PMCAMx-2015 evaluation over Europe against 1 AERONET and MODIS aerosol 2 optical depth measurements

1 Institute of Chemical Engineering Sciences (ICE-HT/FORTH), Platani, P.O. Box 1414, Patras, 26504, 7 Greece 8 2 Department of Chemical Engineering, University of Patras, University Hill, Patras, 26504, Greece 9 3 Department of Environment, University of the Aegean, University Hill, Mytilene, 81100, Greece 10 4 Department of Chemical Engineering, Carnegie Mellon University, Pittsburgh, PA 15213, USA 11 12 Correspondence to: Spyros N. Pandis (spyros@chemeng.upatras.gr) 13

almost no effect on the estimated single scattering albedo (SSA) but the aerosol water content  Geosci. Model Dev. Discuss., doi:10.5194/gmd-2015-225, 2016 Manuscript under review for journal Geosci. Model Dev. Published: 18 January 2016 c Author(s) 2016. CC-BY 3.0 License. of stations in each region is shown in Table 1. Some AERONET stations in the domain of interest did 1 not have available Level 2 AOD data for the period of interest while all data from three stations 2 (OHP_OBSERVATOIRE in South France, FORTH_CRETE in Crete, Greece, and ATHENS_NOA in 3 Athens, Greece) have been excluded after the dust coarse particle rejection filtering.

4
The polar-orbiting MODIS monitors global aerosol properties from two satellites: Terra and 5 Aqua (Salomonson et al., 1989). MODIS employs 36 channels from 0.412 to 14.2 μm, has a wide swath 6 of 2,330 km, and observes every part of the globe at least once daily. The default resolution for aerosol 7 retrieval is 10x10 km 2 (Levy et al., 2009). Each data set retrieved by MODIS is associated with a 8 Quality Assurance Confidence (QAC) flag which ranges from 0 (no confidence) to 3 (highest 9 confidence). For increased spatial coverage we use the union of Terra and Aqua MODIS AOD 10 retrievals with QAC ≥ 1. We employ the MODIS Level 2 Collection 5.1 aerosol datasets. AOD 11 retrievals are provided at seven wavelengths (470,550,660,870,1,200, 1,600, 2,100 nm) over water 12 surface and four wavelengths (470, 550, 660, 2,100 nm) over land. In this study we focus on the 550 nm 13 values. Figure 2 presents the geographical distribution of the available MODIS AOD measurements 14 during the period of interest (1-29 May 2008) over Europe. The average number of retrievals is 12±9.

15
The maximum number of retrievals is 65 in areas in the North Atlantic.

16
Dust emissions from the Sahara are not included in the PMCAMx emissions used here and the 17 focus of this study is on periods and regions in which Saharan dust does not contribute significantly to 18 the AOD. To exclude periods with high dust levels and to focus on the rest of the anthropogenic and 19 biogenic aerosol components, MODIS AODs are filtered. Over water we employ the dust coarse 20 particle rejection filter of Barnaba and Gobbi (2004). According to this filter, AOD values greater than 21 0.3 also corresponding to coarse mode fraction higher than 0.3 are assumed to be dust-influenced 22 periods. Over land we only use the AOD values which correspond to Angstrom Exponent values exceeding 0. 9 (Schuster et al., 2006). The above filters discard 16% of MODIS AOD values over land 1 and 0.4% over water.

2
The evaluation of the MODIS AODs at 550 nm for the land algorithm was performed following 3 the approach of Remer et al. (2005) and Levy et al. (2007b). The collocated data were sorted according 4 to the AERONET measurements. The resulting data were partitioned into bins of 100 points and then 5 averaged. At higher optical depths since the data became sparser we employed 25 points for each bin.

6
The regression results of the collocated AODs, prior to binning had a slope of 1.05. 73% of the 8,331 with the expected accuracy. The highest quality flag QAC = 3 provides the closest match but including the QAC = 2 and 1 retrievals results in only a minor reduction of accuracy while increasing significantly 11 the size of the dataset (Table S2).
Previous studies have shown that MODIS AOD retrievals have an expected error of ±(0.05 + 13 0.15AOD AERONET ) over land and ±(0.03 + 0.05AOD AERONET ) over water (Chu et al., 2002;Remer et al., 14 2005;Levy et al., 2007aLevy et al., ,b, 2010Anderson et al., 2013). Table S3 summarizes the values of the      Table 2. 94% of the monthly mean AOD values fall inside the expected MODIS error envelope over land (Fig. 5a). Over the whole domain 13 the PMCAMx monthly mean AODs have a mean error of 0.05 and a fractional bias of -16% compared 14 to the MODIS monthly mean AODs (Table 2). measurement periods. The PMCAMx monthly mean AODs had a mean error of 0.03 and a fractional 18 bias of 4% compared to the AERONET monthly mean AODs (Table 1) (Table 3).

9
Russia, Belarus, and Ukraine: PMCAMx reproduces well (0.14 predicted versus 0.15 measured) the average AOD observations at the 5 AERONET stations in this region (2 in West Russia, 1 in Belarus, 1 11 in Ukraine, and 1 in Crimea) ( Table 1). The model has a similar good performance against the MODIS retrievals (0.12 predicted versus 0.13 retrieved) ( Table 2). As a result, the monthly mean PMCAMx measured in the station of Chibolton). The monthly mean PMCAMx AODs have a mean error of 0.04 23 compared to MODIS with a small tendency towards overprediction (14%). 90% of the monthly mean May 2008 and there is evidence that they may be underpredicted. However, errors in relative humidity 23 or cloud contamination could be also responsible for these discrepancies (Anderson et al., 2013).
In the first test the absolute humidity was increased uniformly by 5%, while maintaining the 23 maximum relative humidity in cloud-free regions at 99%. The PMCAMx monthly mean AOD increased 24 on average by 13% (Fig. S2). The increases ranged from 7% in Turkey and Northern Africa to 31% in the North Atlantic. This AOD change can explain a significant part of the base case discrepancies which 1 cause a fractional error of PMCAMx 22% versus AERONET and 33% versus MODIS.

2
In another test the diameter of all particles was increased by 20%. 72% of the PMCAMx 3 monthly mean AOD values changed by less than 0.01. The average increase of the monthly mean AOD 4 was 1% (ranging from 0.3% in the Black Sea to 4% in the UK and Ireland).

5
In a third sensitivity test we assumed that BC was always externally mixed with the other 6 components in each size range, forming pure BC spheres. 73% of the PMCAMx monthly mean AOD 7 values changed by less than 0.01 in this test. The average change of the monthly mean AOD was 8 negligible (< 0.5%).  The details of the evaluation results differ, as expected, depending on the use of either the datasets, but also to the MODIS AOD retrieval uncertainties. The major conclusion is that PMCAMx 23 can reproduce the observed AODs for this period with little bias (-16% for MODIS and +4% for sites and airborne measurements from several flights over central and northern Europe.

4
The AOD performance of PMCAMx against the MODIS retrievals is excellent (absolute 5 fractional bias less than 15% and fractional error less than 35%) in the Iberian Peninsula, UK/Ireland, 6 central Europe, Russia-Belarus-Ukraine, Turkey-northern Africa. It is good (absolute fractional bias less 7 than 25% and fractional error less than 35%) in East Europe, the Balkans, and over the Mediterranean, 8 the North Atlantic, and the Black Sea. Finally, its performance is average (absolute fractional bias less 9 than 50% and fractional error less than 55%) in the relatively clean area of North Europe and the South 10 Atlantic. The performance is more or less similar against AERONET with the exception of a few areas 11 with only one or two AERONET stations. The average performance against the AERONET 12 measurements is considered using the above criteria excellent and against MODIS it is the borderline 13 between good and excellent.
14 The above PMCAMx performance suggests that overall the model does a good job in 15 reproducing the fine aerosol sulfate, organics, nitrate, and sea-salt levels over Europe during the 16 evaluation period. Its major weaknesses appear to be potential overpredictions of sulfate and/or organics 17 over North and East Europe, underprediction of sulfate over the Balkans, and underprediction of fine 18 sodium chloride, sulfates, or organics in the southern Mediterranean and South Atlantic. However, these 19 discrepancies are quite sensitive to the relative humidity fields predicted by WRF. In a sensitivity test 20 the average predicted AOD increased by 13% (ranging from 7 to 31% depending on the area) for a 21 uniform 5% change in RH. On the other hand, the details of the fine PM size distribution and the black   Atmosph. Res., 114-115, 38-69, 2012. Murphy, B.N. and Pandis, S. N.: Simulating the formation of semivolatile primary and secondary 23 organic aerosol in a regional chemical transport model, Environ. Scienc. & Technol., 43, 4722-24 4728, 2009. multiphase multicomponent inorganic aerosols, Aquat. Geochem., 4, 123-152, 1998.