Observations for Carbon Data Assimilation Scott Doney Woods Hole Oceanographic Institution Where does the “data” come from for “data assimilation”? Atmospheric CO 2 data initial conditions innovation terms error covariance terms Land and ocean carbon data flux estimates for priors process/mechanistic information Data is NOT a black box!!

Atmospheric Inverse Modeling of CO 2 Concentration (observed samples) Transport (modeled) Sources & Sinks (solved for) +=

How data enters the problem (1) Variational Assimilation Adjust model state “ x ” (atmospheric CO 2 field) to minimize cost function J : Deviation of x from “background” Deviation of x from “observations” So where do we get: y 0 (t) data for innovation (model-data misfit) R data error covariance B model error covariance

Some Issues to Ponder Atmosphere CO 2 drastically under-sampled original design for marine background air mostly discrete surface samples NWP deg. freedom O(10 7 ); 6 hourly observations O( ) CO 2 weekly observations O( ) Small signal on large background variability surface North-south difference ~2.5 ppmv zonal continent to land contrast <1 ppmv measurement precision accuracy in time & across stations and networks reduce systematic biases!!

Representativeness of data y 0 “footprint” of observation & mismatch with model grid local heterogenity or point sources aliasing of unresolved frequencies/wavenumbers (e.g., diurnal cycle) data selection (i.e., exclude “unrepresentative” observations) Some Issues to Ponder

CO 2 Concentration in the Outer Damon Room, NCAR Mesa Lab, 2/7 – 2/9/06 CCSM Climate Working Group Board of Trustees Reception National Science Board Breakfast ASP Reviews

Multiple Time/Space Scales

In-situ Atmospheric Observing Network Discrete surface flasks (~weekly) Continuous surface (hourly) observatories Tall towers continuous (hourly) Aircraft profiles (~weekly)

Surface Observatory

Seasonal CO 2 : Continental vs Oceanic Sites CO 2 seasonal cycle attenuate, but still coherent, far away from source/sink region Peak-trough amplitude of seasonal cycle ~ 30 ppmv (~10%)

TURC/NDVI Biosphere; Takahashi Ocean ; EDGAR Fossil Fuel [U. Karstens and M. Heimann, 2001] Weather Cycles: CO 2 is variable 10 ppmv [LSCOP, 2002] Modeled: one day in July On  T~5 days,  x ~ 1000 km CO 2 variability in the boundary layer~ 10 ppmv (3 %)

Diurnal CO 2: Highly variable in boundary layer Diurnal cycle of photosynthesis and respiration  > 60 ppmv (20%) diurnal cycle near surface Varying heights of the planetary boundary layer (varying mixing volumes) Calm night: stably stratified boundary layer Well-mixed PBL

Vertical Profiles (free troposphere) DATE

Aircraft Campaigns COBRA 2000

Surface Fluxes => Atmospheric CO2

Orbiting Carbon Observatory (Planned Fall 2008 launch) Estimated accuracy for single column ~1.6 ppmv 1 x 1.5 km IFOV 10 pixel wide swath 105 minute polar orbit 26º spacing in longitude between swaths 16-day return time

How data enters the problem (2) transport fluxes Deviation of initial conditions from “prior” Deviation of x from “observations” Deviation of fluxes from “prior” Separating transport, initial conditions & surface fluxes Analysis at time i => forecast at time i+1 4D Variational methods: adjust initial conditions to better match future data initial conditions

NEE = GPP – R eco NEE is measured at the tower Ecosystem Respiration typically based on nighttime NEE & air temperature & ? R eco = R hetero + R auto GPP Photosynthesis (Daytime only) CO 2 Adapted from Gilmanov et al.

Eddy-Flux Towers Vertical velocity (mean and anomaly) CO2 concentration (mean and anomaly) Vertical CO2 flux

FluxNet Tower Sites

Diurnal and Seasonal Cycle

Time-scale character of carbon fluxes Diurnal Seasonal 1.Variability is at a maximum on the strongly forced time scales 2.They have an annual sum of ~0 3.Modeling the carbon storage time scales is much more difficult

Tall-Tower Footprint (Wang, 2005) South-North Direction (km) West-East Direction (km) Barren Shrubland Lowland Shrub Wetland Forested Wetland Mixed BL Deciduous Mixed Dec/Con Mixed Coniferous Aspen Grassland Pine Urban/Dev/Ag Water

Forest Inventory Analysis: Slow Process Observations Plot-scale measurement of carbon storage, age structure, growth rates: 170,000 plots in US! Allows assessment of decadal trends in carbon storage

Air-Sea CO 2 Flux Estimates Takahashi -undersampling & aliasing of surface water  pCO2 -transfer velocity k empirically derived from wind speed relationships F = k s(pCO 2 air - pCO 2 water ) = ks  pCO 2 ThermodynamicsKinetics transfer velocity

Gas Transfer Velocity

Generating a Global Flux Map

Chavez et al. In-Situ Sensors and Autonomous Platforms Moorings/drifters (  pCO 2, pH, DIC, NO 3 ) Riser et al. Profiling ARGO floats (also AUVs, caballed observatories)

-Global baseline (hydrography, transient tracers, nutrients, carbonate system) -Improved analytical techniques for inorganic carbon and alkalinity ( ±1-3  mol/kg or 0.05 to 0.15%) -Certified Reference Materials -Data management, quality control, & public data access JGOFS/WOCE global survey (1980s and 1990s)

Ocean Inversion Method -Ocean is divided into n regions (n = 30, aggregated to 23) -Basis functions for ocean transport created by injecting dye tracer at surface in numerical models

Basis functions are model simulated footprints of unit emissions from a number of fixed regions Estimate linear combination of basis functions that fits observations in a least squares sense.  Inversion is analogous to linear regression footprintsfluxesobs Premultiply both sides by inverse of A Simple Inversion Technique estimated fluxes