Statistical metrics used in the evaluation of the CHIMERE results compared
to observations are defined as follows.
Mean bias (MB)=1n∑i=1n(mi-oi)Normalized mean bias (NMB)=1n∑i=1n(mi-oi)o¯Root mean square error (RMSE)=1n∑i=1n(mi-oi)2Normalized root mean square error(NRMSE)=1n∑i=1n(mi-oi)2o¯Correlation coefficientR=∑i=1n(mi-m¯)(oi-o¯)∑i=1n(mi-m¯)2∑i=1n(oi-o¯)2,
where mi and oi are the modeled and observed concentrations at
time i, respectively, and m¯ and o¯ their average over a given
period.
Sulfate and SO2
Sulfate daily concentrations in Paris are given in Fig. 2. The variability
of sulfate (as of nitrate) during the PARTICULES campaign has been discussed
in details in Bressi et al. (2013). Fine
(PM2.5) sulfate concentrations range between 0.4 and 5.0 µg m-3
(one high value at 8.7 µg m-3), with an average of 2.0 µg m-3 over the studied period (1 April–10 September). The
episodes with highest concentrations are associated with air masses
originating from the north/northeast, also noted by
Bressi
et al. (2013), Petetin et al. (2014) and
Petit et al. (2015). Despite a faster
SO2-to-sulfate conversion due to higher OH levels in summer, lower
concentrations are measured during that season due to a combination of lower
SO2 emissions and a dominant marine regime, with relatively clean air
masses originating from west and southwest and slightly more polluted ones
from northwest.
Observed and modeled daily averaged concentrations (left panel),
diurnal profiles (middle panel), and monthly concentrations (right panel).
MOD-nodep results are only shown for NH3. Note: CHIMERE monthly
concentrations are computed including only days with available observational
data. For particulate matter observations, only daily values are available.
During the period of available data (152 days in spring and summer),
NH3 levels are high enough to fully neutralize both sulfate and
nitrate, as indicated by the linear regression of NH4+ vs.
NO3-+ 2SO42- daily concentrations in the fine mode that
gives a slope of 1.01, a y intercept of -0.20 ppb, and a correlation
coefficient (r2) of 0.97 (n=150; see Fig. S1 in the Supplement). Note
that plotting all major cations
(Na++ NH4++ K++ 2Ca2++ 2Mg2+) against all
major anions (NO3-+ 2SO42-+ Cl-) leads to a slope
of 1.03, a y intercept of +0.13 ppb, and a correlation of 0.97,
demonstrating the neutrality of our fine aerosol.
Statistical results of modeled vs. measured concentrations are reported in
Table 1. The model partially reproduces the day-to-day variability of sulfate
concentrations (r=0.59), but overestimates concentrations, with a NMB of
+48 % and a NRMSE of 74 %. This does not appear to be related to a too
high SO2-to-sulfate conversion since SO2 concentrations are
significantly overestimated in Paris, by about a factor of 3 (Table 1). This
is also suggested by the simulated S ratio. This indicator – defined as the
ratio of SO2 over SO2+SO42-, all concentrations being
expressed in µg m-3
(Hass et al., 2003; Pay et al.,
2012) – allows us to assess how strongly oxidized is a plume containing
sulfur. High S ratios are found in air masses containing freshly emitted
SO2, while low S ratios are associated with older air masses in which
more SO2 has already been converted into sulfate. The observed and
simulated S ratios are shown in Fig. 3 (the SO2+SO42- time
series is shown in Fig. S4 in the Supplement). In the MOD simulation,
CHIMERE clearly overestimates the S ratio (average value of 0.54 vs. 0.34
for the observations); i.e., the simulated air masses contain too much
freshly emitted SO2 compared to reality. Such a high bias on SO2
concentrations is not expected but does not appear representative of the
CHIMERE performance at a larger scale. Considering the SO2 observations
available at nine urban background sites (AIRPARIF operational network) in the
region of Paris, NMB are lower, ranging from +24 to +160 %. As a large
part of SO2 is emitted by point sources, the dilution effect in a 3 × 3 km cell remains a well-known uncertainty source at stations potentially
impacted by plumes coming from nearby industrial facilities. However, in our
case, large SO2 industrial point sources are relatively far from our
background urban station, and emissions from non-point sources (i.e., on-road
transport and residential sectors) remain important in the center of Paris,
which suggests potential errors on the Paris agglomeration emissions
(overestimation of total emissions, wrong vertical allocation) and/or the
BLH. Indeed, the average SO2 diurnal profile shows maximum
discrepancies (up to a factor of 4.8) during the transition from a
convective to a nocturnal boundary layer. This transition occurs too early
in the model (see Fig. S3 in the Supplement), which likely explains a
noticeable part of the bias for SO2. Conversely, the sulfate
overestimation may be due to errors during the transport of air masses from
northeastern Europe.
Observed and modeled (with – MOD case – and without – MOD-noIDF
case – emissions over the Paris region) daily S ratio in Paris.
Statistical results at our urban background sites over the whole
period (all statistical metrics are defined at the beginning of Sect. 4; MO
is the observed concentration mean, N the data coverage).
Species
Case
MO
MB
NMB (%)
RMSE
NRMSE (%)
R
N (%)
NH3* (ppb)
MOD
4.0
-3.0
-75
3.9
99
0.42
64
MOD-noIDF
-3.1
-79
4.1
103
0.39
64
MOD-nodep
-1.8
-46
3.2
82
0.45
64
HNO3* (ppb)
MOD
0.3
+0.5
+195
0.8
320
0.56
81
MOD-noIDF
+0.3
+120
0.6
219
0.36
81
SO2* (ppb)
MOD
0.5
+1.0
+194
1.6
303
0.38
83
MOD-noIDF
-0.1
-20
0.9
170
0.25
83
Ammonium (µg m-3)
MOD
1.2
+0.4
+35
0.9
70
0.84
54
MOD-noIDF
+0.3
+23
0.8
64
0.84
54
Nitrate (µg m-3)
MOD
2.1
+0.4
+19
2.2
109
0.81
54
MOD-noIDF
+0.0
+1
2.1
101
0.81
54
Sulfate (µg m-3)
MOD
2.0
+1.0
+48
1.5
74
0.59
54
MOD-noIDF
+0.9
+42
1.4
69
0.61
54
F-NHx (ppb)
MOD
5.5
-4.1
-75
4.7
87
0.51
37
MOD-noIDF
-4.4
-80
5.0
92
0.48
37
S ratio
MOD
0.3
+0.2
+60
0.3
73
0.46
48
MOD-noIDF
-0.1
-29
0.2
55
0.33
48
GR (ppb ppb-1)
MOD
12.6
-11.4
-90
14.2
112
0.37
36
MOD-noIDF
-11.2
-88
14.0
111
0.33
36
TNH3 (ppb)
MOD
6.4
-3.6
-56
4.4
70
0.43
37
MOD-noIDF
-3.9
-61
4.7
74
0.40
37
TNO3 (ppb)
MOD
1.1
+0.8
+71
1.3
123
0.78
47
MOD-noIDF
+0.3
+31
1.1
97
0.79
47
* Statistics based on hourly data (otherwise, daily data are used).
Ammonia
Temporal variability
Daily averaged concentrations and diurnal profiles of NH3 are given in
Fig. 2. The model results will be discussed in the next section. According
to the review of Reche et al. (2012), NH3
concentrations in worldwide urban environments range between 0.4 and 63.6 ppb, thus spanning over 2 orders of magnitude. On a logarithmic scale, the
average concentration of 4.0 ppb measured in Paris over the whole period is
roughly in the middle range of this range. It is also consistent with the
values obtained in other European cities: 4.4 ppb in Aveiro (Portugal,
August–May), 5.2 ppb in Roma (Italy, May–March), 5.5 ppb in Münster
(Germany, May–June), 3.2 in Thessaloniki (Greece, year), 3.9–10.6 in
Barcelona (Spain, July and January), and 3.1 ppb in Schiedam (the Netherlands,
winter) (Reche et al., 2012, and references therein).
NH3 concentrations in Paris show a large variability (illustrated by a
standard deviation of 2.8 ppb) with several elevated NH3 episodes in
late spring and early summer (hourly concentrations reaching up to 18.5 ppb
in June), moderate concentrations in late summer, and lower ones in autumn
and winter. On average, the observed NH3 diurnal profile (Fig. 2) is
rather flat, with slightly increasing concentrations in the morning leading
to a maximum at 10:00–13:00 UTC. Concentrations decrease in the afternoon up
to a minimum at 20:00 UTC. The diurnal variability of NH3 depends on
many factors, including the strength of local emission sources, the dry
deposition, the evolution of the BLH, the formation of NH4NO3
during the night promoted by larger RH, and its thermodynamically driven
evaporation during the daytime
(Wichink Kruit et al., 2007). The
daytime increase may be partly due to this volatilization of
NH4NO3.
Influence of temperature
Figure 4 shows the NH3 concentrations in function of the temperature.
Both appear clearly linked in Paris, the highest episodes occurring
concomitantly with the warmest conditions (see the meteorology evaluation in
the Supplement, Sect. S2). The lower sensitivity to temperature in the
model will be discussed later. Such a relation between NH3 and the
temperature has already been observed in other cities
(e.g., Perrino et al., 2002;
Gong et al., 2011; Reche et al., 2012). Temperature
and RH strongly influence the equilibrium constant governing the
partitioning of inorganic compounds between the gas and aerosol phases, with
higher NH3 concentrations expected when the temperature is high and the
RH is low due to the volatilization of NH4NO3. In addition,
several NH3 emission sources may be enhanced by high temperature,
including the agricultural (e.g., volatilization of fertilizer) or biological
sources.
Daily observed (in black) and modeled (in blue) NH3
concentrations vs. temperature in Paris (for the model, only days with
available observations are plotted).
The link between NH3 and temperature can be illustrated by the early
July episode when, in parallel with the temperature increase between 30 June
and 2 July, the NH3 baseline progressively increases in Paris, up to
18.5 ppb at the hourly scale (the maximum over the whole FRANCIPOL period).
A part of the NH3 increase is likely due to evaporation of
NH4NO3, but in early July a similar episode is observed for
TNH3, which means that an additional NH3 source is present. The
NH3/ TNH3 ratios are shown in Fig. 5. The experimentally determined
TNH3 is clearly dominated by NH3 that has a contribution around
55–99 % (83 % on average). Negative artifacts on the NH4+
filter measurements cannot be excluded (in particular during summertime),
but increasing NH4+ concentrations by 50 % has a very limited
impact (NH3 contributions ranging in that case around 45–99 %,
78 % on average).
Daily NH3 / TNH3 ratios in observations (points) and
simulations (solid lines).
Influence of traffic NH3
Several studies have previously addressed the question of the NH3
emitted by the traffic in urban areas, although with more or less contrasted
and definitive conclusions depending on the city (e.g.,
Perrino et al., 2002; Gong et al.,
2011). The difficulty notably arises from the short lifetime of NH3
that can quickly deposit on the ground, be diluted or converted into
NH4+. In Paris, the diurnal profile does not show any peak at
morning and evening rush hours, even during periods of lower agricultural
emissions (e.g., August and September; too few data in winter). This suggests
that traffic emissions are probably a relatively minor source during our
study. This is supported by the low correlation of black carbon (BC) (mainly
emitted by the traffic) and NH3 concentrations measured at the LHVP
site (r=0.20 over the whole period). However, it is worth noting that
during the end of June episode, the hourly time series shows some morning
peaks (above an increasing background line likely due to the advection of
agricultural NH3) that may be associated to traffic NH3 emissions,
as illustrated by the increased correlation with BC (r=0.60 between the 21 June and 3 July) (Fig. 6). No similar situation is observed during the rest
of the campaign. In Roma, Perrino et al. (2002) observed
high levels of NH3 at curbside sites with a diurnal profile clearly
influenced by traffic emissions. However, due to the combined action of dry
deposition, dilution after emissions as well as the conversion into
particulate NH4+ (with sulfates and/or nitrates), these
concentrations were severely reduced at the urban background scale (about a
factor of 5) and the traffic profile type had disappeared. As a result, our
urban background conditions may have prevented us from accurately assessing
the potential impact of traffic emissions on ambient NH3
concentrations. Investigating the NH3 diurnal variability at the SIRTA
site, Petit et al. (2015) noticed a bimodal
traffic-like variation but only during spring and not during summer and
winter, suggesting that these variations may be related to processes other
than traffic.
Observed BC (in red) and NH3 (in black) hourly concentrations
at LHVP during the end of June.
Influence of agricultural NH3
As previously mentioned, NH3 is emitted by both agricultural and
non-agricultural sources. The former clearly dominates at the national
scale, as well as at the scale of the Paris region (which includes the rural
areas surrounding Paris), while the latter dominates at the scale of the
city itself (which includes only urban areas). Considering the role of
NH3 in the formation of NH4NO3 and the important contribution
of this aerosol compound to the PM2.5 pollution in Paris, it is of
major importance to assess the relative contribution of both types of
sources to the NH3 urban background in the city. Answering that
question would ideally require additional NH3 observations in Paris and
its surroundings in order to quantify the increment associated to local
sources. Without such observations, it is not possible to quantitatively
investigate the NH3 budget in Paris.
However, based on the available observations, we argue in this section that
among all NH3 emission sources, agriculture is probably the main driver
of the day-to-day variability of NH3 concentrations in Paris during the
time of the campaign (from spring to autumn) (in conjunction with the
thermodynamic equilibrium that drives the partitioning between the gas and
aerosol phases).
This is mainly supported by the NH3 (and TNH3) seasonal
variations. Although incomplete (due to missing observations in winter and
early spring), the NH3 seasonal pattern shows a maximum in spring and
early summer, moderate concentrations in late summer, and a minimum in
autumn. Such a seasonal pattern has been already reported in several studies
(e.g., Reche et al., 2012;
Skjøth et al., 2011). A roughly similar variability is expected for the
fertilizer applications, yet this emission source represents around 40 %
of the total agricultural source at the national scale, and this
contribution appears even higher around the Paris region
(Hamaoui-Laguel et al., 2014; see in particular their
Fig. 2a and b). The observed increase of NH3 with temperature is also
compatible with this source, as increased temperature favors fertilizer
evaporation (e.g., Hamaoui-Laguel et al., 2014).
Conversely, none of the non-agricultural emission sources are expected to be
particularly intense during this time of the year. This was discussed for
traffic-related emissions in the last section. Some NH3 may also be
emitted by biomass burning (for residential heating) but these emissions
are, in any case, low in spring and summer. Emissions from sewage and waste
disposal as well as emissions from other biological sources may also
contribute to NH3 levels. Interestingly, these latter sources may be
influenced by temperature, as are the NH3 concentrations measured in
Paris (see Fig. 4). However, if they dominate, one would not expect such a large
difference in concentrations between late May, early June and August (when
temperatures were comparable). Additionally, in this case, one would also
expect higher NH3 concentrations during stagnant conditions, which is
in contradiction with the low correlation between BC and NH3 (given
that such stagnant conditions lead to an accumulation of BC). The NH3
diurnal profile shows very limited variations along the day, which is
consistent with the idea of a strong NH3 background originating from
agricultural sources around the Paris region. All these elements thus
suggest that the agricultural source (and more precisely the fertilizer
application) drives a larger part of the NH3 day-to-day variability in
Paris than the other emission sources.
Geographical origin of the highest NH3 episodes
In this section, we investigate the geographical origin of the air masses
associated with elevated NH3 episodes. Back trajectories during the 10 days of highest NH3 concentrations (daily averages above 9.2 ppb, the
95th percentile of all daily values) are presented in Fig. 7a. Most
NH3 episodes are associated with moderate winds at altitude (particles
being released at 500 m a.g.l. in FLEXTRA simulations), air masses at D-1
(1 day before reaching Paris) being located in a radius of 50–400 km from
Paris. A noticeable exception is found on 9 July in the morning (around 06:00 UTC) when the wind suddenly changes direction (from southeast to southwest)
and speed (getting much stronger, with air masses originating from Spain at
D-1) while NH3 concentrations increase. Interestingly, some of the
highest NH3 episodes (e.g., 10 July) are associated with oceanic air
masses (excepted to be relatively clean) that have spent only a limited time
over land, which suggests the presence of intense NH3 emissions in the
corresponding regions (Normandy). The trajectory analysis suggests that air
masses with high NH3 concentrations do not appear to originate from a
particular geographical region. Instead, the highest episodes appear linked
to more diffuse NH3 emissions in the northern part of France,
associated with anticyclonic conditions with high temperature and moderate
winds. This is in accordance with Petit et
al. (2015) that suggest, based on NH3 measurements at the SIRTA
suburban site (southwest of Paris), a diffuse regional NH3 source, in
particular during summer (in spring, some high NH3 episodes associated
to east/northeast/southeast winds are also noticed, but without any clear pattern).
Back trajectories at D-1 (1 day before reaching Paris) associated
with highest (a) NH3 (left panel) and (b) HNO3
(right panel) episodes (highest episodes being selected according to daily
concentrations above the 97th percentile of all daily measurements, i.e.,
9.2 and 0.9 ppb for NH3 and HNO3, respectively). For clarity, only
back trajectories of seven particles around the center of Paris are plotted, each
6 h (i.e., 28 back trajectories per day).
Model results
As shown in Fig. 2, NH3 concentrations are significantly underestimated
by the CHIMERE model with a NMB of -75 % (see statistical results in
Table 2). This negative bias affects not only the intense peaks but also the
baseline concentrations. In their evaluation of the CALIOPE-EU modeling
system, Pay et al. (2012) reviewed the statistical results of various
regional models over Europe (during a whole year for most models). As our
study does not cover a whole year, statistical results are not directly
comparable, but figures still shed light on the relative performance of our
CHIMERE simulation. The negative bias in our study is in the range of those
reported by Pay et al. (2012) where NMB varied from -82 to -15 %. Our
RMSE (3.9 ppb) is among the best values reported by Pay et al. (2012)
(1.6 ppb for the CALIOPE-EU model and 7.6–10.6 ppb for the six other
models), as well as the correlation (0.42 vs. 0.05–0.56). Nevertheless, the
CHIMERE model dramatically fails to reproduce the strong spring and summer
episodes (and consequently the seasonal variation) during which negative
biases on daily concentrations can exceed a factor of 10, despite a monthly
distribution of emissions peaking between March and May (spring fertilizer
application).
The similar results obtained in the MOD and MOD-noIDF cases indicate that
most of the simulated NH3 originates from outside the region of Paris.
Concentration maps show that simulated NH3 concentrations closely
follow the spatial distribution of emissions, with maximum levels over
Brittany, northern France, and Benelux. Due to both dilution and deposition,
NH3 concentrations quickly decrease with distance from these source
regions. However, the simulated NH3 lifetime appears high enough to
allow imports over the region of Paris. As an illustration, highest
simulated concentrations in the city (4.5 ppb, the 29 April) result
from the advection of air masses from eastern Brittany and southwest during
the month of maximum emissions (according to monthly factors applied to
emissions).
Comparing observations and model results at the MONTSOURIS meteorological
station, a negative bias for temperature (-1.6 ∘C) and a positive
bias for RH (+5.9 % in absolute) (see Sect. S2 in the Supplement) is
noted. This favors the formation of NH4+ and thus decreases
gaseous NH3 in TNH3. However, correcting these errors in the
ISORROPIA model (i.e., replacing the simulated temperature and RH values with
measured values, without modifying TNH3, TNO3, and TS
concentrations) does not fill the gap with observations (the average
NH3 concentrations increasing by only 7 % on average). Errors may be
larger close to the deliquescence point where the influence of RH is
stronger. The deliquescent RH (DRH) of NH4NO3 and
(NH4)2SO4 at 298 K are 61.8 and 79.9 %, respectively
(Seinfeld and Pandis, 2006). A mixture of both salts will have a
DRH between these two extreme values. Focusing on days when RH ranges
between 60 and 80 % (i.e., close to the deliquescent point of the mixture),
the average NH3 increase is even lower (6 %). It reaches 14 % when
considering RH between 60 and 65 %. In any case, the impact remains
limited. As shown in Fig. 5, the fraction of NH3 in TNH3 simulated
by CHIMERE is highly variable, ranging from less than 5 % to about 90 %,
in contradiction with observations which show a clear gas-phase reservoir
during spring and summer (at around 60–100 %). The already mentioned
overestimation of SO42- in CHIMERE (see Sect. 4.1) may directly
reduce the amount of NH3 available in the gas phase. However, the bias
on TNH3 is only reduced to -56 % (against -76 % for NH3
alone), which indicates that only a minor part of the negative bias on
NH3 can be explained by an erroneous partitioning between both gas and
aerosol phases (including errors related to SO42-).
Although not likely the main NH3 source (see Sect. “Influence
of agricultural NH3”), the
traffic can also contribute to the NH3 urban background levels in
Paris. However, in the TNO-MP inventory, these traffic emissions are missing in
the Paris region (but not outside this region) (see Table S3 in the
Supplement), which may induce an underestimation of modeled NH3
concentrations. The contribution of traffic to ambient NH3 levels in
urban environments is highly variable from one city to another, as
illustrated by the NH3 / (NH3+NOx) emission molar ratios that
range from a few percent (Yao et al., 2013) to a few
tens of percent (Bishop et al., 2010),
which are due to differences in the vehicle fleet (Carslaw
and Rhys-Tyler, 2013). Several sensitivity tests were performed with added
NH3 traffic emissions, derived from the NOx traffic emissions with
NH3 / (NH3+NOx) conversion factors in the range of the values
given in the literature: 1, 6, 12, and 18 % (not shown). Such additional
emissions reduce the bias, but do not improve the correlation between model
and measurements. In particular, they induce a clear increase of NH3
concentrations during the morning and evening rush hours, which is not in
agreement with the observed diurnal profile. These results thus prevent us
from concluding on the importance of these traffic emissions on NH3
urban background levels.
A large part of the model errors probably arises from the representation of
NH3 air–surface exchanges (agricultural emissions and deposition) in
the CHIMERE model. This representation is by far too simplistic in several
respects: (i) the parameterization of NH3 dry deposition is
uni-directional and does not take into account the compensation with
emissions; (ii) the agricultural emissions are temporally disaggregated
based on monthly, day-of-the-week, and diurnal factors without taking into
account any environmental factor (e.g., air temperature, soil moisture,
agricultural practices) known to influence some NH3 emissions (e.g., the
volatilization of fertilizers). This likely explains the much lower
NH3-temperature correlation obtained in the model in comparison with
observations (r=0.52 against 0.72 in observations), as illustrated in Fig. 4. In light of our comparison, the parameterization of the NH3
emissions in CHIMERE cannot represent the high spatiotemporal variability
of NH3 concentrations, and in particular fails in reproducing the large
NH3 peak values observed during the campaign. Indeed, these emissions
result from very complex mechanisms in which numerous environmental
parameters are involved, including the amount of nitrogen fertilizers used
over the land; temperature, moisture, and pH of the soil; the amount of
soluble carbon; the soil disturbance and compaction; and fertilization methods
(Ma et al., 2010, and references therein). More
elaborated parameterizations of NH3 bi-directional fluxes have been
proposed to better handle emission and deposition processes in CTMs
(Massad et al., 2010; Zhang
et al., 2010; Pleim et al., 2013). Hamaoui-Laguel et
al. (2014) have simulated more realistic NH3 emissions over France
during the spring 2007 by combining the one-dimensional mechanistic model
VOLT'AIR (Garcia et al., 2011;
Génermont and Cellier, 1997) with agricultural practice and soil data.
They have shown a spatial variability of NH3 emissions mainly driven by
the soil pH and the types and rates of fertilization, while the temporal
variability was rather driven by meteorological conditions and fertilization
dates. Compared to the EMEP inventory (quite similar to TNO-MP for NH3
emissions), the emissions computed with the VOLT'AIR mechanism appear lower
over Brittany (in the west of France) and higher over the north of
France (around a factor of 2–3). This would suggest a possible
underestimation of agricultural NH3 emissions close to the Paris
region.
Dry deposition of NH3 and wet deposition of NH4+ represent
the two major sinks for NH3 and NH4+, respectively; the first
is dominant near emission sources whereas the second dominates at a
larger scale (Asman et al., 1998). Uncertainties
in the parameterization of both dry and wet deposition in the CHIMERE model
may also partly explain the NH3 underestimation. Results from the
MOD-nodep sensitivity test (with no NH3 dry deposition) allow assessing
an upper bound of uncertainties related to dry deposition. On average, more
than half of the NH3 reaching Paris is deposited in the MOD case, as
illustrated by the increase of NH3 concentrations by a factor of 2.2
when deposition is removed. The diurnal profile indicates that deposition in
CHIMERE more strongly affects nighttime concentrations, likely due to the
shallow boundary layer. Daytime concentrations are also affected but
approximately 2 times less than nighttime ones. Note that typical
deposition velocities simulated by CHIMERE are around 0.3 cm s-1,
although it can substantially vary in time and space. Despite the
unrealistic character of this sensitivity test (dry deposition being one of
the dominant NH3 sinks), this appears not sufficient to increase
concentrations towards observed ambient levels (NMB of -46 %). Thus,
deposition does not appear as the major source of error in the CHIMERE
simulated NH3.
Conclusions on ammonia
Our NH3 urban background measurements in Paris have highlighted several
intense episodes in late spring and early summer. These episodes occur
during anticyclonic conditions with high temperature, expected high
agricultural emissions and moderate winds enabling an accumulation of
NH3 and a subsequent advection over the city. We argued that the
observed NH3 seasonal pattern supports the idea of a NH3 day-to-day variability mainly driven by the agricultural source, in
association with the thermodynamic equilibrium controlling the gas–aerosol
partitioning.
CHIMERE simulations show a significant negative bias for NH3, both for
the baseline concentrations and the intense episodes. Errors in the
partitioning of TNH3 between the gas and aerosol phases (due to errors
in modeled SO42-, NO3-, or local meteorology) as well as
uncertainties for deposition can only explain a minor part of the bias.
Thus, the simulated NH3 concentrations appear mainly affected by
uncertainties in emissions, and in particular the lack of dynamical
treatment of agricultural emissions as a function of environmental factors
(temperature, etc.) in the CHIMERE model (the annual total emissions being
simply disaggregated with a monthly profile).
Nitric acid
Temporal variability
Daily concentrations and the diurnal profile of HNO3 are shown in Fig. 2. Over the whole period, the average HNO3 concentration is 0.25 ppb.
Several moderate episodes are observed in spring and early summer, with
daily concentrations up to 1.2 ppb at the beginning of July. This leads to a
seasonal pattern characterized by higher values in spring/summer compared to
autumn/winter. Such temporal variations are expected in urban environments
close to NOx emissions due to both the higher OH triggered HNO3
production in summer and the higher temperatures (as well as the lower RH)
that diminish its condensation into particulate NO3-. They are
also consistent with those found in other urban studies
(Cadle et al., 1982; Cadle, 1985, in
Warren, Michigan, USA; Solomon et al.,
1992, in Los Angeles, California, USA; Perrino et al., 2002,
in Roma, Italy).
In Paris, the highest HNO3 episodes are associated with high
temperatures and low-to-moderate wind speeds at the ground (see Fig. 7b). These conditions
increase the atmospheric stratification and the residence time of NOx
emissions over the agglomeration and allow for more efficient HNO3
formation via the NO2+OH reaction. This is confirmed by the fact that
many HNO3 peaks follow BC episodes; these episodes are often due to
stagnant conditions allowing the accumulation of the BC emitted by the
traffic.
Hourly concentrations of HNO3 at LHVP and wind speed, RH and
temperature during early June 2010 (left panel), and associated 48 h
back trajectories (one point every 24 h) colored by the day of arrival
(i.e., red is for 6 June).
This is illustrated during the first days of June in Fig. 8. 1 June is
characterized by low wind speed but cloudy conditions that decrease the
photooxidation rate of NOx. During the next 2 days, stronger winds
(above 3 m s-1) and increasing temperatures are observed, associated
with a moderate increase of HNO3 concentrations. A much larger increase
in HNO3 concentrations is observed on 4 and 5 June concomitantly
with high temperatures (up to 30 ∘C) and slow winds. Such stagnant
conditions during the night allow the accumulation of NO2, as shown by
the NO2 measurements at an AIRPARIF station located right next to the
LHVP site (not shown). In the early morning of 4 (5) of June,
NO2 concentrations reach 83 (110) ppb and fall below 20 ppb during
the afternoon. As for NH3, no additional HNO3 measurements are
available upwind of Paris, which prevents us from quantitatively assessing
the importance of local formation vs. imports. However, this specific situation of
early June supports the idea of a strong local formation of HNO3. Some
HNO3 is also probably (slowly) advected by northeasterly winds but the
strong photochemically driven diurnal variation observed during these days
(where concentrations reach 1.5 ppb in the afternoon) suggests that this
contribution is minor in comparison to the local formation. The episode ends
concomitantly with a significant decrease in temperature and an increase in
wind speed (thus favoring the dispersion).
The diurnal profile shows maximum HNO3 concentrations in the afternoon
at around 14:00–18:00 UTC (Fig. 2). On average, the ratio between daytime
and nighttime HNO3 concentrations is close to a factor of 2 (despite
the development of the convective boundary layer in the afternoon). A slight
decrease of HNO3 is found at around 06:00 UTC, which may be explained by
dew formation processes that allows the absorption of water-soluble gases
such as HNO3
(Mulawa et al.,
1986; Parmar et al., 2001; Pierson et al., 1988), although no data are
available to address this hypothesis.
Daily HNO3 / TNO3 ratios.
HNO3 accounts for 51 % of TNO3 on average (Fig. 9) but this
fraction appears highly variable. The lowest HNO3 / TNO3 ratios (a few %) are observed during cold days in mid-May when daily temperatures
fall below 8 ∘C (see Fig. S2 in the Supplement), while the highest
ratios occur during early summer, with values up to 96 %. The correlation
between the HNO3 / TNO3 ratio and the temperature is 0.82, which
illustrates the impact of temperature on the thermodynamic equilibrium.
Despite rather high temperatures, low ratios (below 40 %) are also
observed on specific periods during summer, particularly in August. Such a
pattern may be due to higher measurement uncertainties occurring for low
TNO3 concentrations, closer to the detection limit (roughly around 0.1 ppb for HNO3). In August, ratio values below 40 % indeed correspond
to HNO3 and TNO3 concentrations below 0.2 and 0.7 ppb,
respectively.
Model results
HNO3 concentrations are significantly overestimated by CHIMERE, with a
NMB of +195 % and an NRMSE of 320 %, especially at midday when the
bias can reach a factor of 4 (as illustrated by the diurnal profile in Fig. 2). The correlation is moderate (r=0.56) when considering hourly
concentrations but is slightly higher with daily values (r=0.68).
Several uncertainties may explain the discrepancies between observed and
simulated HNO3 concentrations: (i) uncertainties in NOx emissions
at both local and regional scales, (ii) uncertainties in the thermodynamic
equilibrium (i.e., the errors on either the other inorganic compounds or the
ISORROPIA model itself; Fountoukis and Nenes, 2007) that
determine the distribution between gas and aerosol phases, (iii) uncertainties in the OH concentrations that directly influence the
conversion of NO2 into HNO3, (iv) uncertainties in the HNO3
deposition, and (v) errors in the transport. At the European scale,
uncertainties in NOx emissions are estimated to be around 30 %
(Deguillaume et al., 2007; Konovalov et al.,
2006) and are thus much lower than the errors obtained for modeled
HNO3. Over the Paris agglomeration, NOx emissions from the TNO-MP
inventory used in our model have been evaluated during summer 2009 based
on aircraft measurements in the Paris plume, showing no significant bias
(Petetin et al., 2015). Dry deposition plays an
important role in the HNO3 budget, and corresponding parameterizations
incorporated in the CHIMERE model have been poorly evaluated so far. In
fact, an underestimated deposition rate in CHIMERE may partly explain the
positive bias on HNO3. In CHIMERE, HNO3 deposition velocities are
typically below 1.5 cm s-1, which appears on the lower end of the
values reported in the literature (Brook et al., 1999).
However, due to a lack of appropriate data, this hypothesis remains
difficult to assess. Finally, significant errors in the transport pattern
remain unlikely given the good correlations obtained on nitrates between the
observations and the model. The next subsections aim to investigate in more
details the uncertainties related to the simulated thermodynamic equilibrium
and OH radical.
Uncertainties associated with thermodynamic equilibrium
Bias and RMSE are much lower for TNO3 (NMB of +71 %, NRMSE of
121 %) than for HNO3, because the CHIMERE model overestimates the
HNO3 / TNO3 fraction (on average 68 % for the model against 51 %
observed from experimental data during the period with available
observations of NO3- and HNO3). Partitioning errors may
derive from uncertainties in the ISORROPIA thermodynamic model (e.g., model
formulation, chemical compounds included, activity coefficients treatment)
or in its input data. Apart from CHIMERE, the ISORROPIA model is used in
many other CTMs, including LOTOS-EUROS (Schaap
et al., 2008), REM-CALGRID (Stern, 2003), CAMx, FARM, or CMAQ.
It has been validated in various studies based on comparisons with
observations (Moya et al., 2001) or against other widely
used thermodynamic models (Nenes et
al., 1999; Carnevale et al., 2012). From these studies, several uncertainty
sources emerge: The hypothesis (used in ISORROPIA) of an instantaneous
equilibrium between gas and aerosol phases
(Aan de Brugh et al., 2012) is without
incidence for our study, since the CHIMERE model treats the evolution of
inorganic compounds concentrations through a dynamic approach (see Sect. 3.1). The absence of sodium, chloride, and other crustal species (Ca2+,
K+, Mg2+) in our simulations may also induce errors in the system
(Fountoukis and Nenes, 2007), but the contribution of this
crustal material remains low in the Paris region, about 5 % on average from
1 April to 10 September (with a percentile 95 at 13 %), as previously noted
by Bressi et al. (2013). This low contribution of crustal species is
confirmed by the ion balance obtained when considering only ammonium,
nitrate,
and sulfate: NH4+ vs. NO3-+ 2SO42- (all species
expressed in ppb) gives a slope of 1.01, a y intercept of -0.20 and a
correlation r2=0.97 (see Fig. S1 in the Supplement).
Therefore, errors in the modeled partitioning are most likely due to errors
in the other inorganic compounds involved in the HNO3–NO3-
equilibrium. In particular, the large negative bias on NH3 described in
Sect. 4.2 can potentially lead to an underestimation of the NH4NO3
formation and consequently to an overestimation of HNO3. A sensitivity
test has been performed for that purpose with the ISORROPIA model running
alone (i.e., not coupled with CHIMERE) fed by the concentrations previously
obtained with CHIMERE for inorganic species except for NH3 for which
measurements were taken into account. This approach changes HNO3 concentrations, for instance a decrease of 29 % in May. However, the
significant positive bias in HNO3 in summer persists (HNO3
concentrations decrease by only 11 % between June and August), mainly
because during summer NH4NO3 concentrations are very small and
HNO3 is the major TNO3 component due to the relatively high
temperatures.
Uncertainties associated with OH concentrations
Assuming that (i) the NO2+OH reaction is likely the dominant direct
homogeneous pathway for HNO3 formation during the summertime period,
(ii) a significant bias is observed for modeled TNO3, and (iii) the
maximum discrepancies between measurements and modeled HNO3 are found
during midday, uncertainties in simulated OH could explain a substantial
part of the errors on HNO3. Many studies have attempted to quantify
uncertainties in sources and sinks of OH, traditionally through the direct
comparison between observations and calculations from detailed chemistry
schemes (in box models) fed by ancillary observations of various parameters
(e.g., VOC, NOx, and O3 concentrations, photolysis rates). In such
exercises, uncertainties in daytime OH concentrations usually remain below a
factor of 2 (see Kanaya et al., 2007, for a review, where
the ratio of simulated to observed daytime OH concentrations ranges between
0.5 and 1.5). During summertime,
Michoud et al. (2012) observed a very
low overestimation (5 %) of simulated OH concentrations in Paris using the
Master Chemical Mechanism (MCM) chemistry scheme. However, these results
need to be taken as a lower end of OH uncertainties in CTMs where
constraints are applied on neither long-lived compounds nor photolysis
rates. This is especially true in an urban environment where concentration
gradients of compounds impacting on the OH budget are strong.
In order to assess the influence of OH on HNO3 formation, a sensitivity
test (hereafter designated by MOD-OHx0.5) has been performed over a period
of 35 days in June/early July by artificially reducing OH concentrations.
This is technically performed by decreasing by a factor of 2 the HOx
(HOx= OH+HO2+ RO2) formation yields (i.e., the
stoichiometric coefficient) in several (initiation) reactions, including the
photolytic destruction of O3, formaldehyde, acetaldehyde, glyoxal, and
methyl glyoxal. OH and HNO3 concentrations are then compared to the
reference MOD case in Fig. 10. On average, concentrations of OH and
HNO3 are reduced by -36 and -16 %, respectively. The changes in
NOx concentrations remain below 3 %, which means that only a minor
fraction of NOx is oxidized within Paris. These decreases are even
larger during midday when they reach -42 and -25 %, respectively. Over
midday, the bias between measured and modeled HNO3 is reduced to
+113 % (compared to +154 % in the MOD case). Uncertainties in the OH
radical may thus explain a significant part of the CHIMERE errors on
HNO3.
HNO3 and OH hourly concentrations (left panel) and diurnal
profiles (right panel) at the LHVP site.
Conclusions on HNO3
HNO3 concentrations experimentally determined in Paris show several
intense peaks in late spring and early summer that coincide with high air
temperatures and low to moderate wind speeds. The share between local
production and imports remains difficult to assess precisely, but local
HNO3 may represent a major source during some specific time-limited
episodes. However, uncertainties persist, and the CHIMERE errors are
unfortunately too high to help the investigation of HNO3 origin.
Indeed, the model largely overestimates measured HNO3 concentrations,
approximately by a factor 3, with the highest biases observed in the middle
of the day. The negative bias between measured and modeled NH3
explains a part of the poor model performance for HNO3 but still fails
to explain errors during summertime when TNO3 is mostly in the gas
phase. Uncertainties in NOx emissions are much lower than errors
obtained on HNO3and cannot explain the results of the model.
Uncertainties related to the dry deposition of HNO3 cannot be assessed
and could contribute to the discrepancies given by the model. Finally, a too
large NO2-to-HNO3 conversion through an overestimation of the OH
radical concentrations in CHIMERE could also contribute to the large
modeled overestimation of HNO3 formation. Indeed, due to the absence
of appropriate validation, uncertainties in simulated OH still remain high
in CHIMERE (probably more than a factor of 2) and reducing OH sources have
shown to lead to a significant decrease in OH and HNO3 concentrations,
in particular during the afternoon when NO2 photooxidation (as well as
the HNO3 bias) is at its maximum.
Aerosol nitrate formation
Results of the CHIMERE simulations
Fine particulate pollution with high NO3- contents in Paris
consists of intense (up to 16 µg m-3 in late spring) and
time-limited (a few days) episodes associated with continental wind regimes.
Very low levels of nitrate are observed during periods with marine (clean)
air masses and during summertime (due to volatilization). Despite the large
errors previously highlighted for both NH3 and HNO3, the CHIMERE
model provides relatively good results for nitrate with a NMB of +19 %
and a correlation of 0.81, but still with a large NRMSE (109 %). As
previously mentioned, in the framework of the PARTICULES campaign,
PM2.5 chemical constituents have also been measured at three rural
sites around the Paris region. Results have been analyzed in terms of local
and imported contributions by Petetin et al. (2014)
who found that imported sulfate was slightly underestimated by CHIMERE
(-17 %) while the local production of sulfate was overestimated
(+32 %), leading at the end to a moderate negative bias (-17 %). For
nitrates, they found a similar but larger error compensation identified
between imported and local production (bias of +63 and -109 %,
respectively), leading to a bias in Paris of +23 %. More details can be
found in Petetin et al. (2014) (e.g., statistical
results in Table 7).
It is worth noting that the positive bias highlighted here on the urban
background concentrations in Paris should partly originate from experimental
(negative) artifacts. The model may underestimate NO3- if the
experimental data are corrected for semi-volatile losses. The semi-volatile
particulate matter (SVPM) can be deduced from the difference between
TEOM-FDMS and TEOM PM2.5 concentrations. If we attribute all that SVPM
to NH4NO3, the bias between measured and modeled NO3-
becomes -48 %. This corresponds to an upper bound of the bias since SVPM contains
not only NH4NO3 but also semi-volatile organic aerosol (OA). Semi-volatile
OA may contribute the most to SVPM, as suggested by the higher correlation
of SVPM with OA in comparison with NH4NO3 (0.59 vs. 0.32).
In conclusion, the either positive or negative bias in simulated nitrates
and ammonium remains relatively small in comparison with the biases reported
previously for precursor species. Such a result is not intuitive and cannot
be trivially explained. An interesting point to illustrate is the possible
error compensation related to the saturation condition that needs to be
achieved to allow the formation of nitrates. This condition is defined as
(Ansari and Pandis, 1998)
TNO3TNH3-2TS>K,
with K being the equilibrium constant that depends on various parameters,
including temperature and RH. It is obvious here that the errors in
TNO3 and TNH3 can partly compensate each other. On average, the
left-hand term is 3.6 and 2.5 ppb2 based on observations and
simulation, respectively, which corresponds to a NMB of -31 %, thus much
lower than the NMB affecting the different species (+71, -56 and
+48 % for TNO3, TNH3, and TS). This result thus suggests that
the formation of nitrates is slightly less thermodynamically favored in the
model than in the reality, which would be consistent with a moderate
negative bias in nitrates. Due to possible artifacts, our data set does not
allow a complete assessment of the nitrate formation. It would be useful in
the near future to evaluate the CHIMERE model with artifact-free
measurements (for instance with aerosol mass spectrometer or aerosol
chemical speciation monitor).
Gas ratio and limiting species for nitrate formation
The GR has been proposed to assess which species among NH3
and HNO3 is the limiting reactant for NH4NO3 formation
(Ansari and Pandis, 1998). It is defined as follows (with
concentrations expressed in ppb):
GR=TNH3-2TSTNO3.
GR values above 1 indicate a regime mainly limited by HNO3 (i.e.,
NH3-rich regime) in which there is enough NH3 to neutralize both
sulfate and nitrate. Conversely, a GR between 0 and 1 indicates that there
is enough NH3 to neutralize sulfate but not nitrate, while negative
GR corresponds to a NH3-poor regime in which NH3 amounts are
insufficient to neutralize even sulfate. Nonlinear PM responses to
inorganic concentration changes are expected at GR near unity (Ansari and
Pandis, 1998).
Observed and modeled daily GR.
As shown on Fig. 11, daily GR measurements are available only from the end
of May (no NH3 observations before) until the beginning of September
(no aerosol observations after). During that period, experimentally
determined daily GR values are highly variable (ranging between 2.8 to 56.3)
but always remain above unity (12.6 on average), thus indicating that a
large amount of ammonia is available for neutralizing nitric acid.
Observed GR may be affected by negative artifacts of nitrate filter
measurements (Sect. 2.1). If we assume here that all the SVPM is
NH4NO3 (see Sect. 4.4.1), one can calculate an artifact-corrected
GR with both evaporated NH4+ and NO3- added to measured
TNH3 and TNO3, respectively. Compared to the previous GR, the
artifact-corrected GR is reduced to an average value of 7.3 (the median is
3.5), thus still well above 1. In addition, as noticeable amounts of OA are
expected to be included in the evaporated portion, this artifact-corrected
GR has to be considered as a lower estimate of the actual GR values. The
nitrate formation in Paris thus appears mainly limited by HNO3. Over
Europe, Pay et al. (2012) have also observed GR above
1 in several regions (e.g., Switzerland, Italy, Austria, inland regions of
Spain and Denmark; no data in France), but taking into account observations
restricted to regional background stations (i.e., enriched by agriculture
(NH3) emissions instead of traffic (NOx) emissions). In our study,
we show that such a NH3-rich regime is also observed within a large
megacity like Paris. Considering the high NOx emissions in the Paris
megacity, such a result is counterintuitive but may be explained (as
previously mentioned in Sect. 4.3.2) by a too slow NOx-to-HNO3
conversion rate compared to the efficient dispersive conditions.
In the CHIMERE model, the negative bias for TNH3 and the positive
biases for TNO3 and SO42- result in a significant
underestimation of modeled GR. On average, the model simulates a GR
slightly above unity (1.2). Daily values continuously alternate between the
NH3-rich and NH3-poor regimes with 48 % of simulated daily
values remaining below unity (47 % considering the whole data set). The
diurnal profile given by CHIMERE indicates that the GR regime changes within
a single day, the lowest GR values (below 1) being simulated at 12:00 UTC
(between the maximum TNO3 occurring at 08:00 UTC and the minimum
TNH3 simulated at 15:00 UTC). Therefore, due to significant errors in
gaseous precursors (and to a lesser extent in sulfate), the CHIMERE model
fails half of time at correctly simulating the HNO3-limited regime for
nitrate formation in Paris on a daily basis.
Sensitivity to perturbations
The GR value alone does not allow predicting the sensitivity of nitrate
formation with respect to changes in gas precursor concentrations. This is
due to the inability of GR to take into account both the need for the
atmosphere to be saturated with NH3 and HNO3 (which acts as a
threshold effect; see formula 6 in Sect. 4.4.1) and the influence of
temperature and RH. Additional information can be given by the sensitivity
coefficient Sx (Takahama et al., 2004) of nitrate
formation, defined as
Sx=ΔNO3NO3xΔx,
where ΔNO3 refers to the change in nitrate concentrations
obtained after a Δx change of the parameter x (e.g., temperature, RH,
TNH3, TNO3, or TS).
The ISORROPIA thermodynamic model is used here to compute the sensitivity
coefficient Sx as a function of various decreases (-10, -25, -50, and
-90 %) in TNH3 and TNO3 concentrations. This zero-dimension model
requires fives inputs – temperature, RH, and TNO3, TNH3, and TS
concentrations – and computes the gas–aerosol partitioning coefficient for
TNO3 and TNH3 compounds. Also note that the analysis is local, as
it is performed for the observed and simulated set of parameters at the
urban background site. Decreasing the concentration of TNO3 (or
TNH3) leads to a change in its partitioning between both the gas and
aerosol phases. This change not only depends on the concentration of the
family species which is altered but also on the value of all the other
parameters of the system. Thus, the CHIMERE errors in the different input
parameters propagate to the gas–aerosol partitioning coefficient, which can
potentially lead to an erroneous sensitivity of nitrates to a change in
TNO3 or TNH3. Calculations are performed for both the measurements
and the model; i.e., all inputs are taken from the observations and the
model, respectively, at the urban background site. In each case, the
(observed or simulated) concentrations of TNH3 or TNO3 are
decreased and the sensitivity coefficient is computed to quantify the impact
of this change on the nitrate concentrations. Sensitivity coefficient
results and corresponding GR are shown in Fig. 12.
Sensitivity coefficient Sx of nitrate formation due to
different changes (-10, -25, -50, and -90 %) in TNH3 and
TNO3 concentrations (left panel) and resulting GR (right panel) during
the period from 15 May to 10 September 2010. Experimental data (OBS) are in
black, modeled data (MOD) in blue. Box plots indicate 5th, 25th, 50th, 75th,
and 95th percentiles.
For the experimental data, we do observe a quite similar sensitivity of
nitrate formation for changes either in TNH3 or in TNO3
concentrations, with median sensitivity coefficients around 1 (i.e., close to
a linear response). Considering the high GR values (except for the -50 and
-90 % TNH3 cases that lead to negative GR), such a result with
similar responses to both precursors changes appears quite counterintuitive
in light of the above definition of GR. However, first, the GR approach
considers free NH3, while the sensitivities are calculated with respect
to total NH3. Second, as already mentioned, the formation of nitrates
requires the saturation condition to be achieved (see Eq. 6). So for
large GR values, but small TNO3 and free NH3 values, nitrate
formation will be sensitive to both TNO3 and TNH3. Note that the
equilibrium constant K (and thus the nitrate sensitivity) also depends on
temperature and RH; this is illustrated in Fig. S6 in the Supplement where
the same sensitivity tests are performed after decreasing the temperature by
10 ∘C and increasing the RH by 0.20 in observations, which leads
to STNO3 (still close to 1) much higher than STNH3 (below 0.5 for
-10 and -25 % of TNH3), in accordance with the NH3-rich regime
given by GR.
The CHIMERE nitrate response to TNO3 changes is approximately linear
(i.e., STNO3 close to 1), in reasonable agreement with observations.
However, the model highly overestimates the sensitivity to TNH3
changes, with median STNH3 up to 2.5 for moderate NH3 decreases
while observations show (as for TNO3 changes) a linear response to
TNH3 changes (i.e., STNH3 around 1). The model is able to reproduce
the observed response only when NO3- formation is severely
NH3-limited (negative GR) and when the aerosol nitrate formation is
prevented (which corresponds to the -90 % TNH3 case).
These results have serious implications on the use of the CHIMERE model for
emissions reduction scenarios. As TNH3 concentrations are closely
linked to NH3 emissions, they show that the benefits (in terms of fine
aerosol concentrations) of reducing these emissions would likely be
overestimated by the model, in particular for moderate reductions (below
-50 %). In addition, in terms of dynamical evaluation, changes in NH3
emissions in the next years may potentially degrade the CHIMERE performance
on the simulation of NH4NO3 in Paris if the issues raised here are
not addressed. This is an important conclusion for the use of the CHIMERE
model (in that configuration and input data).