455
Remote Sensing and Modeling of Global Evapotranspiration
NPP (Running et al. 2004). Remote sensing inputs include MOD12Q1 land cover
(Friedl et al. 2002), MOD13A2 EVI (Huete et al. 2002), MOD15A2 LAI (Myneni et
al. 2002), and the 0.05° albedo from MOD43C1 (Schaaf et al. 2002).
19.4.1 SPatially SMoothing coaRSe ReSolution
gMao into ModiS ReSolution
The resolution for GMAO meteorological data is too coarse for a 1-km 2 MODIS
pixel. Zhao et al. (2005) found that, in the Collection 4 MODIS GPP/NPP algorithm
(MOD17), each 1-km pixel falling into the same 1.00° × 1.25° GMAO grid cell inherited the same meteorological data, creating a noticeable GMAO footprint (Zhao et
al. 2005) (Figure 19.1a and c). Such treatment may be acceptable on the global or
regional scale, but it can lead to large inaccuracies at the local scale, especially for
terrain with topographical variation or located in relatively abruptly climatic gradient zones. To enhance the meteorological inputs, Zhao et al. (2005) have nonlinearly
interpolated the coarse resolution GMAO data to the 1-km 2 MODIS pixel level based
on the four GMAO cells surrounding a given pixel. Theoretically, this GMAO spatial
interpolation improves the accuracy of meteorological data for each 1-km 2 pixel,
because it removes the abrupt changes from one side of a GMAO boundary to the
other. In addition, for most World Meteorological Organization (WMO) stations,
spatial interpolation reduced the RMSE and increased the correlation between the
GMAO data and the observed WMO daily weather data for 2000–2003, suggesting
that the nonlinear spatial interpolation considerably improves GMAO inputs. For the
RS-ET process, this method was also adopted to enhance the GMAO quality at the
MODIS pixel level.
19.4.2 teMPoRally inteRPolating contaMinated oR MiSSing ModiS data
Similar to MODIS GPP/NPP datasets, the input MODIS datasets have gaps caused
by cloud contaminations or other unfavorable atmospheric conditions (Zhao et al.
2005). These problems are solved by temporally filling input data gaps as proposed
by Zhao et al. (2005).
The 8-day MODIS LAI (MOD15A2) (Myneni et al. 2002) and 16-day MODIS
EVI (MOD13A2) (Huete et al. 2002) contain some cloud-contaminated or missing
data. According to the MOD15A2 quality assessment scheme provided by Myneni
et al. (2002), Fraction of Photosynthetically Active Radiation (FPAR)/LAI values
retrieved by the main algorithm (i.e., radiation transfer process, denoted as RT) are
most reliable, and those retrieved by the backup algorithm (i.e., the empirical relationship between FPAR/LAI and NDVI) are less reliable, because the backup algorithm is employed mostly when cloud cover, strong atmospheric effects, or snow/
ice is detected. The LAI retrievals by the backup algorithm have low quality and
should not be used for validation and other studies (Yang et al. 2006). The missing
or unreliable LAI, NDVI, and EVI at each 1-km MODIS pixel are temporally filled
based on their corresponding quality assessment data fields as proposed by Zhao
et al. (2005). The process entails two steps (Zhao et al. 2005) (Figure 1a and c).
If the first (or last) 8-day LAI (16-day NDVI, EVI) is unreliable or missing, it will
Remote Sensing and Modeling of Global Evapotranspiration
NPP (Running et al. 2004). Remote sensing inputs include MOD12Q1 land cover
(Friedl et al. 2002), MOD13A2 EVI (Huete et al. 2002), MOD15A2 LAI (Myneni et
al. 2002), and the 0.05° albedo from MOD43C1 (Schaaf et al. 2002).
19.4.1 SPatially SMoothing coaRSe ReSolution
gMao into ModiS ReSolution
The resolution for GMAO meteorological data is too coarse for a 1-km 2 MODIS
pixel. Zhao et al. (2005) found that, in the Collection 4 MODIS GPP/NPP algorithm
(MOD17), each 1-km pixel falling into the same 1.00° × 1.25° GMAO grid cell inherited the same meteorological data, creating a noticeable GMAO footprint (Zhao et
al. 2005) (Figure 19.1a and c). Such treatment may be acceptable on the global or
regional scale, but it can lead to large inaccuracies at the local scale, especially for
terrain with topographical variation or located in relatively abruptly climatic gradient zones. To enhance the meteorological inputs, Zhao et al. (2005) have nonlinearly
interpolated the coarse resolution GMAO data to the 1-km 2 MODIS pixel level based
on the four GMAO cells surrounding a given pixel. Theoretically, this GMAO spatial
interpolation improves the accuracy of meteorological data for each 1-km 2 pixel,
because it removes the abrupt changes from one side of a GMAO boundary to the
other. In addition, for most World Meteorological Organization (WMO) stations,
spatial interpolation reduced the RMSE and increased the correlation between the
GMAO data and the observed WMO daily weather data for 2000–2003, suggesting
that the nonlinear spatial interpolation considerably improves GMAO inputs. For the
RS-ET process, this method was also adopted to enhance the GMAO quality at the
MODIS pixel level.
19.4.2 teMPoRally inteRPolating contaMinated oR MiSSing ModiS data
Similar to MODIS GPP/NPP datasets, the input MODIS datasets have gaps caused
by cloud contaminations or other unfavorable atmospheric conditions (Zhao et al.
2005). These problems are solved by temporally filling input data gaps as proposed
by Zhao et al. (2005).
The 8-day MODIS LAI (MOD15A2) (Myneni et al. 2002) and 16-day MODIS
EVI (MOD13A2) (Huete et al. 2002) contain some cloud-contaminated or missing
data. According to the MOD15A2 quality assessment scheme provided by Myneni
et al. (2002), Fraction of Photosynthetically Active Radiation (FPAR)/LAI values
retrieved by the main algorithm (i.e., radiation transfer process, denoted as RT) are
most reliable, and those retrieved by the backup algorithm (i.e., the empirical relationship between FPAR/LAI and NDVI) are less reliable, because the backup algorithm is employed mostly when cloud cover, strong atmospheric effects, or snow/
ice is detected. The LAI retrievals by the backup algorithm have low quality and
should not be used for validation and other studies (Yang et al. 2006). The missing
or unreliable LAI, NDVI, and EVI at each 1-km MODIS pixel are temporally filled
based on their corresponding quality assessment data fields as proposed by Zhao
et al. (2005). The process entails two steps (Zhao et al. 2005) (Figure 1a and c).
If the first (or last) 8-day LAI (16-day NDVI, EVI) is unreliable or missing, it will