Space (spatial resolution)/time (temporal resolution) trade:
The implications for remote sensing analysis of the Earth's surface is that the data needs to have a high
spatial and temporal resolution with a good revisiting frequency to capture the important changes that
is happening on Earth such as the Planet Labs Dove satellites orbiting daily instead of having the Landsat
revisit every 16 days which would miss important changes of the Earth whereas the Planet Labs Dove
satellites can capture every change and data of the Earth's surface. This is essential for remote sensing
to reduce the effects of space-time trade. Other implications include having the data widely accessible
and available so that anyone can contribute to analyzing what is happening on our planet with the
scientific and technological approaches to properly analyze the Earth using remote sensing. Moreover,
having multiple conversations being used in the orbit whether they are commercial or multi-agency is an
implication for remote sensing analysis since each of the satellites can do various bands and resolutions
which is beneficial in capturing all sorts of data and changes of the Earth's surface. Another way for
remote sensing analysis is to have the satellites closer to the target of interest such as the Lidar satellite
measuring the distance between the instrument from the object to do height mapping for a region
which would provide more accurate results in terms of constructing models of the digital surface. Lastly,
remote sensing analysis has established new analysis methods to better understand the data collected
about the Earth's surface. The two analysis methods that are effective today in remote sensing is the
physics-based analysis and data science analysis. These two analyses are essentially the scientific and
technological methods which nowadays come hand-in-hand instead of treating them as separate
analysis methods since technology is vital when doing the scientific studies of remote sensing which is
better in terms of analysis of the Earth's surface.
Image transformations VI (Vegetation Indices), PCA (Principal Component Analysis)
VI (Vegetation Indices)
NDVI
Green vegetation has high NDVI value whereas water has negative NDVI value. NDVI values are between
-1 and 1. Green grass has positive NDVI value but its NDVI value is lower than dense forest NDVI value.
Water appears dark in the image because its NDVI value is negative/low whereas green vegetation
appears white in the image since its NDVI value is positive/high.
PCA (Principal Component Analysis)
Image classification (supervised, unsupervised, and OBIA)
, Supervised:
In the supervised classification of a satellite remote sensing multi-spectral image, there are three steps
in order to get the supervised classification which is the training stage, the classification stage, and the
output stage. The first step, the training stage, is when the training classes are established for the region
of interest according to the spectral signatures for each class in order to achieve the statistics needed for
analysis. For example, each land cover class represents each training class such as having the waterbody
associated with a training class with waterbody spectral data and vegetation associated with a training
class with vegetation spectral data. When doing the training data, the classes should be chosen in
homogenous areas and they should be established throughout the image in order to get better accuracy
and representativeness of the classes when doing the classification. The accuracy and representation of
the classes statistically can also be achieved by collecting only 10n to 100n pixels with n being the
member of spectral bands of the classification. The second step of supervised classification is the
classification stage which is when a classification algorithm can be chosen to identify the training sites
based on the spectral relationships of each pixel to alike categories. The three common classification
algorithms are the minimum distance to means, parallelepiped, and maximum likelihood. Minimum
distance to means is where it categorizes the pixels to each category by Euclidean distance or "round the
block" distance. Parallelepiped is when the pixels are categorized based on their brightness values.
Maximum likelihood is when the pixels are categorized based on their probability. The last step of
supervised classification is the output stage which it will show the classification for each of the classes in
the form of maps or tables of the data.
OBIA (object-based image analysis):
Explain how object-based image classification is different from per-pixel classification methods as
applied to satellite remote sensing image data.
Object-based image classifications is different from per-pixel classification methods in terms of satellite
remote sensing image data since they can capture the spectral and spatial information of the image and
that they are able to calculate the statistics of the image due to having multiple homogenous pixels or
image objects available such as calculating the mean of the image object whereas the per-pixel
classification only has single pixel information in which it can only compute the spectral or pixel
information. Object-based image classification is also different from per-pixel classification methods
since they stay away from the salt-and-pepper phenomena since it can have images with a high spatial
resolution which are able to capture more kinds of data analysis whereas the per-pixel classification
cannot avoid the salt-and-pepper effect which leads to limitations in certain analysis such as defining the
spectral complications of dividing the land features made by human and by nature into their separate
classes while object-based image classifications are able to solve these types of issues. Another way that
object-based image classifications differ from per-pixel classifications is that it can compute
circumstantial information about the image object such as determining their size, shape, and texture
whereas the per-pixel classifications cannot identify contextual information about the pixels since it
doesn't take into account of the spatial autocorrelation. Lastly, the object-based image classification is
different from per-pixel classification since it is able to distinguish and identify each class with its
associated pixels so that it will not lead to the overlapping of the pixel values and result in greater
The implications for remote sensing analysis of the Earth's surface is that the data needs to have a high
spatial and temporal resolution with a good revisiting frequency to capture the important changes that
is happening on Earth such as the Planet Labs Dove satellites orbiting daily instead of having the Landsat
revisit every 16 days which would miss important changes of the Earth whereas the Planet Labs Dove
satellites can capture every change and data of the Earth's surface. This is essential for remote sensing
to reduce the effects of space-time trade. Other implications include having the data widely accessible
and available so that anyone can contribute to analyzing what is happening on our planet with the
scientific and technological approaches to properly analyze the Earth using remote sensing. Moreover,
having multiple conversations being used in the orbit whether they are commercial or multi-agency is an
implication for remote sensing analysis since each of the satellites can do various bands and resolutions
which is beneficial in capturing all sorts of data and changes of the Earth's surface. Another way for
remote sensing analysis is to have the satellites closer to the target of interest such as the Lidar satellite
measuring the distance between the instrument from the object to do height mapping for a region
which would provide more accurate results in terms of constructing models of the digital surface. Lastly,
remote sensing analysis has established new analysis methods to better understand the data collected
about the Earth's surface. The two analysis methods that are effective today in remote sensing is the
physics-based analysis and data science analysis. These two analyses are essentially the scientific and
technological methods which nowadays come hand-in-hand instead of treating them as separate
analysis methods since technology is vital when doing the scientific studies of remote sensing which is
better in terms of analysis of the Earth's surface.
Image transformations VI (Vegetation Indices), PCA (Principal Component Analysis)
VI (Vegetation Indices)
NDVI
Green vegetation has high NDVI value whereas water has negative NDVI value. NDVI values are between
-1 and 1. Green grass has positive NDVI value but its NDVI value is lower than dense forest NDVI value.
Water appears dark in the image because its NDVI value is negative/low whereas green vegetation
appears white in the image since its NDVI value is positive/high.
PCA (Principal Component Analysis)
Image classification (supervised, unsupervised, and OBIA)
, Supervised:
In the supervised classification of a satellite remote sensing multi-spectral image, there are three steps
in order to get the supervised classification which is the training stage, the classification stage, and the
output stage. The first step, the training stage, is when the training classes are established for the region
of interest according to the spectral signatures for each class in order to achieve the statistics needed for
analysis. For example, each land cover class represents each training class such as having the waterbody
associated with a training class with waterbody spectral data and vegetation associated with a training
class with vegetation spectral data. When doing the training data, the classes should be chosen in
homogenous areas and they should be established throughout the image in order to get better accuracy
and representativeness of the classes when doing the classification. The accuracy and representation of
the classes statistically can also be achieved by collecting only 10n to 100n pixels with n being the
member of spectral bands of the classification. The second step of supervised classification is the
classification stage which is when a classification algorithm can be chosen to identify the training sites
based on the spectral relationships of each pixel to alike categories. The three common classification
algorithms are the minimum distance to means, parallelepiped, and maximum likelihood. Minimum
distance to means is where it categorizes the pixels to each category by Euclidean distance or "round the
block" distance. Parallelepiped is when the pixels are categorized based on their brightness values.
Maximum likelihood is when the pixels are categorized based on their probability. The last step of
supervised classification is the output stage which it will show the classification for each of the classes in
the form of maps or tables of the data.
OBIA (object-based image analysis):
Explain how object-based image classification is different from per-pixel classification methods as
applied to satellite remote sensing image data.
Object-based image classifications is different from per-pixel classification methods in terms of satellite
remote sensing image data since they can capture the spectral and spatial information of the image and
that they are able to calculate the statistics of the image due to having multiple homogenous pixels or
image objects available such as calculating the mean of the image object whereas the per-pixel
classification only has single pixel information in which it can only compute the spectral or pixel
information. Object-based image classification is also different from per-pixel classification methods
since they stay away from the salt-and-pepper phenomena since it can have images with a high spatial
resolution which are able to capture more kinds of data analysis whereas the per-pixel classification
cannot avoid the salt-and-pepper effect which leads to limitations in certain analysis such as defining the
spectral complications of dividing the land features made by human and by nature into their separate
classes while object-based image classifications are able to solve these types of issues. Another way that
object-based image classifications differ from per-pixel classifications is that it can compute
circumstantial information about the image object such as determining their size, shape, and texture
whereas the per-pixel classifications cannot identify contextual information about the pixels since it
doesn't take into account of the spatial autocorrelation. Lastly, the object-based image classification is
different from per-pixel classification since it is able to distinguish and identify each class with its
associated pixels so that it will not lead to the overlapping of the pixel values and result in greater
