Transcript Slide 1

Wang, J., T.W. Sammis, and V. P. Gutshcick, 2006. Inferring complex patterns
of surface flux and atmospheric circulation via remote sensing. Flux
Measurements in Difficult Conditions, a Specialist Workshop. Boulder,
Colorado, USA, 26-28 January 2006.
Inferring complex patterns of surface flux and atmospheric circulation via remote sensing
Junming Wang, Vincent P. Gutschick, Theodore W. Sammis
[email protected], New Mexico State University, Department of Agronomy and Horticulture
mm d-1
Start
Introduction
Update H for each pixel
Lack of energy closure in eddy covariance
measurements casts doubt upon our understanding
of both atmospheric and biophysical control of
surface fluxes. Poor closure may arise from
limitations in instrumental responses, as well as from
advection and other complex air circulations in
inhomogeneous terrain. Inhomogeneity increases
with spatial extent sampled by towers, and thus it
increases under the troublesome stable conditions.
Improved understanding of atmospheric circulations
will be aided if we can quantify the spatial pattern of
surface scalar fluxes of sensible and latent heat.
However, the spatial pattern of fluxes is poorly
inferred from tower flux measurements, which
comprise a convolution over the pattern (as well as
having the compromised accuracy that is the subject
of the inquiry). Inversion of the flux to spatial pattern
is an ill-conditioned problem that defies useful
solution. More direct estimates of spatial patterns in
fluxes may be obtained using remote sensing. One
remote sensing method of sufficiently high spatial
resolution for LE (ET) is a version of the surface
energy balance method (modified from
[Bastiaanssen et al., 1998]). Spatial patterning of ET
is evident at 90-m resolution (see “Results”). We
also offer an extension to CO2 flux estimation, using
ET to infer stomatal conductance that may be used
to correct radiation-use efficiencies. With satellite
data, the energy-balance methods are limited to
daytime conditions, but the method could be used
with finer spatial and temporal sampling by deploying
inexpensive infrared thermocouples.
Calculate u* at
weather station
N
H/H<0.1
Calculate wind
speed at 200 m
Y
End
Calculate zm for
each pixel
Calculate u
each pixel

for
Calculate dT
for each pixel
Calculate rah
for each pixel
Figure 7. Evapotranspiration (ET) estimated from remote sensed hourly ET for
Jornada (desert area), Las Cruces, New Mexico on a summer day of 2001. The ET
map shows ET spatial variability in the absence of terrain variability. The resolution is
90 m by 90 m and the scale is 50 km by 35 km (Height by width).
Calculate H for
each pixel
Figure 1. Flux measurement
towers. Picture from
http://public.ornl.gov/ameriflux/
Results
Calculate stability
parameter for each pixel
Remote sensing model
The ET map shows the Jornada rain track (Figure 7). ET distribution was quite
different among the pixels from differences in land cover.

Correct u according to
stability parameter for each
pixel
Correct rah according to
Objective:
the parameter for each pixel
1) Develop a remote sensing model to estimate spatial ET distribution and give
guidelines to set up flux towers or test if a ET measurement at a location can
represent the values at that area,
N
mm h-1
Figure 4. Atmospheric correction for H.
2) Use spatial ET data to infer CO2 flux,
3) Sense high spatial and temporal variation in ET and CO2 Fluxes.
Method:
Start
The other site is an Ameriflux site, the flux tower site at
Blodgett Forest, CA. The site is situated in a ponderosa pine
plantation within a mixed-evergreen coniferous forest, located
adjacent to Blodgett Forest Research Station. The plantation is
relatively flat, and contains a homogenous mixture of 5 to 7
year old (in 1997) ponderosa pine with other trees and shrubs
(http://public.ornl.gov/ameriflux/). ASTER satellite data on
August 13, 2001 was obtained and is processed to test if the
tower ET measurement represents the values at its area.
Remote sensing model
Satellite inputs: surface
A Remote Sensing ET model (RET) written in
c++ program language was developed. The
model can estimate ET in 90 m  90 m
resolution using ASTER satellite and local
weather data. ASTER data was obtained from
NASA Earth Observing System Data Gateway
(http://redhook.gsfc.nasa.gov/~imswww/pub/ims
welcome/). The model general flowchart is
shown in Figure 2.
temperature and reflectance.
Local weather inputs: solar
radiation, humidity and wind
speed
Rn=f(solar radiation, reflectance,
Figure 5. Jornada Experimental Range
landscape. ARS pilot Michael René Davis
flies the Cessna over the Jornada Range.
Photo by Scott Bauer. Latitude: 32.50 N.
Longitude: 106.75 W.
http://www.ars.usda.gov/is/graphics/photos/
aug01/k9536-2.htm
The tower location.
humidity)
At the Blodgett tower, the ET from RET calculation is 0.40 mm h-1 vs. 0.38 mm h-1
with the tower measurement at 11:30am local time (Central America time zone, GMT
time 19:30pm) on August 13, 2001 (Figure 8). The tower measurement represents
the fluxes along the 270 m distance in the southwest direction of the tower. The wind
direction in daytime is from southwest to northeast. The average flux of the pixels at
this distance was 0.36 mm h-1 and standard deviation was 0.03 mm h-1. However, the
tower measurement does not represent other sides’ ET flux. The surrounding pixels
in other sides (total 5) had average daytime ET of 0.57 mm h-1 and standard
deviation was 0.04 mm h-1.
NDVI=f(reflectance)
Inputs
The inputs include wind speed, humidity, and
solar radiation data at the local weather station
and satellite data products from ASTER,
including ground surface reflectance and
temperature. The reflectance has a resolution of
15 m  15 m for the bands 1 to 3 (Visible and
Near-infrared bands) and 30 m  30 m for the
bands 4 to 9 (Shortwave Infrared bands). The
temperature data has a resolution 90 m  90 m.
Each satellite scene covers an area of 60 km by
60 km. The reflectance data were averaged over
90 m  90 m to match the temperature data
resolution.
G=f(NDVI,
solar
radiation ，
reflectance)
H=f(NDVI, temperature,
reflectance, solar radiation, wind
speed)
LE=Rn-H-G
Output daily or hourly ET
End
Outputs
The spatial ET (mm d-1 or mm h-1) is the output
from the model. The resolution is 90 m  90 m.
Model theory
Rn  Rns  Rnl
Figure 2. The RET model flow
chart.
Rn is net radiation (W m-2),
Rns is net short-wave radiation (W m-2),
Rnl is net long-wave radiation (W m-2).
Rns  (1   ) Rs
where:
 is surface albedo (calculated from reflectance),
Rs is incoming solar radiation measured at the
local weather station (W m-2).
2
c
0.4
C is a function of NDVI (normalized difference
vegetation index) (Figure 3, Bastiaanssen et al.,
1998 )
3
y = -2.70 x + 3.98 x - 1.64 x
- 0.11 x + 0.41
2
R = 0.66
0.6
G  c  Rn
0.2
0
NDVI is calculated as follows:
3  2
NDVI 
3  2
0
where α3 and α2 are the reflectance data of
bands 3 and 2 respectively.
H
4
  c p  dT
rah
Where  is the air density (mol
cp is the
specific heat of air (29.3 J mol-1 ºC-1), dT is the
near surface temperature difference (K), rah is
the aerodynamic resistance to heat transport (s
m-1), where dT is calculated according to dT
values at a hot and a cold spot (Wang et al.,
2005). H calculations need atmospheric
correction (Figure 4).
m-3),
0.2
0.4
0.6
NDVI
0.8
1
Figure 3. Relationship of C = soil heat flux/net
radiation, with NDVI. Data are from Clothier et al.
(1986), Choudhury (1989), Kustas and Daughtry
(1990), Van Oevelen (1993) and Bastiaanssen et al.
(1998).
Using ET to infer CO2 flux
Premises:
1) Both A (CO2 assimilation) and E (or ET) arise from the
combination of physiological and meteorological control. At a
given meteorological condition (solar flux density; air temperature,
humidity, and pressure; windspeed; downwelling TIR flux density
or effective sky temperature), physiological control sets stomatal
conductance gs, leaf temperature TL, A, and E. The relationships
are set by three process equations:
A) leaf energy balance, in which gs, along with meteorological
variables, sets TL;
B) stomatal control program, taken as well represented by the
Ball-Berry equation (Ball et al., 1987; see Gutschick and
Simonneau, 2002): gs = m A hs / Cs + b; m, b = robust empirical
constants (m very near 10 for all unstressed leaves of plants with
dominant C3 pathway of photosynthesis), hs and Cs = relative
humidity and CO2 mixing ratio at surface of leaf, under the leaf
boundary layer;
C) enzyme-kinetic equation for A, well represented by the model
of Farquhar et al. (1980 ff.) - at light-saturation (ca. 70-80% of all
assimilation), A = Vc,max(Ci-Γ)/(Ci+KCO), where Vc,max =
photosynthetic capacity (A at light- and CO2-saturation; set by
enzyme investment that acclimates to environment under genetic
constraints; predictable for vegetation types), Γ=compensation
partial pressure of CO2, a function only of temperature and O2
partial pressure, as also is KCO, an effective Michaelis constant.
These three transcendental equations can be solved
simultaneously by numerical methods (Gutschick, unpubl.)
2) Water stress has its dominant effect on the stomatal control
parameter, m, in the short term (days); for long stress, the effect
extends to photosynthetic capacity, Vc,max. Both of these types of
physiological changes alter both A and E. Effectively, water stress
reduces gs, consequently reducing leaf-internal CO2 partial
pressure and the radiation-use efficiency of leaves; A decreases at
constant photon flux density.
3) We can predict spatial and temporal variations in A from
observable spatial and temporal variations in E, if the relationship
of A to E is sufficiently strong and also robust. By “robust”, we
mean that it has only moderate dependence upon initial
physiological variations (some plants have m moderately above or
below 10 when unstressed, while their value of intercept b differs
in the opposite direction: high m, low b).
We simulated gs, TL, A, and E under the same meteorological
conditions, representative of Eastern deciduous forests1, while
varying 1) m, with b constant - representing moderate-term water
stress of increasing degree; 2) Vc,max , with m and b at original
values - representing different species of different intrinsic growth
rate, OR variations in nitrogen stress; 3) Vc,max with low m = 5 representing the same species under water stress; 4) initial m
(from 12 to 8), with b countervarying from 0.02 to 0.04 representing species differing in stomatal control.
Figure 8. Blodgett area ET map processed by RET
model for August 13, 2001. The resolution is 90 m by
90 m.
ET to infer CO2 flux
1) The curve of A (CO2 assimilation) vs. E (ET) is smooth and
relatively flat (Figure 9). The low slope is expected, in that
reduction of gs affects E much more than it affects A; gs is almost
all of the resistance for water vapor transport while it is a minor
part of the resistance for CO2 transport.
Figure 6. The flux tower site at Blodgett
Forest, CA.
Latitude: 38° 53' 42.9" N
Longitude: 120° 37' 57.9" W
http://public.ornl.gov/ameriflux/
Sensing high temporal and spatial ET flux
Supplementing satellite imagery with local sensing of surface
temperatures:
The temporal and spatial resolution of satellite imagery (in the critical
thermal infrared) is necessarily modest, at 1 km x 1 day (MODIS) or
120m x 16 days (ASTER). Even daily resolution requires modelling of
the diurnal time course of ET to estimate daily total ET. The 16-day
resolution of ASTER misses short-term events that dominate in drier
environments.
The most challenging data to fill in temporally and spatially is the
thermal infrared data. In contrast, downwelling shortwave and
longwave radiation varies little spatially in the absence of clouds and
both can be extrapolated from sparse station measurements. Similar
considerations hold for the other principal meteorological drivers, air
temperature and humidity and windspeed. Vegetation interception of
radiation and also momentum transfer depend upon plant structure,
which changes slowly in almost every landscape.
Infrared thermcouples such as marketed by Apogee Instruments®
can sense surface radiative temperature very accurately and minimal
drift, while being relatively inexpensive and readily logged. The useful
models have separate voltage outputs for target and body temperatures,
allowing for correction of the apparent target temperature. While it is not
practical to cover a large landscape with such sensors, a single sensor
at a key site will allow accurate estimation of the diurnal course of ET.
Sensors at key locations on a large transect will enable the testing of
models of the spatial distribution of vegetation distribution and ETaltering water stress.
2) Consequently, A variations are predictable from variations in E.
The effect of errors in estimating E upon errors in estimating A are
modest. At moderately high stress (m reduced to 5), a 10%
relative error in E (which is now at 60% of its original value)
generates a 6% error in estimated A (which is at 79% of its
original value).
3) Variations in Vc,max exert stronger control over both A and E,
and the A-E relation is steeper. The same curve does NOT apply
as for water stress. One must use a different curve to predict A
from E, so that one must know the different species (readily done,
with field work) or that the variations within one species
correspond to variations in nutrient stress (stress is relatively
unlikely in natural conditions, though likely in agricultural
conditions; in natural conditions, high N stress leads to
replacement by other species competitively).
4) Variations among species in initial m values (optimistic vs.
pessimistic stomata, we may say) do not alter the A-E curve at all.
In the Figure 9, the curve of linked m-b variations is visually
indistinguishable from simple variations in m alone.
Consequently, there is promise in using remotely-sensed ET to
estimate spatial and temporal variations in landscape CO2 fluxes.
Relation of A<-->E under stress
2
A = -0.8146E + 7.8264E + 0.8165
25
A (umol m-2 s-1)
Study sites:
Two sites are chosen to be studied. One is the Jornada
Experimental Range in New Mexico (Figure 5). The site has
mesquite and other shrubs. ASTER satellite data was obtained
for September 17th 2001. Two days before this scene it rained.
This data set is processed here to determine the ET distribution
after rain.
2
R = 0.9997
20
15
mBB varies
10
Vcmax varies,
mBB=10
5
Vcmax varies,
mBB=5
0
Initial mBB=12,
b=0.02
0
2
4
6
Poly. (mBB varies)
E (mmol m-2 s-1)
Figure 9. CO2 assimilation, A, and
ET (E in the graph) relations under
varying physiological stress.
Sensing high spatial and
temporal ET
This work is being developed.
Conclusions
The RET model is capable to map spatial ET with a resolution 90 m by 90 m. The
RET model is capable to give guidelines to set up ET measurement towers and
check that if a ET measurement at a location represents the values at that area.
Spatial ET can be used to infer the spatial CO2 fluxes.
Sensing high spatial and temporal ET is possible using satellite data and high spatial
and temporal-density infrared data.
References
Bastiaanssen, W. G. M., M. Menenti, R. A. Feddes, and A. A. M. Holtslag, 1998: A remote sensing surface energy balance algorithm for land
(SEBAL). 1. Formulation. J. Hydrol., 212/213, 198-212.
Kustas, W. P., & Norman, J. M. (2000). A two-source energy balance approach using directional radiometric temperature observations for sparse
canopy covered surfaces. Agronomy Journal, 92(5), 847–854.
Wang, J., T.W. Sammis, C.A. Meier, L.J. Simmons, D.R. Miller, and Z. Samani. 2005. A modified SEBAL model for spatially estimating pecan
consumptive water use for las cruces, new mexico. 15th Conference on Applied Climatology. Hilton Savannah DeSoto, Savannah, Georgia. 20-24
June, 2005.
Acknowledgments:
Research partially funded by New Mexico Agricultural Experiment Station, Rio Grande Initiative and USDA Forest Service.