identifying sources of uncertainty and observational ...that goal, we intend to perform a...
TRANSCRIPT
Identifying Sources of Uncertainty and Observational Constraints to Improve the Simulation of Long Term Carbon Sequestration in Northern Wisconsin
Brett Raczka1, Kenneth Davis1, Michael Dietze2 1The Pennsylvania State University, 2Boston University
Support: US Department of Energy’s Office of Science through the Northeastern Regional Center of the National Institute for Climate Change Research INTERFACE Exchange Travel Grant sponsored by the National Science Foundation and Purdue University
BackgroundA significant portion of uncertainty related to climate changeis attributed to the terrestrial system feedback upon the climate. The Ecosystem Demography Model (ED2), a terrestrial biosphere model (TBM), is uniquely suited to helpdiagnose, quantify and predict the net exchange of carbonbetween the atmosphere and land, thereby specifying the sign and magnitude of the climate feedback. ED2 uses sizeand age structured vegetation equations to capture the competition that a single tree experiences for light and nutrients. This gap-level competition is important for simulating successional growth and long term carbon dynamics that other ‘big-leaf’ models may miss. Our goal is to identify the most signifcant sources of parametric uncertainty within ED2 by performing a 100 year simulation at Willow Creek, Wisconsin (1901-2010). Tothat goal, we intend to perform a sensitivity analysis to identify the most signficant sources of uncertainty and thenincorporate both existing and synthetic observations that bestreduce the uncertainty.
How well does the CRU-NCEP meteorology match the tower observed meteorology?
Contact: [email protected]
Initial Vegetation Conditions
Meteorology
We examine the differences in magnitude by aggregating the30 minute resolution tower meteorology to 6 hour resolution to compare the common years (1998-2006).
Willow Creek 10 km width
0 3750 7500 11250 ft.
Legend
This map is a user generated static output from an Internet mapping site and is for generalreference only. Data layers that appear on this map may or may not be accurate, current, orotherwise reliable. THIS MAP IS NOT TO BE USED FOR NAVIGATION.
Scale: 1:39,868
In general the CRU-NCEP meteorology is likely more conducive for vegetationgrowth given increased levels of shortwave radiation (+19%) and increasedprecipitation (+10%). The impact of these biases on carbon dynamics has notyet been quantified for this simulation.
Ecosystem Demography Model:Parameterization
There are over 40 parameters per plant species that have the potentialto influence carbon dynamics. The table below presents a subset ofparameters that initial testing indicates is important for carbonsequestration, forest demography, and forest succession.
How sensitive is carbon sequestration to initial conditions and meteorology?
Willow Creek Default Simulation: (1901-2010)
Additional Findings/Conclusions:-ED2 requires additional parameter tuning to reduce unreason-able understory growth and die-off of initialized tree canopy. Based upon extremely high stem density and low carbon balancefor specific cohorts, sapling and seedling mortality need to be increased.
-Performing a 700 year run with recycled NACP meteorologyalso simulates intense understory growth and reduction in lateDBF LAI (1.0) from 1900-1950. However, late DBF begins torecover by 1950, then surpasses early DBF LAI shortly after2100.
-Differences in model setup have not significantly changed carbon dynamics from the default run including 1) June vs January start date 2) choice of canopy sub-module 3) most recent model version ED2.2 -The CRU-NCEP meteorology is signficantly biased for somevariables and the model is sensitive to meteorology resolution. This motivates the need to correct for meteorologybiases.
Acknowledgements: We thank Bjorn Brooks (University of Illinois) for providingCRU-NCEP meteorology, Simon Goring (University of Wisconsin-Madison) forproviding the witness tree data, and for Ryan Knox (MIT), Marcos Longo (Harvard University) and Naomi Levine (Harvard University) for providing program code which was modified to create the ED2 figures.
The Willow Creek site is located within the Chequa-megon National Forest of northern Wisconsin. The present day tree stand (70-80years old) within a 0.5 kmradius of the flux towerconsists primarily of sugar maple/basswood (75%) andgreen ash/red oak (20%). However, the surroundingarea (see WISCLAND land cover map to right) is com-posed of aspen, elm and fir species. This region is ideal for study because it is part of an Amerifluxnetwork of sites within the Chequamegon Ecosystem-Atmosphere Study, andhas experienced both climate change and disturbance.
8
7
1
2
3
0 0
1
00
1
2
3
4
5
6
7
8
9
maples birches elms basswood hemlock tamarack cedar ironwood spruce
spec
ies
occ
ure
nce
1900 Willow Creek Species Distribution The vegetation conditions areestimated for the year 1900from witness tree data thatprovides the species, tree diameter and aggregate stemdensity for two trees at eachlocation. Eleven locationswithin 16 km2 of the sitecoordinates were used toinitialize the model. We assume that the site is composed of 11 patchespopulated by 2 cohorts each consisting of the type and stemdensity observed.
0
3
7
1
11
0
2
4
6
8
10
12
early conifer
late conifer
early DBF mid DBF late DBF
PFT
occu
renc
e
CRU-NCEP meteorology data (1901-2010) was used to drive atmospheric condtions within ED2. A 1 degree grid cell with 6 hour temporal resolutioncorresponding to the site coordinates, was adapted to drive the 13 variables:radiation (5), temp, precip, mixing ratio, pressure, wind (2), CO2, and elevation.
0 50 100 150 200 250 300 3500
50
100
150
200
250
Days
W/m
2
Willow Creek Meteorology, Daily Mean Short Wave Radiation (1998-2006)
CRU NCEPNACP (Tower)
2 4 6 8 10 12 14 16 18 20 221.5
2
2.5
3
3.5
4
4.5
5
5.5x 10
6 hour average (LST)
kg/m
2 /sec
Hourly Mean of Growing Season Precipitation Rate (1998-2006)
CRU NCEP
NACP (Tower)
0 0.1 0.2 0.3 0.4 0.5 0.60
5
10
15
20
25
30
35
[m2/m3]
Ele
vatio
n [
m]
LAI Height Profile
LAI=5.3
Late Conifer (0.14)Early DBF (1.30)Mid DBF (0.18)Late DBF (3.69)
0.1 0.2 0.3 0.4 0.5 0.60
5
10
15
20
25
30
35
[m2/m3]
Ele
vatio
n [m
]
LAI Height Profile LAI=4.8
Late Conifer (0.17)Early DBF (4.09)Mid DBF (0.00)Late DBF (0.55)
1900 1905 1910 1915 1920
1
2
3
4
Time
LAI [m2] by PFT
0.1 0.2 0.3 0.4 0.5 0.60
5
10
15
20
25
30
35
[m2/m3]
Ele
vatio
n [m
]
LAI Height Pro�leLAI=4.6
Late Conifer (0.25)Early DBF (4.30)Mid DBF (0.00)Late DBF (0.05)
1900 1950 2000
1
2
3
4
5
Time
LAI [m2] by PFT
The figures below are July snapshots of the leaf area index (LAI) profile of thesimulated tree canopy. Parameters were taken from short term Willow Creekruns (<10 years) and were not optimized for the century time-scale. The stemdensity was increased from the witness tree observations to bring in line withtypical mature canopy stem density and LAI.
Despite a full overstory canopy (LAI=5.3, left figure) a surge in early DBFsapling growth occurs almost immediately (middle figure). The overstory midand late DBF species die off within the first 40 years. The understory saplinggrowth matures and merges with the existing canopy by 2010 (right figure). Unrealistic understory stem density coupled with low carbon balance indicates mortality and/or seedling growth parameters require additional tuning.
The figures below compare the default simulation (11 locations, 22 cohorts) with a run that was initialized from a wider sampling (in space) of the witnesstree data (30 locations, 60 cohorts) Although the photosynthesis and respiration processes are very similar(middle and right figure), the default simulation predicts 25% more carbon sink(~590 gC/m2/yr) compared to the other run (~470 gC/m2/yr).
The figures below compare simulations that use the observed tower meteorology (NACP) recycled throughout the simulation time period. The firstsimulation uses 30 minute resolution meteorology, whereas the second simu-lation averages the meteorology to 6 hour resolution. The coarser resolutionmeteorology increases carbon sequestration (+50%, middle figure) and unlikethe default simulation, supports the growth of late DBF (left figure).
1903 1916 1930 1944 1957 1971 1984 1998
0
10
KgC
/m
Cumulative NEE (Net Ecosystem Exchange)
default,
1902 1916 1930 1943 1957 1971 1984 19980
0.5
1
1.5
2
2.5
3
KgC/
m2 /y
ear
GPP (Gross Primary Productivity)
default (22 cohorts)
more sampling (60 cohorts)
1902 1916 1930 1943 1957 1971 1984 19980
0.5
1
1.5
2
2.5
3
KgC/
m2 /
year
RE (Ecosystem Respiration)
default (22 cohorts)
more sampling (60 cohorts)
0.1 0.2 0.3 0.4 0.50
5
10
15
20
25
30
35
[m2/m3]
Ele
vatio
n [m
]
LAI Height ProfileLAI=5.3
Late Conifer (0.32)Early DBF (2.44)Mid DBF (0.00)Late DBF (2.55)
1900 1950 2000
1
2
3
Time
LAI [m2] by PFT
1902 1916 1930 1943 1957 1971 1984 1998
0
5
KgC
/m2
Cumulative NEE, NACP Meteorology
30 min. resolution
6 hour resolution
Future Work:
1902 1916 1930 1943 1957 1971 1984 19980
0.5
1
1.5
2
2.5
KgC
/m2 /y
ear
GPP, NACP Meteorology
30 min. resolution
6 hour resolution
-Once reasonable carbon dynamics are achieved the PredictiveEcosystem and Analyzer software (PECAN) (www.pecanproject.org) will be used to identify the most import-ant parameters for determining 100 year carbon sequestration.
-A pre-calibration technique will be used to identify what current/synthetic observations are most useful for reducing parametricuncertainty.
SW radiation LW radiation Precip Temp Mixing Ratio Pressure WindW/m2 W/m2 kg/m2/sec Kelvin kg/kg Pascal meter/sec
CRUNCEP 165 283 2.48*10-5 278.13 0.0052 97900 3.02Tower (NACP) 138 295 2.26*10-5 278.48 0.0062 95200 3.15
% bias 19% -4% 10% -0.13% -17% 2% -4%
1998-2006 Average Values
Met Product
PFTS mort1 mort2 mort3 seedling mortality r_fract
root turnover
rate
growth respiration
factor
storage turnover
rate
quantum efficiency
dark respiration
factor
units year-1 year-1 year-1 year-1 mol CO2/ mol photon
Early DBF* 1.0 20 6.14*10-3 0.95 0.30 5.77 0.00 0.62 0.08 0.02Mid DBF 1.0 20 3.81*10-3 0.95 0.30 5.08 0.00 0.62 0.08 0.02Late DBF 1.0 20 4.28*10-3 0.95 0.30 5.07 0.00 0.62 0.08 0.02Late Conifer 1.0 20 2.36*10-3 0.95 0.30 3.80 0.45 0.00 0.08 0.02
Descriptiondensity
dependent mortality
density dependent mortality
density indpendent mortality
% of seed that dies
% of storage
carbon to seed
fine root turnover
% daily carbon gain
lost to growth respiration
carbon assimilate turnover
Farquhar (photosynthesis)
leaf respiration coefficient
Willow Creek Select Parameter Settings (Default Values)
http://dnr.wi.gov/maps/gis/datalandcover.html
WISCLAND LAND COVER MAP
*DBF = deciduous broadleaf forest