When one walks into the wilderness, it is useful to know what vegetation is helpful and what vegetation is harmful. What can you use for your stew and what will lead to you not waking up the next morning? Even if the tree is not necessarily useful to oneself, animals and other organisms could depend upon its products or overall structure. This could help determine the location of certain ecosystems. If one animal is there another one will follow as it goes with the predator prey model. If there are too many of these trees then there are too many of prey 1 and then there would soon overtime, be too many of predator 1. Maybe predator 1 is invasive and reproduces exponentially to the point that it becomes disastrous for the ecosystem and the plant that started it all. Maybe as this predator 1 destroys tree 1, a medicinal herb 1 thrives off of tree 1’s shade and moisture. This herb could be an expensive medicine in one country and one of their greatest exports. What if as these herbs die off due to the overpopulation of the predator the country raises the price of this live saving medicine 1. Then country 2 needs it as well, but cannot afford the price. There is now unrest and war. Everyone dies the end. *Delete later to something more serious, I needed to just get words on the page with low pressure*
If one can monitor these ecosystems from afar and not directly interfere with the animals or vegetation, there will be ways to implement safe guards while keeping the environment wild.
That is the appeal of safaris and game preserves. It is acres of land, but closed off and monitored for poachers. With vast amounts of land such as this, UAS systems are extremely beneficial to cover this ground efficiently and quickly. In the case of lions and the unknown of Australia’s wildlife, it does a proper risk assessment and management of sending researchers out into the field to do the same analysis a drone can do. *Improper flow please fix later*
In this paper we explore the Australian landscape of a forest full of various species of Eucalyptus, one specie of willow, and other trees along with bare soil. Employing the software ENVI certain methods of analysis were studied and developed to provide a thorough understanding of the landscape. Taking all the data in at once can be overwhelming so this is digested in steps.
Theory:
Reflectivity of vegetation in the infrared:
If humans could see all wavelengths and not just the visible spectrum then vegetation would be red and not green. This is due to the high reflectance of infrared light given off this vegetation.
*Draw an example of a spectrum maybe even from Radiometry class if that has been okayed*
Structure of a tree leaf or sample:
At different wavelengths in the electromagnetic spectrum there are different characteristics of the vegetation that have the largest influence on the radiance and reflectance data.
*Write out what was from class about how pigment in the visible range is the main determining factor, but leaf structure is the determining factor in the near infrared.* (Environmental Remote Sensing, Jan. 2025)
Regions of interest:
Regions of interest describe a mapping of an area of the land where the specie in question would be located. Although the author did not need to produce the ROI’s used this was the procedure in which it was done. (Reads ROI and the polygon part).
Pictures of the different tree train images and selecting the proper wavelengths.
Resizing the Raster:
Raster is an outdated term for the image itself. The new version of ENVI has errors with the packages they install from time to time so consistently there was a problem with doing a spectral analysis and it gave the error of a spatial analysis.
*Display error message*
This then led to the sponsor supplying the raster subset data in this folder.
Now there are two given bands for this information: raw radiance and MNF components. Raw radiance is the ______ and MNF stands for Minimum Noise Fraction where _________. These bands go in order of importance of how much it influences the picture output by ENVI. Bands ___ ___ for the raw radiance matter the most and the end ones could be thrown away with not much data being lost. The same goes for bands 9 11 17 for the MNF components matter the most. Throwing away the less important bands saves processing power that can be crucial to large datasets.
Classification:
Now our data is trained enough to be able to classify our data to the tree species and bare soil in question.
Now with these new subsets we can perform a maximum likelihood calculation.
*Shows all the pictures with the classification data for raw radiance and MNF…I’m not sure how to have all of the pictures without dumping them. I want to give an analysis of each.*
This displays how the pixels can be mixed with other tree species pixels. The purple bare soil case demonstrates how problematic an impure sample can be to one’s analysis. This is the first example of how an impure sample can be confused with the other tree species. This can be contrasted with the AMea specimen where that is one specimen of tree.
Using this maximum likelihood we can then produce a confusion matrix.
Side note: there should be an infographic of how each file feeds into each other.
At first the author was confused by the confusion matrix. *Delete later to something more serious.* Now, why has the author chosen to include these results? Null results or common errors being published along side with significant results should be encouraged for two reasons: (1) it keeps science open and transparent as omission can skew the narrative then (2) this eases the problem of reproducibility in science. Now with that sentiment, let us investigate the dud confusion matrices to conclude how one could know that simply by looking at the results that there has been an error.
The first example has the tree training data being used instead of the tree test data. This error led to the ROI’s claiming to be 100 percent accurate while there were two ROI’s in question. In a basic numbers perspective, the most casual observers can see how this is wrong.
Now we move to the second example where ROI’s were mapped to each other, but it was still the tree train data.
This produced an accuracy of 19 percent which was dangerously low. If this was true then the person classifying the data could have randomly characterized each ROI and gotten better results.
Lots of Honey 