422
Multiscale Hydrologic Remote Sensing: Perspectives and Applications
18.4 EVALUATION OF THE MICROWAVE VEGETATION INDICES
To assess or evaluate these new microwave vegetation indices, we need to compare
them with ground-based studies or other satellite-derived vegetation indices at the
regional and global scales. Ground measurements can provide the accurate vegetation information and are critical for quantitatively assessing microwave interactions
with the vegetation properties; however, there are only a few such data sets available
from specific field campaigns for a limited range of land cover types (Jackson et al.
2005). The measurements are commonly obtained at points or at very high resolutions and are not comparable to the spatial scale of the passive microwave satellite
measurements. Because of the complexity of natural surfaces, especially for coarseresolution passive microwave measurements at tens of kilometers, we do not fully
understand how to scale small-scale relations to satellite footprints. Therefore, for
the purposes of this initial study, we chose to compare the new MVIs with the widely
accepted optical sensor vegetation index (NDVI). We compared the general global
patterns and the seasonal phenology of NDVI over a 1-year period (2003) to the
MVIs to assess their potential in global vegetation monitoring. The global map of the
land cover types in the International Geosphere-Biosphere Program (IGBP) scheme
(http://modis-land.gsfc.nasa.gov/landcover.htm) is used to interpret the analyses.
18.4.1 data
One year of AMSR-E level 3 brightness temperature data from 1 January to 31
December 2003, obtained from the National Snow and Ice Data Center (http://nsidc
.org/daac/amsre/), were analyzed. This product is the 25 km × 25 km grid data resagrid
resampled from AMSR-E level 2A brightness temperature data into a global EASE-GRID
projection (http://nsidc.org/data/docs/daac/ae_land3_l3_soil_moisture.gd.html). Only
the descending pass data (night pass) were used to minimize the errors that could result
from the physical temperature differences between vegetation and soil.
It has been demonstrated that radiofrequency interference (RFI) has a significant
impact on the retrieval of land surface geophysical properties from satellite microwave
observations (Li et al. 2004; Njoku et al. 2005). While RFI has not been well characterized, strong RFI may result in an irregular frequency gradient of the observed brightness temperatures. Except for snow-covered areas, the observed microwave signals at the
higher frequency are generally greater than that observed at the lower frequency when
there is no significant atmospheric effect. A negative frequency gradient for the satellite observations has been utilized to identify strong RFI signals (Li et al. 2004; Njoku
et al. 2005). However, an effective technique for identifying weak RFI signals, which
may contribute to significant errors in estimating land surface properties, has not been
developed yet. For the MVIs, RFI may result in the estimated MVIs being out of range.
Therefore, evaluating the A and B parameters derived from the different frequency pairs
against their normal observed range, A > 0 and 0 < B < 1, at each frequency pair can be
used to partially eliminate the pixels that are contaminated.
The general characteristics of the estimated MVIs described in Section 18.3.2
and the effects of strong RFI on the MVIs allow us to establish certain criteria for
quality control in interpreting the MVIs, which are summarized in Table 18.1. The
Multiscale Hydrologic Remote Sensing: Perspectives and Applications
18.4 EVALUATION OF THE MICROWAVE VEGETATION INDICES
To assess or evaluate these new microwave vegetation indices, we need to compare
them with ground-based studies or other satellite-derived vegetation indices at the
regional and global scales. Ground measurements can provide the accurate vegetation information and are critical for quantitatively assessing microwave interactions
with the vegetation properties; however, there are only a few such data sets available
from specific field campaigns for a limited range of land cover types (Jackson et al.
2005). The measurements are commonly obtained at points or at very high resolutions and are not comparable to the spatial scale of the passive microwave satellite
measurements. Because of the complexity of natural surfaces, especially for coarseresolution passive microwave measurements at tens of kilometers, we do not fully
understand how to scale small-scale relations to satellite footprints. Therefore, for
the purposes of this initial study, we chose to compare the new MVIs with the widely
accepted optical sensor vegetation index (NDVI). We compared the general global
patterns and the seasonal phenology of NDVI over a 1-year period (2003) to the
MVIs to assess their potential in global vegetation monitoring. The global map of the
land cover types in the International Geosphere-Biosphere Program (IGBP) scheme
(http://modis-land.gsfc.nasa.gov/landcover.htm) is used to interpret the analyses.
18.4.1 data
One year of AMSR-E level 3 brightness temperature data from 1 January to 31
December 2003, obtained from the National Snow and Ice Data Center (http://nsidc
.org/daac/amsre/), were analyzed. This product is the 25 km × 25 km grid data resagrid
resampled from AMSR-E level 2A brightness temperature data into a global EASE-GRID
projection (http://nsidc.org/data/docs/daac/ae_land3_l3_soil_moisture.gd.html). Only
the descending pass data (night pass) were used to minimize the errors that could result
from the physical temperature differences between vegetation and soil.
It has been demonstrated that radiofrequency interference (RFI) has a significant
impact on the retrieval of land surface geophysical properties from satellite microwave
observations (Li et al. 2004; Njoku et al. 2005). While RFI has not been well characterized, strong RFI may result in an irregular frequency gradient of the observed brightness temperatures. Except for snow-covered areas, the observed microwave signals at the
higher frequency are generally greater than that observed at the lower frequency when
there is no significant atmospheric effect. A negative frequency gradient for the satellite observations has been utilized to identify strong RFI signals (Li et al. 2004; Njoku
et al. 2005). However, an effective technique for identifying weak RFI signals, which
may contribute to significant errors in estimating land surface properties, has not been
developed yet. For the MVIs, RFI may result in the estimated MVIs being out of range.
Therefore, evaluating the A and B parameters derived from the different frequency pairs
against their normal observed range, A > 0 and 0 < B < 1, at each frequency pair can be
used to partially eliminate the pixels that are contaminated.
The general characteristics of the estimated MVIs described in Section 18.3.2
and the effects of strong RFI on the MVIs allow us to establish certain criteria for
quality control in interpreting the MVIs, which are summarized in Table 18.1. The