Ini$a$ve overview
08/19/11
Talk Outline
•
Background
•
Data rescue and databank
•
Mul6plicity of data products
•
Common assessment through consistent
performance benchmarking
•
Data and product portal development
Background
•
Several independent global es6mates of Land
Surface Air Temperatures
–
Independently produced
–
Gross agreement on all 6mescales
–
Process metadata in some cases missing
–
Much of the work was undertaken in late 1980’s/
early 1990’s -‐ technology and expecta6ons have
changed since then
–
Monthly large scale gridbox averages only (generally)
Examples of temperature reconstruc$ons
IPCC Fourth Assessment Report (2007), Figure 3.1
Annual anomalies of global land-surface air temperature (°C), 1850 to 2005, relative to the 1961 to 1990 mean for CRUTEM3 updated from Brohan et al. (2006). The smooth curves show decadal variations … The black curve from CRUTEM3 is compared with those from NCDC (Smith and
Reynolds, 2005; blue), GISS (Hansen et al., 2001; red) and Lugina et al. (2005; green). See http://
www.ipcc.ch/
ISI Congress, Dublin, August 26th 2011 5
Why the need to revisit?
•
Global surface temperature records are key line of evidence
•
Backed by many other indicators (ice loss, humidity increases,
sea level increases etc.)
BUT:
•
Link from sta6on data to global averages is indirect (sta6s6cal
problem – reconstruc6on depends on methodology used)
•
Data holdings are dispersed with poor provenance
•
Few user tools: access difficult for non-‐specialists
•
Concerns over: evolving sta6on networks; changes in
measurement technique / sta6on loca6on / recording
prac6ce (oYen undocumented); urban “heat island” effects;
etc. etc.
•
Increasing role of “blogosphere” & cri6ques outside peer-‐
Inhomogenei$es: annual mean min temp
at Reno, Nevada, USA
Issues not well understood …
•
Not all commentators / policymakers understand the
issues:
The big ques$on
•
Can we create a process that leads to a suite of well
characterized es6mates of land surface temperatures
that can be used to answer scien6fic and societal
demands of the 21
stCentury?
–
Open and transparent
–
Becer understanding of fundamental instrument
performance and measurement proper6es
–
Consistent performance evalua6on
–
User tools
–
Not just monthly at the largest scales. Daily, sub-‐daily,
How the ini$a$ve was started
•
2010 UK Met Office Submission to World Meteorological
Organiza6on Commission for Climatology
– Call for crea6ng new suite of products to meet 21st Century
demands / expecta6ons
•
September 2010 ins6ga6on workshop, UK Met Office,
Exeter UK
– 80 interna6onal experts including climate scien6sts,
metrologists, sta6s6cians, soYware engineers
– White papers posted online and public comments solicited
– Agreed project outline and governance structure
– Agreed outcomes published in Bull. Amer. Met. Soc. doi:
Databank Progress
•
Working Group ins6gated
–
Data rescue task team
–
Provenance and version control task team
•
Development version posted
–
hcp://www.gosic.org/
GLOBAL_SURFACE_DATABANK/GBD.html
•
First version release and accompanying
Stage 0 data in hardcopy / image
format
•
Millions of rescued images in the NOAA
Foreign Data Library
•
2000+ boxes of data in the NCDC library
•
Holdings in other libraries and repositories,
par6cularly former colonial powers
•
Holdings literally ropng away or seen as a
nuisance in many countries
Data currently digi$zed
•
Numerous public holdings
–
NCDC
–
UCAR
–
RIHMI
–
Regional and na6onal holdings
•
Some restricted for commercial or geopoli6cal
Stage 2 – common format
Stage 3
•
Same format as stage 2
•
One unique version for each sta6on –
recommended version for most users
•
Forms basis for crea6on of analogs (see later)
•
Protocols used in merging sources are as yet to
be finalized
•
Provenance tracking will ensure an unbroken
chain to earlier stages
Partnerships essen$al
• Bring together exis6ng efforts, augment and ensure pull through. e.g. ACRE ( Atmospheric Circula6on Reconstruc6ons over the Earth ) project (hcp://
www.met-‐acre.org/), Interna6onal Environmental Data Rescue Organisa6on
(www.iedro.org) and other na6onal / interna6onal programs.
• Pursue innova6ve approaches (crowdsourcing building upon success of oldweather.org etc.)
• Build on ICOADS model for sea surface temperatures (hcp://
icoads.noaa.gov/) – easy submission and access to data to recognized global
repository
• Recognize key partners and contribu6ons
• And you can also help …
– www.surfacetemperatures.org/databank
– hcp://www.surfacetemperatures.org/databank/DataSubmission-‐Stage1-‐Guidance.pdf?
acredirects=0
Mul$plicity of data products
•
Structural uncertainty is the key
– Raw data is far from traceable to interna6onal measurement
standards
– Data artefacts are numerous and have myriad causes
– Metadata is patchy at best
– Data is discrete in both space and 6me
– No “how to” … rather very many cases of “it may work …”
– Mul6ple subjec6ve decisions required even in automated
procedures (thresholds, periods, test type etc.)
– Different approaches may have different strengths and
weaknesses
• Need mul6ple independent efforts with different choices
– Sta6on selec6on
– Time and space resolu6on
– Quality control choices
– Homogeniza6on decisions
– Averaging procedures
• Should not just be climate scien6sts as need broad range of
approaches
– Sta6s6cians, metrologists, soYware engineers, ci6zen scien6sts etc.
• Dis6nct approaches pinpoint key uncertain6es so redundancy is
of fundamental scien6fic value …
• … at least providing the underlying algorithm is understood
Benchmarking and assessment
•
With real world data we do not have the luxury of
knowing the truth – we cannot measure
performance of a specific method or closeness to
real world truth of any one data-‐product.
•
Instead focus on performance of underlying
algorithms
•
Consistent synthe6c test cases, simula6ng real world
Benchmarking progress
•
Working group and task teams ins6gated
•
Create c.10 analog-‐error-‐worlds
– Climate model basis (maintains plausible far field correla6on
structure) tweaked with real sta6on climate characteris6cs
– Add in random and systema6c errors to approximate the real
world error structures which may exist
– Algorithm assessment based upon ability to recover original
data
– Error structures should enable answering a range of ques6ons /
assump6ons regarding the true error to avoid over-‐tuning
•
Analogs to be made available Nov. 2012 based upon version 1
Benchmarking example
•
For USHCN (lower 48 states)
•
Step changes and white noise
– 4 iden6cal error structure on 2 20th Century runs, control and
emission halt runs from different models
– No errors (do no harm)
– Predominantly frequent small breaks (hard)
– Perfect metadata, large breaks (easy)
•
100 member perturbed ensemble of the NCDC pairwise
algorithm
•
Considera6on solely of 6meseries and trends
•
Analyses that follow are for predominantly frequent small
Area averaged $meseries and trends
Open issues on data product
provision
•
Data formats
•
Degree of user interac6on
•
Ability to create graphical and tabular output
on the fly
•
Limited progress to date
–
Largely a reflec6on that this data provision is
some way down the road?
Progress to date summary
•
Steering commicee set up
–
Terms of reference
–
World Meteorological Office and The Interna6onal
Environmetrics Society sponsored (seeking Interna6onal
Bureau of weights and measures)
•
Working groups on databank and benchmarking ac6ve
–
Databank prototype made public and data sources coming
in.
•
Implementa6on Plan published
•
Progress documented on ini6a6ve website at
You can help make this a success
•
Help to find raw data sources
•
Come up with novel ways of analyzing the
data
•
Partake in the benchmarking exercise
•
Help to make a data portal a reality
Background materials
•
Links to other relevant organiza6ons (finite)
–
hcp://www.met-‐acre.org/
-‐ ACRE – data rescue,
reanalyses, end-‐user engagement
–
hcp://iedro.org/
-‐ IEDRO – data rescue
–
hcp://icoads.noaa.gov/
-‐ ICOADS – marine
databank
–
hcp://www.data-‐rescue-‐at-‐home.org/
-‐ data
Ecological Phenological Health & Disease Reinsurance Climate Monitoring Model Validation Environmental
Assessments Extremes, Impacts & Risks Water resources Agricultural Forestry Energy Marine operations Fisheries Cultural landscapes and
built heritage Education Global historical reanalysis
56 realisations every 6 hours, 2⁰ x 2 ⁰ spatial resolution
O p en Pl atfo rm Me t O ffi ce -T SB -I B M PR EC IS D ow nsca lin g H ig he r R eso lu tio n Citizen Science Crowd Sourcing
(eg. Oldweather.org
Data rescue@home)
Global surface weather observations over the last