Inﬂuence of satellite-derived photolysis rates and NO x emissions on Texas ozone modeling

Abstract. Uncertain photolysis rates and emission inventory impair the accuracy of state-level ozone (O3) regulatory modeling. Past studies have separately used satellite-observed clouds to correct the model-predicted photolysis rates, or satellite-constrained top-down NOx emissions to identify and reduce uncertainties in bottom-up NOx emissions. However, the joint application of multiple satellite-derived model inputs to improve O3 state implementation plan (SIP) modeling has rarely been explored. In this study, Geostationary Operational Environmental Satellite (GOES) observations of clouds are applied to derive the photolysis rates, replacing those used in Texas SIP modeling. This changes modeled O3 concentrations by up to 80 ppb and improves O3 simulations by reducing modeled normalized mean bias (NMB) and normalized mean error (NME) by up to 0.1. A sector-based discrete Kalman filter (DKF) inversion approach is incorporated with the Comprehensive Air Quality Model with extensions (CAMx)–decoupled direct method (DDM) model to adjust Texas NOx emissions using a high-resolution Ozone Monitoring Instrument (OMI) NO2 product. The discrepancy between OMI and CAMx NO2 vertical column densities (VCDs) is further reduced by increasing modeled NOx lifetime and adding an artificial amount of NO2 in the upper troposphere. The region-based DKF inversion suggests increasing NOx emissions by 10–50% in most regions, deteriorating the model performance in predicting ground NO2 and O3, while the sector-based DKF inversion tends to scale down area and nonroad NOx emissions by 50%, leading to a 2–5 ppb decrease in ground 8 h O3 predictions. Model performance in simulating ground NO2 and O3 are improved using sector-based inversion-constrained NOx emissions, with 0.25 and 0.04 reductions in NMBs and 0.13 and 0.04 reductions in NMEs, respectively. Using both GOES-derived photolysis rates and OMI-constrained NOx emissions together reduces modeled NMB and NME by 0.05, increases the model correlation with ground measurement in O3 simulations, and makes O3 more sensitive to NOx emissions in the O3 non-attainment areas.


CAMx modeled profile-based OMI retrieval 4
The OMI-retrieved tropospheric NO 2 vertical column density (VCD) used in this study is 5 calculated via Eq. (S1) (Bucsela et al., 2013), In Eq. (S3), CAMx vci represents the CAMx modeled NO 2 VCD at each model layer (i), and 2 CAMx vctot is the CAMx modeled total tropospheric VCD. The AMF which contains the a priori 3 GEOS-Chem modeled profile is now merged with the CAMx modeled VCD. 4 The way of removing the a priori GEOS-Chem modeled profile via applying AKs is carried 5 out by generating the CAMx modeled profile-based AMF CAMx as shown in Eq. (S4), using AMF CAMx to replace AMF GEOSChem in Eq. (S1) and then creating a CAMx modeled profile-8 based OMI tropospheric NO 2 VCD (V c(CAMx) ). However, this procedure can only be realized in 9 the inversion process by comparing the AKs applied CAMx VCD ( 2 predicted NO C ) and original OMI 10 retrieved VCD (V c(GEOSChem) ).
When applying 2 predicted NO C to the direct scaling method (Martin et al., 2003;Tang et al., 2013) in Eq.
where all AMF GEOSChem are removed, and the original V c(GEOSChem) becomes V c(CAMx).

11
There is an alternative way to create V c(CAMx) instead of applying OMI AKs, which is to use 12 the CAMx modeled profile directly in the OMI retrieval process. In this case, the error of interpolating AKs values into the CAMx layer can be avoided, and the CAMx profile-based OMI 1 retrieval can be calculated directly and viewed. In this study, we have created a CAMx profile-2 based OMI product that uses a CAMx profile in the retrieval process for the AMF calculation 3 and planned to use this new OMI retrieval product at the beginning for the inversion study.

4
However, we find that the CAMx profile-based OMI overestimates NO 2 VCD by approximately 5 30% compared to the original OMI retrieval using a GEOS-Chem profile (Fig. S1, right). We  This may reduce the AMF values (Eq. S4) because instrument sensitivity related SW is much 10 higher in the upper troposphere than in the boundary layer and thus increases the total retrieval 11 quantity. Unfortunately, there are no corresponding measurement data available to validate the 12 CAMx and GEOS-Chem profiles in Fig. (S1), but similar bias has been found in the CAMx 13 modeled NO 2 profile compared to the DC-8 and P-3 aircraft NO 2 measurements (Fig. 8)

18
The NASA OMI high resolution product used in this study shows reduced NO 2 in the rural areas,

19
while enhanced NO 2 in the urban, compared to the NASA standard retrieval, version 2 (Tang et 20 al., 2013); however, it still shows more smeared-out pattern than the CAMx modeled NO 2 VCD 21 (Fig. S3a). The CAMx simulations with the a priori NO x emission inventory created in Tang et al.

22
(2013) shows larger NO 2 VCD in the cities, while lower NO 2 VCD in the rural places than OMI 23 ( Fig. S3b). Reducing the reaction rate constant of the reaction OH + NO 2 by 25% in the CB05 1 chemical mechanism increases the NO x lifetime, makes more NO x transport to the rural, and 2 enhances around 3% NO 2 VCD on average in the inversion region, but the impact is small (  where the adjustments in the aviation sector are relatively more sensitive to the emission 22 uncertainty, ranging from 3.9 to 4.6 when emission uncertainty increases from 50% to 100%. It 23 seems to offset against area and nonroad sector which the scaling factors reduce from 0.6 to 0.5 1 (Fig. S2 middle). However, the inversion becomes insensitive to the emission uncertainties in the 2 sector-based DKF inversion case II when merging aviation into the area and nonroad sector ( Fig.   3 S2 bottom), indicating the DKF inversion in case II is more stable and less responsive to the 4 uncertainty matrices than that in case I.  contributors to all five VOC species (Table S1).

12
Since all modeled ETH, ETHA, ISOP, TOL, and XYL are from the primary emissions, a direct 13 scaling (DS) inversion method that adjusts VOC emissions based on the ratios between modeled 14 VOC and PAMS measured VOC is applied here. The inversion is conducted on a regional basis, 15 which means the scaling factor calculated from the measurement data in one region only applies 16 to adjust the emissions in that region. Therefore, due to the availability of observations, the five 17 chosen VOC species emissions are adjusted in only three regions, DFW, HGB, and BPA.

18
The scaling factors generated from the inversions vary significantly in different regions

19
( The temporal variations of the five VOC species (Fig.S4) show that the discrepancies between 10 observed VOCs and the a priori modeled VOCs are significantly reduced by using the a 11 posteriori emissions. The inverted ETHA emission improves modeled R 2 and reduces modeled 12 NMB and NME by 0.5 and 0.1, respectively (Table S3). The inversed ETH shows increased R 2 13 and 0.13 reduced NMB, but no improvement in the modeled NME against ground measurement 14 (Table S3); however, it shows 0.4 reductions in both modeled NMB and NME against P-3 15 measured data (Table S4). The inverted ISOP emissions reduce approximately 20% NMB and 16 NME in ground ISOP simulation (Table S3), but no improvements are found compared against 17 aircraft measurement (Table S4). The modeled NMB in the inversed TOL is reduced by 18 approximately 0.4 (Table S3) compared against PAMS and 0.13 compared against P-3 (Table   19 S4), while the modeled NME has not been improved. The inversed XYL shows increased R 2 and 20 around 0.2 reduced modeled NMB and NME compared to ground measurement (Table S3) and 21 0.02 reduced modeled NMB and NME compared to aircraft measurement (Table S4). However, 22 no improvements are found in the model performance of simulating ground-level NO 2 (Table S5) ,

23
and there is a slight decreasing, around 0.01, of modeled NMB and NME in ground-level O 3 1 simulations using the inverted VOC emissions (Table S6).       Figure S4. Comparisons of monthly averaged daily variation between observed (black) and 3 modeled VOC species using the a priori (red) and the a posteriori (blue) VOC emission 4 inventory over all monitoring sites.