sensitive to the healthy vegetation compared to the visible portion of the spectrum
due to photosynthetic activities, the ratio between the infrared and visible bands
provides an indicator to distinguish between the healthy vegetation and damaged
ones (Smith 1997; Frazier and Page 2000; Sun et al. 2011; Shrestha and Di 2013).
Vegetation Indices (VI) are primary examples of spectral band ratioing which
provide an indicator of vegetation condition. The flood crop damage assessment is
performed by examining VI before and after the flood in a specific location (or pixel)
and applying the VI difference into a relation model to compute crop damage and
acreage loss. The general idea behind using VI to assess the crop damage is that as it
indicates the greenness (health) of the vegetation, decrease of VI after the flood event
indicates possible damage to the crop, and these differences in VI could be further
quantified to assess actual crop loss.
VI-based crop loss assessment methods, compared to classification-based
methods, are much more efficient and robust. By transforming these daily products
into weekly/bi-weekly composite products, VI, to some extent, can reduce abnormalities in data such as cloud cover effects. It is a process of combining multiday VI
time series (7 days, 10 days, 14 days, etc.) and selecting a single best (usually
maximum) VI value to represent the entire composite VI timeline. The most often
used composite data sets are 10-day Maximum Value Composite (MVC) products
(Holben 1986; Chen et al. 2004). It is a pixel-by-pixel method where the VI value for
each pixel is examined over 10 days period of time, and the maximum VI is selected
as a representative VI for that entire 10 days period. These composite products
improve data quality and help ensure the elimination of impurities with at least one
clear observation within the composite time series; however, if all observations
within the composite period contain noise (cloud cover due to continuous rain for
10 days), this method could be ineffective. Additionally, the temporal resolution of
these multiday composite products becomes coarse and may not be efficient to
capture and monitor crop vegetation index variants within a short period of time,
especially for daily-level change analysis. For example, to address the crop phenological changes within a growing season would require examining daily-level VI
time series. Similarly, assessing the crop loss due to the oversupply of water
(Yu et al. 2013) would also require performing daily VI assessments. These limitations of composite methods can be, however, addressed by statistical and/or
mathematical-based models to reduce the anomalies in a daily VI time series.
These models utilize some version of curve fitting methods to first identify the
anomaly observations in the time series and then smooth these values based on the
fitting algorithm. Some of these statistical approaches include Fourier-based, asymmetric function, threshold-based, and iterative-based fitting (Chen et al. 2004).
Various VI-utilizing remote sensing data have been applied for crop yield managements such as normalized difference vegetation index (NDVI), vegetation condition index (VCI), and temperature condition index (TCI) (Prasad et al. 2006).
NDVI uses a ratio between the infrared and red spectral bands to derive an index
that measures the greenness (growth condition) of the vegetation cover. Equation
16.1 shows the mathematical formula to calculate the NDVI:
334
R. M. Shrestha and M. S. Rahman
due to photosynthetic activities, the ratio between the infrared and visible bands
provides an indicator to distinguish between the healthy vegetation and damaged
ones (Smith 1997; Frazier and Page 2000; Sun et al. 2011; Shrestha and Di 2013).
Vegetation Indices (VI) are primary examples of spectral band ratioing which
provide an indicator of vegetation condition. The flood crop damage assessment is
performed by examining VI before and after the flood in a specific location (or pixel)
and applying the VI difference into a relation model to compute crop damage and
acreage loss. The general idea behind using VI to assess the crop damage is that as it
indicates the greenness (health) of the vegetation, decrease of VI after the flood event
indicates possible damage to the crop, and these differences in VI could be further
quantified to assess actual crop loss.
VI-based crop loss assessment methods, compared to classification-based
methods, are much more efficient and robust. By transforming these daily products
into weekly/bi-weekly composite products, VI, to some extent, can reduce abnormalities in data such as cloud cover effects. It is a process of combining multiday VI
time series (7 days, 10 days, 14 days, etc.) and selecting a single best (usually
maximum) VI value to represent the entire composite VI timeline. The most often
used composite data sets are 10-day Maximum Value Composite (MVC) products
(Holben 1986; Chen et al. 2004). It is a pixel-by-pixel method where the VI value for
each pixel is examined over 10 days period of time, and the maximum VI is selected
as a representative VI for that entire 10 days period. These composite products
improve data quality and help ensure the elimination of impurities with at least one
clear observation within the composite time series; however, if all observations
within the composite period contain noise (cloud cover due to continuous rain for
10 days), this method could be ineffective. Additionally, the temporal resolution of
these multiday composite products becomes coarse and may not be efficient to
capture and monitor crop vegetation index variants within a short period of time,
especially for daily-level change analysis. For example, to address the crop phenological changes within a growing season would require examining daily-level VI
time series. Similarly, assessing the crop loss due to the oversupply of water
(Yu et al. 2013) would also require performing daily VI assessments. These limitations of composite methods can be, however, addressed by statistical and/or
mathematical-based models to reduce the anomalies in a daily VI time series.
These models utilize some version of curve fitting methods to first identify the
anomaly observations in the time series and then smooth these values based on the
fitting algorithm. Some of these statistical approaches include Fourier-based, asymmetric function, threshold-based, and iterative-based fitting (Chen et al. 2004).
Various VI-utilizing remote sensing data have been applied for crop yield managements such as normalized difference vegetation index (NDVI), vegetation condition index (VCI), and temperature condition index (TCI) (Prasad et al. 2006).
NDVI uses a ratio between the infrared and red spectral bands to derive an index
that measures the greenness (growth condition) of the vegetation cover. Equation
16.1 shows the mathematical formula to calculate the NDVI:
334
R. M. Shrestha and M. S. Rahman