capacity in determining the accurate acreage with the resolution of remote sensing
data. The identification of crop types using satellite data remains a technical
challenge due to the diversity of cropping systems – crop types, crop varieties,
management practices, and field sizes (Song et al. 2017). Agricultural landscapes
are too complex to accurately classified each cropland with satellite remote
sensing in use (Wu and Li 2012). Acreage estimation cannot be done directly
using pixel counting due to misclassification and the existence of mixed pixels
(Gallego 2004).
2. Mismatched crop mapping timing: Optical remote sensing is often affected by
cloud cover (Hale et al. 1999; Allen et al. 2002). High-resolution satellite remote
sensing has limited frequency of revisits. The timing of acquired remotely sensed
data may not be well fit for the classification of certain crops. For example,
soybean and corn have a 2-week difference in Iowa State. This is a useful
signature for distinguishing these two types of crops. However, the revisit of
Landsat is more than 2 weeks apart. One clouded image acquisition may lead to
the completely missed pair images for distinguishing the two crops.
3. No distinguishable spectral signatures for accurately identifying crop types: The
spectral signatures for certain crops are still hard to find due to their similarity
with other crops or surroundings.
4. Complexity of crop classification training: The current method of applying
classification method with remote sensing is mostly based on supervised
approach. The classifier is often trained with uncorrected spectral measurements
that lead to the limited applicability of trained classifier. The trained classifier is
applicable to the similar images with the similar time or even just the scene where
the training samples are collected. There are still technical barriers preventing the
classifier to be trained once and be applicable to all images of the same sensor.
Several recent advancements of remote sensing technologies may help in elevating the role of remote sensing in crop acreage estimates and eventually evolve as an
operational method. These advancements and their impacts are summarized as
follows.
1. Improved spatial resolution: Satellite remote sensing is reaching a submeter
resolution. This will make it possible to accurately identify cropland with detailed
spectral and textual signatures. The advancement of computing technologies,
especially cloud computing, allows the paralleled speedup of processing large
volume of data, which makes it possible to classify images of extremely high
resolution.
2. Improved temporal resolution: The temporal resolution of the extremely highspatial-resolution remote sensing is also significantly improved. Daily revisits are
common through the constellation of many satellites. This would help in solving
the timing problem of image acquisition. Cloud coverage may be eliminated to a
great degree due to the frequent revisits.
3. Beyond spectral signatures: The resolution of Radar remote sensing is also
significantly improved. The all-weather capability of Radar will significantly
eliminate the problem of cloud coverage. The intensity signal of RADAR and
184
E. G. Yu and Z. Yang
data. The identification of crop types using satellite data remains a technical
challenge due to the diversity of cropping systems – crop types, crop varieties,
management practices, and field sizes (Song et al. 2017). Agricultural landscapes
are too complex to accurately classified each cropland with satellite remote
sensing in use (Wu and Li 2012). Acreage estimation cannot be done directly
using pixel counting due to misclassification and the existence of mixed pixels
(Gallego 2004).
2. Mismatched crop mapping timing: Optical remote sensing is often affected by
cloud cover (Hale et al. 1999; Allen et al. 2002). High-resolution satellite remote
sensing has limited frequency of revisits. The timing of acquired remotely sensed
data may not be well fit for the classification of certain crops. For example,
soybean and corn have a 2-week difference in Iowa State. This is a useful
signature for distinguishing these two types of crops. However, the revisit of
Landsat is more than 2 weeks apart. One clouded image acquisition may lead to
the completely missed pair images for distinguishing the two crops.
3. No distinguishable spectral signatures for accurately identifying crop types: The
spectral signatures for certain crops are still hard to find due to their similarity
with other crops or surroundings.
4. Complexity of crop classification training: The current method of applying
classification method with remote sensing is mostly based on supervised
approach. The classifier is often trained with uncorrected spectral measurements
that lead to the limited applicability of trained classifier. The trained classifier is
applicable to the similar images with the similar time or even just the scene where
the training samples are collected. There are still technical barriers preventing the
classifier to be trained once and be applicable to all images of the same sensor.
Several recent advancements of remote sensing technologies may help in elevating the role of remote sensing in crop acreage estimates and eventually evolve as an
operational method. These advancements and their impacts are summarized as
follows.
1. Improved spatial resolution: Satellite remote sensing is reaching a submeter
resolution. This will make it possible to accurately identify cropland with detailed
spectral and textual signatures. The advancement of computing technologies,
especially cloud computing, allows the paralleled speedup of processing large
volume of data, which makes it possible to classify images of extremely high
resolution.
2. Improved temporal resolution: The temporal resolution of the extremely highspatial-resolution remote sensing is also significantly improved. Daily revisits are
common through the constellation of many satellites. This would help in solving
the timing problem of image acquisition. Cloud coverage may be eliminated to a
great degree due to the frequent revisits.
3. Beyond spectral signatures: The resolution of Radar remote sensing is also
significantly improved. The all-weather capability of Radar will significantly
eliminate the problem of cloud coverage. The intensity signal of RADAR and
184
E. G. Yu and Z. Yang