1. ecosystem Modelling And Scaling
infrasTructure (eMAST)
- Where models and data become one
Presentation by Brad Evans based on contributions by Colin
Prentice, Michael Hutchinson, Gab Abramowitz, Ben Evans,
Rhys Whitley, Julie Pauwels
5. eMAST’s objectives 2013-2015
DELIVER research data infrastructure to
integrate TERN (and other) data streams on the
National Computing Infrastructure
ENABLE data assimilation, model evaluation and
accreditation and ecosystem model optimization
DRIVE advances in ecosystem science, impact
assessment and land management
6. Driving science questions
CARBON: How much CO2 is exchanged? How
much carbon can be stored and where?
WATER: What drives water use by ecosystems,
and runoff in rivers?
CLIMATE CHANGE: How does it change the
rules?
LAND MANAGEMENT: What will work, in a
changing climate?
7. More driving science questions
FIRE: What are the risks? How can they be
mitigated?
CLIMATE FEEDBACKS: How will ecosystem
changes influence the exchanges of carbon,
water and energy with the atmosphere?
BIODIVERSITY: What species are threatened?
Where are likely refugia? Is there a tipping
point?
8. What eMAST is delivering
High-resolution data products: climate, canopy
conductance, water use, primary production
Tools for interpolation, downscaling, upscaling,
hindcasting, forecasting
A state-of-the-art data assimilation system for
ecosystem model optimization
Software for model evaluation (based on PALS)
Top-level ecosystem drivers and targets for
models
11. ANUClimate
A NEW approach to interpolating our national network
0.01 degree climate surfaces
Who? Professor Mike Hutchinson (ANU)
12. Climate data sets (1 km)
Tmin
Tmax
vp
P
daily
✔
1970-2011
✔
✔
✔
monthly
✔
1970-2011
✔
✔
✔
✔
mean
monthly
pan
evap.
wet
days
✔
✔
✔
✔
✔
✔
solar
rad.
wind
speed
✔
✔
13. ANUClimate
When? Delivery timeline…
Complete set of Climate and
Bioclimatic data available on
RDSI
RDSI opendap netCDF CF
& Metadata store complete
= public release
Data starts propagating
to RDSI*
ADVANCED USER ACCESS
DOI’s NOT YET AVAILABLE = NO PUBLISH
30 Nov 2013
24 Dec 2013
31 Jan 2013
*Currently experiencing delays in RDSI allocation – delays in the Raijin cloud roll out etc…
14. ANUClimate
What is different?
• Improved ‘background-anomaly-interpolation’
approach
•
•
•
Temperature and both positive and zero rainfall
can be effectively interpolated by the thin plate
splines method - with adaptive capacity !
Monthly means, topographically corrected yield
influence of atmospheric processes and terrain
Significant improvement over both direct (nonanomaly) and current anomaly approach
• Coastal proximity: A new ‘proximity to coast’
modifier captures marine perturbation of
climate
15. ANUClimate
What can we expect?
• Temperature estimates improved by around
25% compared to Jones et al. 2009 (RMSE
cross validation)
• Precipitation estimates a modest, but
significant, improvement (7-15% RMSE cross
validation)
The model makes no further improvement on
accuracy beyond the 1km mark !
19. ecosystem Production in Space and
Time: ePiSaT
eMAST: How does gross primary
productivity (GPP) vary in space and time
across Australia?
Colin: How can we ‘simply’ estimate
GPP across Australia?
What data does TERN provide that
might be useful for addressing this
research question?
20. User workflow: ePiSaT GPP
Choose the ePiSaT
model from the TERN
portal
Produce continental
scale estimates of GPP
and evaluate them
Obtain OzFlux data via
the TERN/ OzFlux
portals
Obtain climate (eMAST)
and satellite data
(AusCover) to scale the
ePiSaT parameters
Run the ePiSaT model –
generate estimates of
ecosystem parameters,
evaluate them
24. OzFlux: Flux partitioning
1
Respiration
R=
1
Data filtering:
Removal of outliers
etc.. Gap filling of
PAR (PPFD) for GPP
Amax = - 2
Quantum
3
Rectangular
Hyperbole
3 parameter
1
Assimilation
2
2
Efficiency
2
Φ=
3
3
FC = R -
Amax * Φ I
Amax +Φ I
25. ePiSaT v 1.0 : Tower GPP
GPP =
Amax * I
Amax + C
Where: Amax is the maximum rate of carboxylation, I is PAR (PPFD) and C = parameter
3 from the rectangular hyperbola described in the previous slide
26. ePiSaT v 1.0 : Map GPP
GPP = fAPAR *I* LUE
Where: fAPAR is the fraction of absorbed photosynthetic active radiation,
I is PAR (PPFD) and LUE is light use efficiency derived from the relationship of
Tower GPP (previous slide) and fAPAR and I.
ePiSaT v 2.0 : Map GPP
GPP = fAPAR *I* LUE*WUE*Trange
Where: fAPAR is the fraction of absorbed photosynthetic active radiation,
I is PAR (PPFD) and LUE is light use efficiency derived from the relationship of
Tower GPP (previous slide) and fAPAR and I. WUE and Trange are derived similarly.
30. Plant trait surfaces
•
•
•
•
•
•
•
•
Leaf nitrogen
Leaf phosphorus
Specific leaf area
Leaf area
Maximum plant height
Photosynthesis per leaf
area
Photosynthesis per leaf
dry mass
Leaf stomatal
conductance
Dr. Rhys Whitley
34. Summary: Data-model fusion tools
Data assimilation collaboration with NEON and NCAR, CSIRO,
Macquarie University and the Australian National University
- ACEAS workshop on data assimilation early 2014
eMAST : An R-Package ‘emast’ for the computation and visualization of
bioclimatic indices
ePiSaT : Collaboration with OzFlux and AusCover to model Gross
Primary Production across the landscape, another R-Package ‘ePiSaT’
-ACEAS worskshop on SPEDDEXES
Protocol for the Analysis of Land Surface Models (PALS) for evaluation
of data and models
35. The future of eMAST
Continue delivery of our key datasets through
the RDSI, Data Discovery, Visualization &
Exploitation… consolidation of our tools and
porting them to Raijin.