intercomparing products from different sensor systems to test consistency. Both
direct and indirect validation provides a comprehensive knowledge about the
accuracy of these products and level of uncertainties that may results due to input
data and modeling errors.
Direct validation results for the MODIS LAI/FPAR over vegetation types
representative of all the major biome types suggest that the product provides
reasonable estimates of LAI for most cover types and land use types (Garrigues
et al. 2008; Huang et al. 2006; Kauwe et al. 2011; Pisek and Chen 2007; Sea et al.
2011; Tan et al. 2005; Yang et al. 2006). The MODIS LAI/FPAR products are
categorized as a Stage 2 land validated product (http://landval.gsfc.nasa.gov/
ProductStatus.php?ProductID=MOD15) that have the following characteristics: (a)
LAI accuracy of 0.5 LAI units (uncertainty of 0.66 LAI), FPAR accuracy of 0.1;
(b) spatial resolution from 500 m to 1 km; (c) temporal frequency from 4 days to
monthly (Yang et al. 2006). Direct validation of the GEOV-1 products also shows
a satisfactory agreement with field observations. An indirect validation implementing a scaled version of the MODIS algorithm to derive an LAI dataset from
AVHRR shows satisfactory agreement with the MODIS and CYCLOPES LAI
products at a range of spatial resolutions and field data (Ganguly et al. 2008a). The
Landsat based LAI products are not rigorously validated, however an indirect
validation with MODIS shows comparable results (Ganguly et al. 2012).
Figure 2.4 briefly demonstrates the results obtained from validation exercises
performed with the AVHRR, MODIS and GEOV suite of LAI products.
2.7 Concluding Remarks
Current scientific research and application studies have demonstrated the usefulness of physically derived LAI/FPAR products at local-to-regional scales; however, there are certain limitations in physically based approaches. First, data
measurement uncertainties from different sensors can impact the retrieval of a
biophysical product. Data uncertainties mostly result from calibration ambiguities,
current state of the atmospheric correction algorithm and other effects introduced
by solar/view angle corrections. Second, global retrievals of LAI/FPAR products
utilize land cover classification maps. Classification inaccuracies are a critical
source of error in the LAI retrieval process, especially for those regions undergoing dynamic land cover change (e.g. changes from herbaceous to woody biomes). There are intrinsic limitations in the retrieval algorithms that mostly include
(1) accurately modeling the uncertainty of the input reflectances and incorporating
the variability in model and input uncertainties with biome types; (2) incorporating
a better understory reflectance characterization in simulating the soil reflectance
behavior and (3) using constrained definitions of leaf spectral properties as defined
by the broad biome types. Finally, a global validation of coarse-to-fine resolution
2 Green Leaf Area and Fraction of Photosynthetically
55
direct and indirect validation provides a comprehensive knowledge about the
accuracy of these products and level of uncertainties that may results due to input
data and modeling errors.
Direct validation results for the MODIS LAI/FPAR over vegetation types
representative of all the major biome types suggest that the product provides
reasonable estimates of LAI for most cover types and land use types (Garrigues
et al. 2008; Huang et al. 2006; Kauwe et al. 2011; Pisek and Chen 2007; Sea et al.
2011; Tan et al. 2005; Yang et al. 2006). The MODIS LAI/FPAR products are
categorized as a Stage 2 land validated product (http://landval.gsfc.nasa.gov/
ProductStatus.php?ProductID=MOD15) that have the following characteristics: (a)
LAI accuracy of 0.5 LAI units (uncertainty of 0.66 LAI), FPAR accuracy of 0.1;
(b) spatial resolution from 500 m to 1 km; (c) temporal frequency from 4 days to
monthly (Yang et al. 2006). Direct validation of the GEOV-1 products also shows
a satisfactory agreement with field observations. An indirect validation implementing a scaled version of the MODIS algorithm to derive an LAI dataset from
AVHRR shows satisfactory agreement with the MODIS and CYCLOPES LAI
products at a range of spatial resolutions and field data (Ganguly et al. 2008a). The
Landsat based LAI products are not rigorously validated, however an indirect
validation with MODIS shows comparable results (Ganguly et al. 2012).
Figure 2.4 briefly demonstrates the results obtained from validation exercises
performed with the AVHRR, MODIS and GEOV suite of LAI products.
2.7 Concluding Remarks
Current scientific research and application studies have demonstrated the usefulness of physically derived LAI/FPAR products at local-to-regional scales; however, there are certain limitations in physically based approaches. First, data
measurement uncertainties from different sensors can impact the retrieval of a
biophysical product. Data uncertainties mostly result from calibration ambiguities,
current state of the atmospheric correction algorithm and other effects introduced
by solar/view angle corrections. Second, global retrievals of LAI/FPAR products
utilize land cover classification maps. Classification inaccuracies are a critical
source of error in the LAI retrieval process, especially for those regions undergoing dynamic land cover change (e.g. changes from herbaceous to woody biomes). There are intrinsic limitations in the retrieval algorithms that mostly include
(1) accurately modeling the uncertainty of the input reflectances and incorporating
the variability in model and input uncertainties with biome types; (2) incorporating
a better understory reflectance characterization in simulating the soil reflectance
behavior and (3) using constrained definitions of leaf spectral properties as defined
by the broad biome types. Finally, a global validation of coarse-to-fine resolution
2 Green Leaf Area and Fraction of Photosynthetically
55