H9+V5+-+Western+US

Labs 11/12

I chose the H9 V5 tile, which encompasses much of the western US including California, Arizona, New Mexico, Nevada, and parts of Colorado. The reason for my interest in this area is because of my interest in variability in forest fires forced by climate change. I suspect that this region will only experience increased devastation as a result of a drier climate, coupled with deforestation and misuse of forested lands. I want to investiage the changes in the fire season and also inspect changes within the land use data to see if a mass change in land use could explain those changes.

Below is my MODIS tile, classified by land use / land type. The dark magenta is the pacific ocean and the brighter orange colour is the land, although the colours are so bright that it is hard to differentiate between the different classes.

The goal of this assignment was to construct a land use change map (using NDVI) from three different seasons - as you can see, there are quite a few major changes that occur within the span of each year in this area. In the spring, you notice that most of the California coast is deep green, an indication of lush vegetation which in some of the higher alititude areas, you see water, a possible indicator of snow still on the mountain tops. It is not a surprise that the highly-irrigated areas of California that are used for agriculture are green while the areas surrounding it are desert and obviously don't receive a lot of rain. You can see that as we progress throught he seasons that the brown areas indicating very little vegetation grow drastically to the point where the fall picture is essentially dominated by the desert conditions. The green areas also become less saturated and more sparsely distributed indicating a period of drought (SUMMER!!). For this assignment, we look into using anthromes to direct class studies to see which classes dominated our MODIS tiles. Below are the classes and % of pixels covered by this class.

The top five contributors to my MODIS tile are
 * Class # || Class Type || Number of Pixels in Area || % of pixels covered ||
 * 11 || Urban ||  ||   ||
 * 12 || Dense ||  ||   ||
 * 21 || Rice Villages ||  ||   ||
 * 22 || Irrigated Villages ||  ||   ||
 * 23 || Cropped Pastoral Villages ||  ||   ||
 * 25 || Rainfed Villages ||  ||   ||
 * 26 || Rainfed Mosaic Villages ||  ||   ||
 * 31 || Residential Irrigated Cropland ||  ||   ||
 * 32 || Populated Irrigated Cropland ||  ||   ||
 * 33 || Populated Irrigated Cropland ||  ||   ||
 * 34 || Populated Rainfed Cropland ||  ||   ||
 * 35 || Remote Croplands ||  ||   ||
 * 41 || Residential Rangelands ||  ||   ||
 * 42 || Populated Rangelands ||  ||   ||
 * 43 || Remote Rangelands ||  ||   ||
 * 51 || Populated Forests ||  ||   ||
 * 52 || Remote Forests ||  ||   ||
 * 61 || Wild Forests ||  ||   ||
 * 62 || Sparse Trees ||  ||   ||
 * 63 || Barren ||  ||   ||

Then, we compared our MODIS tile to anthromes, and the results in the standard deviations of NDVI values are graphed below:
 * Rank || Class Type || Percent Covered ||
 * 1 || 32 ||  ||
 * 2 || 34 ||  ||
 * 3 || 35 ||  ||
 * 4 || 42 ||  ||
 * 5 || 43 ||  ||





The graphs above indicate an obvious seasonality in the NDVI values, which makes sense because the main class types dominating my MODIS tile are rangelands and croplands, which will vary in their vegetation depending on what time of year it is. However, you can see in the standard deviation graph that some vary much more than others in their NDVI values but their means (seen in the second graph) are suprisingly close. This suggests that the standard deviations are not only one-way, but that the areas that get much greener also get much less green depending on the time of year, in order for them to average out to be similar to the classes that don't vary as much.

---**BEGIN LANDSAT***---

The first step in finding out landsat anniversary images was to make a list of all the dates that our landsat tile was available with less than 10% cloud cover. Then, we looked at each image after it had been atmospherically corrected:

Next, we applied filters to subsets of our FLAASH corrected images, some of my favorites are below:
 * [[image:1998_FLAASH_WIKI.jpg width="537" height="408" caption="1998 FLAASH Image"]] || [[image:1998_FLAASH_spectral.jpg caption="1998 FLAASH Spectral Profile"]] ||
 * [[image:2010_FLAASH_WIKI.jpg width="553" height="417" caption="2010 FLAASH image"]] || [[image:2010_FLAASH_spectral.jpg caption="2010 FLAASH Spectral Profile"]] ||


 * [[image:1998_HighPass.jpg width="323" height="352" caption="1998 High Pass"]] || [[image:1998_HighPass31.jpg width="324" height="336" caption="1998 High Pass - 31 Kernel"]] || [[image:1998_HighPass5.jpg width="378" height="326" caption="1998 High Pass - 5 Kernel"]] || [[image:1998_Median3.jpg width="370" height="322" caption="1998 Median - 3 Kernel"]] ||

For the duration of my work with LANDSAT data, this will be my subset. I wanted to pick something that had all types of land use classes, the high pass filters allow for nice varification and distinction of the agricultural lands from the mountainous regions and also the areas that have been slowly washes away from the freeze/melt cycle typical of areas in this altitude. The median filter did a nice job of smoothing the image and reducing noise.

Below, for Lab 9, is a link to the PC analyses for both 2010 and 1998. However, at the time, I did use two different areas to see if the bands so dominant in the first area were still dominant when faced with a different terrain. You will see that although the first band provides a great deal of clarity, band 2 is the best for analyzing details.


 * Lab 9 PC Analysis**


 * [[image:before_1998.jpg width="509" height="419" caption="Before 1998"]] || [[image:after_1998.jpg width="512" height="423" caption="After 1998"]] ||
 * [[image:before.jpg width="519" height="489" caption="Before 2010"]] || [[image:after.jpg width="517" height="493" caption="After 2010"]] ||
 * [[image:EigenValues_1998.jpg caption="Eigen Values 1998"]] || Mysteriously missing the eigen values for 2010.... let the investigation begin. ||
 * [[image:statistics_1998.JPG width="698" height="857" caption="1998 Statistics"]] || [[image:min_max_stats.JPG width="711" height="755" caption="Min/Max 2010 Stats"]] ||
 * || [[image:stdev_stats.JPG width="800" height="658" caption="Standard Deviation 2010 Statistics"]] ||

We were asked to look at the difference between the 'before' true colour image and the Inverse PC image: There isn't much of a change in these images, in my opinion - they are off by a few shades and the before picture contrasts better with the agricultural lands (darker plots) but they both emphasize the salt and mineral washes (white streams) very well and both show elevation very well in the mountainous regions. Granted, the first three bands made up nearly 100% of the variability in both cases, so there isn't much to be left out.
 * [[image:truecolor_before_1998.jpg width="609" height="551" caption="1998 True Colour Imager"]] || [[image:inversePC_1998_WIKI.jpg width="586" height="557" caption="1998 Inverse PC Image"]] ||

The next pictures are from my PC Band 3 from Change Detection -- this is the band which I think shows the change most clearly. Next, we have the Two Colour Multiview which is suprisingly diverse -- it is showing changes in places where I don't think changes can occur - bare rock. However, when I do the classification process, we will see if possibly there were forests there are some point or some kind of vegetation, or possibly even a cloud shadow that then 'changed' by the time the second image came around. I will also zoom in to the two colour multiview on the agricultural area to show the changes seasonally and anually that happen in this area - new crops, crop rotation, fallow land rotation. Finally, there is the NDVI image -- I know it is supposed to look different but for some reason, one of the NDVI images from change detection was purely white (every time I tried to run it). So, I don't know what the problem is with that.