22nd October D4L2Kerridge1 Data Comparison Methods Lecture by B.Kerridge Rutherford Appleton Laboratory, UK ESA-MOST DRAGON 2 PROGRAMME Advanced Training Course in Atmospheric Remote Sensing th October 2009, Nanjing University

22nd October D4L2Kerridge2 Contents 1.Introduction 2.Radiances 3.Ozone 4.Aerosol 5.Surface networks 6.Summary

22nd October D4L2Kerridge3 1. Introduction – Why are data comparisons needed? 1.Comparison of observation with theory is fundamental. 2.Validation of satellite data is also essential →inform users of data attributes and quality – Which approaches are used? –Quantitative comparison of new data set with existing data sets of established attributes and quality (ie resolution, precision & accuracy) –Spatial & temporal distributions and variances –Biases and standard deviations between new and established data –Sensor-to-sensor, sensor-to-model or -analysis – What level of stringency is necessary (ie precision and accuracy)? –Atmospheric variability ← lifetime –Quality of established observing techniques What (scientific or other) value are the new observations required to add? First detection ↔ well-established measurement techniques / sensors →Determine level of sophistication

22nd October D4L2Kerridge4 2. Radiances / reflectances –The measured quantities (L1 data)

22nd October D4L2Kerridge5 Wavelength / nm Reflectance GOME coverage v55 v87 v67 v16 AATSR channels Comparison of GOME-1/ATSR-2 & SCIA/AATSR reflectances SCIA spectra

22nd October D4L2Kerridge6 Inter-comparison requires: –Accurate co-location of imager/spectrometer –Spectral averaging of SCIA/GOME-1 –Spatial averaging of AATSR/ATSR-2 AATSR / SCIA / GOME spatial coverage SCIAMACHY GOME 80 km 40 km Comparison of GOME-1/ATSR-2 & SCIA/AATSR reflectances

22nd October D4L2Kerridge7 Comparison of AATSR, ATSR-2, GOME-1 and SCIAMACHY 670nm Reflectances Two different orbits on 15th December 2002

22nd October D4L2Kerridge8 3. Ozone –A retrieved quantity

22nd October D4L2Kerridge9 Direct comparison of MIPAS O 3 with HALOE IR solar occultation Solar occultation: –High vertical resolution, precision & accuracy –Sparse geographical coverage HALOE: –79 co-locations between 22/07/02-14/12/02 –Accuracy: 30-60km 6%; 15-30km 20% –Co-location: <250km on same day

22nd October D4L2Kerridge10 High latitude Subtropics Individual HALOE Profiles

22nd October D4L2Kerridge – 50 hPa: +5 – +15% (+/- 20%) >50hPa: +ve biases and increases in RMS HALOE sampling weighted to N. mid. lats; only 10% in tropics Ensemble of HALOE Profiles

22nd October D4L2Kerridge12 Factors Limiting Direct Comparisons Precision: –Direct satellite comparisons underestimate MIPAS precision, due to: Imperfect co-location cf atmospheric variability Differences in representation of O 3 vertical profile Precision of the correlative sensor factored in Accuracy: –Vertical structure of bias partly reflects different vertical resolutions of respective sensors

22nd October D4L2Kerridge13 Direct Assimilation of MIPAS Data Ozone: July 10 th 2003, 850K level Direct assimilation equivalent to Kalman Smoother, but far lower computational cost. Uses isentropic advection as modelling constraint. → No chemistry or vertical advection Field allowed to adjust continually towards the observations. → Analysis unbiased relative to I/P data. Applied to MIPAS ozone, methane and water vapour products (Feb. to July 2003 v4.61/2) M.Juckes, RAL

22nd October D4L2Kerridge14 MIPAS Ozone Comparisons with Other Sensors Mean [ppmv] Standard error [ppmv] Assimilation allows comparison against other observations which are not coincident. Biases (left) found to be small in the lower and mid-stratosphere. Standard error (right) is <10% in most of the stratosphere. Using a range of independent instruments helps establish confidence. Height [km] In both graphs, the outer limit of shading is 10% of sample mean profile, inner boundary of shading 1%, transition 5%. 10%

22nd October D4L2Kerridge15 Assimilation of MLS ozone profiles: comparison of analyses with ozone sondes 1 st July – 30 th Sept 2007 Ozone data actively assimilated into ECMWF operational system: CTRL: SCIAMACHY total column (KNMI) MLS: SCIAMACHY total column + MLS O 3 profile → Agreement with ozonesonde profiles <200hPa improved by assimilating MLS R.Dragani, ECMWF

22nd October D4L2Kerridge16 TES a priori profile DIAL or sonde profile TES averaging kernel Comparison of TES Retrieved O 3 Profiles with Ozonesondes & Airborne DIAL

22nd October D4L2Kerridge17 TES O 3 – Ozonesondes Worden et al.

22nd October D4L2Kerridge18 TES O 3 – Ozonesondes (contd.) TES V1 TES V2

22nd October D4L2Kerridge19 Airborne DIAL Measurements DIAL profiles ozone simultaneously above & below DC-8 aircraft DIAL profiles <0.15 o (lat/long) from TES profile averaged & interpolated to TES p grid. Missing data in DIAL profile → TES a priori DIAL: accuracy <10% (2 ppbv) Vertical resolution of 300 m. Richards et al

22nd October D4L2Kerridge20 TES O 3 – DIAL TES V2 TES V3 – Care needed in comparisons near tropopause, where O 3 vertical gradient strong. – Using AKs, TES bias in upper troposphere seen to be reduced in V3

22nd October D4L2Kerridge21 Comparison of OMI integrated ozone column to ECMWF analyses 70-day period from 12 Aug 2005 Assimilation of operational obs: –ozonesondes –aircraft –satellite IR rads –SBUV/2 partial columns –SCIAMACHY columns –MLS ozone profiles –OMI data: –No sunglint –SZA< 84 deg –Coincidence criterion: within ± 1 hour of analysis time Analysis by S.Migliorini, U.Reading

22nd October D4L2Kerridge22 Total column ozone comparisons Total column should be compared to ECMWF analyses first interpolated to OMI pixel locations Integrated column: Retrieved profile:

22nd October D4L2Kerridge23 OMI ozone layer averaging kernels a < 1 15 Oct UTC nadir pixel

22nd October D4L2Kerridge24 OMI total column ozone

22nd October D4L2Kerridge25 Simulation with ECMWF 3-D analysis

22nd October D4L2Kerridge26 Integrated Column Differences (%) For 70 day’s data: [OMI-ECMWF] bias % and RMS 5.2 %

22nd October D4L2Kerridge27 4. Aerosol

22nd October D4L2Kerridge28 Height-Integrated Aerosol Aerosol optical thickness (AOT),  ( ), at = 550 and 870 nm Cloud screening & handling of surface BRDF critical for aerosol retrieval from satellite sensors

22nd October D4L2Kerridge29 Comparisons of AATSR & SEVIRI with MODIS and MISR  MODIS & MISR on EOS-Terra  Day-time equator crossing time similar to Envisat (~10am)  Terra ascending node in daytime whereas Envisat descending, →Observing times at high latitude quite different to AATSR.  SEVIRI on MSG samples hourly, though viewing geometry a fixed function of geographical location

22nd October D4L2Kerridge30 Impact of coincident sampling on AATSR – MODIS Comparison in Sept’04 Cor=.64 Cor=.48Cor=.56 SD=.16 Cor=.71 SD=.08

22nd October D4L2Kerridge31 C=0.86 C=0.53 C=0.80C=0.45 SEVIRI MODIS MISR SEVIRI aerosol optical thickness comparisons with MODIS & MISR July 2005

22nd October D4L2Kerridge32 5. Ground-based networks

22nd October D4L2Kerridge33 Aeronet AErosol RObotic NETwork. Standardized instruments, calibration & processing. →Geographically distributed observations of aerosol spectral optical depths & retrieved products

22nd October D4L2Kerridge34 Aeronet Each ground station has CIMEL sun-photometer – ’s: 340, 380, 440, 500, 675, 870, 940 and 1020 nm –Widths: 2nm at 340nm, 4 at 380nm, others 10nm. –Narrow FOV ~1 o Direct sun measurement every 15 mins AOT from direct sun extinction → Beer’s law –Corrected for Rayleigh scat. + trace-gas abs. –Rel. acc. cf other photometers <0.004 –Abs. acc. < [Angstrom coefficient fitted to AOT at 440, 500, 675, 870nm]

22nd October D4L2Kerridge35 Satellite grid-box containing AERONET station identified. For each day, AOT (and Angstrom coeff.) extracted within ±20km from this satellite pixel. Aeronet measurements for each station extracted within ±30 minutes of satellite measurements. For both Aeronet and satellite sensor, no. of valid retrievals, means and standard deviation stored Minimum of 4 matches required. If SD of either satellite or Aeronet exceeds 0.15, comparison discounted. Satellite – Aeronet Comparison

22nd October D4L2Kerridge36 SEVIRI – Aeronet individual time series comparison

22nd October D4L2Kerridge37 SEVIRI Aeronet MODIS MISR SEVIRI – Aeronet time series comparison for North Atlantic Ocean

22nd October D4L2Kerridge38 SEVIRI – Aeronet: statistical comparison C=.2 SD=.07 C=.84 SD=.04 C=.38 SD=.12 C=.81 SD=.07

22nd October D4L2Kerridge39 SEVIRI – Aeronet: Europe & Africa C=.64 SD=.06 C=.36 SD=.15 – SEVIRI performance reasonable over sea and coasts – High-reflectance land sites (Africa) are problematic

22nd October D4L2Kerridge40 –Representativeness –point location vs satellite field of view –Only five aerosol types in retrievals –aerosol composition varies continuously –Cloud flagging may be too stringent –upper limit on allowed AOT set too low –High land surface reflectance not modelled well Some Factors Limiting Aeronet Comparisons

22nd October D4L2Kerridge41 Network for the Detection of Atmospheric Composition Change (NDACC) >70 high-quality stations observing stratosphere and upper troposphere → impact of stratospheric changes on troposphere and on global climate LIDAR profiles: Raman lidar - water vapor Differential Absorption Lidar (DIAL) - O 3 Backscatter lidars - aerosol Raman and Rayleigh lidars - temperature Microwave radiometers : ozone, water vapor, and ClO profiles UV/VISIBLE SPECTROMETERS: column ozone, NO 2 (OClO and BrO) FTIR SPECTROMETERS: column ozone, HCl, NO, NO 2, ClONO 2, and HNO 3 DOBSON/BREWER: column ozone SONDES: ozone and aerosol profiles UV SPECTRORADIOMETERS: UV radiation at the ground

22nd October D4L2Kerridge42

22nd October D4L2Kerridge43 Network of ground-based FTS Near-IR solar absorption spectrometry 4,000–14,000 cm−1 at 0.02 cm−1 resolution Analysis with standard algorithm (GFIT) –Non-linear least squares → scale profile → CO 2, CH 4, O 2 & other columns O 2 to convert column densities to pressure- weighted column average mixing ratios Complementary surface in-situ at each site To usefully constrain global carbon budget – directly, and through validation of satellite column measurements – precision of 0.1% required! SCIAMACHY & GOSAT validation → Stringent accuracy reqs. eg: - solar tracking - correction for source fluctuations - FTS instrument line shape -permanently monitored with HCl cell in solar beam. - spectroscopic parameters - retrieval algorithms Total Column Carbon Observing Network - TCCON

22nd October D4L2Kerridge44 – Variability reflects atmospheric lifetimes: CH 4 ~10yrs, N 2 O >100yrs – Correction needed for stratospheric column variability Sherlock et al

22nd October D4L2Kerridge45 6. Summary –Comparison of observation with theory is fundamental –Validation of satellite data also essential –Important to compare like-with-like as far as possible →Apply observation operators to model fields and in assimilation →Account for sensor vertical smoothing & a priori also in comparisons with profiles observed at higher resolution –Techniques of increasing sophistication are being used →to compare with models and other observations →to quantify value added in assimilation –Production of long-term satellite data-sets on the essential climate variables will depend on surface networks for ground truth.