machine classifiers. The object-oriented classifications have been used for veryhigh-spatial-resolution imagery.
Supervised classification requires the training samples of different crop types.
The ground truth samples can be collected from the field, fine-spatial-resolution
images, or from the same remote sensing image used for classification. Classification
accuracy depends on the quality of training samples and separability among the
different crops. In the early era of remote sensing, a single image was normally used
in classification due to the cost. When a single image is used, the classification
accuracy entirely depended on the spectral and spatial feature. Sometimes different
crop types may share a similar spectral feature on the acquisition date, and the
classifier may not be able to separate them. Since medium-resolution remote sensing
images are freely available today, images acquired from different crop growing
stages have been used and improved classification accuracy. Crop types that show
similar spectral features in a day could be different on other days. Therefore,
temporal information is valuable for improving classification accuracy. Crop type
classification may also be affected by the pixel’s spatial resolution. In the United
States, the sizes of crop fields are large, and thus Landsat 30-m resolution is good
enough for mapping the crop types for each field. However, in Africa, Asia, and
Europe, the field sizes are much smaller; remote sensing imagery at a finer spatial
resolution is needed to avoid the mixture of different land cover types in a pixel.
Crop type classification using remote sensing has been operational in the United
States. The US Department of Agriculture (USDA) National Agricultural Statistics
Service (NASS) has produced the Cropland Data Layer (CDL) over CONUS every
year since 2008. For the earlier years (1997–2007), CDL maps are available for the
selected states. CDL was produced using a decision tree classifier (Boryan et al.
2011). Classification results were assessed for each crop. The classification accuracy
varies. Major crops such as corn, soybean, and wheat have much higher accuracy
than smaller crops. The CDL data are available through the CropScape portal
(https://nassgeodata.gmu.edu/CropScape/). The CropScape provides spatial subset,
analysis, and mapping functions. CDL for the entire CONUS can also be
downloaded through USDA NASS (https://www.nass.usda.gov/Research_and_
Science/Cropland/Release/index.php). NASS produces CDL within the season and
releases it in the early next year. Note that even remote sensing classification
methods are mature; mapping crop types at early growing season is still very
challenging. Using crop rotation pattern from the previous years can help the early
season crop type mapping (Hao et al. 2020).
2.3.2 Crop Phenology Mapping
Accurate spatiotemporal information about crop progress and condition during the
growing season is critical for crop management and yield estimation (Walthall et al.
2012). The amount of yield loss realized during a drought year is dependent on the
crop growth stage when water stress occurs. Crop progress provides information
16
F. Gao
Supervised classification requires the training samples of different crop types.
The ground truth samples can be collected from the field, fine-spatial-resolution
images, or from the same remote sensing image used for classification. Classification
accuracy depends on the quality of training samples and separability among the
different crops. In the early era of remote sensing, a single image was normally used
in classification due to the cost. When a single image is used, the classification
accuracy entirely depended on the spectral and spatial feature. Sometimes different
crop types may share a similar spectral feature on the acquisition date, and the
classifier may not be able to separate them. Since medium-resolution remote sensing
images are freely available today, images acquired from different crop growing
stages have been used and improved classification accuracy. Crop types that show
similar spectral features in a day could be different on other days. Therefore,
temporal information is valuable for improving classification accuracy. Crop type
classification may also be affected by the pixel’s spatial resolution. In the United
States, the sizes of crop fields are large, and thus Landsat 30-m resolution is good
enough for mapping the crop types for each field. However, in Africa, Asia, and
Europe, the field sizes are much smaller; remote sensing imagery at a finer spatial
resolution is needed to avoid the mixture of different land cover types in a pixel.
Crop type classification using remote sensing has been operational in the United
States. The US Department of Agriculture (USDA) National Agricultural Statistics
Service (NASS) has produced the Cropland Data Layer (CDL) over CONUS every
year since 2008. For the earlier years (1997–2007), CDL maps are available for the
selected states. CDL was produced using a decision tree classifier (Boryan et al.
2011). Classification results were assessed for each crop. The classification accuracy
varies. Major crops such as corn, soybean, and wheat have much higher accuracy
than smaller crops. The CDL data are available through the CropScape portal
(https://nassgeodata.gmu.edu/CropScape/). The CropScape provides spatial subset,
analysis, and mapping functions. CDL for the entire CONUS can also be
downloaded through USDA NASS (https://www.nass.usda.gov/Research_and_
Science/Cropland/Release/index.php). NASS produces CDL within the season and
releases it in the early next year. Note that even remote sensing classification
methods are mature; mapping crop types at early growing season is still very
challenging. Using crop rotation pattern from the previous years can help the early
season crop type mapping (Hao et al. 2020).
2.3.2 Crop Phenology Mapping
Accurate spatiotemporal information about crop progress and condition during the
growing season is critical for crop management and yield estimation (Walthall et al.
2012). The amount of yield loss realized during a drought year is dependent on the
crop growth stage when water stress occurs. Crop progress provides information
16
F. Gao