Micrometeorological Methods Used to Measure
Greenhouse Gas Fluxes:
The Challenges Associated with Them, at the Local to
Global Scales
Dennis Baldocchi
University of California, Berkeley
JASON GHG Emissions Monitoring
Summer Study (June 16-18, 2010)
La Jolla, CA
Methods To Assess Terrestrial Carbon Budgets
at Landscape to Continental Scales, and Across
Multiple Time Scales
GCM Inversion
Modeling
Remote Sensing/
MODIS
Physiological Measurements/
Manipulation Expts.
Eddy Flux
Measurements/
FLUXNET
Forest/Biomass
Inventories
Biogeochemical/
Ecosystem Dynamics
Modeling
From point to globe via integration with remote sensing
(and gridded metorology)
century
forest
inventory
plot
Forest/soil inventories
decade
Temporal scale
year
month
Landsurface remote sensing
Eddy
covariance
towers
week
tall
tower
observatories
remote sensing
of CO2
day
hour
local 0.1
1
plot/site
From: Markus Reichstein, MPI
10
100
1000
10 000 global
Countries EU
Spatial scale [km]
Challenges in Measuring Greenhouse Gas Fluxes
• Measuring/Interpreting greenhouse gas flux in a quasicontinuous manner for days, years and decades
• Measuring/Interpreting fluxes over Patchy, Microbiallymediated Sources (e.g. CH4, N2O)
• Measuring/Interpreting fluxes of Temporally Intermittent
Sources (CH4, N2O, O3, C5H8)
• Measuring/Interpreting fluxes over Complex Terrain
• Developing New Sensors for Routine Application of Eddy
Covariance, or Micrometeorological Theory, for trace gas
Flux measurements and their isotopes (CH4, N2O,13CO2,
C18O2)
• Measuring fluxes of greenhouse gases in Remote Areas
without ac line power
Flux Methods Appropriate for Slower Sensors, e.g. FTIR
F  w' c'   w (c up  c dn )
• Relaxed Eddy Accumulation
• Modified Gradient Approach
• Integrated Profile
• Disjunct Sampling
z
z
c z
Fc ~ Fs
 sz
1
F
u(  c   background )dz
x0
Eddy Covariance
• Direct Measure of the Trace Gas Flux Density between the
atmosphere and biosphere, mole m-2 s-1
• Introduces No Sampling artifacts, like chambers
• Quasi-continuous
• Integrative of a Broad Area, 100s m2
• In situ
ESPM 228 Adv Topics Micromet & Biomet
Eddy Covariance,
Flux Density: mol m-2 s-1 or J m-2 s-1
F  a ws ~ a  w ' s '
 c mc pc
s

 a ma Pa
ESPM 228 Adv Topics Micromet & Biomet
Eddy Covariance Tower
Sonic Anemometer, CO2/H2O IRGA,
inlet for CH4 Tunable diode laser spectrometer &
Meteorological Sensors
24 Hour Time Series of 10 Hz Data,
Vertical Velocity (w) and Methane (CH4) Concentration
4
W m/s
2
0
-2
-4
0
1
2
3
4
5
6
7
8
9
5
x 10
5
4.5
CH 4 ppm
4
3.5
D164, 2008
3
2.5
2
1.5
0
1
2
3
4
5
seconds
Sherman Island, CA: data of Detto and Baldocchi
6
7
8
9
5
x 10
Non-Dispersive Infrared Spectrometer, CO2 and H2O
Open-path , 12.5 cm
Low Power, 10 W
Low noise, CO2: 0.16 ppm; H2O: 0.0047 ppth
Low drift, stable calibration
Low temperature sensitivity: 0.02%/degree C
LI 7500
Measuring Methane with Off-Axis Infrared Laser Spectrometer
Closed path
Moderate Cell Volume, 400 cc
Long path length, kilometers
High power Use:
Sensor, 80 W
Pump, 1000 W; 30-50 lpm
Low noise: 1 ppb at 1 Hz
Stable Calibration
Los Gatos Research
LI-7700 Methane Sensor,
variant of frequency modulation spectroscopy
Open path, 0.5 m
Short optical path length, 30 m
Low Power Use: 8 W, no pump
Moderate Noise: 5 ppb at 10 Hz
Stable Calibration
Power Spectrum defines the Frequencies to be Sampled
Power Spectrum

w' w'   S ww ( )d
0
Co-Spectrum

F  w' c'   S wc ( )d
0
ESPM 228 Adv Topics Micromet & Biomet
Signal Attenuation:
The Role of Filtering Functions and Spectra
• High and Low-pass filtering via Mean Removal
– Sampling Rate (1-10Hz) and Averaging Duration (30-60
min)
• Digital sampling and Aliasing
• Sensor response time
• Sensor Attenuation of signal
– Tubing length and Volumetric Flow Rate
– Sensor Line or Volume averaging
• Sensor separation
– Lag and Lead times between w and c
ESPM 228 Adv Topics Micromet & Biomet
Comparing Co-spectra of open-path CO2 & H2O sensor and closed-path CH4 sensor
Co-Spectra are More Forgiving of Inadequate Sensor Performance than Power Spectra
Because there is little w-c correlation in the inertial subrange
M.Detto and D. Baldocchi
Co-Spectra is a Function of Atmospheric Stability:
Shifts to Shorter Wavelengths under Stable Conditions
Shifts to Longer Wavelengths under Unstable Conditions
Detto, Baldocchi and Katul, Boundary Layer Meteorology, accepted
Zero-Flux Detection Limit, Detecting Signal from Noise
F  w ' c '  rwc w c
rwc ~ 0.5
ch4 ~ 0.84 ppb
co2 ~ 0.11 ppm
Methane Sensor
0.1
0.2
0.3
0.4
0.5
Mean: 1897.4277
StdDev: 0.8411
Std Err 0.0219
1910
w
m/s
U*
m/s
Methane Lab Calibration
1920
1900
1890
1880
260
280
300
320
340
360
380
400
Time
ESPM 228 Adv Topics Micromet & Biomet
Fmin, CH4
Fmin, CO2
nmol m-2 s-1 mmol m-2 s-1
0.125
2.1
0.275
0.25
4.2
0.55
0.375
6.3
0.825
0.5
8.4
1.1
0.625
10.5
1.375
Flux Detection Limit, v2
Based on 95% CI that the Correlation between
W and C that is non-zero
0.035 mmol m-2 s-1, 0.31 mmol m-2 s-1 and 3.78 nmol m-2 s-1 for
water vapour, carbon dioxide and methane flux,
Detto et al, in prep
Most Sensors Measure Mole Density, Not Mixing Ratio
Formal Definition of Eddy Covariance, V2
F   a ws   a  w' s'  w c  w'  c '  w c
ESPM 228 Adv Topics Micromet & Biomet
Webb, Pearman, Leuning Algorithm:
‘Correction’ for Density Fluctuations when
using Open-Path Sensors
ma  c
 v ma  c
'
Fc  w'  c ' 
w'  v  (1 
) w' T '
mv  a
 a mv T
ESPM 228 Adv Topics Micromet & Biomet
Raw <w’c’> signal, without density ‘corrections’,
will infer Carbon Uptake when the system is Dead and Respiring
Fwpl
<w'c'>
dead annual grassland
10
F (mmol m-2 s-1)
5
0
-5
-10
-15
-20
180
181
182
183
Day
ESPM 228 Adv Topics Micromet & Biomet
184
185
methane correction term for <wt>, nmol m-2 s-1
methane correction term for water vapor, nmol m-2 s-1
Density ‘Corrections’ Are More Severe for CH4 and N2O:
This Imposes a Need for Accurate and Concurrent Flux Measurements of H and LE
20
18
16
14
12
10
8
6
4
2
0
0
2
4
6
-2 -1
wq mmol m s
8
10
120
100
80
60
40
20
0
0
0.1
0.2
0.3
-1
wt K m s
0.4
Annual Time Scale, Open vs Closed sensors
Hanslwanter et al 2009
AgForMet
ESPM 228 Adv Topics Micromet & Biomet
Towards Annual Sums
Accounting for Systematic and Random Bias Errors
• Advection/Flux Divergence
• U* correction for lack of adequate turbulent mixing at night
• QA/QC for Improper Sensor Performance
– Calibration drift (slope and intercept), spikes/noise, a/d off-range
– Signal Filtering
• Software Processing Errors
• Lack of Fetch/Spatial Biases
– Sorting by Appropriate Flux Footprint
• Change in Storage
• Gaps and Gap-Filling
ESPM 228 Adv Topic Micromet & Biomet
Systematic and Random Errors
3
signal
random error
2
1
0
f(t)
-1
-2
-3
0
1
2
3
4
5
6
7
time
2.0
1.5
signal
bias error
1.0
0.5
f(t)
0.0
-0.5
-1.0
-1.5
0
1
2
3
4
5
6
7
time
1.5
1.0
0.5
0.0
f(t)
-0.5
signal
systematic bias
-1.0
-1.5
0
1
2
3
4
5
time
ESPM 228 Adv Topic Micromet &
Biomet
6
7
Random Errors Diminish as We Measure Fluxes Annually
and Increase the Sample Size, n
Oak Ridge, TN
Temperate Deciduous Forest
1997
15
10
5
Tall Vegetation, Undulating Terrain
-5
-10
-15
-20
-25
-30
-35
-40
100
150
200
250
300
350
Vaira Grassland 2001
Day-Hour
15
10
5
-1
50
0
-2
0
Fc (mmol m s )
Fwpl (mmol m-2 s-1)
0
Short Vegetation, Flat Terrain
-5
-10
-15
-20
-25
0
50
100
150
200
Day/Hour
250
300
350
Systematic Biases and Flux Resolution:
A Perspective
• FCO2: +/- 0.3 mmol m-2 s-1 => +/- 113 gC m-2 y-1
• 1 sheet of Computer paper 1 m by 1 m: ~70 gC m-2 y-1
• Net Global Land Source/Sink of 1PgC (1015g y-1): 6.7 gC m-2 y-1
The Real World is Not Kansas, which is Flatter than a Pancake
Eddy Covariance in the Real World
Cloud
Cloud
ESPM 228 Adv Topics Micromet & Biomet
Diagnosis of the Conservation Equation for C for Turbulent Flow
u j ' c '
dc c
c
c
c
u ' c '  v ' c '  w ' c '

u v w  



dt t
x
y
z
x j
x
y
z
I
III
II
I: Time Rate of Change
II: Advection
III: Flux Divergence
ESPM 228 Adv Topics Micromet & Biomet
Daytime and Nightime Footprints over an Ideal, Flat Paddock
F
0
z
F
F   dz  Const
z
Detto et al. Boundary Layer Meteorology, conditionally accepted
Estimating Flux Uncertainties:
Two Towers over Rice
Detto, Anderson, Verfaillie, Baldocchi, unpublished
Examine Flux Divergence
Detto, Baldocchi and Katul,
Boundary Layer Meteorology, conditionally accepted
Underestimating C efflux at Night, Under Tall Forests, in Undulating Terrain
10
5
-1
-2
mmol m s
CO2 Flux Density
0
-5
-10
-15
-20
-25
0
4
8
12
16
20
24
time (hours)
Ne: measured (-4.84 gC m-2 day-1)
Ne: computed (-5.09 gC m-2 day-1)
Fwpl+Storage: measured (-5.96 gC m-2 day-1)
Fwpl: measured (-6.12 gC m-2 day-1)
Baldocchi et al., 2000 BLM
Losses of CO2 Flux at Night: u* correction
Wheat, Columbia River Valley, Oregon
8
Canopy Respiration
8 to 13 C
7
r ² 0.325
Fco2 (mmol m-2 s-1)
6
5
4
3
2
1
0
0.0
0.2
0.4
0.6
u* (m s-1)
ESPM 228 Adv Topic Micromet & Biomet
0.8
1.0
Systematic Biases are an Artifact of Low Nocturnal Wind Velocity
Friction Velocity, m/s
Annual Sums comparing Open and Closed Path Irgas
Hanslwanter et al 2009 AgForMet
ESPM 228 Adv Topics Micromet & Biomet
Biometric and Eddy Covariance C Balances Converge after Multiple Years
Gough et al. 2008, AgForMet
Contrary Evidence from Personal Experience:
Crops, Grasslands and Forests
Boreas 1994
Hourly averages
600
800
wheat
500
700
r2=0.98
slope=1.05
Old Jack Pine
r2=0.93
slope=0.93
LE+H+S+G (W m-2)
600
H+LE (W m-2)
400
300
200
500
400
300
200
100
100
0
-100
-100
0
-100
-100
0
100
200
300
Rn-G (W m-2)
400
500
300
400
500
600
700
800
400
Coefficients:
b[0] 3.474
300
b[1] 1.005
r²
coefficients:
b[0] 5.55
b[1] 0.94
r ² 0.926
0.923
LE+H+G
E + H + G + S + Ps (W m )
200
Vaira Grassland, D296-366, 2000
800
-2
100
Rn (W m-2)
Temperate Deciduous Forest
600
0
600
400
200
100
200
0
0
0
200
400
600
800
-100
-100
0
100
200
-2
-2
Rnet (W m )
Rnet (W m )
300
400
Many Studies Don’t Consider Heat Storage
of Forests Well, or at All, and Close Energy Balance when they Do
Lindroth et al 2010 Biogeoscience
Haverd et al 2007 AgForMet
ESPM 228 Adv Topic Micromet & Biomet
FLUXNET: From Sea to Shining Sea
500+ Sites, circa 2009
Limits and Criteria for Network Design for
Treaty Verification
• Can We Statistically-Sample or Augment
Regional, Continental and Global Scale C
Budgets with a Sparse Network of Flux
Towers?
• If Yes:
– How Many Towers are Enough?
– Where Should the Towers Be?
– How Good is Good Enough with regards to Fluxes?
– How Long Should We Collect Data?
How many Towers are needed to estimate mean NEE, GPP
and assess Interannual Variability, at the Global Scale?
Green Plants Abhor a Vacuum, Most Use C3 Photosynthesis,
so we May Not need to be Everywhere, All of the Time
We Need about 75 towers to produce Robust and Invariant Statistics
Probability Distribution of Published NEE Measurements, Integrated Annually
0.07
0.06
0.05
mean: -182.9 gC m-2 y-1
std dev: 269.5
n: 506
p(x)
0.04
0.03
0.02
0.01
0.00
-1500
-1000
-500
0
FN (gC m-2 y-1)
Baldocchi, Austral J Botany, 2008
500
1000
1500
Interannual Variability of the Statistics of NEE is Small across a sub-network of 75 Sites
Mean NEE Ranges between -220 to -243 gC m-2 y-1
Standard Deviation Ranges between 35.2 and 39.9 gC m-2 y-1
0.25
FLUXNET Network, 75 sites
2002: -220 +/- 35.2 gC m-2 y-1
2003: -238 +/- 39.9
2004: -243 +/- 39.7
2005: -237 +/- 38.7
0.20
p(NEE)
0.15
0.10
0.05
0.00
-1000
-800
-600
-400
-200
0
NEE (gC m-2 y-1)
200
400
600
Interannual Variability in GPP is small, too,
across the Global Network (1103 to 1162 gC m-2 y-1)
2002: 1117 +/- 74.0gC m
2003: 1103 +/- 67.8
2004: 1162 +/- 77.0
2005: 1133 +/- 70.1
-2 -1
y
0.25
FLUXNET Network, 75 sites
p(GPP)
0.20
0.15
0.10
0.05
0.00
0
500
1000
1500
2000
2500
3000
3500
GPP (gC m-2 y-1)
Assuming Global Arable Land area is 110 106 km2,
Mean Global GPP ranges between 121.3 and 127.8 PgC/y
Precision is about +/- 7 PgC/y
Ecosystem Respiration Scales Tightly with Ecosystem Photosynthesis,
But Is with Offset by Disturbance
4000
Undisturbed
Disturbed by Logging, Fire, Drainage, Mowing
3500
FR (gC m-2 y-1)
3000
2500
2000
1500
1000
500
0
0
500
1000
1500
2000
2500
FA (gC m-2 y-1)
Baldocchi, Austral J Botany 2008
3000
3500
4000
Net Carbon Exchange is a Function of Time Since Disturbance
Conifer Forests, Canada and Pacific Northwest
1000
800
FN (gC m-2 y-1)
600
400
200
0
-200
-400
-600
1
10
100
Stand Age After Disturbance
Baldocchi, Austral J Botany, 2008
1000
C Fluxes May Vary Interannual on 7 to 10 year Time Scales
Walker Branch Watershed, TN: 1981-2001
CANOAK
1
year
130 days
nSnee/nee
0.1
0.01
7 years
0.001
0.0001
0.0001
0.001
0.01
Frequency (1/day)
0.1
1
Upscale NEP, Globally, Explicitly
1. Compute GPP = f(T, ppt)
2. Compute Reco = f(GPP,
Disturbance)
3. Compute NEP = GPP-Reco
Leith-Reichstein Model
GPP  min  f MAT , g P  

1  e a1 a2 15C
1  e  k P 

min  GPP15C 
, GPP1000mm 
a1  a2 MAT
 k 1000mm 
1

e
1

e


4000
Undisturbed
Disturbed by Logging, Fire, Drainage, Mowing
3500
FLUXNET Synthesis
Baldocchi, 2008, Aust J Botany
FR (gC m-2 y-1)
3000
2500
2000
Reco = 101 + 0.7468 * GPP
1500
1000
Reco, disturbed= 434.99 + 0.922 * GPP
500
0
0
500
1000
1500
2000
2500
FA (gC m-2 y-1)
3000
3500
4000
<NEE> = -222 gC m-2 y-1
This Flux Density Matches FLUXNET (-225) well, but S NEE = -31 PgC/y!!
Implies too Large NEE (|-700 gC m-2 y-1| Fluxes in Tropics
Ignores C losses from Disturbance and Land Use Change
Explicit Climate-Based Upscaling
Under Represents Disturbance Effects
FLUXNET Database
0.14
0.12
0.10
FLUXNET
Global Map
pdf
0.08
0.06
FLUXNET Under
Samples Tropics
0.04
0.02
0.00
-1400 -1200 -1000
-800
-600
-400
-200
0
NEE (gC m-2 y-1)
Statistically Sampling and Climate Upscaling Agree
<NEE: FLUXNET> = -225 +/- 164 gC m-2 y-1
<NEE 0% dist: sinusoidal> = -222 gC m-2 y-1
200
400
600
To Balance Carbon Fluxes infers that Disturbance Effects May Be Greater than Presumed
<NEE> = -4.5 gC m-2 y-1
S NEE = -1.58 PgC/y
UpScaling Tower Based C Fluxes with Remote Sensing
LIDAR derived map of Tree location and Height
Hemispherical Camera
Web Camera
Upward Looking Camera
Annual Grassland, 2004-2005
1.0
Reflectance
0.8
Oct 13, 2004
Oct 27, 2004
Nov 11, 2004
Jan 5, 2005
Feb 2, 2005
Apr 1, 2005
Mar 9, 2005
May 11, 2005
Dec 29, 2005
0.6
0.4
0.2
0.0
400
500
600
700
800
Wavelength (nm)
Falk, Ma and Baldocchi, unpublished
ESPM 111 Ecosystem Ecology
900
1000
Ground Based, Time Series of Hyper-Spectral Reflectance Measurements,
in Conjunction with Flux Measurements Can be Used to Design Future Satellites
Remote Sensing of NPP:
Up and down PAR, LED, Pyranometer, 4 band Net Radiometer
LED-based sensors are Cheap, Easy to Replicate and
Can be Designed for a Number of Spectral Bands
Spectrally-Selective Vegetation Indices Track Seasonality of C Fluxes Well
Ryu et al. Agricultural and Forest Meteorology, in review
Vegetation Indices can be Used to Predict GPP with Light Use Efficiency Models
Ryu et al. Agricultural and Forest Meteorology, in review
UpScaling of FluxNetworks
What We can Do:
Is Precision Good Enough for Treaties?
Xiao et al 2010, Global Change Biology
Map of Gross Primary Productivity Derived from Regression Tree Algorithms
Derived from Flux Network, Satellite Remote Sensing and Climate Data
Xiao et al 2010, Global Change Biology
Xiao et al 2010, Global Change Biology
Net Ecosystem C Exchange
spring
summer
autumn
Xiao et al. 2008, AgForMet
winter
Upscale GPP and NEE to the Biome Scale
area-averaged fluxes of NEE and GPP were -150 and 932 gC m-2 y-1
net and gross carbon fluxes equal -8.6 and 53.8 TgC y-1
Jingfeng Xiao and D Baldocchi
Take-Home Message for Application of Eddy
Covariance Method under Non-Ideal
Conditions
•Routine Flux Measurements Must Comply with Governing
Principles of Conservation Equation
•Design Experiment that measures Flux Divergence and
Storage, in addition to Covariance
•Networks need more Sites in Tropics and Distinguish C3/C4
crops
•Networks need Sites that Cover a Range of Disturbance
History
•Network of Flux Towers, in conjunction with Remote Sensing,
Climate Networks and Machine Learning Algorithms has
Potential to Produce Carbon Flux Maps for Carbon Monitoring
for Treaty, with Caveats and Accepted Errors
Additional Background Material
Sampling Error with Two Towers
Hollinger et al GCB, 2004
Gap-Filling Inter-comparison Bias Errors
Moffat et al., 2007, AgForMet
ESPM 228 Adv Topic Micromet & Biomet
Root Mean Square Errors with Different Gap Filling Methods
Moffat et al., 2007, AgForMet
ESPM 228 Adv Topic Micromet & Biomet