Flux Measurement and Advanced Modeling

1.) ACME07 regional flux workshop - 24-25 July, 2008, Boulder, CO

Confirmed attendees: Britt Stephens, Teresa Campos, Dave Moore, Ankur Desai, Will Ahue

Pending: Steve Aulenbach (maybe Fri), Dave Schimel, Russ Monson

Location:

7/24 - TIIMES meeting space, FL-2 3rd floor

7/25 - RAF meeting room, Jeffco (tentative)

Agenda: (forthcoming)

Topics:

• Data updates - remote sensing, sfc met, airborne met, CO2, CO, O2, reanalysis, particles
• Modeling activities
• Research goals of each group
• Future funding opportunities
• Publications and presentations - maybe one in the near future?

2.) Ecological Data Assimilation Group - wiki

June 22-23, 2008, Niwot Ridge

Presenters: Ken Davis, Dave Schimel, Dave Moore, Ankur Desai

Notes (modified agenda below with notes from the 10 June 2008 telecon interspersed). Action items noted in RED. Questions in BLUE.

A pile of interesting recent research papers related to ecological data assimilation with flux towers are included at the bottom in the attachments section. Feel free to include others. (Ken I included the Qaife paper that Dave M mentioned). We should make sure students are aware of relevant and recent research in the field. Also our talks may want to highlight a result or two from these. I'm tempted to include a greatest hits of such in my talk.

At the end of the module we would like students to

1. Know the group of techniques typically called data-assimilation or model-data fusion
2. Understand the principles underlying each technique and the strengths and weaknesses of each.
3. Understand how data-assimilation can be used with flux measurements on an ecosystem scale.
4. Understand how data-assimilation of flux measurements could be combined with regional estimates of carbon and water exchange
5. Understand how data quantity and quality impacts the results of data-assimilation experiments.

Tuesday, July 22

• 9:00 AM-10:30 AM Ken Davis Model-Data Assimilation (Introduction) 10:30 AM-10:45 AM Break
  • What is data assimilation?
  • • least-squares fit
    • fit the model parameters to the observations
    • define terms - model parameters, model structure, model state variables, model prediction
  • Historical development of ecological data assimilation
    • model construction - piecing together process-level understanding
    • process-level measurements and experiments to determine model parameters
    • popularization of whole-system flux measurements
    • few-variable analyses of flux measurements
    • model-data "fusion" or "data assimilation"
      • meteorological - model structure and parameters are known, but state is chaotic - alter the model state variables to keep close to the observations through time
      • ecological - model structure and parameters are not well known. State varies slowly (e.g. soil carbon, above ground biomass).

• • What purposes are served by model-data assimilation?
    • create a best fit to a data set
    • test model structure - can model A, B, or C best match a data set?
    • test data sources - do various data yield consistent constraints on model parameters? do they constrain essential processes?
    • evaluate model complexity - how many variables are needed to simulate a given data set?
    • **develop predictive skill - what will fluxes be in the future?
    • challenge of encompassing the processes essential for prediction
      • disturbance (fire, insects, wind, humans), succession and regrowth, nutrients and hydrology - not just light and temperature
    • (link to Moore's talk)
  • What role do flux towers play in carbon cycle/biogeochemical research?
    • lack of predictive capacity for the carbon cycle story
    • complementary methods and spatial/temporal scales story - need to understand sites and the globe
    • possible approaches for merging information from multiple measurements
    • (link to Desai's talk)
    • what is the value of prediction in climate change management?
  • How do we do data assimilation?
    • global vs. local methods
    • pdfs vs. best values of parameters
    • solving for state variables vs. parameters
    • error sources and impact on likelihood or cost functions
      • distribution, homo/heteroskedasticity, spatial or temporal correlation
    • (link to Schimel's talk)
      
      

• 10:45 AM-12:00 PM Dave Schimel Model-Data Assimilation (Statistical Foundation)
  • Stats behind data assimilation
  • Bayes theorem / Gaussian assumptions
  • Errors in models and observations, systematic and random, covariances, etc...
  • ACTION ITEM: Need to confirm with Dave S (or others) if this will work - Ken to contact
  • Dave - what do you intend to present - does this fit?
    
  • If Dave is unable or wants to tag team, we could have: Doug Nychka, Dave Baker, Dan Ricciuto (not local), Nathan Urban (not local, Ken's postdoc), Andrew Richardson (may have already left by then) - to provide more stats background...
• 12:00 PM-1:30 PM Lunch
• 1:30 PM-2:45 PM Dave Moore Model-Data Assimilation (Current Science: Getting the Most Out of Data at Flux Tower Sites)
  • ACTION ITEM: Detailed draft / powerpoint from Dave M
  • Limitations of measurements and utility of ecosystem models in research
  • Key Ecosystem models used currently - role of model complexity
  • Use of data assimilation to
    • Estimate fluxes (e.g. DALEC)
    • Estimate parameters (e.g. SIPNET)
  • Using different data streams – types of data we might want to use
  • CO2 only vs CO2/H2O fluxes using SIPNET
  • Incorporating of data limitations into assimilation strategies
  • General model for data assimilation using ecosystem models at Flux sites - 20 site parameterization (network) - feeds into Ankur's sptial talk
  • Using an ecosystem model constrained with data to project future change
  • MODIS and Remote sensing approaches
  • Advantages and limitations
• 2:45 PM-3:00 PM Break
• 3:00 PM-4:15 PM Ankur Desai Model-Data Assimilation (Spatial Perspective)
  • ACTION ITEM: More detailed draft from Ankur (maybe as powerpoint)
  • Rehash of data assimilation from an observationalist view (regional focus)
    • What's the point of flux tower observations?
    • Why a regional focus?
    • Notes from my Ameriflux07_Desai.pdf and Desai-AOS907-Sp2008.ppt
  • State vs parameter estimation
    • Doug Nychka’s Carbon data assim talk notes on this are good
    • Noise in flux data (link to Richardson talk)
  • Regional perspective – flux tower mesonet approach - various upscaling
    • Cheas-Sipnet - upscaling - role of parameters and micrometeorology
    • ChEAS? -ED - Role of disturbance and things you can't see at flux sites
    • ChEAS? -TRIFFID (D Ricciuto) - What is interannual variability and its coherence?
    • Cook/Sulman water table work - MODIS/TREES - carbon + hydrology
    • ACME has a way to integrate flux towers and other data
  • Introduction to tomorrow’s hands on assignment
  • Discussion

Wednesday, July 23

• 9:00 AM-12:00 PM Hands-On: Model-Data Assimilation I
  • Part 1: Introduction to SIPNET (build on ASP version)
  • Open input files
  • Run Model
  • Change parameters
  • Sensitivity tests
  • Set up groups for data assimilation experiments
    • Let's students in groups of ~4 choose an experiment
    • ACTION ITEM (Dave M): Computing setup - we should have Excel and Matlab, in addition to a text editor, shell, C compiler
    • Note: Macs have shell, text editor, and C compiler (gcc) out of the box
    • Do we have enough helpers? Dave M suggests Laura ? from Russ' lab (new postdoc learning the model)
    • Possible experiments listed below

• 12:00 PM-1:30 PM Lunch
• 1:30 PM-4:30 PM Hands-On: Model-Data Assimilation II
  • 1. Variability in parameters depending on how much data you use
    • a. Random removal of data
    • b. Assimilation of each year simultaneously
    • c. Assimilation of different numbers of years
    • d. Contrast day/night vs daily aggregations
  • 2. Regional BGC experiment (http://www.cgd.ucar.edu/~dmoore1/regionalBGC)
    • a. Set up half daily assimilation of CO2/H2O fluxes
    • b. Choose one or more scaling parameters
    • c. Compare model output to regional estimates of GPP and NEE
  • 3. Compare parameters from more than one site - ChEAS? , Harvard, etc...
  • 3. Gap filling
  • 4. Assimilate different data streams into model ???? (e.g., NACP datasets, ChEAS? datasets) - H2O? and CO2
  • 5. Predicting future
  • 6. Model structure / parameter sensitivity - turn on/off model components
  • From Ken: I would be interested in having the tutorial explore different problems that can be addressed with flux data assimilation if we can do this in a day (ha ha). E.g. mapping of fluxes, model structural evaluation, gap filling, prior for atmospheric inversions, info content of data sources. And I could try to create an overview of this in my presentation. I would support using as a base the Sipnet tutorial that Ankur et al presented last summer for our ASP colloquium.
• 4:30 PM-6:00 PM Free Time
• 6:00 PM Dinner