Preparing Ocean Observations for Reanalysis
Nick Rayner, Met Office Hadley Centre 5th International Conference on Reanalyses, Rome, 13th-17th November 2017 Material provided by John Kennedy, Clive Wilkinson and Holly Titchner
www.metoffice.gov.uk © Crown Copyright 2017, Met Office Evolution of the use of ocean measurements in reanalysis SST from ships, buoys and satellites Statistical analysis: globally- Atmosphere-only Sea ice complete forcing reanalysis concentration data from charts and satellites
Temperature and salinity profiles from Ocean-only Assimilation MBTs, XBTs, reanalysis bottles, CTDs and Argo
www.metoffice.gov.uk © Crown Copyright 2017, Met Office Evolution of the use of ocean measurements in reanalysis SST from ships, buoys and satellites Statistical analysis: globally- Weakly-coupled Sea ice complete forcing reanalysis concentration data from charts and satellites
Temperature and salinity profiles from MBTs, XBTs, Assimilation bottles, CTDs and Argo
www.metoffice.gov.uk © Crown Copyright 2017, Met Office Evolution of the use of ocean measurements in reanalysis SST from ships, buoys and satellites In all cases, Sea ice whatever the concentration length of the from charts Fully-coupled Assimilation reanalysis, to and satellites reanalysis achieve a Temperature climate-quality and salinity result, the same profiles from MBTs, XBTs, preparation is bottles, needed CTDs and Argo
www.metoffice.gov.uk © Crown Copyright 2017, Met Office Categories of work needed to prepare observations for reanalysis
• Data assembly (see e.g. Peter Thorne’s presentation) • Data Rescue • Quality Control • Achieving consistency: bias correction or homogenisation • Uncertainty quantification • Feedback analysis
www.metoffice.gov.uk © Crown Copyright 2017, Met Office Quality control
www.metoffice.gov.uk © Crown Copyright 2017, Met Office Not-so-subtle errors in SST data
60N FAIL
PASS Latitude 0N
60S April 1973 0C SST 30C www.metoffice.gov.uk © Crown Copyright 2017, Met Office The subtle dangers of quality control
Background field
www.metoffice.gov.uk © Crown Copyright 2017, Met Office Effective use of backgrounds to trap errors
noisy sensor buoy run aground
buoy picked up by a ship www.metoffice.gov.uk Atkinson et al, 2013: JGR, doi:10.1002/jgrc.20257 © Crown Copyright 2017, Met Office Achieving internal consistency: bias correction / homogenisation
www.metoffice.gov.uk © Crown Copyright 2017, Met Office Aspects of internal consistency
Time
ships with evolving SST measurement methods satellites buoys surface
bottles CTDs MBTs XBTs Argo
Climate quality reanalysis requires observations that are consistent: surface Depth • in time and sub- surface • between different components of the observing system for one variable Where observations for different variables are brought together need to remove relative biases between observing system components addressing different variables www.metoffice.gov.uk © Crown Copyright 2017, Met Office April 1905 April 1935
Ocean data for coupled reanalysis: April 1965 April 1985 HadIOD Contains platform ID, position, time, depth, Maximumplatform & instrument depth of type, observed temperature data& salinity, in any provenance location information andApril a unique 1995 April 2010 ID, together with quality flags, bias corrections and throughuncertainty time estimates
>4000m www.metoffice.gov.uk Atkinson, et al 2014: JGR, doi:10.1002/2014JC010053 © Crown Copyright 2017, Met Office Ways of achieving consistency
• Compare everything and develop empirical corrections, relative to a chosen reference • Risks picking the wrong reference and biasing the whole system
• Understand each data source physically and correct according to its own biases
A Call for New Approaches to Quantifying Biases in Observations of Sea Surface Temperature. Kent et al. (2017) BAMS https://doi.org/10.1175/BAMS-D-15-00251.1
www.metoffice.gov.uk © Crown Copyright 2017, Met Office Ways of achieving consistency
• Compare everything and develop empirical corrections, relative to a chosen reference • Risks picking the wrong reference and biasing the whole system
• Understand each data source physically and correct according to its own biases • Then compare to everything else and check consistency
www.metoffice.gov.uk © Crown Copyright 2017, Met Office Ways of achieving consistency
• Compare everything and develop empirical corrections, relative to a chosen reference • Risks picking the wrong reference and biasing the whole system
• Understand each data source physically and correct according to its own biases • Then compare to everything else and check consistency
• But this requires good metadata, which is often lacking
• However, this allows potential propagation of error structure
• Let the reanalysis handle it – still requires good understanding and metadata www.metoffice.gov.uk © Crown Copyright 2017, Met Office Corrections to ship SST in HadIOD
Ensemble of estimated Engine Room 0.2 Intake measurement biases
Ensemble of estimated biases in SST measurements made using buckets to sample water -0.2
-0.4
1900 1950 2000
www.metoffice.gov.uk © Crown Copyright 2017, Met Office Use of HadIOD in historical ocean reanalyses (1900-2010)
See poster by Chunxue Yang1, Simona Masina2,3 and Andrea Storto2 1ISAC-CNR, Italy; 2CMCC, Bologna, Italy; 3INGV, Bologna
www.metoffice.gov.uk © Crown Copyright 2017, Met Office Uncertainty quantification
www.metoffice.gov.uk © Crown Copyright 2017, Met Office This is what we have
December 27-31 1961
This is what we want
© Crown copyright Met Office Available observations do not uniquely define the past
© Crown copyright Met Office The Ensemble Generator
First, generate a range of plausible bias adjustments to the data
Bias adjustment
© Crown copyright Met Office Reject in situ ensemble members that disagree with ARC ATSR
ARC ATSR
Constrained 20-member HadSST3 ensemble
1000 member ensemble Reduced to ~10 members
© Crown copyright Met Office Blending satellites - daily
AVHRR ATSR BLEND
© Crown copyright Met Office Blending satellite and in situ - pentad SATELLITE IN SITU BLEND
© Crown copyright Met Office From one realisation of the in situ bias adjustments, produce 10 interchangeable realisations of the broad-scale reconstruction
Bias adjustment Broad-scale reconstruction © Crown copyright Met Office GUESS EOFS
project on to OBSERVATIONS
AT EACH TIME STEP
BROAD-SCALE RECONSTRUCTION
EOF1 EOF2 & EOF3 time series of EOF4 weights of EOFs
© Crown copyright Met Office Bayesian PCA EOF1 EOF2 Weights EOF3 of EOFS EOF4 project on to
OBS
AT EACH LOCATION
NEW EOFs
© Crown copyright Met Office Bayesian PCA NEW EOFS
project on to OBSERVATIONS
AT EACH TIME STEP
BROAD-SCALE RECONSTRUCTION EOF1 & EOF2 EOF3 time series of weights of EOFs EOF4
© Crown copyright Met Office Bayesian PCA Then, from each of the 10 realisations of the broad-scale reconstruction, we can create an ensemble of interchangeable local OIs of the residuals from that reconstruction
Bias adjustment Broad-scale Local OI of reconstruction residuals © Crown copyright Met Office Drawing samples from Local OI
© Crown copyright Met Office One random selection from the analyses of the residuals gives us one of our realisations of HadISST2
Bias adjustment Broad-scale Local OI of reconstruction residuals © Crown copyright Met Office Pick 10 such random paths to span the total uncertainty in the analysis and provide an ensemble of interchangeable versions of HadISST2
Bias adjustment Broad-scale Local OI of reconstruction residuals © Crown copyright Met Office HadISST2 ensemble is under- dispersive Uncertainties in Alternatives to method climatology
Alternatives to Changes to filling of Alternate near neighbour spatial gaps between methods infilling Atlas Climatologies/NIC charts in SH Small changes to current Filling of method spatial/temporal Changes to Alternate gaps current method methods Bias Concentration infilling adjustment (when only extents are known) Uncertainties in Structural current method uncertainty AARI ice charts Reference Input data data set sources
Others? SIBT1850 instead Others? Passive of Walsh and microwave Chapman Compilation
Future: Other passive NIC ice charts instead of microwave passive microwave products AMSR-E generate sea during overlap period passive ice ensemble microwave Data rescue
www.metoffice.gov.uk © Crown Copyright 2017, Met Office Example Documents: Christian Salvesen Archive, Edinburgh Example Document: Logbook Norvegia - 4 December 1928 Vestfold Archive, Sandefjord
4-Hourly Wind direction Wind force Weather Sea state Pressure Temperature Whaling Records – Archive of Sea Mammal Research Unit, University of St. Andrews 1. 3/2/1931 – No ice 68⁰S, 177⁰E – wind west, fresh
2. 28/2/1930 – Icepack 66.25⁰S, 179.30⁰E – wind south, fresh Note: a separate form for each whale captured. Imaging of historical Southern Ocean observations
Antarctic & Southern Ocean Research vessels, whaling vessels, commercial shipping (sail and steam) 1900-1950 National Maritime Museum Valparaíso, Chile 22,000 UK National Meteorological Archive 21,570 Sea Mammal Research Unit, Scotland 14,890 Christian Salvesen Archive, Scotland 2,730 Maritime Museum, Mariehamn Finland 20,300 Vestfold Archive, Sandefjord, Norway 30,500 Total 111,990
When added to previous work total 137K images and estimated up to 7M observations
Clive Wilkinson Summary
• Data assembly • Data Rescue • Quality Control • Achieving consistency: bias correction or homogenisation • Uncertainty quantification • Feedback analysis
www.metoffice.gov.uk © Crown Copyright 2017, Met Office