Sea surface temperatures with a certain degree of uncertainty Chris Merchant 2 July 2015 RMS DA SIG, ECMWF www.reading.ac.uk
Acknowledgements ESA Sea Surface Temperature Climate Change Initiative / ARC Fidelity and Uncertainty in Climate data records from Earth Observation (FIDUCEO) www.fiduceo.eu Emma Woolliams and Jon Mittaz (National Physical Laboratory) Gary Corlett (Leicester) Nick Rayner and John Kennedy (Hadley Centre) Claire Bulgin (Reading) 2
Metrology Worldwide trade/manufacturing Public health and safety Scientific Research Requires data that are: Stable over time so that scales and references aren t changing Insensitive to the method of measurement so the result doesn t depend on how you make the measurement Uniform worldwide so you can build the wings in France and the fuselage in Spain Based on references that can improve methods will improve over time as new technologies are available harmonisation should not be at the expense of improvements
Earth Observation Climate is difficult to measure. Requires data that is: Stable over time particularly for multi-decadal applications so data can be compared across decades meaningfully Insensitive to the method of measurement so data from different sensors (and techniques) can be combined Uniform worldwide so data from different space agencies can be combined Based on references that can improve methods will improve over time as new technologies are available harmonisation should not be at the expense of improvements
GUM is recommended No. 11 Measurement Good Practice Guide A Beginner's Guide to Uncertainty of Measurement Stephanie Bell www.bipm.org/en/publications/guides/gum.html Issue 2 The National Physical Laboratory is operated on behalf of the DTI by NPL Management Limited, a wholly owned subsidiary of Serco Group plc
Viewpoint of this talk Earth Observation community can improve as a measurement science, particularly in our approach to uncertainty information We should do this by taking advantage of the hard-won insights, vocabulary and methods from metrology Metrology is the science of measurement We are making measurements We should apply the principles of the science of measurement Benefits: logical rigour, reduction in ambiguity, better communication, More informed use of data generated some satellite SSTs may be usable as reference data for calibration of models and re-analysis systems but a present, potential users have little idea about the relative qualities of alternative datasets 6
Two definitions Error Concept: How different is the measured value from the (unknown) true value of the measurand? WRONGNESS Uncertainty Concept: Given the measured value, what range of values is it reasonable to attribute to the measurand? DOUBTFULNESS These are the correct scientific definitions and also match the usage of normal people Only scientists say things like the error in this value is +/-X and think it makes sense
More terms Standard uncertainty usual quantification of uncertainty as standard deviation of the estimated distribution from which errors are drawn Random effect a source of errors that are uncorrelated between repeated measured values note: errors can be random (uncorrelated); uncertainty cannot be random (or systematic) Systematic effect a source of correlated errors that you could correct for if you understood it note: this is NOT the same as a bias In EO data and in NWP fields there are effects in between these limits, e.g., locally systematic 8
Cascade of uncertainty det. noise, digitisation calibration, geolocation propagation (random, systematic), inversion propagation (rand., syst., locally syst.), sampling extra-/interpolation, smoothing L0 ê L1 ê L2 ê L3 ê L4
Key utility of metrology Link to SI has its importance Metrology gives methods for propagating uncertainty through the process cascade traceably Rigour all effects 10
Must consider all sources of uncertainty for long- term records Rela;ve effect of category of error source Random Locally systema;c Systema;c (single sensor) Systema;c (series) L2 L3 full res hi res global instant daily decadal total u stability
Importance of accurate SST observations Sea surface temperature prescribed for atmospheric re-analyses key variable constraining coupled re-analyses prescribed for climate model investigations (e.g, AMIP) widely used in model evaluation Relevant processes some examples air-sea heat flux exchanges oceanic uptake of CO2 SST anomalies provide memory for seasonal forecasting teleconnections through atmospheric circulation (e.g. ENSO) location and timing of tropical convection (e.g. MJO) Subtle errors in SST matter to people 12
20 o S to 20 o N mean The differences in simulated tropospheric temperature trends between model runs [forced with] HadISST1 and Hurrell [NOAA OI v2] are significant compared to the trends themselves, focusing attention on the uncertainty in the SST datasets. Flanaghan et al., 2014 Journal of Geophysical Research: Atmospheres Volume 119, Issue 23, pages 13,327-13,337, 11 DEC 2014 DOI: 10.1002/2014JD022365 http://onlinelibrary.wiley.com/doi/10.1002/2014jd022365/full#jgrd51871-fig-0004 13
Aims for SST Climate Data Record: 1. INDEPENDENT of in situ SST measurements 2. Of useful, quantified ACCURACY and SENSITIVITY 3. With context-sensitive UNCERTAINTY estimates 4. Harmonised to provide useful STABILITY 5. Able to be linked to the longer HISTORICAL RECORD 6. Generated by a ROBUST, SUSTAINABLE processing system in short delay mode These aims drive the approach of the project and products created SST CCI Phase-II
SST CCI data: 1 month per second Visualisation: Guy Griffiths, Reading E-Science Centre http://dx.doi.org/10.6084/m9.figshare.1246151 SST CCI Phase-II SST / K
ATSR-series SST 0.2m minus drifting buoys Merchant et al, 2012, JGR SST CCI Phase-II "Not another SST dataset!" NOAA 2 December 2014 Page 16
How uncertainties are communicated in SST CCI products SST CCI Phase-II
How uncertainties are communicated in SST CCI products SST CCI Phase-II
How uncertainties are communicated in SST CCI products SST CCI Phase-II
How uncertainties are communicated in SST CCI products SST CCI Phase-II
Random ( Noise in SST) Propagation of NEDT (scene dependent and instrument state dependent) through retrieval process (context dependent) SST CCI Phase-II
Random Systematic (large scale) SST CCI Phase-II
Random ( Noise in SST) Locally systematic (limitations of retr.) Optical Radiometry for Ocean Climate Measurements (EMPS Vol 47) Chapter 4.3 SST CCI Phase-II
Random Systematic (large scale) Locally systematic Total uncertainty (skin SST) SST CCI Phase-II
Dual view 3 and 2 channel estimates have different uncertainty distributions D3 D2
How to validate uncertainty? If we had perfect validation data (in situ), then
How to validate uncertainty? But drifting buoys have calibration uncertainty ~0.2 K
AATSR D2 SST at 0.1 deg lat-lon based on coefficients SST uncertainty estimate / ck
How should uncertainty be presented? Just give me one number (60%) Total uncertainty (35%) Confidence interval (25%) Separate out main components of uncertainty (20%) A probability distribu;on of error would be nice. (15%) Ensemble, please. (5%) Result of SST CCI User Survey, 2010
How should uncertainty be presented? Ensemble, please. Significant request from major users at SST CCI user consultation in 2014 (after discussing uncertainty concepts and issues for two days)
FIDUCEO approach det. noise, digitisation, calibration, geolocation propagation (random, systematic), inversion propagation (rand., syst., locally syst.), sampling extra-/interpolation, smoothing L0 ê L1 ê L2 ê L3 ê L4
FIDUCEO approach easy-fcdr, all main components of uncertainty FIDUCEO CDRs uncertainty traceable to FCDR FIDUCEO L3 uncertainty traceable to L2 CDR L0 ê L1 ê L2 ê L3 ê L4
FIDUCEO FCDRs DATASET NATURE POSSIBLE USES AVHRR FCDR HIRS FCDR MW Sounder FCDR Meteosat VIS FCDR Harmonised infra-red radiances and best available reflectance radiances, 1982-2016 Harmonised infra-red radiances, 1982-2016 Harmonised microwave BTs for AMSU-B and equivalent channels, 1992 2016 Improved visible spectral response functions and radiance 1982 to 2016 SST, LSWT, aerosol, LST, phenology, cloud properties, surface reflectance Atmospheric humidity, NWP re-analysis, stratospheric aerosol Atmospheric humidity, NWP re-analysis Albedo, aerosol, NWP re-analysis, cloud, wind motion vectors,
FIDUCEO CDRs DATASET NATURE USE Surface Temperature CDRs UTH CDR Albedo and aerosol CDRs Aerosol CDR Ensemble SST and lake surface water temperature From HIRS and MW, 1992-2016 From M5 7 (1995 2006) 2002-2012 aerosol for Europe and Africa from AVHRR Most of climate science model evaluation, reanalysis, derived/ synthesis products.. Sensitive climate change metric, re-analysis Climate forcing and change, health Climate forcing and change, health
What will be new about FIDUCEO data? Uncertainty-quantified FCDR At all data set scales (from pixel level in product through to multi-annual stability) there is adequate quantification of error distributions to propagate uncertainty across all data transformations (especially to CDR) accounting for error correlation structures, in a traceable, metrologically defensible manner Uncertainty-quantified CDR Uncertainty information in product that (i) discriminates more and less certain data, (ii) is validated as being realistic in magnitude, (iii) is traceable back to the FCDR uncertainty information
Key points Progress has been made in improving uncertainty information in SST datasets (www.esa-sst-cci.org) Some SSTs have better quoted uncertainty than most in situ sources Learning how to validate uncertainty estimates, with success in some cases Extending and rolling out metrology-based principles to other historic satellite datasets in FIDUCEO (www.fiduceo.eu)
Questions Will these approaches convince users that some, identifiable satellite data could count as reference measurements? Who should do the reference identification data producers or leave it to re-analysis users? Views on ensemble satellite SST records? Anyone keen to be a trail-blazer user for SST CCI Phase II SSTs or FIDUCEO FCDR/CDRs? 37