Algorithm improvements for Dutch OMI NO2 retrievals (towards v3.0) Folkert Boersma, Jos van Geffen, Bram Maasakkers, Henk Eskes, Jason Williams, and Pepijn Veefkind

1. Spectral fitting (lead: Jos van Geffen) OMI NO2 slant columns higher than GOME-2 & SCIAMACHY (similar to findings N. Krotkov) A difference of about 0.5x1015 molec/cm2 is expected due to difference in observation time So what else is different? Well a lot: fitting window, included reference spectra, polynomial degree, wavelength calibration approaches OMI GOME-2 SCIAMACHY fit window [nm] 405 – 465 425 – 450 426.5 – 451.5 secondary trace gases O3, H2O O3, H2O, O2-O2 pseudo-absorbers Ring degree of polynomial 5 3 2 slant column processing * KNMI / NASA BIRA-IASB *) Data assimilation, separation of tropospheric and stratospheric column, and conversion to total columns for all at KNMI.

A. Improve solar reference spectrum Update of high-resolution solar spectrum: - better representation of OMI ITF (measured) - better Ring-spectrum and I0-corrected reference spectra New solar spectrum improves effective λ-calibration With old parameterized ITF Now OMI slit function instead of parameterized broadened Gaussian High-resolution solar spectrum from Kurucz but through updated OMI slit function RMS of fit: -17% NO2 slant columns: -6% Other reference spectra otherwise identical to OMNO2A

B. Improve other reference spectra NO2: still Vandaele et al. [1998] but with new ITF O3: Bogumil et al. [2000] instead of obscure WMO-1975 H2O: HITRAN-2012 intead of HITRAN-2004 O2-O2: Now included! LQW: Now included! Stronger amplitude in new NO2 cross section! Now OMI slit function instead of parameterized broadened Gaussian High-resolution solar spectrum from Kurucz but through updated OMI slit function O2-O2: Thalman and Volkamer LQW: Pope and Fry RMS of fit: -30% NO2 slant columns: -11%

Including liquid water Fit results for cloudy pixels are not affected much when including liquid water absorption. Including liquid water absorption in the NO2 fit is a good idea. All these processes are interacting on one another Strong feature for clear water pixel !

Including LQW and O2-O2 improves the fit More realistic ozone columns Lowest RMS when including both LQW and O2-O2 All these processes are interacting on one another Including LQW and O2-O2 reduces NO2 SCDs by 0.3 1015 molec.cm-2

C. Improve λ-calibration window to map I to I0 Identify λ-calibration window that minimizes RMS/NO2 fit error RMS of fit: -32% NO2 slant columns: -13% Now OMI slit function instead of parameterized broadened Gaussian High-resolution solar spectrum from Kurucz but through updated OMI slit function O2-O2: Thalman and Volkamer LQW: Pope and Fry Best results for 409-428 nm Previously: 408-423 nm

Summary of the spectral fitting improvements Statistical NO2 fitting error reduces from 1.3 to 1.0 x 1015 Good consistency with QDOAS (thanks Isabelle De Smedt) Difference between 425-450 nm and OMI window 0.2-0.4 x 1015 Reduction of 1.0-1.3 x 1015 Now OMI slit function instead of parameterized broadened Gaussian High-resolution solar spectrum from Kurucz but through updated OMI slit function O2-O2: Thalman and Volkamer LQW: Pope and Fry

2. A new framework for DOMINO retrievals The 3-D Tracer Model is at the heart of DOMINO retrieval for: data assimilation to estimate stratospheric NO2 columns a priori NO2 profile shapes input to AMF temperature corrections DOMINO v1 and v2: based on TM4 with 3° × 2° DOMINO v3: based on TM5 on 1° × 1° Bram Maasakkers, M.Sc Motivation for re-coupling: TM4 no longer supported/developed TM5 v3.0 is a benchmarked (Huijnen et al., 2010) model Allows 1° × 1° Employs up-to-date emission inventories more representative for the OMI-era Differences in layering, emissions,

2. Re-coupling worked (3° × 2°) TM4-based TM5-based Differences in layering, emissions White areas: negatives With new TM5: 27% fewer negatives 20-30 Oct 2004

2. Re-coupling worked (3° × 2°) Lower emissions in TM5 TM4-based Differences in layering, emissions Fewer negatives in TM5-based retrieval 20-30 Oct 2004

2. Updating to TM5 at 1° × 1° We improve the retrieval by using TM5 a priori profiles at 1o×1o instead of 3o×2o Better resolved a priori profile shapes lead to a better understanding of pollution gradients observed from space TM4-based Differences in layering, emissions Madrid Madrid

2. Updating to TM5 at 1° × 1° TM4-based 20-30 Oct 2004 Differences in layering, emissions 20-30 Oct 2004 Average tropospheric NO2 2°×3°: 0.31 1015 molec./cm2 1°×1°: 0.31 1015 molec./cm2

2. Updating to TM5 at 1° × 1° Sharper contrast between urban and rural profile shapes will resolve some of the previously found underestimations over hotspot regions (Ma et al., 2013; Shaiganfar et al., 2011) TM4-based Lower AMFs over hotspots Differences in layering, emissions 3x higher surface NO2 for Barcelona in 1x1 vs. 3x2

2. DOMINO with TM5 at 1° × 1° vs. at 3° × 2° Barcelona +22% TM4-based Differences in layering, emissions Some of the urban-rural contrast is indeed a consequence of the coarse a priori profiles

3. Systematic error in O2-O2 cloud pressures Retrieved O2-O2 slant column depends on atmospheric temperature profile LUT for O2-O2 based on fixed AFGL mid-latitude summer T-profile O2-O2 slant columns require T-correction since ‘true’ T-profile may differ strongly from mls-profile Credits: Russ Dickerson, Johan de Haan, Bram Maasakkers Acknowledgment: Johan de Haan

3. Corrections to O2-O2 cloud pressures TM4-based Differences in layering, emissions

3. Impact on NO2 retrievals TM4-based Differences in layering, emissions Higher clouds  more screening  lower AMFs

3. Conclusions Improved NO2 slant columns are smaller by 1.0-1.3 1015 molec.cm-2 with 30% lower fitting residuals Coupled DOMINO to TM5 v3: 27% fewer negatives Improved resolution for a priori profile leads to increases over hotspots (+20%) and stronger contrast urban-rural Temperature-correction for cloud pressures relevant over polluted areas (TROP)OMI NO2 team from 2014 onwards:

Differences between TM4 and TM5 Emissions TM4: POET 1997

Integrate over pressure, not altitude Express number density O2 as a function of pressure and temperature Retrieved slant column will depend on 1/T, but LUT slant columns do not…