Australia’s National Carbon Accounting System - Land Cover ... · NCAS Landcover change project...
Transcript of Australia’s National Carbon Accounting System - Land Cover ... · NCAS Landcover change project...
Australia’s National Carbon Accounting System
- Land Cover Change Project
Department of Climate Change, CSIRO, RMIT, Others
Monitoring at continental scale…
…to identify timing and location of an event…
…forming time-series forest extent information…
NCAS Landcover change project(NCAS- National Carbon Accounting System)
• For Australia, land based emissions are significant and a remotesensing approach for area of landcover change was adopted
• Assessment of Landcover (“tall woody/other”) change for period 1972-2001 (2002) (2004) (2005) (2006)
{1972,1977,1980,1985,1988,1989,1991,1992,1995,1998,2000,2001,2002,2004,2005}
• Approximately 3690 (4000) (4400) (4800) (5400) Landsat TM and MSS scenes occupying 10 (11) (12) (13) (14) “time” slices
• Groups - AGO, CSIRO, RMIT, PrivateEnterprise + other state and federal agencies
• benefit/cost 16:1 (ICE)
• Improved estimates of LULUCF – kyoto
• {forest change, trend, plantation type, urban extent, sparse covers}
National Carbon Accounting System…
• “Ground truth”
• Analysis
• Processing
• Validation
Quantitative methods are applied – scalable
• Remote sensing applications development since late 1980’s
• Terra bytes of data (manageable)
• Few CPU years of compute power – use clusters
• Semi-automated (with QA embedded)
• Project partners: public – private mix
• approach is scalable (geographic extent, sensor resolution)
Some figures and comments…
Also for Natural resource management:eg Change in area + trends…
Forest trends / disturbance
Hot colours = -ve trend
Cool colours= +ve trend
Supporting Natural Resource Management Communities
Supporting Natural Resource Management Communities…
1. Pilot areas to establish methodology2. Specification of manuals, QA* and contractual arrangements3. Landsat scene selection4. Rectification base5. Calibration base6. Time series rec/cal7. Training and validation data8. Analysis to derive indices9. Individual year decision boundaries => p(woody/non-woody)10. Time series classification11. Attribution (all change -> human induced change)12. Validation, continuous improvement
Phase 1 - Project stages (1999-2001)“Establishing continental system”
Phase 2 – (2002-2005)
Method responses to “Continuous improvement”
• Reducing operator effort – streamlining
• Identification and reduction of errors (eg. Operator variability, interpretation errors)
• Increasing processing efficiencies
• new training data targeted in areas where interpretation was a problem
• Increased number of discriminant vectors
• Automatic determination of classifier parameters
• Terrain illumination correction
• Spatial context added to time series model
• Texture measures
Phase 3 – (2006, 2008)
“Possible improvements”
• Increase time series density• 2nd order time series models (eg n-f-n very unlikely
configuration in short term)• More sophisticated spectral modelling (y|l), classifier –
mixture modelling, mlp
• Terabytes of data• Approximately 1 CPU year to run present time series models• Would need order(s) of magnitude increase in processing
power
• MPI, beta version running,• 1600 iridium cpu cluster => 3 orders of magnitude
1. Systematic acquisition of high resolution data and API
2. Local knowledge + Continuous improvement – on each regular update, those areas or dates identified by users as requiring attention are re-examined, in the light of local information provided by user
Validation
Groups - AGO, CSIRO, RMIT, PrivateEnterprise + other state and federal agencies
• Local: by pilot study (as part of method development)
• Systematic: acquisition of high resolution data and API – By 1:1000000 mapsheet, error matrix + meta data describing types of (omission and commission) errors most commonly occuring(RMIT)
Sources of data used for training
Resources
• http://www.climatechange.gov.au/ncas/publications/index.html
Comments on validation
• Pilot study validation of methods (as part of methods development).
• Independent, systematic validation by API of random sites acrossAustralia
– Forest extent
• Independent systematic validation by API and field work– Plantation type
• Phase 1 validation prompted series of continuous improvement steps
• Pilot study validation of methods (as part of methods development).
• Independent, systematic validation by API of random sites for multiple dates (at least two dates) across Australia
– Forest extent and change– Sparse covers