Building from the bottom-up and learning as we go: data requirements for upscaling ecosystem function


Download 5.29 Mb.
Sana15.01.2020
Hajmi5.29 Mb.
  • CarbonFusion meeting, 4 or 5 June 2008
  • Building from the bottom-up and learning as we go:
  • data requirements for upscaling ecosystem function
  • Paul Stoy1*, Mathew Williams1
  • 1 School of GeoSciences, University of Edinburgh, UK
  • Jon Evans2, Colin Lloyd2
  • 2 Center for Ecology and Hydrology, Wallingford, UK
  • Ana Prieto-Blanco3, Mathias Disney3
  • 3 Department of Geography, University College London, London, UK
  • Gaby Katul4, Mario Siqueira4, Kim Novick4, Jehn-Yih Juang4, Ram Oren4
  • 4 Nicholas School of the Environment and Earth Sciences, Duke University, USA
  • Intro
  • Motivation
  • 2) Examples
  • Duke sites
  • Tundra site
  • IC
  • 3) Summary
  • What
  • models
  • need
  • Motivation
  • ‘The Leuning 7’ [after Liu and Gupta (2007)]
  • LSMs consist of 7 components:
  • 1) the system boundary, B
  • 2) inputs, u
  • 3) initial states, x0
  • 4) parameters, θ
  • 5) model structure, M
  • 6) model states, x and
  • 7) outputs, y
  • 9) How should the (FLUXNET) flux data be processed?
  • 10) What ancillary data (including EO) can and should be used?
  • Motivate these q’s using the upscaling challenge
  • Intro
  • Motivation
  • 2) Examples
  • Duke sites
  • Tundra site
  • IC
  • 3) Summary
  • What
  • models
  • need
  • The challenge:
  • 3.5
  • 0.0
  • 0 200m
  • LAI
  • Oren et al., (2006) GCB
  • …interpreting ecosystem
  • function from dynamic
  • EC measurements.
  • Example: The Duke FACE
  • Site (PP) measures a footprint with relatively low LAI.
  • NEEA would be ca.
  • 50 g C m-2 y-1 if the tower
  • was located centrally
  • How do we move from leaf to tree to tower to region?
  • N gradient
  • Intro
  • Motivation
  • 2) Examples
  • Duke sites
  • Tundra site
  • IC
  • 3) Summary
  • What
  • models
  • need
  • The challenge (continued):
  • Oishi et al., (in press) AFM
  • The adjacent DBF ecosystem (HW) has:
  • wet & dry subplots, multiple species, LAI variability
  • 95% peak s.w.f.
  • 50% peak s.w.f.
  • sapflux
  • Litter
  • baskets
  • Intro
  • Motivation
  • 2) Examples
  • Duke sites
  • Tundra site
  • IC
  • 3) Summary
  • What
  • models
  • need
  • A small part of a complicated landscape
  • Juang et al., (2007) WRR
  • Stoy et al., (2007) GCB
  • Intro
  • Motivation
  • 2) Examples
  • Duke sites
  • Tundra site
  • IC
  • 3) Summary
  • What
  • models
  • need
  • MODIS GPP algorithm for PP
  • Heinsch et al., (2006) IEEE-TGRS
  • ENF or MF?
  • Savanna?
  • Observational bias
  • (remote sensing)
  • plays a central role
  • for modelling & measurement
  • Intro
  • Motivation
  • 2) Examples
  • Duke sites
  • Tundra site
  • IC
  • 3) Summary
  • What
  • models
  • need
  • Sources of bias (tundra)
  • Burba et al., (2008) GCB
  • Asner et al., (2003) GEB
  • Intro
  • Motivation
  • 2) Examples
  • Duke sites
  • Tundra site
  • IC
  • 3) Summary
  • What
  • models
  • need
  • Left: LAI map of
  • Abisko Tundra (AT)
  • With ½ hr. footprint
  • Right: pdf of tower-measured (daily, black) vs. footprint NDVI
  • Intro
  • Motivation
  • 2) Examples
  • Duke sites
  • Tundra site
  • IC
  • 3) Summary
  • What
  • models
  • need
  • ‘De-biasing (?)’ using a footprint model
  • Intro
  • Motivation
  • 2) Examples
  • Duke sites
  • Tundra site
  • IC
  • 3) Summary
  • What
  • models
  • need
  • Upscaling = preserving information?
  • Stoy et al. (in review) Ecosystems
  • Finding spatial averaging operator(s) that preserve fine-scale information content (IC)
  • [via Shannon Entropy, Kullback-Liebler divergence]
  • IC for parameter space analysis?
  • Intro
  • Motivation
  • 2) Examples
  • Duke sites
  • Tundra site
  • IC
  • 3) Summary
  • What
  • models
  • need
  • NEE Residual Spectrum (mg C m-2 s-1)2
  • 2000.5 2001 2001.5 2002 2002.5 2003 2003.5
  • Year
  • Time Scale (y)
  • Time Scale (y)
  • 10-4
  • 10-3
  • 10-2
  • 10-1
  • 100
  • 10-4
  • 10-3
  • 10-2
  • 10-1
  • 100
  • 2000.5 2001 2001.5 2002 2002.5 2003 2003.5
  • H
  • D
  • W
  • M
  • Y
  • H
  • D
  • W
  • M
  • Y
  • Wavelet half plane model residual analysis: Duke PP and HW
  • Color = residual energy
  • Suggestions for LSMs
  • Problems for upscaling and models
  • Observational bias Measurement bias (and random error)
  • Potential for de-biasing using additional ecological information
  • Future directions / needs for FLUXNET
  • The ‘super site’ concept (e.g. IMECC)
  • Ray’s 20 ecosystems?
  • - We need temporal and spatial data for:
  • Ecosystem structure and
  • (with parameters), function
  • - How much?
  • - Probably just enough to describe
  • ecosystem change over time.
  • How does flux ‘resonate’ with climate?
  • Intro
  • Motivation
  • 2) Examples
  • Duke sites
  • Tundra site
  • IC
  • 3) Summary
  • What
  • models
  • need
  • Intro
  • Motivation
  • 2) Examples
  • Duke sites
  • Tundra site
  • IC
  • 3) Summary
  • What
  • models
  • need
  • Few high frequency
  • (bi-monthly or less)
  • Differences among
  • Veg/climate types
  • We need PFTs after
  • The bi-monthly t.s.
  • Questions?
  • Funding: NERC (IPY)
  • Intro
  • Motivation
  • 2) Examples
  • Duke sites
  • Tundra site
  • IC
  • 3) Summary
  • What
  • models
  • need
  • Early
  • Season
  • Improvement
  • PLIRT
  • gapfilling
  • model
  • (Burba GCB ’08?)
  • Adding data increases confidence
  • State (t)
  • (Shaver et al.
  • Parameters)
  • Initial
  • Forecast
  • State (t+1)
  • g C m-2
  • Cumulative
  • Obs (t+1)
  • Forecast (t+1)
  • Assimilation
  • 77±3
  • 127±2
  • 140±3
  • 168±13
  • model
  • (PLIRT)
  • (Ensemble Kalman Filter)
  • Intro
  • Motivation
  • Model
  • 2) Methods
  • Site
  • Meas
  • Movie
  • 3) Results
  • Model
  • Data
  • assimilation
  • c) FLUXNET

Download 5.29 Mb.

Do'stlaringiz bilan baham:




Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2020
ma'muriyatiga murojaat qiling