Introduction the GSICS Correction Fred Wu (NOAA) Tim Hewison (EUMETSAT) For GSICS Research Working Group
Introducing GSICS Correction GSICS Research Working Group (GRWG) D. Blumstein, C. Cao, S.-R. Chung, D. Doelling, M. Gunshor, P. Henry, T. Hewison (vice chair), X. Hu, D. Kim, M. König, J. Lafeuille, J. Liu, P. Minnis, A. Okuyama, J. Privette, B.-J. Sohn, Y. Tahara, D. Tobin, X. Xiong, L. Van de Berg, X. Wu (chair), P. Zhang, and Y. Zhang CMA, CNES, EUMETSAT, JMA, KMA, NASA, NESDIS, NIST, UW, WMO
Outline GSICS Correction for GEO Based on AIRS/IASI Other Products Programmatic Considerations Theoretical Basis Analysis Other Products Summary What is produced, how and why Find proper applications Help us to improve and tailor
Outline GSICS Correction for GEO Based on AIRS/IASI Other Products Programmatic Considerations GEO-AIRS/IASI Hierarchical structure for ATBD Theoretical Basis Analysis Other Products Summary
GEO-AIRS/IASI GEO AIRS/IASI An early priority Members are operators of meteorological satellites Coordinated by WMO and assisted by other agencies All have GEO AIRS/IASI Hyperspectral Sounder on LEO Versatile for collocation in space and spectrum Highly accurate and mutually consistent An early priority Make all GEOs consistent with AIRS/IASI Why GEO-LEO has been the focus
Hierarchical Structure for ATBD GEO-AIRS/IASI is one of many inter-cal Other orbits Other instruments Other spectral regions Within GEO-AIRS/IAI, each pair can be unique Instrument characteristics Agency priority Hierarchical Structure for Algorithm Theoretical Basis Document (ATBD) The need for ATBD’s and hierarchical structure to manage them
Hierarchical Structure for ATBD Details for Specific Instrument Pairs Procedures for Specific Inter-Calibration Class General Implementation Options The need for ATBD’s and hierarchical structure to manage them Basic Principles
Hierarchical Structure for ATBD Can build all inter-calibration on common principles And minimise differences between instrument pairs For maximum consistency Modular Different GSICS partners can work on different instrument pairs Provides traceability Include version number for each process, option, dataset Integration with review cycle Simplifies documentation Based on common principle, with specific details for each instrument pair Further details of benefits
Outline GSICS Correction for GEO Based on AIRS/IASI Other Products Programmatic Considerations Theoretical Basis Inter-calibration Collocation Transformation Analysis Other Products Summary
Inter-Calibration Calibration – Quantification of instrument responses to known signals Inter-Calibration – Quantification of instrument responses to signals defined by the reference instrument Differ in source of known signals Requires identical observing conditions: time, location, spatial response, spectral response, and viewing geometry. In reality … There are generally accepted definitions for calibration. We define inter-calibration accordingly.
Masks, flags, … SRFs, PSFs, … Correcting GSICS Correction Correction Coeffs Comparison Data Collocated Data IUT Lvl 1 Data Re-Cal Data Plots and Tables Orbit Prediction Colloc. Criteria IUT Level 1 Data Ref Level 1 Data Reports Monitoring Diagnosing Collocation Transformation Products Users Analysis A series of processes to minimize and/or account for any and all differences due to observing conditions Blue – input data Yellow – process Green – intermediate products Pink – output products
Collocation Three objectives of GSICS Requirements for Collection Quantify the differences – magnitude and uncertainty Correct the differences – empirical removal Diagnose the differences – root cause analysis Requirements for Collection Results are valid for the collocated data only Assumed, often implicitly, to be valid for the rest Variety of the collocated data is important Single pixel collocations anywhere within the GEO field of regard be collected continuously over long term for all bands. Save as many collocations as practical First, three specific objectives of GSICS. Note that we would find the root cause before correcting the differences. In reality, especially during satellite operation, we either don't have the time or ability to find the root cause of all the differences, so we correct empirically first and diagnose some of the root causes later. In that regard we don't necessarily want to eliminate all the differences, only those we are reasonably sure. Since we know only the collocated data and we have to generalize to all data, we need variety in the collocated data, which is condensed in the 2nd bullet. Single pixel vs. large area, all FOR vs. nadir only, continuous vs. fixed time, long term vs. short period or intermittent, all band vs. single or those interested. Note that sun-synchronous LEO pass GEO nadir at fixed time of day. 3rd bullet is data management.
Collocation Time Location Azimuth Angle Zenith Angle From Telemetry Threshold depends on refresh rate and size of data Location Operational geolocation Azimuth Angle Archived Zenith Angle geo_zen-leo_zen < threshold – penalize at small angle sec(geo_zen) - sec(leo_zen) < threshold – penalize at larger angle cos(geo_zen)/cos(leo_zen)-1 < threshold Three choices of geometric alignment. Some considerations of threshold follow.
Collocation Not much of an issue for window channels …
Collocation 13.3 um … than for absorptive channels. Empirical correction is helpful, although one cannot depend on that too much since this correction depends on the lapse rate
Collocation Transformation Products Users Analysis Masks, flags, … SRFs, PSFs, … Correcting GSICS Correction Correction Coeffs Comparison Data Collocated Data IUT Lvl 1 Data Re-Cal Data Plots and Tables Orbit Prediction Colloc. Criteria IUT Level 1 Data Ref Level 1 Data Reports Monitoring Diagnosing Collocation Transformation Products Users Analysis Will explain each of these processes.
Spatial Transformation GEO FOV, may be square or overlapping Non-uniform features LEO FOV, relative to GEO FOV Collocation FOV, may depend on GEO and LEO FOVs Collocation environment, may depend on time window and wind speed Basic requirement is to average the 3-by-3 GEO FOV for comparison with LEO FOV. But environment stdv matters.
Spectral Transform MTSAT-1R 6.8-um AIRS blacklist ch. SRFs of Gap channels SRFs of AIRS SRF of MTSAT SRF of super channel consists of AIRS and gap channels Relatively straightforward for IASI. AIRS has spectral gaps due to design and operation failure. JMA designed algorithm to find the R_v where AIRS measurements are not available. Weights of AIRS ch. Weights of gap ch.
Outline GSICS Correction for GEO Based on AIRS/IASI Other Products Programmatic Considerations Theoretical Basis Analysis Monitoring Diagnosing Correcting and impacts Other Products Summary Tim will present this part
Standard Radiance, ISTD Analysis Basic Comparison method: Weighted Linear Regression* of all collocated radiances within defined period Error bars = Variance In general Slope ≠ 1 Scene-dependent biases Mean difference will depend on sample population Define standard radiance Calculate bias at ISTD and uncertainty * Reduced Major Axis – under investigation Standard Radiance, ISTD Bias
Collocation Transformation Products Users Analysis Masks, flags, … SRFs, PSFs, … Correcting GSICS Correction Correction Coeffs Comparison Data Collocated Data IUT Lvl 1 Data Re-Cal Data Plots and Tables Orbit Prediction Colloc. Criteria IUT Level 1 Data Ref Level 1 Data Reports Monitoring Diagnosing Collocation Transformation Products Users Analysis Will explain each of these processes.
Monitoring MTSAT-AIRS/IASI Monitoring Example from JMA website MTSAT-1R – AIRS/IASI Time Series of Bias at 220, 250, 290K and lots more…
Monitoring GOES12-AIRS
Monitoring Meteosat9-IASI IR3.9-IR12.0: Small, stable Biases <0.2K ± 0.05K IR13.4: Larger Bias ~-1K -0.05K/mnth+Jump Time series of brightness temperature differences between MSG2-IASI for typical clear-sky radiances. Error bars represent statistical uncertainty on each mean bias (may be very small).
Collocation Transformation Products Users Analysis Masks, flags, … SRFs, PSFs, … Correcting GSICS Correction Correction Coeffs Comparison Data Collocated Data IUT Lvl 1 Data Re-Cal Data Plots and Tables Orbit Prediction Colloc. Criteria IUT Level 1 Data Ref Level 1 Data Reports Monitoring Diagnosing Collocation Transformation Products Users Analysis Will explain each of these processes.
Analysis – Defining GSICS Correction GSICS Corrected radiance from GEO operational product a, b from weighted regression Coalesce collocations over Period ~ 1 month IGSICS a b ISTD ΔISTD IGEO
Analysis – Correcting EUMETSAT routinely run prototype inter-calibration of MSG-IASI Results published on webpage for Inter-calibration Services : http://www.eumetsat.int/Home/Main/Access_to_Data/IntercalibrationServices This webpage also allows access to coefficients required to apply GSICS Correction Users can implement this as change in calibration coefficients
Analysis – Correcting GSICS Correction Coefficients Contain best estimate of relationship between Instruments’ radiance and Reference Includes regression coefficients needed to apply GSICS Correction Draft NetCDF Format defined
Collocation Transformation Products Users Analysis Masks, flags, … SRFs, PSFs, … Correcting GSICS Correction Correction Coeffs Comparison Data Collocated Data IUT Lvl 1 Data Re-Cal Data Plots and Tables Orbit Prediction Colloc. Criteria IUT Level 1 Data Ref Level 1 Data Reports Monitoring Diagnosing Collocation Transformation Products Users Analysis Will explain each of these processes.
Analysis – Diagnosing Diagnosis is performed by offline Examples Investigating anomalous results Generating reports and recommendations Examples Meteosat-9 Ice Contamination of IR13.4 channel GOES-13 Imager 13.3 m channel cold bias GOES-11/12 Sounder Channel 15 bias GOES Midnight calibration anomaly MODIS SRF errors Will be discussed at request? Covered elsewhere.
Diagnosing Ice Contamination of Meteosat-9 IR13.4 Meteosat-9-IASI Bias In 13.4µm channel increasing by ~0.5K/yr Recovers after decontamination Theory: Ice on optics Bias jumps ~0.5K at decontamination Bias increases as ice builds up Time series of radiances relative biases between IR13.4 channel of Meteosat-9(MSG2)-IASI for Standard Radiance scenes. Blue crosses indicate results from individual satellite orbits. Red circles show monthly means. Error bars represent statistical uncertainty on each mean bias. Solid blue lines show trends before and after the decontamination procedure of December 2008.
Diagnosing Ice Contamination of Meteosat-9 IR13.4 Meteosat-9-IASI Bias In 13.4µm channel increasing by ~0.5K/yr Recovers after decontamination Theory: Ice on optics Ice absorption model Changes SRFs Transmission spectra of ice layers of different thicknesses (black): 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0 µm layers. Spectral Response Functions of Meteosat-8 infrared channels (red).
Diagnosing Ice Contamination of Meteosat-9 IR13.4 Meteosat-9-IASI Bias In 13.4µm channel increasing by ~0.5K/yr Recovers after decontamination Theory: Ice on optics Ice absorption model Changes SRFs Biases non-window channels Modelled by ~0.6µm ice Brightness temperature bias modelled by modifying Meteosat-9’s SRF by the absorption of different thicknesses of ice.
Diagnosing Ice Contamination of Meteosat-9 IR13.4 Meteosat-9-IASI Bias In 13.4µm channel increasing by ~0.5K/yr Recovers after decontamination Theory: Ice on optics Ice absorption model Changes SRFs Biases non-window channels Modelled by ~0.6µm ice Consistent with gain changes Gain Changes (%) in Meteosat-9 IR channels during decontaminations and estimated ice thickness (µm) from transmission model.
Possible shift of MODIS 6.7μm SRF causing a 3K cold bias . Collocated MTSATM- MODISM TB Collocated IASI- MTSATM Bias corrected MTSAT TB MTSAT = MTSATM-Bias Bias estimate Bias = MTSAT-MTSATM The 'm' stands for measured one. Without it, they are referred to as theoretically calculated values from IASI measurements. Because the direct comparison of MODIS data with IASI is not possible, I utilized MTSAT as a kind of surrogate of IASI. Of course I suspected that MODIS WV has a problem. Followings are procedures: - By applying the SRFs of MODIS and MTSAT to IASI spectrum, we obtained a theoretical relationship between MODIS and MTSAT. - We know that the MTSAT WV calibration is good (from JMA web site). But, we tried to correct any bias from collocated MTSAT-IASI data so that MTSAT is equivalent to IASI. In other words, IASI information has been transferred to MTSAT. - Now because collocated MTSAT-IASI data, and bias between MTSAT and IASI are available, we can create IASI-suggested MODIS WV TB vs. measured value (MODIS**m). Theoretical TB calculations MODIS = a + b (MTSAT) OR MODIS = a + b (MTSATM+Bias) B.J. Sohn (Seoul National Univ., Seoul, Korea; sohn@snu.ac.kr)
(a) Theoretical TB relationship (b) MTSAT bias correction Correlation Coefficient: 0.98 Mean Bias: -0.12 Slope: 1.00 Intercept Point: 0.07 RMSE: 0.43 June 2007 171 Points MODIS=-5.83+1.03MTSAT Correlation Coefficient: 0.99 Cold Bias: 3.06 Slope: 0.97 Intercept Point: 9.62 RMSE: 0.47 610 Points Mean Bias: -0.07 Slope: 0.96 Intercept Point: 10.68 RMSE: 0.46 (a) Theoretical TB relationship (b) MTSAT bias correction (c) Cold bias of MODIS WV (d) 11 cm-1 SRF shift Results show about 3K cold bias, corresponding to 11 cm-1 SRF shift. - Cold bias seems to be variable with the season (but about same 11 cm-1 SRF shift)
Preliminary Impacts – Bias Before 3K Bias After ~0K Bias The first major deliverable to the user community is the GSICS correction algorithm for geostationary satellites. The user applies the correction to the original data using GSICS provided software and coefficients. The correction adjusts the GOES data to be consistent with IASI and AIRS. The above figure shows the difference between observed and calculated brightness temperatures (from NCEP analysis) for GOES-12 channel 6 before and after the correction, respectively. The bias is reduced from 3 K to nearly zero. 39
Impact of Bias on Cloud Top Height GSICS Correction applied to IR13.4 channel of Meteosat-9 to evaluate impact bias on meteorological products for case studies based on MSG images taken during November 2008 IR13.4 used in CTH algorithm to correct heights of semi-transparent clouds Impact of change is quite small: Bias=6.1 hPa, SD=27 hPa. Because CO2 slicing uses ratio of differences in radiances between window + CO2 channels in clear and cloudy sky More impact found for cloud detection using IR10.8 – IR13.4 test: With corrected IR13.4 radiances this test finds 2.9% more cloud However, cloud detection schema uses more threshold tests Overall change 0.4% more cloud
Impact on Global Instability Index Global Instability Index (GII) simultaneously retrieves several convective indices Total Precipitable Water (TPW) GII sensitive to biases in IR13.4 channel Provides information on mid-troposphere temperature, which is compared to low-level temperature. GII most sensitive to this bias in areas with high surface temperatures Bias correction does not affect TPW K and Lifted index → "more stable" Most impact is on data coverage: Without bias correction algorithm found no solution for 7.6% of pixels. With bias correction: 3.1% K-Index over Southern Africa from GII algorithm from Meteosat-9 before (left) and after (right) applying GSICS Correction to IR13.4 channel. The correction improves the data availability greatly over the hot land surface (dark grey areas in left panel), while only introducing small changes (~0.1 K as area average) to the instability index.
Outline GSICS Correction for GEO Based on AIRS/IASI Other Products Programmatic Considerations Theoretical Basis Analysis Other Products Visible LEO-LEO Other references Summary What to expect Get involved early
Inter-cal of GEO Visible Channel Challenges – lack of common target Viewing geometry (especially critical for visible) Viewing domain (statistics) Spectral response function
Inter-cal of GEO Visible Channel
Inter-cal of GEO Visible Channel Challenges – lack of common target Viewing geometry (especially critical for visible) Viewing domain (statistics) Spectral response function Strategy Relative Deep Convective Clouds (DCC) Moon Absolute Calibrated radiometer Desert – Expanding SADE for Asian and American GEOs
Inter-cal of GEO Visible Channel Lakes 0.5 deg Google Map 17km
LEO-LEO Simultaneous Nadir or Conical Overpass (SNO & SCO) From inter-comparison to inter-calibration SRF
Other references Compare with other references Earth targets NWP models
Summary Developed Inter-Calibration methodology Following hierarchical ATBD Applied to GEO IR Imagers Using LEO hyperspectral sounders as reference Demonstrated generation and impacts of product Bias Monitoring GSICS Correction Coefficients Root Cause Diagnosis of calibration anomalies Examples of Impact Ready for testing by wider user community To provide feedback