integration of biosphere and atmosphere observations yingping wang 1, gabriel abramowitz 1, rachel...
Post on 04-Jan-2016
223 Views
Preview:
TRANSCRIPT
Integration of biosphere and atmosphere observations
Yingping Wang1, Gabriel Abramowitz1, Rachel Law1,
Bernard Pak1, Cathy Trudinger1, Ian Enting2
1CSIRO Marine and Atmospheric Research2 University of Melbourne
Objective
• Land surface model (LSM) as a key component in models for climate or weather predictions;
• In LSM, we represent land biosphere by biome types, and assume that vegetation in each biome type has a set of parameters. Values of most parameters are commonly provided by a lookup table.
• The objective: obtain the best estimate of those parameters in the lookup table using multiple types of data, eg atmospheric concentration and eddy fluxes.
The Carbon Cycle Data Assimilation Scheme (Rayner et al. 2005)
• A biosphere model calculates C flux for a given set of parameters at 2o by 2o;
• Transport model maps the flux to concentration;• Adjoints of both model are available and used in
the optimization;• Cost is calculated as the squared mismatch in
concentration• 57 parameters are optimized with 500
concentration obs per year for 20 years• Estimates of all 57 parameters were estimated
using least square
The Carbon Cycle Data Assimilation Scheme (Rayner et al. 2005)
Key findings:
•Only the ratio of NEP/NPP is well constrained
•Model errors important. Vcmax ranges from 160 mol m-2 s-1 for deciduous shrub to 8 mol m-2 s-1 for C4 grassland!!
Some errors can not be accounted for by parameter tuning
Use the improved CBM (CABLE)
Eight parameters varied within their reasonable ranges
Grey region shows PDF of ensemble predictions
From Abramowitz et al. 2008
Model and model errors
error obsprojectionobs
error randomerror model
state
matrix transition
P
S
wuP
SΦ
P
S
t
t
tt
tt
t
t
t
vHZ
1
1
eparm: parameter error, model calibration
erep: representation error, increasing model resolution
esys: systematic error (statistical model)
)(teeeu sysrepparmt
How big are those errors?
Abramowitz et al. 2006Averaging window size (day)
Parameter error
Systematic error
Random error
But we need estimates of all parameters for global vegetations
• Flux tower and most ecological measurements (except remote sensing) has small spatial coverage;
• Parameter values at similar spatial scales of global climate models should be estimated from fluxes at that scale;
• Atmospheric inversion can provide flux estimates at that spatial scales and a good diagnosing tool.
TRANSCOM III Results (Gurney et al. 2004)
• Obs: monthly mean [CO2] at 75 sites; and uncorrelated;
• Prior uncertainties based on CASA fluxes and uncorrelated
• 94 land regions in CSIRO transport model, and were aggregated to 11 regions;
• 11 transport models, > 2o by 2o.
TRANSCOM III Results (Gurney et al. 2004)
Month Month
Map of covariance/CCAM grid
1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0
1 0
2 0
3 0
4 0
5 0
6 0
7 0
8 0
9 0
- 0 . 3
- 0 . 2
- 0 . 1
0 . 0
0 . 1
0 . 2
0 . 3
0 . 4
0 . 5
0 . 6
Correlation matrix
1-16: Australia
20-30 south America
57-62: Europe
85-95: North America
Combining top down and bottom up
-150 -100 -50 0 50 100 150
-50
0
50
Top down
Coarse resolution
Globally consistent
Results sensitive to priors
Concentration to flux
Bottom up
Fine resolution,
Potentially large error
Results sensitive to parameters
Parameter to flux
Rayner et al. submitted
Wang et al. unpublished
To estimate key parameters in biosphere model
1. Use eddy flux, remote sensing and other ecological measurements to calibrate a process model, and use a statistical model to account for systematic and random errors;
2. Use a biophysical model to provide prior estimates of fluxes and covariance;
3. Use the atmospheric data and other data to retrieve land surface fluxes;
4. Concentration-> global flux -> global parameters and use other estimates at regional scale if possible.
Recent developments
• CO2 satellites will be launched in 2009;
• New technique is being develop to estimate surface fluxes at finer resolution (ca 4o by 8o monthly);
• But the estimates are sensitive to background covariance of fluxes, data error covariance and other assumptions;
• If only fluxes are estimated, we always have more unknown than number of measurements, and lack of predictive capability
• We need estimates of parameters
Using atmospheric data as a diagnosis tool: Southern Hemisphere
South Pole
Blue: obs, green: model, red: CASA
Contribution of source from each semi-hemisphere
Data: GLOBALVIEW-CO2 (2003)
From: Law et al 2006
Southern tropical fluxes
Saleska et al., Science, 302, 1554-1557, 2003
Tapajos, Brazil
0-30oS
Tapajos, Brazil
Seasonality in model opposite to observed. Model seasonality dominated by photosynthesis, observed by respiration
From: Law et al 2006
Model results
Use atmospheric data as a diagnosis tool: Northern Hemisphere site
Blue: obs
Green: CABLE
Red: CASA
Data: GLOBALVIEW-CO2 (2003)
Barrow Ulaan Uul
Mauna Loa Cape Rama
Figure 1. Comparison of the modelled monthly mean concentration by CCAM with CABLE (green) or CCAM using the carbon fluxes as calculated using CASA model (red) with the observed (blue) at four land stations at different latitudes. The latitudes are: 71.32 oN for Barrow; 44.45 oN for Ulaan Uul, 19.5 oN for Mauna Loa and 15.08 oS for Cape Rama.
Integration: top down and bottom up
Concentration and isotopes
Atmospheric inversion
LSM (parameter)
+ stat modelEco data
Prior flux and variance Global flux LSM +stat model
Global parameters
Other regional estimates
top related