458
Multiscale Hydrologic Remote Sensing: Perspectives and Applications
basis, and FLUXNET data are available for download from the Oak Ridge National
Laboratory Distributed Active Archive Center (http://daac.ornl.gov/FLUXNET/fluxnet
.shtml). These flux towers cover all typical land cover types and climates. This long-term
tower data will help scientists validate the regional and global ET results. The indirect
ET validation effort with precipitation and MODIS GPP at the regional and global scales
will also continue to help in the development of global ET product. More work will be
done with watershed mass balance and gridded hydrologic model–based measures of
regional ET (for example, see the work of Hamlet et al. 2007). The RS-ET algorithm has
been validated at 19 AmeriFlux eddy covariance tower sites (Table 19.2, Figure 19.5),
which cover six typical land cover types and a wide range of climates.
19.5.2 validation ReSultS
For each tower, ET was estimated using two different sets of meteorological data:
(1) integrated meteorological data derived from the half-hour observations at flux
tower sites and (2) the GMAO meteorological data at a 1.00° × 1.25° resolution. ET
was calculated for the vegetated 3 × 3 1-km 2 MODIS pixels surrounding each site
driven by the preprocessed GMAO and MODIS data and averaged across all pixels.
These averages were then compared with the tower ET observations.
For site observations of ET and meteorology, the half-hourly data provided by
the tower researchers were aggregated into daily data without using additional quality control, and there was no gap filling for these data to maintain the integrity and
originality of the observations. Since the observed water vapor fluxes are the sum of
the plant transpiration and soil evaporation and it is not possible to separate the two
fluxes using standard flux tower data, only the total ET estimates by RS-ET were
compared with the observed total ET.
Figure 19.6 shows the comparison of the annual mean MODIS LE estimated
using the ET algorithm to observations at 19 eddy covariance flux tower sites (Mu et
al. 2007a). The correlation coefficients between the LE observations and estimates
are R = .86 (p < .00001) when the algorithm is driven by tower-specific meteorology
(Figure 19.6a) and R = .86 (p < .00001) for the global GMAO meteorology (Figure
19.6b). The relative error between the 8-day averaged LE estimates driven by GMAO
and tower meteorology is 14.3%, indicating that meteorology plays an important role
in the accuracy of the RS-ET algorithm. In addition to predicting the annual ET, the
RS-ET algorithm captures seasonal variation, for example, at Duke_pine in North
Carolina and Barrow in Alaska as depicted in Figure 19.7.
19.5.3 unceRtainty in Site-BaSed validationS
When driven by tower meteorological data, the average RMSE of the 8-day latent
heat fluxes over the 19 flux towers by RS-ET against ET observations was 27.3 and
29.5 W ∙ m –2 , respectively, driven by GMAO meteorology. The average correlation
coefficient between the ET estimates and observations for the 8-day results was 0.72
with GMAO meteorological data and 0.76 with tower meteorological data. The existing biases between the ET estimates and the ET observations may be influenced by
the following:
Multiscale Hydrologic Remote Sensing: Perspectives and Applications
basis, and FLUXNET data are available for download from the Oak Ridge National
Laboratory Distributed Active Archive Center (http://daac.ornl.gov/FLUXNET/fluxnet
.shtml). These flux towers cover all typical land cover types and climates. This long-term
tower data will help scientists validate the regional and global ET results. The indirect
ET validation effort with precipitation and MODIS GPP at the regional and global scales
will also continue to help in the development of global ET product. More work will be
done with watershed mass balance and gridded hydrologic model–based measures of
regional ET (for example, see the work of Hamlet et al. 2007). The RS-ET algorithm has
been validated at 19 AmeriFlux eddy covariance tower sites (Table 19.2, Figure 19.5),
which cover six typical land cover types and a wide range of climates.
19.5.2 validation ReSultS
For each tower, ET was estimated using two different sets of meteorological data:
(1) integrated meteorological data derived from the half-hour observations at flux
tower sites and (2) the GMAO meteorological data at a 1.00° × 1.25° resolution. ET
was calculated for the vegetated 3 × 3 1-km 2 MODIS pixels surrounding each site
driven by the preprocessed GMAO and MODIS data and averaged across all pixels.
These averages were then compared with the tower ET observations.
For site observations of ET and meteorology, the half-hourly data provided by
the tower researchers were aggregated into daily data without using additional quality control, and there was no gap filling for these data to maintain the integrity and
originality of the observations. Since the observed water vapor fluxes are the sum of
the plant transpiration and soil evaporation and it is not possible to separate the two
fluxes using standard flux tower data, only the total ET estimates by RS-ET were
compared with the observed total ET.
Figure 19.6 shows the comparison of the annual mean MODIS LE estimated
using the ET algorithm to observations at 19 eddy covariance flux tower sites (Mu et
al. 2007a). The correlation coefficients between the LE observations and estimates
are R = .86 (p < .00001) when the algorithm is driven by tower-specific meteorology
(Figure 19.6a) and R = .86 (p < .00001) for the global GMAO meteorology (Figure
19.6b). The relative error between the 8-day averaged LE estimates driven by GMAO
and tower meteorology is 14.3%, indicating that meteorology plays an important role
in the accuracy of the RS-ET algorithm. In addition to predicting the annual ET, the
RS-ET algorithm captures seasonal variation, for example, at Duke_pine in North
Carolina and Barrow in Alaska as depicted in Figure 19.7.
19.5.3 unceRtainty in Site-BaSed validationS
When driven by tower meteorological data, the average RMSE of the 8-day latent
heat fluxes over the 19 flux towers by RS-ET against ET observations was 27.3 and
29.5 W ∙ m –2 , respectively, driven by GMAO meteorology. The average correlation
coefficient between the ET estimates and observations for the 8-day results was 0.72
with GMAO meteorological data and 0.76 with tower meteorological data. The existing biases between the ET estimates and the ET observations may be influenced by
the following: