The following was another addition to the discussion boards for the SAIT BGIS program.
·
The
Normalized Difference Vegetation Index (NDVI) is a commonly used vegetation
index in the remote sensing community. What are some other vegetation
indexes? What platform/sensor were they developed for and what is their
application?
There are many different types of vegetation indexes
besides the NDVI. Many are similar, such
as the Soil-Adjusted Vegetation Index, which is used to remove soil brightness
and can be performed with any instruments that provide red and near-infrared
bands (Wikipedia,
2021). While some
indices are similar to NDVI, others work differently are provide important
environmental variables such as FAPAR (Fraction of absorbed photosynthetically
active radiation) (Wikipedia,
n.d.). While this can be directly measured on a local
scale, on a more broad scale it can be estimated using MODIS and AVHRR sensor
on board satellites (Wikipedia,
n.d.). A more recent index that has been published is the
Broadband Green-Red Vegetation Index, initially the tracking of similar data
was limited to only a few satellites (Gaofei
Yin et al., 2022), but this index publishing in
the Journal of Remote Sensing in 2022 suggests that a similar index can be
derived from MODIS data.
·
What
are the similarities and differences between Principal Components Analysis and
the Tasseled Cap transformation?
Both Principle Component Analysis and Tasseled Cap transformations
seek to find correlations between bands within the image to reduce redundancy (SAIT,
n.d.). The result of
this is an image with less bands, but which show maximum variability (SAIT,
n.d.).
There are two key differences between the two transformations. The first is that Tasseled Cap can only be
run on Landsat data, which Principal component analysis can be run on any
multi-band image (SAIT,
n.d.). The second is
that while they both utilize eigen vectors to complete the transformation, Principal
Component Analysis uses eigen vectors unique to the scene being transformed making
is a local image transformation (SAIT,
n.d.). The Tasseled
Cap transformation is said to be a global transformation as it always uses the
same eigen vectors (SAIT,
n.d.)
·
What
are some possible inputs to the supervised/unsupervised classification
process? What are some ways to determine the most effective inputs to
this process?
For both of these types of classifications selecting
the algorithm will be an input that needs to be chosen (James
B Campbell & Randolph H Wynne, 2012), although each of them will have
different methodologies. The type of classification,
whether pixel or object based will need to be determined as well (Adrian
Faraguna & Jay Reid, n.d.)
For supervised classification some of the other inputs
we need to choose are the classification schema we are to use and the training
sites for that schema (Adrian
Faraguna & Jay Reid, n.d.).
For unsupervised classification we do not need to
choose training sites, but we do need to specify how many classifications we
are seeking in our final product (Adrian
Faraguna & Jay Reid, n.d.).
While there is no definitive set of decisions that can
be followed to give the absolute best results when it comes to classification (James
B Campbell & Randolph H Wynne, 2012), there are some things to
consider to help guide your decisions.
For supervised classifications, perhaps the single
most important aspect is the selection of training data, it has been demonstrated
that this single aspect has a much larger impact on the accuracy of the
classification than the classification algorithm does (James
B Campbell & Randolph H Wynne, 2012).
When choosing the training data your areas should be large enough to capture
the spectral signature of the area, but they should be small enough to maintain
a single spectral signature (James
B Campbell & Randolph H Wynne, 2012).
These training areas should be away from edges and should not cover wild
spectral signature swings, as these will both confuse the classification system
(James
B Campbell & Randolph H Wynne, 2012).
Our lab manual suggests the minimum size to be N+1 pixels for the maximum
likelihood algorithm and Intro to Remote Sensing suggests training data should
consist of at least 100 total pixels for Landsat data.
For unsupervised data one of the biggest issues is matching
the classified data with the informational classes (James
B Campbell & Randolph H Wynne, 2012), because of this it is best to
choose a number of classes in excess of your actual end result (Adrian
Faraguna & Jay Reid, n.d.).
Having more classes that can be merged to form the desired information
classes will be more desirable than not having enough separation (Adrian
Faraguna & Jay Reid, n.d.)
In general, the most effective inputs are going to
rely on an analysis of the image being classified, a strong knowledge of the
area and the experience of the analyst.
Knowing what the output will look like and having a good high resolution
multi-spectral image to use as an input will all be beneficial to the process
of classification.
Works Cited
Adrian Faraguna & Jay Reid. (n.d.). Lab
Manual for the Display and Processing of Remote Sensing Datasets using ArcGIS
Pro and Pix4d. SAIT.
Gaofei Yin, Aleixandre
Verger, Adrià Descals, Iolanda Filella, & Josep Peñuelas. (2022, March 19).
A Broadband Green-Red Vegetation Index for Monitoring Gross Primary
Production Phenology. Journal of Remote Sensing.
https://spj.sciencemag.org/journals/remotesensing/2022/9764982/
James B Campbell &
Randolph H Wynne. (2012). Introduction to Remote Sensing (5th ed.).
Guilford Publications.
SAIT. (n.d.). GEOS451
Module 5—Lecture 9—Image Processing. Spring 2022 - GIS Data Capture II
(GEOS-451-O4A).
https://learn.sait.ca/content/enforced/520161-202140GEOS-451-O4A/GEOS451_Mod4_ImageProcessing1.pdf?_&d2lSessionVal=AtC78OPJVmSPaM1ud9XxLCFOv&ou=520161
Wikipedia. (n.d.). Fraction
of absorbed photosynthetically active radiation. Wikipedia. Retrieved July
10, 2022, from
https://en.wikipedia.org/wiki/Fraction_of_absorbed_photosynthetically_active_radiation
Wikipedia. (2021, June
5). Soil-adjusted vegetation index. Wikipedia.
https://en.wikipedia.org/wiki/Soil-adjusted_vegetation_index
Comments
Post a Comment