Impact of 3D Cloud Structures on the Atmospheric Trace Gas Products from UV-VIS Sounders - PowerPoint Presentation

1. Impact of 3D Cloud Structures on the Atmospheric Trace Gas Products from UV-VIS SoundersvEGU2021 Session AS3.16(28.04.2021)Huan Yu1 (huan.yu@aeronomie.be), Claudia Emde2, Arve Kylling3, Michel Van Roozendael1, Bernhard Mayer2, Kerstin Stebel3,Ben Veihelmann41Royal Belgian Institute for Space Aeronomy (BIRA), Brussels, Belgium2Meteorological Institute, Ludwig-Maximilians-University (LMU), Munich, Germany3Norwegian Institute for Air Research (NILU), Kjeller, Norway4ESA-ESTEC, Noordwijk, the Netherlands1

2. Operational retrievals of tropospheric trace gases from space-borne instruments are based on 1D radiative transfer, and neglect the effects of 3D cloud features.3D effects become significant when the spatial resolution of the instruments is close to or better than the dimensions of cloud features. Therefore, measurements from S5P, S5 and S4 sensors are strongly influenced by 3D cloud effects.ApproachMonte Carlo radiative transfer (MYSTIC-ALIS): simulation of spectra for realistic 3D model atmospheresApplication of NO2 retrieval algorithm on simulated data: estimation of retrieval error due to 3D cloud scattering Sensitivity study to investigate which parameters can be extracted from measurement information to identify pixels with possible 3D biasIdentification of 3D effect on NO2 retrieval from real satellite observationsMitigation strategy for such pixelsIntroduction2

3. 3D radiative transfer model MYSTIC - Monte carlo code for the phYSically correct Tracing of photons In Cloudy atmospheres (Mayer 2009, www.libradtran.org) Polarized radiative transfer (Emde et al., 2010)VROOM: variance reduction methods (Buras and Mayer, 2011) ⇒ radiance calculations for strongly peaked scattering phase functions ALIS method (Emde et al., 2011) ⇒ very efficient high spectral resolution calculations complex topography (Mayer et al., 2010) spherical geometry (Emde and Mayer, 2007) box air mass factors in 3D domain NO2 retrieval algorithmDOAS approach: 1) SCD from DOAS spectral fit; 2) SCD/VCD conversion using AMF. AMF calculation uses the independent pixel approximation, which accounts for cloud scattering using cloud fraction, cloud pressure and cloud albedo from two cloud retrieval algorithms: O2-O2 (Acarreta, etl al, 2004) and O2 A-band (FRESCO, Koelemeijer et al., 2001) In this study, the error of the NO2 retrieval is mainly from the simplified cloud correction used in the AMF calculation (~20%) and from the effect of 3D cloud structure.Satellite observationsTROPOMI NO2 productVIIRS cloud product (reflectance, cloud top height, cloud optical thickness, cloud shadow flag, etc.)3

4. Clear-sky pixels in vicinity of box-cloudsBase case settings:CloudCloud base at 2km altitudeCloud top at 3km altitudeCloud droplet effective radius 10μmOptical properties from Mie calculationsGeometryNadir observation geometry1×1km2 square field-of-viewSolar zenith angle: 50ºOthersNO2 profiles: a polluted and a clean profile Surface albedo: 0.05No aerosolshadowin-scattering4AMF retrieval bias over clear scene is mainly due to 3D clouds.The bias is more significant for polluted profile than for clean profileCloudyxxClear

5. Sensitivity study – retrieval bias 5NO2 AMF retrieval bias (with polluted NO2 profile) as a function of distance from cloud edge for various SZAsLargest AMF bias in the clear scene as a function of solar zenith angle / surface albedo / cloud optical thickness / cloud geometrical thickness / cloud bottom heightThe significant retrieval bias is mainly from the pixels in the cloud shadow.In the cloud shadow, the bias strongly depends on SZA, surface albedo, COT from neighbouring pixels.Difference between AMF corrected by O2-O2 and FRESCO cloud is within ~20%.

6. Sensitivity study – influence of spatial resolution6Binning the spectrum by a factor of 3/5/7/9/11/13/15  the size of footprint = 3/5/7/9/11/13/15 kmThe bias as a function of spatial resolutionThe dependence of the AMF bias is linked to the change of cloud shadow fraction.clearPartly cloudycloudy

7. Dependence on cloud shadow fraction(CSF)Size of satellite pixel = sShadow area (in satellite footprint) = x0-s/2 + h·tan(θ0)Shadow fraction = (x0-s/2 + h·tan(θ0)) / sCloud shadow fraction is calculated based on cloud height, solar zenith angle, size of satellite pixel and distance from the edge of cloud.Independent pixel approximation can be used to estimate AMF retrieval bias and AMF over the cloud shadow band:AMFbias=CF*AMFbiascloud+CSF*AMFbiasshadowAMF=(1-CF-CSF)*AMFclear+CF*AMFcloud+CSF*AMFshadowxyθ0hshadowcloudx00x0-s/2x0-s/27Pixels with partly clear, partly cloudy and partly clear but in the cloud shadow

8. Dependence on slant cloud optical thicknessxyθ0shadowcloud  Slant cloud optical thickness(SCOT): integrated abundance of cloud optical thickness from the Sun through the atmosphere to the ground SCOT =  Bias of NO2 AMF retrieval (the largest bias for each scenario) as a function of slant COT8

9. Observational evidence of 3D cloud effects – case study9Zoom-in VIIRS RGB image with TROPOMI footprintAverage of TROPOMI NO2 VCD (using FRESCO cloud correction) for selected region for each rowNO2 in the shadow is ~20% lower than NO2 from neighbouring pixels In real observation, “true” NO2 VCD is unknown, the error of NO2 retrieval is from many sources.Large retrieval bias is from large SZA / thick and high clouds from neighbouring pixels.cloudyshadowclear

10. Observational evidence of 3D cloud effects – statistical analysis10Grouping of NO2 biases: NO2(center) – NO2(neighbour)Data: TROPOMI: NO2 VCD (quality value>0.95)VIIRS: cloud top height, cloud optical thickness, cloud shadow maskPeriod: October 2018 and March 2019Region: 5º-14ºE, 48º-54ºN (Germany)Binning the following parameters used to quantify the cloud shadow effect ParameterBin bordersMaximum cloud height of neighbour pixels (NCH)0, 3000, 7000, 20000Maximum cloud optical depth of neighbour pixels (NCODM)0, 3, 5, 10, 20, 50, 300Cloud fraction neighbour pixels (NCF)0, 0.1, 0.2, 0.3, 0.4, 0.5, 1.0Cloud shadow fraction center pixel (CSF)0.05, 0.10, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.5, 1.0The difference between the center pixel NO2 VCD and the average of the NO2 VCD in the cloud free neighbours for various parameter bins as given in the left table. (Each point is the average of 5 or more pixels)Assumption: clouds are the main reason for the variations in the NO2 VCD over the cloud shadow bandFor high NCH cases, NO2 bias increases with NCODM  linked to cloud shadow effects.

11. 11Estimation of 3D effect on NO2 retrieval -- approachDevelopment of parameterizations based on relationships identified from the sensitivity study for simulation with 2D box-cloud to quantify the 3D cloud effects on the NO2 retrieval. Ideally, the bias can be estimated as a function of all related parameters, stored in a look-up table (LUT). Since there is limited amount of 3D radiative simulation, the estimation is simply described as a formula: : NO2 profile height (PH), the latitude below which resides 75% of the integrated tropospheric NO2 profile : fraction of the integrated NO2 above the cloud (from neighboring pixels) in total NO2 column: cloud top height from neighboring pixels: logarithm of slant cloud optical thickness : cloud shadow fraction, , are quadratic polynomials, and the coefficients of the polynomials are obtained by fitting the retrieval bias in the cloud shadow for the simulation with 2D box clouds from the sensitivity study.F1(x) = 5.14 – 0.88x + 0.043x2 F2(x) = 0.81 + 0.42x – 0.061x2 F3(x) = 0.047 + 1.76x – 2.54x2Note: surface albedo is not taken into account here, only the cases with surface albedo of 0.05. 

12. 12Estimation of 3D effect on NO2 retrieval – application to synthetic dataSCOT is calculated at finer resolution (1.2km×1.2km) based on cloud information from the simulation input. (resolution of satellite domain is 7km×7km) CSF is a fraction of pixels with SCOT>1 & COT<3 in satellite domain (affected by neighbour clouds, but not cloudy pixels) Averaged SCOT is an average of SCOT for pixels with SCOT>1 & COT<3.CTH: median CTH value for a south neighbor pixel (satellite domain)SZA=60°, VZA=0°, albedo=0.05, SAA=13°/353°

13. 13Estimation of 3D effect on NO2 retrieval – application to real observationNO2 profile: TM5 CTMCloud top height (CTH): VIIRS CTH productCloud shadow fraction (CSF): by comparing to the averaged VIIRS reflectance (band M3) from the neighboring clear pixelsCloud optical thickness (COT): (VIIRS COT is not available in this case) based on the correlation between COT and effective CF from 1D simulation. Positive correlation between the NO2 difference from the retrieval and the estimated retrieval bias, and the estimation biases higher.The amount of data is very limited.

14. Summary and outlook3D radiative transfer effects are important for trace gas retrievals from satellite UV-VIS sensors (e.g. S4, S5P, S5), especially for polluted conditions in the cloud shadow (up to ~100%).Sensitivity study for box cloud simulations: 3D cloud effect highly relies on NO2 profile, slant cloud optical thickness, cloud shadow fraction, and surface albedo.3D effects are identified over the cloud shadow band in TROPOMI observations.3D effects are approximately quantified as a function of CSF, SCOT, CTH and NO2 PH.Future work: more precise correction of 3D cloud effects on NO2 retrieval by AI exploration of synthetic data, and application on the real measurements.This work was funded by ESA (3DCATS project 4000124890/18/NL/FF/gp)14