Spatial Disaggregation of Particulate Matter Emission Inventory in the Metropolitan Area of Aburrá Valley for Air Quality Modelling

13 In this paper a local emission inventory for PM10 and PM2.5 is presented that has been developed 14 using a top-down spatial disaggregation of the official emission inventory for the Metropolitan 15 Area of the Aburrá Valley in Colombia. The local emission inventory was evaluated using the 16 LOTOS-EUROS Chemical Transport Model in a high-resolution simulation, and compared with 17 the global emission inventory EDGAR. A detailed analysis of the model using the local emission 18 inventorywas performed. The results showed a considerable improvement inmodel performance 19 when the local emission inventory was used in comparison to the global emission inventory. 20


Temporal disaggregation
To be able to use the AMVA emission information in a simulation model, it is necessary to expand it with a temporal 117 profile. The temporal profile distributes a yearly total emission over seasons (months), days (work days or weekends), 118 and hours of the day. For road-traffic emissions, a daily profile following the traffic density for a working day in 119 the metropolitan area was taken from (UPB and AMVA, 2017). This profile has an hourly resolution, as shown in Figure 4. Industrial emissions can have a strong variability within a day, but since no detailed information is available,

123
Apart from a temporal profile, a simlation model also requires a spatial disaggregation. The result is a map of 124 emission intensities that shows spatial differences in emission strengths; the total sum should equal the inventory data.

125
The AMVA inventory was disaggregated over the where , is the road segment in the grid cell , is the total length of road segments in each grid cell, and is the 141 total number of grid cells. Figure 5 shows the simplified road network map used for the on-road spatial disaggregation.

142
The point-source emissions were distributed on the grid using their known location, obtained from the official 143 emissions inventory (UPB and AMVA, 2017).  Figure 6: LOTOS-EUROS model nested domains for Metropolitan Area of Aburrá Valley assessment. 166 The Sistema de Alerta Temprana del Valle de Aburrá (SIATA, www.siata.gov.co) is a sensor network that pro-  The PM 2.5 and PM 10 equipment consists of Met One Instruments BAM-1020 and BAM-1022 monitors using a beta 171 ray attenuation method to measure airborne PM concentration levels (Hoyos et al., 2019). In this study, the PM 10 and 172 PM 2.5 stations selected for validation should have at least 70% data coverage for the periods of interest.

173
Three different metrics are used to compare observations from ground stations with simulations of the LOTOS-

175
• The mean fractional bias (MFB) normalizes the bias between observation and simulations using division by the average of the model and observation before taking the sample mean (Boylan and Russell, 2006): where is the number of observations, is the model simulation output, and is the observation.

178
• The root mean square error (RMSE) represents the sample standard deviation of the differences between predicted values and observed values (Zhang, Roussel, Boniface, Cuong Ha, Frappart, Darrozes, Baup and Calvet, 2017): The RMSE penalizes a high variance as it gives errors with larger absolute values more weight than errors with 179 smaller absolute values (Chai and Draxler, 2014).

180
• The last metric used is the correlation factor (CF), which shows how the values from one data set (simulations) relate to the value of a second data set (observations). The correlation coefficient is calculated following: where the overline denotes a sample mean over the elements.

182
Using the disaggregation methodology described in Section 2.3 with the data presented in Section 2.1, a local  The AMVA inventory has a much higher spatial resolution (1x1 km) than EDGAR (10x10 km). Although this does 198 not necessarily means an improvement in accuracy, a higher resolution does allow a more detailed spatial representation The disaggregated AMVA inventory provides a more detailed representation of the city's traffic network. In this 207 inventory, it was possible to differentiate the main vehicular artery that traverses the valley from south to north-east.

208
The largest share of emissions is concentrated in the center of the city of Medellín (largest urban hub in the metropolitan 209 area), and along its Southern borders with Envigado, Sabaneta, and Itagui (see Figure 5), a location characterized by

213
In terms of total emissions for the region, the EDGAR inventory estimates a total PM 2.5 emission from road traffic 214 that is approximately 18 times lower than the estimate in the by AMVA inventory. The lower total suggest that the upstream and local emissions inventories (Gonzalez et al., 2017).  Valley, which is actually mainly a residential or even rural area.

266
While both LE-AMVA and LE-EDGAR are able to capture the daily cycle, the correlation factors CF shown in Figure   267 13 are lower than 0.5 what is usually declared as needed for good correlation (Chang and Hanna, 2004;Shaocai, Brian, 268 Robin, Shao-Hang and E., 2006;Boylan and Russell, 2006). The low CF values arise because the representation of the     Valley at a high resolution of 1 km × 1 km.

299
The emission distribution factors for traffic emissions were calculated using a top-down methodology based on the 300 road density, since actual traffic intensities are hardly available. For industrial point sources, actual locations are used.

301
The higher resolution has led to a more detailed spatial representation of emissions. Despite the simple methodology, 302 the AMVA inventory represents accurately the known hot-spots and high emissions regions for both on-road and point-

312
The results highlight the importance of detailed emissions information in regions where the global inventories are 313 not accurate, as is the case for Colombia. Even simple methodologies as the one employed here could strengthen the 314 capacity to represent and understand the dynamical behaviour of air pollution in complex cities.

315
An interesting future work, which is outside the scope of this paper, would be to implement data assimilation 316 techniques to improve the model performance and correct model uncertainties in the emissions inventory and mete-317 orological fields. The new high-resolution disaggregated AMVA inventory will support ongoing efforts to quantify 318 exposure to air pollution in Medellín and surrounding area.  s00703-003-0070-7.