Evaluation of WRF-CHIMERE coupled models for the 1 simulation of PM 2 . 5 in large East African urban conurbations . 2

Urban conurbations of East Africa are affected by harmful levels of air pollution. The paucity of local 12 air quality networks and the absence of capacity to forecast air quality make difficult to quantify the real level of 13 air pollution in this area. The chemistry-transport model CHIMERE has been coupled with the meteorological 14 model WRF and used to simulate hourly concentrations of Particulate Matter PM2.5 for three East African urban 15 conurbations: Addis Ababa in Ethiopia, Nairobi in Kenya, and Kampala in Uganda. Two existing emission 16 inventories were combined to test the performance of CHIMERE as an air quality tool for a target monthly period 17 of 2017 and the results compared against observed data from urban and rural sites. The results show that the 18 model is able to reproduce hourly and daily temporal variability of aerosol concentrations close to observations 19 both in urban and in rural environments. CHIMERE’s performance as a tool for managing air quality was also 20 assessed. The analysis demonstrated that despite the absence of high-resolution data and up-to-date biogenic and 21 anthropogenic emissions, the model was able to reproduce 66-99% of the daily PM2.5 exceedances above the 22 WHO 24-hour mean PM2.5 guideline (25 μg m) in the three cities. An analysis of the 24-hour mean levels of 23 PM2.5 was also carried out for 17 constituencies in the vicinity of Nairobi. This showed that 47% of the 24 constituencies in the area exhibited a low air quality index for PM2.5 in the unhealthy category for human health 25 exposing between 10000 to 30000 people/km to harmful level of air contamination. 26 27 28

Oceanic and Atmospheric Administration (Powers, 2017;Skamarock, 2008).   Initial and boundary conditions for the external coarse domain at 18×18 km were obtained from the NCEP FNL 146 (Final) Operational Global Analysis data (Wu, 2002). Boundary condition for the first (6×6 km) and second (2×2 147 km) nest domains were taken from the respective parent domains using the two-way-nesting approach. The 148 process enables the lateral conditions for the internal domains to be calculated from the outputs of the respective 149 parent domains at lower resolution at every time step of the simulation.

151
The Land use option chosen for the simulations was based on NOAH (Tewari, 2004) while the WRF Single-    The configuration adopted in this work uses initial and boundary conditions from the global three-dimensional

179
Horizontal and vertical diffusion is calculated using the approach suggested by Van Leer (1979) and the 180 thermodynamic equilibrium ISORROPIA model (Nenes, 1998) is used for the particle/gases partitioning of semi-181 volatile inorganic gases. The model permits calculation of the thermodynamical equilibrium between sulphates, 182 nitrates, ammonium, sodium, chloride and water dependent upon temperature and relative humidity data.

189
Secondary organic aerosols (SOAs), including biogenic and anthropogenic precursors, have been modelled in 190 CHIMERE as described by (Pun, 2006). SOAs formation is represented as a single-step oxidation of the 191 precursors, differentiating hydrophilic by hydrophobic SOAs in the partitioning formulation. Finally, biogenic 192 emissions were taken in account within CHIMERE using MEGAN model outputs as described by (Guenther, 193 2006               The coupled WRF-CHIMERE model was run at spatial resolutions of 18×18, 6×6 and 2×2 km for meteorology  Table 2), Kampala (n 5 in Table 2) and Nairobi Airport (n 7 in Table 2).

312
Initially, the validation of CHIMERE focused on Kenya for which hourly concentrations of PM2.5 were taken from 313 two different sites (roadside and rural) from the field sampling campaign described by Pope et al., (2018).

314
Secondly, the same configuration adopted for Kenya was used for Ethiopia and Uganda to test both the

320
The validation process was hindered by the highly variable quantity and quality of available meteorological data.

321
The majority of the weather observations are provided on a 3-hourly basis, with varying amounts of missing data.

322
Despite this the statistical evaluation for WRF has been performed comparing model and observations only when 323 the latter were available. We recall that the objective of this work aims to test the performances of a modelling system for the simulation of air quality concentrations for East Africa, updating and using the available input data 325 available and assessing the possible adoption of these tools for air quality policy making at this extent of the data.  (Table 3).

340
The results of the statistical analysis show that WRF is capable of reproducing the mean levels of surface  (Table 3).

383
The hourly surface temperature and relative humidity are shown in Figure 4 for the three ground weather stations 384 closest to the centre of the three cities: Addis-Bole (n.1 in Figure 3a), Kampala Station (n.5 in Figure 3b) and 385 Nairobi (n.7 in Figure 3c).

387
The temperature range observed at the three stations was between 9 and 27° C for the Addis Bole Station, 16 and 388 31° C for Kampala and 16 and 33° C for Nairobi. By inspection of Figure 4, it can be seen that the WRF model 389 is able to reproduce the main diurnal cycle of variation of temperature and relative humidity for the three ground  (Table 4).

532
Values inside this are (shown as red dashed lines in Figure 6) corresponds to an average model performance.

533
Finally values with MFE > 75% and -60% > MFB > +60% represent a poor representation by the model.       Figure 7 it can be seen that CHIMERE is able, in general, to reproduce the daily variation of PM2.5 599 across the simulated period at both sites.

601
The magnitude of the emissions adopted seems to be suitable both for the roadside area of Tom Mboya street and 602 for the rural background site of Nanyuki, with higher agreement shown by the latter. CHIMERE captures only 603 part of the daily peak observed in Tom Mboya Street with comparable magnitude but misrepresents some peaks.

604
In particular it models higher hourly peaks than those observed as previously mentioned in the MFB and MFE 605 analysis above (Figure 8).

607
The misrepresentation of some high peaks in Tom Mboya street is possibly due to a number of different reasons.   highlighting how many areas of the city showed low air quality indexes during the analysed period and the relative 690 population density exposed to PM2.5 pollution (Figure 11).

699
Daily concentrations of PM2.5 modelled by CHIMERE were compared with the number of exceedances of the 700 WHO limit (i.e. 25 µg m -3 ) observed during the simulated period. Figure 10 shows the daily average 701 concentrations for the three cities in the sampling sites used for the validation of the model. It can be seen that

702
Nairobi and Kampala have the highest number of exceedances from the WHO limits (24) followed by Addis

703
Ababa with only 6 observed exceedances. From Table 6 it can be seen that CHIMERE provides sufficient accuracy 704 to detect the exceedances of PM2.5 from the WHO limits. In particular, it was able to detect 67 % of the exceedance 705 for Addis Ababa with only two false positives, 91 % for Kampala and all of the exceedances for Nairobi without 706 any false positives. The Air Quality Index (AQI) represents the conversion of concentrations for fine particles such as PM2.5 to a 713 number on a scale from 0 to 500 (Table 7). The higher the AQI value, the greater the level of air pollution and the 714 greater the health concern. AQI values at or below 100 are generally thought of as satisfactory. When AQI values  Figure   724 11) contains the Tom Mboya Street sampling site (black spot in Figure 11) previously discussed where the WHO 725 limits for PM2.5 have been systematically exceeded during the analysed period. According to the SEDAC 726 population density data this area has population density between 15000 and 30000 people/km 2 exposed to AQI 727 between 151-200 corresponding to unhealthy category for human health.  Nevertheless, using existing data sets, CHIMERE has shown reliability in reproducing both hourly and daily levels 781 of PM2.5 with hourly values largely inside the range of reliability connected with mean fractional bias and error.

782
The model therefore can be adopted as a decision support tool for the management of air quality, as despite the