sensors. All these lead to a heightened requirement on fine geometric correction
and radiometric correction to establish the comparability of values up to pixel
levels across time and sensors. The actual methods and approaches can be
referred to in the previous data preprocessing section for cropland mapping.
3. Condition indicators: Crop growth condition and stages changes over environmental conditions and growing stages. Its health or stage status is related to the
content of chlorophyll representing the coverage of green leaves and density of
leaves (e.g., leaf area index). Crop emits thermal radiance while absorbing red
radiance. These biological and physical characteristics and relationships form the
bases for developing many vegetation indices that enhance the signal of crop
coverage while subduing other signals. Vegetation indices are developed from
these assumptions and theoretical bases. They have been extensively used in
monitoring crop status.
There are many vegetation indices developed in the past decades since the
1970s (Bannari et al. 1995; Silleos et al. 2006; Yang et al. 2009, 2011b). The
comprehensive reviews of vegetation indices can be seen in (Bannari et al. 1995;
Silleos et al. 2006; Basso et al. 2013). Table 10.4 lists selected vegetation indices
that have been applied in crop status monitoring.
4. Denoising and modeling: The calculated crop condition indicators (e.g., NDVI)
may be contaminated and undulated abnormally over time due to the contamination of cloud, haze, and atmospheric conditions. Before they can be used in
evaluating and comparing the crop status, smoothing or model fitting is often
called for to eliminate bad data and enhance the trend. There are several levels of
smoothing/filtering: eliminating extremes, smoothing, and kernel-fitting.
Table 10.4 lists selected algorithms for eliminating, smoothing, or kernel-fitting
time series of vegetation indices. The first level is to eliminate those abnormal that
are way off and highly suspicious as bad records in terms of crop status monitoring. The Best Index Slope Extraction (BISE) is a typical method to eliminate
extreme values during the growth stages. This algorithm assumes that the contamination of cloud or haze causes NDVI to be lower than usual and the drop in a
short period cannot be extreme considering the crop growth. The second level is
to smooth and interpolate the time series using certain smoothing algorithms. The
third level is to use an underlying kernel to fit the curve of vegetation indices over
a growing season. This is typically done for estimating the crop growth stage.
This makes it logical that once a crop at a location reaches a growth stage, it
should not fall back into an early stage. The monotonic increase in pre-peak stage
or the monotonic decrease in post-peak stage should be assured for avoiding such
an illogical estimation to happen.
5. Condition evaluation: To determine what the crop status is, there are several
approaches (Meng and Wu 2008). First, the condition indicators are directly used
to reflect the crop status. In general, these indicators are often designed to be
positively correlated with the crop condition. The higher the crop condition
indicator is, the better the crop condition is. For example, VCI is related to the
water condition of cropland (Kogan 1995; Yang et al. 2011b; Yu et al. 2012a).
10 Crop Pattern and Status Monitoring
187
and radiometric correction to establish the comparability of values up to pixel
levels across time and sensors. The actual methods and approaches can be
referred to in the previous data preprocessing section for cropland mapping.
3. Condition indicators: Crop growth condition and stages changes over environmental conditions and growing stages. Its health or stage status is related to the
content of chlorophyll representing the coverage of green leaves and density of
leaves (e.g., leaf area index). Crop emits thermal radiance while absorbing red
radiance. These biological and physical characteristics and relationships form the
bases for developing many vegetation indices that enhance the signal of crop
coverage while subduing other signals. Vegetation indices are developed from
these assumptions and theoretical bases. They have been extensively used in
monitoring crop status.
There are many vegetation indices developed in the past decades since the
1970s (Bannari et al. 1995; Silleos et al. 2006; Yang et al. 2009, 2011b). The
comprehensive reviews of vegetation indices can be seen in (Bannari et al. 1995;
Silleos et al. 2006; Basso et al. 2013). Table 10.4 lists selected vegetation indices
that have been applied in crop status monitoring.
4. Denoising and modeling: The calculated crop condition indicators (e.g., NDVI)
may be contaminated and undulated abnormally over time due to the contamination of cloud, haze, and atmospheric conditions. Before they can be used in
evaluating and comparing the crop status, smoothing or model fitting is often
called for to eliminate bad data and enhance the trend. There are several levels of
smoothing/filtering: eliminating extremes, smoothing, and kernel-fitting.
Table 10.4 lists selected algorithms for eliminating, smoothing, or kernel-fitting
time series of vegetation indices. The first level is to eliminate those abnormal that
are way off and highly suspicious as bad records in terms of crop status monitoring. The Best Index Slope Extraction (BISE) is a typical method to eliminate
extreme values during the growth stages. This algorithm assumes that the contamination of cloud or haze causes NDVI to be lower than usual and the drop in a
short period cannot be extreme considering the crop growth. The second level is
to smooth and interpolate the time series using certain smoothing algorithms. The
third level is to use an underlying kernel to fit the curve of vegetation indices over
a growing season. This is typically done for estimating the crop growth stage.
This makes it logical that once a crop at a location reaches a growth stage, it
should not fall back into an early stage. The monotonic increase in pre-peak stage
or the monotonic decrease in post-peak stage should be assured for avoiding such
an illogical estimation to happen.
5. Condition evaluation: To determine what the crop status is, there are several
approaches (Meng and Wu 2008). First, the condition indicators are directly used
to reflect the crop status. In general, these indicators are often designed to be
positively correlated with the crop condition. The higher the crop condition
indicator is, the better the crop condition is. For example, VCI is related to the
water condition of cropland (Kogan 1995; Yang et al. 2011b; Yu et al. 2012a).
10 Crop Pattern and Status Monitoring
187