Satellitedriven Gross Primary Production estimates

1
Satellite-driven Gross Primary Production
estimates
Hirofumi Hashimoto (University of Tokyo),
Ramakrishna R. Nemani Jennifer Dungan (NASA
Ames Research Center), Feihua Yang (University of
Wisconsin), Steven W. Running (University of
Montana) hiro_at_fr.a.u-tokyo.ac.jp
3. Satellite Data GPP Estimation
1. Introduction If and when feasible,
completely satellite-driven estimates of gross
and/or net primary production have the advantages
of consistency and reliability that current
assimilation methods lack. In many of the
current, mainly satellite-based GPP/NPP methods
implementing a water stress function has been the
most difficult task. Current MOD17 algorithm
simplifies the process by representing water
stress mainly through vapor pressure deficits.
However, geographic representations of VPD are
very coarse. In this study, we estimated the
spatial patterns of VPD from land surface
temperature. Then, we estimated the GPP only from
satellite data by using MODIS 17 algorithm.
Satellite Data
(1) MODIS 11 Land Surface Temperature (2) MODIS
15 fPAR (fraction of the photosynthetically
active radiation) (3) MODIS 10 Snow Cover (4)
GCIP GOES daily incident PAR (photosynthetically
active radiation)
Algorithm
As GPP calculation algorithm, we used the part of
the MODIS 17 algorithm as following equation.
2. VPD estimation from Land Surface Temperature
Where emax is the maximum light use efficiency
for each biome type, W(VPD) and T(Tmin) are
limiting factors of VPD and minimum
temperature,respectively. Then we can get GPP
from the following equation.
Theory
Granger (1982) showed that VPD can be estimated
from land surface temperature as the result of
the feedback link between the dryness of the air
and land surface moisture status. The dry air
reflects the dry soil moisture status, and the
lack of the water makes more energy for warming
the surface. The feedback link is well realized
in the complementary relationship between actual
evapotranspiration and potential
evapotranspiration. Granger (1982) proposed the
linear relationship between VPD and saturation
vapor pressure at the land surface temperature.
All the biome-specific constant variables are
same as MODIS 17 algorithm. When there is snow,
we set GPP to zero for that pixel. For
estimating Tmin value we also used land surface
temperature by linear regression as following
equation for now.
This equation can be improved by using another
satellite data or other algorithm to estimate the
minimum temperature. GPP is calculated for every
8day period in 2001 over the conterminous USA,
then summed over the year.
Application
We tested this method for the 12 Ameriflux sites
(Figure 1) during 2001. For the land surface
temperature, we used MODIS 11 data for the same
period. Figure 2 shows the scatterplot of the
daytime average VPD against the saturation vapor
pressure at the land surface temperature for each
day.
4. Results
Mapping GPP across the U.S
Figure 1 The locations of the test sites from
Ameriflux sites.
In order to test the stability of the slope of
the line, we used a number primary weather
stations over the U.S. Results of this analysis
are shown in Figure 3. Except over southern
California and parts of Northeast, the slopes
tend to be in a narrow range, suggesting a single
slope may capture much of the variability.
Figure 5 Spatial pattern of the GPP for the
conterminous USA during 2001
Figure 5 is the spatial pattern of the yearly GPP
for the conterminous USA in 2001. Because of
the cloud cover, there are some missing data
(black spot in the Figure 4), but the texture of
the GPP is more realistic than climate based GPP
estimation that can be tile -like texture or too
smoothed pattern.
Figure 2 Scatterplot of the daytime average
VPD against saturation vapor pressure at the land
surface temperature.
GPP Validation using FLUXNET data
We validated the result of the yearly GPP by
using Ameriflux data. Figure 6 shows the
scatteplot of measured GPP at the Ameriflux sites
against estimated GPP. Overestimation appears to
be common in this test. Errors in VPD estimation
and/or minimum temperature, MODIS 17 algorithm,
biome-specific constants, and errors in measuring
GPP at Ameriflux sites are all possible
candidates for further exploration.
Figure 3 Map of regression slopes computed at
primary weather stations across the continental
U.S.

Figure 6 Validation of satellite-derived GPP
using FLUXNET data.
Figure 4 is the comparison of estimated VPD
between this study and interpolation method.
Interpolated method is smoother than our method
because interpolation method usually estimates
the VPD by inverse-distance method by using few
points. As expected, land cover dominated surface
energy exchange is better captured in satellite
method.
5. Conclusions
We attempted to estimate GPP from satellite data
only by using VPD estimation from land surface
temperatures. Our method, though promising,
appears to overestimate GPP in some Ameriflux
sites. We can improve the GPP estimation by (1)
improving VPD estimation method (2) using other
satellite data for minimum temperature (3) using
new version of MODIS 17 biome-specific constants.
Figure 4 Comparison of the estimated spatial
pattern between this study (A) and interpolated
method (B).