Marine Biogeochemical Modeling

Transcription

1 Marine Biogeochemical Modeling Scott Doney (WHOI) NCAR ASP Colloquium 2009 Talk Outline Science Motivation Philosophy of Modeling Model Construction Evaluation Against Data Supported by:

2 Ocean Biological Pump Combined biological processes which transfer organic matter and associated elements to depth -pathway for rapid C sequestration Remove C from surface ocean & atmosphere -turn off bio pump and 200 ppmv increase atm. CO 2 Anthropogenic CO 2 uptake currently controlled by ocean circulation For overview on scope and history of ocean biogeochemical models see: Doney, S.C., K. Lindsay, J.K. Moore, 2003: Global ocean carbon cycle modeling, Ocean Biogeochemistry, ed. M. Fasham, Springer,

3 Dissolved Inorganic Carbon Dissolved Oxygen Distributions of ocean carbon, oxygen & nutrients driven by circulation and sinking & respiration/ remineralization of sinking organic particles

4 Rising Atmospheric CO 2 Ice core data -strong evidence for human causation -highest level in at least last million years Thus human beings are now carrying out a large scale geophysical experiment Revelle and Suess, Tellus, 1957

5 Fate of Anthropogenic CO 2 Emissions ( ) 1.5 Pg C y Pg y -1 Atmosphere 46% Pg C y Pg y -1 Land 29% 2.3 Pg y -1 Oceans 26% Canadell et al. 2007, PNAS;

6 Anthropogenic CO 2 & Ocean Acidification Doney et al. Ann. Rev. Mar. Res Anthropogenic CO 2 Column Inventories (mol m -2 ) Sabine et al. Science 2004

7 CO 2 Effects on CaCO 3 Saturation Ω (aragonite) Ω= [Ca 2+ ][CO 3 2- ] / K sp Δ[CO 3 2- ] = [CO 3 2- ] obs - [CO 3 2- ] sat Present 2100 Orr et al. Nature (2005)

8 Future Climate Projections Major uncertainties: -CO 2 emissions (social, political, economic, geological) -atmospheric CO 2 (carbon sinks, climate-carbon feedbacks) -climate sensitivities (clouds, water vapor) IPCC (2007)

9 Frolicher et al. Global Biogeochem. Cycles 2009

10 Coupled Model Uncertainties Strength of Ocean CO 2 Sink Sensitivity to Climate Warming 20 years of current carbon emissions Friedlingstein et al. J Climate 2006

11 Climate Change Impacts on Primary Productivity Steinacher et al. Biogeosciences Disc. 2009

12 Ocean Climate Responses & Feedbacks CHANGING CLIMATE Higher atmosphere CO 2, altered ocean properties sea-ice & circulation OCEAN CO 2 SINK Ocean acidification BIODIVERSITY BIOGEOGRAPHY PHYSIOLOGY ECOSYSTEM SERVICES Fisheries, tourism, shore protection, BIOGEOCHEMISTRY Carbon & nutrient fluxes, CO 2 & other greenhouse gases ECOSYSTEMS Food webs, energy flow

13 Ocean Climate Responses & Feedbacks OCEAN MITIGATION Ocean iron fertilization, direct CO 2 injection CHANGING CLIMATE Higher atmosphere CO 2, altered ocean properties sea-ice & circulation OCEAN CO 2 SINK Ocean acidification BIODIVERSITY BIOGEOGRAPHY PHYSIOLOGY ECOSYSTEM SERVICES Fisheries, tourism, shore protection, BIOGEOCHEMISTRY Carbon & nutrient fluxes, CO 2 & other greenhouse gases ECOSYSTEMS Food webs, energy flow Doney et al. Deep-Sea Res. II 2009

14 "I am never content until I have constructed a mechanical model of the subject I am studying. If I succeed in making one, I understand; otherwise I do not. - Lord Kelvin "People don't understand the earth, but they want to, so they build a model, and then they have two things they don't understand, - Gerard Roe in The Whale and the Supercomputer by C. Wohlforth Data Complex models Simple models

15 The Interdisciplinary Conundrum Physics (relatively) simple set of (mostly) known governing equations (Navier-Stokes) => complex phenomenon unresolved scales Chemistry mass balance equations chemical fields integrate over time/space variability don t uniquely identify process/mechanism Biology information as stories and conceptual pictures historical/evolutionary contingency no biological Navier-Stokes aggregate complicated dynamics across multiplescales (genes, cells, populations, ecosystems) Humans

16 Why do we build/use models? -Quantitative dynamical framework Are different data sets, rate estimates consistent? -Design of experiments & observing systems What, where and when do we sample? -Hypothesis testing If we add or change X, what happens? -Forecasting What will the ocean look like at some point in the future? -Solving for unknown parameters and rates Given things we can measure, can we estimate can we estimate properties that are difficult to measure?

17 Stocks versus Rates C dc/dt Estimated from data Unknowns

18 What is a model? -regression curves variable Y is a function of other variables X and regression parameters p: Y=f(variables, parameters) Chl = p 0 + p 1 T -forward models (often time dependent) Integrate forward in time to find Y: dy/dt = circulation + f(variables, parameters, forcing) d Phyto /dt = circulation + µ*phyto*(1-e -αe/µ ) -λphyto -inverse models Invert the problem to find parameters from data: Parameters = f(circulation, data, forcing)

19 What is a model? (continued) -diagnostic versus prognostic models In diagnostic models, some variables may be prescribed based on observations: e.g., satellite chlorophyll => ecosystem model (no equation for dchl / dt) -data assimilation Combine model equations and observations in a dynamically consistent fashion e.g., weather prediction analysis = f(model forecast, observations) Prognostic, forward models needed to project into the future

20 From Word Problem to Equations Phytoplankton levels depends on nutrient inputs Perturbations relax back to some stable background level dp dt = µ 1 P 0 C P p dp/dt µ 0 >0 net growth Functional form Logistic model State variable (concentrations) P (mmol C/m 3 ) phytoplankton Parameters µ 0 (1/d) and C p (mmol C/m 3 ) P 0 C p P C p <0 net loss time

21 dp dt = µ N 0 k N + N Simple NPZ Model ( 1 e αe / µ 0 )P g P k P + P Z m P P Nutrient limitation Light limitation Grazing dz dt = ag P Z m Z Z k P + P Mortality dn dt = µ N αe / µ0 P 0 ( 1 e )P + (1 a)g Z + m P P + m Z Z k N + N k P + P Three coupled ordinary differential equations Mass conservation

22 How do you estimate parameters and functional forms? Laboratory & field incubations P-E curves; nutrient uptake curves elemental stochiometry Comparative analysis allometric relationships Tuned or optimized against field data mismatch between parameters and data cross-site comparison Previous models

23 Adding Circulation Control volume u P in P t u P out P t + u P κ 2 P = RHS advection diffusion biological source/sink terms

24 Models have time/space scale limits Global Climate Model Regional Coastal Model Dickey (2003) Computational costs scale as (length) 3 to (length) 4 and (time) Typically get ~2-3 decades in space (more in time) Can not resolve all scales; parameterize sub-grid scale

25 Coupled Eco-biogeochemical Elements Physics (flow field; mixing) equations for resolved flow; level of approximation (e.g., primative equations; quasi-geostrophy) forcing (winds, heat & freshwater fluxes, light, tides) parameterization of unresolved scales (mixing) model architecture (e.g., horizontal vs. isopycnal) Chemistry (CO 2, O 2, nutrient fields) air-sea gas exchange elemental stoichiometries trace metal deposition and scavenging Biology primary production, respiration, remineralization community structure and succession bio-optics, etc.

26 State of the Art Model -Aggregate into trophic levels/functional groups -Rates/processes from limited culture/field studies -Many aspects empirically based -Data poor for validation (rates, grazing, loss terms)

27 Ocean ecology and biogeochemistry are (still) data-driven sciences data model model How do we avoid the trap of: false models tested by inadequate data John Steele

28 Assessing Model Skill Poor Model Skill Stow et al. J. Mar. Systems 2009

29 Good Model Skill Stow et al. J. Mar. Systems 2009

30 Model Modern Air-Sea CO 2 Flux Looks pretty good test Takahashi (2002) χ (chi) by eye Doney et al. Deep- Sea Res. II 2009 Doney et al. J. Mar. Systems 2009

31 Look at the magnitude & structure in model-data residuals Stow et al. J. Mar. Systems 2009 Ducklow et al. Ann. Rev. Mar. Res. 2009

32 Bias Correlation rms Error Doney et al. J. Mar. Systems 2009 Stow et al. J. Mar. Systems 2009

33 Taylor Diagram Taylor J. Geophys. Res. 2001

34 Ocean Carbon Model Intercomparison Project (OCMIP) Mostly analysis of tracer & ocean circulation skill; Data metrics for choosing subset of skillful models Matsumoto et al. (Geophys. Res. Lett, 2006)

35 Gregg et al. J. Mar. Systems 2009

36 Data Assimilation Methods (adjust model prognostic variables to data values as integrate forward in time) -nudging -variational methods -Kalman filter Parameter Optimization (adjust model parameters; multiple iterations of forward model) Friedrichs et al. J. Geophys. Res. Oceans 2007 Gregg et al. J. Mar. Systems 2001

37 Modeling Methods for Marine Environments David M. Glover, William J. Jenkins & Scott C. Doney -data analysis -modeling techniques -ocean examples and applications -MATLAB based demos and code -detailed web notes (and perhaps some day a book) (

38 J. Marine Systems Special Issue on Skill Assessment for Coupled Biological / Physical Models of Marine Systems Vol. 76, Issue 1-2, 2009

39