions. In Part 1 of this paper (Denby et al., 2013) the road dust sub-model was described and assessed for 7 periods at 2 sites and in this second paper the surface moisture sub-model was further

Net mean PM10 concentration

50 45 40 35 30 25 20 15 10 5 0

□ Uncoupled model □ Coupled model ■ Observed

§1 O Qj ■■J 3 N) O

§1 O QJ 00 3 N) O O -J

ID 3 M

N> m 3J 3J 30

V» 3* ra < < <

O O 00 M M M

ro o o o

o n Ji U1 Ol

ID n> 3

o 1 ?»

N> S" o

IV) 9 M 9

o 00 o CO

o N) o o 00 IV) o

01 ■vl

03 N) O O ID

> □0

Fig. 9. Net mean PM1o concentrations for the 18 datasets for which NORTRIP has been applied. Shown are uncoupled (no retention and no salting) and the coupled (including surface moisture retention and salting) model results. Observed net mean concentrations are also shown. * indicates sites where information on salting is available, the other sites use the salting rule model.

described and the complete coupled model was applied to 18 periods from 7 different sites. The surface moisture is shown to be essential for determining the retention of particles on the surface, their subsequent suspension and also for removal of the surface salt through drainage and vehicle spray. The impact of salt on the surface vapour pressure, which directly affects the evaporation

and condensation processes, is also included in the model and this has been shown to have an impact on the modelling results.

The variability of road dust concentrations of PM10 are dependent on several factors including traffic, dispersion conditions, road maintenance activities, road surface wetness and surface dust loading and suspension. The variability due to traffic

Net daily mean 90'th percentile PM10 concentration

Sr 80 £

□ Uncoupled model

□ Coupled model ■ Observed

° So Si

o a o ST

Ml £ I

8 9 2 °

' Jfi J. in

h-» ..

)_» (U

S- ore

O CD O CD

O g <£.

Fig. 10. As in Fig. 9 but showing the net daily mean 9o'th percentile PMio.

Net daily mean PM10 correlation

□ Uncoupled model □ Coupled model

M X isj I M I M I oo oo oo o o

? s si s I

M ÎH o So S, oo s

o SO ai ID s

l«J X <? =

h I V>

K OO O Q)

o ¡o ti ST vl ™ 05

o O, o O,

Fig. 11. As in Fig. 9 but showing the correlation (R2) between modelled and observed net daily mean PMi0 concentrations.

Relative frequency wet road (modelled/observed)

Reference (86 salting events) Salting

Salting but no salt Impact Salting by rule (87 events) Salting by rule (50% salt mass MgCI2) Dry salting No salting Moisture processes No vehicle spray Vehicle spray (x 5) Drainage threshold (x 2) Drainage threshold (x 0.5) No street canyon shadowing No traffic heat flux or turbulence Retention threshold parameters (x 2) Retention threshold parameters (x 0.5) Evaporation threshold parameter (x 2) Evaporation threshold parameter (x 0.5) Drainage mixing efficiency of dust (x 10) Dust processes Suspension rate (x 2) Suspension rate (x 0.5) Road wear rate (x 1.5) Road wear rate (x 0.66) Meteorological data RH (-5%) Temperature (+2 C) Wind speed (x 2) Wind speed (x 0.5) Incoming longwave (+10 W/m2) Using observed moisture for retention

J 0.93 I 0.94

J 0.92

H 0.90

J 0.89

H 0.87 I 0.88

J 0.89

H 0.94

J 0.87

J 0.91

H 0.87

H 0.93

II 0.98

□ 0.99

II 1.03

□ 0.98

□ 1.02

II 0.99

0.96 0.96 0.96 0.96

J 1.02

Fig. 12. Results of the sensitivity analysis of model parameters applied to the Hornsgatan 2010-2011 dataset. Shown is the relative frequency of wet roads (modelled/observed). The reference case is that presented in Figs. 9-11.

Correlation (Rz) of daily mean PM10 concentrations

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Reference (86 salting events) Salting

Salting but no salt impact Salting by rule (87 events) Salting by rule (50% salt mass MgCI2) Dry salting No salting Moisture processes No vehicle spray Vehicle spray (x 5) Drainage threshold (x 2) Drainage threshold (x 0.5) No street canyon shadowing No traffic heat flux or turbulence Retention threshold parameters (x 2) Retention threshold parameters (x 0.5) Evaporation threshold parameter (x 2) Evaporation threshold parameter (x 0.5) Drainage mixing efficiency of dust (x 10) Dust processes Suspension rate (x 2) Suspension rate (x 0.5) Road wear rate (x 1.5) Road wear rate (x 0.66) Meteorological data RH (-5%) Temperature (+2 C) Wind speed (x 2) Wind speed (x 0.5) Incoming longwave (+10 W/m2) Using observed moisture for retention

0.48 0.53 0.53

H 0.29

H 0.46

H 0.35 H 0.35

U 0.32 J 0.30 Z1 0.32

ZJ 0.60 □ 0.58

□ 0.59

□ ( □ <

J o.6:t

0.68 0.69 65 3.67

.66 .66 0.68 .66

Fig. 13. As in Fig. 12 but showing the correlation (R2) of daily mean PMio concentrations.

Mean PM10 concentration (|jg/nri3)

0 2 4 6 8 10 12

Reference (86 salting events) Salting

Salting but no salt impact Salting by rule (87 events) Salting by rule (50% salt mass MgCI2) Dry salting No salting Moisture processes No vehicle spray Vehicle spray (x 5) Drainage threshold (x 2) Drainage threshold (x 0.5) No street canyon shadowing No traffic heat flux or turbulence Retention threshold parameters (x 2) Retention threshold parameters (x 0.5) Evaporation threshold parameter (x 2) Evaporation threshold parameter (x 0.5) Drainage mixing efficiency of dust (x 10) Dust processes Suspension rate (x 2) Suspension rate (x 0.5) Road wear rate (x 1.5) Road wear rate (x 0.66) Meteorological data RH (-5%) Temperature (+2 C) Wind speed (x 2) Wind speed (x 0.5) Incoming longwave (+10 W/m2) Using observed moisture for retention

14 16 18 20

J 13..

J 12.4

J 11.9

] 15.4 14.8 L4.5 H 16.2

14.9 L4.4

] 15.7 □ 16.0 14.5 : i5.3 4.3 14.5 14.8

□ 16.6

H 18.9

U 15.9 U 15.8 I 16.1

.4.3 14.9 14.6

Fig. 14. As in Fig. 12 but showing the mean concentrations of PMio.

can be accounted for by using observed traffic counts. By using NOx as a tracer then the variability due to dispersion is also accounted for. Running the model both with and without surface retention allows the impact of retention on the variability to be estimated and for almost all datasets this was found to be the most dominant source of variability in the emissions. In two cases, notably HCAB 2006-2008, the variability is better explained without the use of surface retention. This indicates that retention and suspension is low at this site or that there may be other sources leading to the variability. In the years 20082010, at this same site, the explained variability due to surface retention was seen to be significant.

The emissions from salt are found in the current model calculations to contribute from 1 to 10% of the total PM10 emissions and this aspect has already been reported in Part 1. In the current version of the model salt is assumed to be suspended at the same rate as dust and to have the same size distribution. This aspect of the modelling still requires attention and a future study will use surface and ambient measurements of salt to further investigate the contribution of salt to PM10. Though sand can be added to the road surface in the model, as demonstrated in Part 1, this aspect has not been addressed further as large uncertainties still remain concerning its contribution to emitted PM10. Further measurement data is required if this aspect of the modelling is to be improved.

As discussed in Part 1 there are significant uncertainties associated with the modelling of the road dust, as well as the use of NOx data to convert emissions to concentrations. There are also uncertainties associated with the modelling of the road moisture. The level of uncertainty of the moisture model on an hourly basis can be interpreted from Fig. 5 where the number of correct hits is found to be around 85% for the two sites Hornsgatan and HCAB. Correctly modelling the surface moisture has most impact on the timing of the road dust emissions and has less impact on the mean or per-centiles (Figs. 6 and 7) as long as the wet road frequency is reasonably predicted. The model appears to be able to represent the surface wetness fairly well but there are clearly periods in the datasets where surface wetness is not well modelled. It is not known if this is the result of input data, e.g. precipitation, or of the model formulation itself.

From the 18 periods modelled, covering 7 different sites, the coupled model provided an average absolute fractional bias for the mean concentrations of 15% and for the daily mean 90'th percentile concentrations of 19%. The coupled model provided an average correlation of R2 = 0.49 whilst the average correlation of the uncoupled model was only 0.16.

We conclude from this study that the coupled road dust and surface moisture model provides an improved methodology for modelling non-exhaust road dust emissions under most conditions. The model is most effective when the road surface is moist over longer periods and significant wear, due to studded tyres, or dust loading is present. If timing of the emissions is of interest, e.g. with forecasting, or percentiles are required then the model provides a significant improvement compared to models based on emission factors only (no retention). Since it is a process based model it also presents a significant step towards the application of non-exhaust emission models as management tools for the study of abatement strategies. The inclusion of salt and its impact on vapour pressure has been demonstrated for NaCl, but the ability to simulate this also provides the mechanism for modelling the impact of dust binding salts which have a larger impact on the vapour pressure.

Substantial efforts still remain to improve the model. As more observational data is gathered concerning road surfaces, dust loading, sanding, removal processes and source characterisations of emissions then so too does the information pool for testing and

developing the model. Future measurement campaigns can make use of the models conceptual structure and can focus on elements that would help improve the model and the underlying processes related to road dust emissions.

Acknowledgements

This work has been carried out within the Nordic Council of Ministers Project NORTRIP (BLS-306-00064) with substantial financial support provided by the Norwegian Climate and Pollution Agency (KLIF 4011009). The measurements at HCAB are funded by the Danish Environmental Protection Agency.

Appendix A. Supplementary data

Supplementary data related to this article can be found at http:/ dx.doi.org/10.1016/j.atmosenv.2013.09.003.

Appendix B. Physical constants and equations used in the model description

The relative humidity of water vapour in air, RHQ, is specified by the ratio of the water vapour partial pressure (ea) and the saturated partial pressure (ea)

RHa = 100

Calculation of the specific humidity, saturated (q*a) or unsaturated (qa), is carried out using

0.622 ea (Pa - 0.378-ea)

where pa is the atmospheric pressure. The Bolton equation fit to the saturated water vapour pressure is given by

ea = 6.112-exp

17.67 -Ta

Ta + 243.5

where Ta is the atmospheric temperature in °C. Calculation of air density from pressure and temperature is given by

Pa RdTK,a

where Rd is the specific gas constant for dry air and is the atmospheric temperature in Kelvin.

Table B.1

Physical constants used in the model

Parameter Description Units Value

Is Latent heat of condensation J kg-1 2.5 x 106

(vapour-water)

^ice Latent heat of sublimation J kg-1 2.8 x 106

(vapour-ice)

^m Latent heat of fusion (water-ice) J kg-1 3.3 x 106

Cp Heat capacity of dry air J kg-1K-1 1005

Rd Specific gas constant for dry air J kg-1K-1 287

s Stefan—Boltzmann constant W m-2 K-4 5.67 x 10-8

k von Karman constant 0.4

U Angular velocity of the Earth rad s-1 7.3 x 10-5

Appendix C. Model variables and parameters

Table C.1

Variables principally related the surface moisture sub-model. For variables related to the road dust sub-model see Part 1, Table A.1.

Variable Units Variable typea Description

groad mm P Water mass on the road surface

sroad mm w.e. P Snow/ice mass on the road surface. Units for ice/snow are in mm w.e. (water equivalent)

groad,drainable mm D Amount of water that may be drained from the road

groad,drainable-min mm IP Non-drainable road water mass

Pg mm hr-1 D Production rate of liquid water on the road surface

Ps mm hr-1 D Production rate of frozen water (ice/snow) on the road surface.

Sg mm.hr-1 D Sink rate of liquid water on the road surface

Ss mm hr-1 D Sink rate of frozen water (ice/snow) on the road surface.

Rain mm IT Amount of liquid precipitation within the model time step Dt

Snow mm w.e. IT Amount of solid precipitation within the model time step Dt

groad-wetting mm ITorD Amount of water applied when wet salting/sanding or during cleaning.

^wetting hr ITorD Timing of the wetting event. Input or derived by salting rules

groad,sprayable-min mm IP Minimum surface moisture level for spray to occur

Rg,spray hr-1 D Rate of road water removal by spray processes

•^0, spray veh-1 IP Basic factor defining the proportion of surface moisture removed with the passage of

one vehicle due to spray processes at the reference speed Vref spray

Vref, spray km hr-1 IP Reference vehicle speed at which fj spray is valid

um spray eff um drainage eff 0-1 IP Efficiency factor for removal of dust and salt mass (m) due to vehicle spray

0-1 IP Efficiency factor for removal of dust and salt mass (m) due to drainage

usnow ploughing eff 0-1 IP Efficiency factor for removal of snow due to snow ploughing

Ta °C IT Atmospheric temperature, usually at 2 m.

K IT Atmospheric temperature in Kelvin, usually at 2 m.

Ts °C P Road surface temperature

Tsub °C IT and D Sub-surface road temperature

Tks K P Road surface temperature in Kelvin

Gs Wm-2 D Surface energy flux

Gsub Wm-2 D Sub-surface energy flux

Rnet,s Wm-2 IT or D Surface net radiation flux

RSin Wm-2 IT Incoming short wave radiation

aroad 0-1 IM Road surface albedo

asnow 0-1 IP Road surface snow albedo

Rl-in Wm-2 D Incoming long wave radiation

RL-out Wm-2 D Outgoing long wave radiation

Hs Wm-2 D Surface sensible heat flux

Ls Wm-2 D Surface latent heat flux

Htraffic Wm-2 D Traffic heat flux to the surface

n 0-1 D or IT Cloud cover fraction

£eff 0-1 D Effective long wave emissivity of the atmosphere

£cs 0-1 D Clear sky long wave emissivity of the atmosphere

£cl 0-1 IP Cloudy sky long wave emissivity of the atmosphere

£s 0-1 IP Long wave emissivity of the surface

broad m IM Total width of the road, from kerb to kerb

blane m IM Lane width

bcanyon m IM Width of the street canyon

hcanyon m IM Height of the street canyon. Two values, one for north and one for south.

ea< e'a Pa D Water vapour partial and saturateda pressure in the atmosphere.

es, e* Pa D Water vapour partial and saturateda pressure on the surface.

esalt, esalt Pa D Water vapour partial and saturateda pressure on the surface for a salt solution.

eice Pa D Vapour pressure for water and ice.

qa q'a D Water vapour specific humidity and saturateda specific humidity in the atmosphere.

qs, q* D Water vapour specific humidity and saturateda specific humidity on the surface.

RHa, RHs and RHssalt D Relative humidity of the atmosphere, on the surface and of the salt solution on the surface.

FF(z) m s-1 IT Wind speed at height z.

^traffic and ^wind s m-1 D Aerodynamic resistance for traffic induced turbulence and wind shear induced turbulence

rT and rq s m-1 D Aerodynamic resistance for temperature and water vapour

Zo, Zt, and Zq m IP Roughness lengths for momentum, temperature and water vapour

traffic s veh- 1 IP Aerodynamic traffic coefficient

'veh m IP Length of vehicle type v

Hveh W m-2 veh-1 IP Surface heat flux from vehicle type v

groad,evap thresh mm IP Threshold value for surface moisture below which evaporation is reduced by reduction of relative humidity

gretention-thresh,source mm IP Threshold value defining the upper limit for retention, above which full surface retention is achieved

gretention-min,source mm IP Threshold value defining the lower limit for retention, below which no surface retention is achieved

Pa Pa IM Atmospheric pressure

Pa kg m-3 D Atmospheric density

Ps kg m-3 IP Road surface density

Cs J kg-1 K-1 IP Road surface specific heat

ks W m-1 K-1 IP Road surface thermal conductivity

DZs m D or IP Sub-surface layer slab depth

Tmelt °C D Melt/freezing temperature of the surface moisture

a Variable types are defined as prognostic (P), diagnostic (D), model input parameter (IP), site specific input metadata (IM) or site specific input time series (IT).

Table C.2

Model parameters for the surface moisture sub-model. For model parameters related to the road dust sub-model see Part 1, Table A.2.

Spray and splash factors Heavy (he) Light (li)

/o,spray (veh-1) Vref,spray (km hr-1) groad,sprayable-min (mm) 6.0 70 0.1 x 10-4 1.0 x 10"4

Drainage parameters Value

gdrainable (mm) 0.5

Ploughing parameters Value

Ploughing efficiency for snow removal Ploughing threshold (mm) 0.8 3

Energy balance parameters Value

groad,evap-thresh (mm) Roughness length (z0) (mm) Snow albedo (asnow) Subsurface parameters 0.02 1 0.4 Ps (kg m-3) 2400 Cs (J kg-1 K) 800 ks (W m"1 K) 2

Traffic turbulent exchange and heat flux Heavy (he) Light (li)

Otrafflc (veh 1) Hveh (W m-2veh-1) 1.0 x 10-2 30 1.0 x 10"3 10

Retention parameters Road Brake

gretention-thresh (mm) gretention-min (mm) 0.1 0.04 1 0.7

Efficiency factors for mass removal

Efficiency parameter Suspendable dust Salt

hploughing-eff hdrainage-eff hspraying-eff 0.001 0.001 0.001 000

Table C.3 Parameters used to define the salting rule model described in Section 2.10.

Salting rule Value Units Comment

parameter

^rule-hour (1) 05:00 hh:mm First time of day when salting can occur

^rule-hour (2) 20:00 hh:mm Second time of day when salting can occur

^rule-delay 0.2 days Minimum allowable time between salting events in days

^rule-window 0.5 days Time window checked ahead (temperature, RH) and behind (precipitation) in days

Trule-min -6 °C Minimum temperature for salting in the forward time window

Trule-max 0 °C Maximum temperature for salting in the forward time window

Precrule-min 0.1 mm hr-1 Salt if precipitation occurs above this level in the forward and behind time window

RHrule-min 95 % Salt if the relative humidity is above this level in the forward time window

grule-min 0.25 mm Dry salt if the surface moisture is above

Msalt(i) salting this value at time of salting

15 g m-2 Mass of salt applied at each application

Saltsolution 0.2 kg litre-1 Salt solution by mass, if 0 then dry salting

References

Berger, J., Denby, B., 2011. A generalised model for traffic induced road dust emissions. Model description and evaluation. Atmos. Environ. 45, 3692—3703. http://dx.doi.org/10.1016/j.atmosenv.2011.04.021.

Bukowiecki, N., Lienemann, P., Hill, M., Furger, M., Richard, A., Amato, F., Prevot, A.S.H., Baltensperger, U., Buchmann, B., Gehrig, R., 2010. PM10 emission factors for non-exhaust particles generated by road traffic in an urban street canyon and along a freeway in Switzerland. Atmos. Environ. 44, 2330— 2340.

Denby, B.R., Sundvor, I., 2012. NORTRIP Model Development and Documentation. Norwegian Institute for Air Research (NILU OR 23/2012). URL: www. nilu.no.

Denby, B.R., Sundvor, I., Johansson, C., Pirjola, L., Ketzel, M., Norman, M., Kupiainen, K., Gustafsson, M., Blomqvist, G., Omstedt, G., 2013. A coupled road dust and surface moisture model to predict non-exhaust road traffic induced particle emissions (NORTRIP). Part 1: road dust loading and suspension modelling. Atmos. Environ. 77, 283—300. http://dx.doi.org/10.1016/j.atmosenv. 2013.04.069.

Denby, B., Snellen, H., 2002. A comparison of surface renewal theory with the observed roughness length for temperature on a melting glacier surface. Bound.-Layer Meteorol. 103, 459—468. http://dx.doi.org/10.1023/A: 1014933111873.

Garratt, J.R., 1994. The Atmospheric Boundary Layer. Cambridge University Press, Cambridge, UK.

Gustafsson, M., Blomquist, G., Gudmundsson, A., Dahl, A., Jonsson, P., Swietlicki, E., 2008. Factors affecting PM10 emissions from road pavement wear. Atmos. Environ. 43, 4699—4702.

Iqbal, M., 1983. An Introduction to Solar Radiation. Academic Press, Toronto, Canada.

Karlsson, M., 2001. Prediction of hoar-frost by use of a road weather information system. Meteorol. Appl. 8 (1), 95—105.

Konzelmann, T., van deWal, R.S.W., Greuell, W., Bintanja, R., Henneken, E.A.C., Abe-Ouchi, A., 1994. Parameterization of global and longwave incoming radiation for the Greenland ice sheet. Global Planet. Change 9,143—164.

Kauhaniemi, M., Kukkonen, J., Härkönen, J., Nikmo, J., Kangas, L., Omstedt, G., Ketzel, M., Kousa, A., Haakana, M., Karppinen, A., 2011. Evaluation of a road dust suspension model for predicting the concentrations of PM10 in a street canyon. Atmos. Environ. 45, 3646—3654.

Möller, S., 2006. VTl-winter Model. Road Condition Model. The Swedish National Road and Transport Research Institute, Linköping (VTl Report 529) (in Swedish). URL: http://www.vti.se/en/publications/winter-model-road-condition-model/.

Morillon, V., Debeaufort, F., Jose, J., Tharrault, J.F., Capelle, M., Blond, G., Voilley, A., 1999. Water vapour pressure above saturated salt solutions at low temperatures. Fluid Phase Equilibr. 155, 297—309.

Norman, M., Johansson, C., 2006. Studies of some measures to reduce road dust emissions from paved roads in Scandinavia. Atmos. Environ. 40, 6154—6164.

Omstedt, G., Bringfelt, B., Johansson, C., 2005. A model for vehicle-induced non-tailpipe emissions of particles along Swedish roads. Atmos. Environ. 39, 6088— 6097.

Pant, P., Harrison, R.M., 2013. Estimation of the contribution of road traffic emissions to particulate matter concentrations from field measurements: a review. Atmos. Environ.. ISSN: 1352-2310 77, 78—97. http://dx.doi.org/10.1016/ j.atmosenv.2013.04.028.

Pay, M.T., Jimenez-Guerrero, P., Baldasano, J.M., 2011. lmplementation of resuspension from paved roads for the improvement of CALlOPE air quality system in Spain. Atmos. Environ. 45, 802—80 .

Sass, B.H., 1997. A numerical forecasting system for the prediction of slippery roads. J. Appl. Meteorol. 36, 801—817.

US EPA AP42, 2006. Chapter 13: Miscellaneous Sources, fifth ed., vol. I http://www. epa.gov/ttn/chief/ap42/ch13/.

Vaa, T., Meland, S., 2005. Fors0k Med Befuktning Med Magnesiumkloridl0sning Pa Gj0vik/Toten. Sluttrapport (In Norwegian). SINTEF Report NR: 2414, Ref: A6755. URL: www.vegvesen.no/_attachment/69710/binary/34080.