427
Microwave Vegetation Indices from Satellite Passive Microwave Sensors
NDVI in the April images. Regions covered by savannas and a deciduous broadleaf
forest exhibit moderate to low MVI B values and moderate to high NDVI and A values.
The north–south orientated stripe of shrubland along the west coast of the continent is
clearly shown in all optical- and microwave-derived vegetation index images for both
April and July. This is characterized by high microwave B values and low NDVI and
microwave A values. The Amazon River system is very distinct in the MVI images.
AMSR-E measurements in both frequencies cannot “see” through the dense Amazon
rainforest vegetation canopy (canopy fraction cover F v < 1), which results in polarization difference measurements close to zero for densely vegetated areas.
The measured microwave signals, Equations 18.14 and 18.15, only reflect the signals from the Amazon River. As a result, the B parameters derived from the bare
ground surfaces could have similar magnitudes as those derived from a dense forest
with the fraction cover F v < 1. It is clear that the derived MVIs can have a nonunique
relation in those regions where the sensor cannot “see” the ground surface in at least
one frequency. This is a limitation of the currently available AMSR-E sensor due to
its vegetation penetration capability. Therefore, it is likely that the technique should
only be used in regions with bare soil or short vegetation.
18.4.2.4 North America
This region shows the most disagreement between the MODIS and AMSR-E vegetation indices. At high latitudes, in northern Canada, the NDVI exhibits a large
change in magnitude and coverage, extending northward significantly from April to
July. However, the MVIs show very little response to the vegetation changes in these
regions. The MVIs from the low-frequency pair (6.925 GHz/10.65 GHz) show a reasonable agreement in pattern and seasonal change within regions of the western and
central portion of the United States but have poor agreement in the eastern to central
part of the United States. The microwave MVIs derived with the high-frequency pair
(10.65 GHz/18.7 GHz) agree well in both pattern and seasonal change in April and
July. This behavior is most likely due to widespread significant RFI in the 6.925 GHz
measurements over the United States (Li et al. 2004; Njoku et al. 2005).
As a further examination of the capability of the microwave vegetation indices in
monitoring seasonal changes of vegetation, we calculated the coefficients of variation for both the NDVI and MVIs for January 1 to December 31, 2003. The coefficient of variation is a measure of the degree of variation (defined as the ratio of the
standard deviation to its mean value). It can be used as an indication of the seasonal
variation of the index relative to its annual mean value during 2003. Figure 18.4
shows the coefficient of variation values for the optical and microwave indices. The
global statistical mean values of the coefficient of variation in 2003 for each major
land cover type from the IGBP database are summarized in Table 18.2. With the
exception of grassland, all of the A values of the MVIs indicate seasonal variations
ranging from 0.08 to almost 0.50 of their annual mean values, which is less than that
observed for NDVI (0.20–0.63). Seasonal variations of the B values range from 0.05
to almost 0.23 of their annual mean values. It was expected that the A values would
exhibit greater seasonal variation than B, because they are affected not only by the
vegetation properties but also by the surface temperature, which would have some
correlation with vegetation growth.
Microwave Vegetation Indices from Satellite Passive Microwave Sensors
NDVI in the April images. Regions covered by savannas and a deciduous broadleaf
forest exhibit moderate to low MVI B values and moderate to high NDVI and A values.
The north–south orientated stripe of shrubland along the west coast of the continent is
clearly shown in all optical- and microwave-derived vegetation index images for both
April and July. This is characterized by high microwave B values and low NDVI and
microwave A values. The Amazon River system is very distinct in the MVI images.
AMSR-E measurements in both frequencies cannot “see” through the dense Amazon
rainforest vegetation canopy (canopy fraction cover F v < 1), which results in polarization difference measurements close to zero for densely vegetated areas.
The measured microwave signals, Equations 18.14 and 18.15, only reflect the signals from the Amazon River. As a result, the B parameters derived from the bare
ground surfaces could have similar magnitudes as those derived from a dense forest
with the fraction cover F v < 1. It is clear that the derived MVIs can have a nonunique
relation in those regions where the sensor cannot “see” the ground surface in at least
one frequency. This is a limitation of the currently available AMSR-E sensor due to
its vegetation penetration capability. Therefore, it is likely that the technique should
only be used in regions with bare soil or short vegetation.
18.4.2.4 North America
This region shows the most disagreement between the MODIS and AMSR-E vegetation indices. At high latitudes, in northern Canada, the NDVI exhibits a large
change in magnitude and coverage, extending northward significantly from April to
July. However, the MVIs show very little response to the vegetation changes in these
regions. The MVIs from the low-frequency pair (6.925 GHz/10.65 GHz) show a reasonable agreement in pattern and seasonal change within regions of the western and
central portion of the United States but have poor agreement in the eastern to central
part of the United States. The microwave MVIs derived with the high-frequency pair
(10.65 GHz/18.7 GHz) agree well in both pattern and seasonal change in April and
July. This behavior is most likely due to widespread significant RFI in the 6.925 GHz
measurements over the United States (Li et al. 2004; Njoku et al. 2005).
As a further examination of the capability of the microwave vegetation indices in
monitoring seasonal changes of vegetation, we calculated the coefficients of variation for both the NDVI and MVIs for January 1 to December 31, 2003. The coefficient of variation is a measure of the degree of variation (defined as the ratio of the
standard deviation to its mean value). It can be used as an indication of the seasonal
variation of the index relative to its annual mean value during 2003. Figure 18.4
shows the coefficient of variation values for the optical and microwave indices. The
global statistical mean values of the coefficient of variation in 2003 for each major
land cover type from the IGBP database are summarized in Table 18.2. With the
exception of grassland, all of the A values of the MVIs indicate seasonal variations
ranging from 0.08 to almost 0.50 of their annual mean values, which is less than that
observed for NDVI (0.20–0.63). Seasonal variations of the B values range from 0.05
to almost 0.23 of their annual mean values. It was expected that the A values would
exhibit greater seasonal variation than B, because they are affected not only by the
vegetation properties but also by the surface temperature, which would have some
correlation with vegetation growth.