Remote Sensing of Leaf, Canopy, and Vegetation Water Contents for Satellite Environmental Data Records - Satellite-based Applications on Climate Change - page 331

respectively. There were highly significant linear relationships between measured
and retrieved LWC and CWC for the three wavelength regions (Hunt et al. 2011a ).
Larger values of retrieved parameters are regularly observed in the analytical
procedures of near-infrared spectroscopy, which are attributed to multiple scatter-
ing (Zhang et al. 2011 ).
Because it is difficult to account for multiple scattering without additional terms
in Eq. 20.1 , the retrieved values of LWC and CWC probably should not be used,
except when the actual values of LWC or CWC are not relevant, such as with
atmospheric correction (Gao et al. 2009 ) or water-spectrum removal (Gao and
Goetz 1994 , 1995 ; Schlerf et al. 2010 ; Ramoelo et al. 2011 ; Wang et al. 2011a ).
Alternative methods for retrieving LWC or CWC with spectroscopy may be
insensitive to multiple scattering within a leaf and other confounding factors (Asner
and Martin 2008 ; Zhang et al. 2011 ). Li et al. ( 2008 ) used a genetic algorithm-
partial least squares regression and obtained very low root mean square errors
(RMSE), but the results did not scale between leaves and canopies. Cheng et al.
( 2011b ) used wavelet transforms to obtain similar results. While the recent research
is promising for the spectroscopic retrieval of LWC and CWC, the results are not
yet applicable for imaging spectroscopy.

20.5 Remotely Sensed Foliar Water Indices

Initially, vegetation indices such as NDVI were utilized because satellite multispec-
tral imagery was acquired as digitized radiances (digital numbers), whereas ground
data were usually measured reflectances. Jackson et al. ( 1983 ) wrote that an ideal
vegetation index should be sensitive to vegetation amount and insensitive to the
following: (1) soil background, (2) surface topography, (3) atmospheric effects, and
(4) solar zenith angle. Furthermore, an ideal vegetation index should be insensitive
to the bidirectional reflectance distribution functions of vegetation and soils (Huete
et al. 2002 ). Today, satellite imagery is frequently corrected for atmospheric effects
and solar-target-sensor geometry to obtain apparent land-surface reflectance (Gao
et al. 2009 ). Even with apparent land-surface reflectances, however, vegetation
indices are still a practical method for monitoring vegetation because (1) atmo-
spheric corrections usually assume a standard atmosphere and (2) surface topogra-
phy and soil background create variation in the apparent reflectances.
Hardisky et al. ( 1983 ) proposed that a normalized difference index using
Landsat Thematic Mapper (TM) bands 4 and 5 was related to the amount of
water in vegetation (Fig. 20.2 ). Kimes et al. ( 1981 ) studied this index but found
TM band 3 and TM band 5 reflectances were very highly correlated ( r ¼

0.97), so
there was no added benefit for using TM band 5 to monitor agronomic variables.
Hardisky et al. ( 1983 ) called NDVI simply the Vegetation Index and called the
normalized difference of TM bands 4 and 5 the Infrared Index. Hunt and Rock
( 1989 ) and Ji et al. ( 2011 ) recommended that the Infrared Index be called instead
the Normalized Difference Infrared Index (NDII, Table 20.2 ) parallel to the univer-
sally accepted NDVI. Hunt et al. ( 1987 ) and Hunt and Rock ( 1989 ) developed and