• CREATE A TRUE COLOR IMAGE FROM A LANDSAT SCENE •
Because of our interest in using geospatial technology tools for observing change, we expect to do a lot of work with true-color composite images. Our Geographic Information Systems software includes a set of tools for creating composite images and, as you might expect, the tools contain configuration settings that influence the appearance of the output image. We feel that it is important to use the same settings every time the task is performed. The quality and optical properties of the source data are all over the place so we don't want to introduce any more variation than is already present. Student Technician, Austin Carter documented the procedure he used when he was assisting in the Biodiversity Center GIS Lab in the summer of 2013.
"In the making of the RGB composite maps we took 3 key steps. The first step was the acquisition of the Landsat images that we used to make the composite maps. The images were provided by USGS at their website usgs.gov. After getting these maps and putting them into ArcMap. Next, we used the tool "composite bands" which makes Landsat images in to raster images by assigning each band a color, which gives the composite the appearance of a true color image. Finally, we had to make sure that the correct bands were displaying the correct color so we changed the bands to a 321 order. This is because the 3 band represents the red color, the green is represented by 2 and the blue is represented by 1. It is important to change these bands because otherwise you will not get a true color image. We also changed the stretch type to standard deviation and if we were using Landsat 7 or 8 we used pan sharpening.
When making RGB composition maps we used a standard deviation stretch type. We believe that this is the best method for removing outliers. Standard deviation is better than percent clip because percent clip is going to remove a defined number of values from the top and bottom of the data pool. This can be ineffective because you can remove too much information or not enough information from the data set. The reason for this is because if you use too high of a value then you could remove more than what the true outliers in the dataset are. If you use too low of a number then you will not remove all of the outliers. Standard deviation fixes this problem because the more outliers there are the more it removes from the dataset. This is why we decided to use a standard deviation of 2 deviations instead of using percent clip.
When deciding how many standard deviations to remove from the data set we decided to remove 2 because removing any more that that would affect the dataset dramatically. This was determined using the 68, 95, 99.7 rule which states that from the mean going in both directions 1 standard deviation contains 68% of the data and 2 deviations contains 95% percent of the data. This means that removing 1 from the end would remove .3% of the data and 2 would remove 5% altogether. If we were to use the default and remove another half then we would remove another 13% of the data totaling around 18% which is a lot of data to remove when trying to get rid of outliers. This is why 2 deviations is the best amount to remove without significantly affecting the data but still removing the outliers."