H26.V5.Landsat

Daniel Russell
 * Path ||  ||   || 126 ||   ||   ||   ||
 * Row ||  ||   || 35 ||   ||   ||   ||
 * **ID** ||  ||   || **Day** || **Month** || **Year** || **% of Cloud Cover** ||
 * LT51260352005163BJC01 ||  ||   || 12 || 6 || 2005 || 5 ||
 * LT51260352006150BJC00 ||  ||   || 30 || 5 || 2006 || 4 ||
 * LT51260352006166BJC00 ||  ||   || 15 || 6 || 2006 || 2 ||
 * LT51260352006230IKR00 ||  ||   || 18 || 8 || 2006 || 0 ||
 * LT51260352007137IKR00 ||  ||   || 17 || 5 || 2007 || 0 ||
 * LT51260352007153IKR00 ||  ||   || 2 || 6 || 2007 || 0 ||
 * LT51260352007265IKR00 ||  ||   || 22 || 9 || 2007 || 0 ||
 * LT51260352009030BJC00 ||  ||   || 30 || 1 || 2009 || 0 ||
 * LT51260352009190IKR00 ||  ||   || 9 || 7 || 2009 || 16 ||
 * LT51260352009270IKR00 ||  ||   || 27 || 9 || 2009 || 3 ||
 * LT51260352010193IKR00 ||  ||   || 12 || 7 || 2010 || 1 ||
 * LT51260352010209IKR00 ||  ||   || 28 || 7 || 2010 || 16 ||
 * LT51260352010257IKR00 ||  ||   || 14 || 9 || 2010 || 10 ||
 * LT51260352010257IKR00 ||  ||   || 14 || 9 || 2010 || 10 ||





= = =**Spatial Enhancements**= The Sobel enhancement makes it easier to distinguish between the various height difference which is very good for the mountainous region where this image lies. It creates what appear to be lines of constant heights. The kernel size is 3x3 and was not adjusted. The image has a %60 image add back option which helped keep the spatial context of the mountainous region. The Laplacian filter seems help distinguish the highest points of mountains quite well and has a very pronounced impact upon the urban area. Property lines are much easier to distinguish with the Laplacian filter and this filter used a 3x3 kernel.

The High pass filter limits the amount of distinction between the higher points on the mountain and the points of lower elevation on the mountainous region. It helps limit the amount of shadows that are present from the original image the kernel used was a 3x3 kernel which further helped limit the shadow problem. The Median pass filter is most like the original image. The image appears to have smoother transitions between the green vegetation and the moutainous region which makes it easier to look at.

=**Multispectral Transformations and Change Detection**=



For the correlation matrix, bands 1, 2 and 3 are all very highly correlated. They have over a 93% correlation with each other. Band 5 and band 6 also show a high correlation with values in the %92. Band 4 is the least correlated to the other bands but has a decent correlation value of 77.3% to that of band 5.

Principle Component (forward Principle Component Rotation)

The first principle component is most like the original subsetted image. The clouds and shadows from the clouds clearly pop out using this first rotation. The second principle component has some interesting results. It helps clear the white clutter that hovers over the urban area (which could be smog from the city) and helps define the urban areas. Roads are much easier to see. Large land separation lines are clearer and more defined; however, the smaller property lines are less clear. The third principle component makes the water in this image appear black including the clouds. The smaller property lines are much easier to view with this principle image. The 4th principle image is where the noise appears to show up. The only parts that stand out are the river and drainage basin as well as a few of the bigger clouds. The places where height drops fairly quickly appear to be the other places of notice in this image. The 5th principal component contains more light areas for regions with less vegetation and the more vegetated regions are darker. The noise in the 6th principal component transformation decreases the definition of the entire urban area and most of the mountainous region. The river somewhat stands out, but everywhere else appears to be static noise.

One thing that is noticeable that happened with inverse principal transformation is that it made the river blue and also made the vegetation on the mountainous region a little duller and harder to see in the areas where vegetation is lower.
 * ~ PC Number ||~ Eigenvalue ||~ %Variability ||
 * 1 || 1319294.87 || 0.754354348 ||
 * 2 || 286362.3717 || 0.163737998 ||
 * 3 || 121318.3672 || 0.069368145 ||
 * 4 || 14821.92859 || 0.008474971 ||
 * 5 || 5164.410534 || 0.002952938 ||
 * 6 || 1944.084691 || 0.0011116 ||
 * Summation || 1748906.032 ||  ||
 * Summation || 1748906.032 ||  ||

=Change Detection with Spears Tools=







Daniel Russell April 21, 2011 Change Detection with Spears Tools ENVI’s Spears Change Detection tools can be extremely beneficial towards evaluating the change over time for a specific region. The region of China that is being investigated has two approximate anniversary dates; one September 14, 2009 and one on September 27, 2010. MODIS data proved that there had been a significant cooling trend from the years 2000-2009 with 2010 having generally higher temperatures than the year of 2009. This fact is important when looking at the change detection images from the Landsat data. The 3 images that were analyzed were a two color multiview image for band 4, the principle component transformation of band 3, and the change in NDVI for the two years of 2009 and 2010. Band 4 for Landsat images operates in the range from .76 micrometers to .90 micrometers and is best used to determine vegetation. When analyzing the multiview image for band 4, there are a few obvious observations that stand out. The first is that the image is primarily red, which means that the image for the year 2009 has higher reflectance values in band 4 than that of 2010. The second is that in the bottom left hand corner of the image is a large area of blue which indicates cloud cover from the 2010 image. The right hand side of the image has a bluer tint than the rest of the image and this is because of the presence of a cloudy haze in the 2009 image which limits the reflectance values for the 2009 image. Most of the darker colors (either blue or red) seemed to be caused by cloud coverage and the shadows that the clouds give, but there are a few areas that can be seen that seem to distinguish some sort of significant change. The northern portion of the image by the river that cuts through the region is an area where red seems to dominate. There is a bit of cloud cover in the 2009 image but there seems to be much higher vegetation in this region for 2009 than there is in 2010. Another portion of the image that seems to be undergoing change is the southern portion of the image around the widest point of the river. The river seems to be fuller in 2010 than in the image of 2009 which could signify a time period of more rain in 2010 than in 2009. The second image that was processed was the change in NDVI. This subtractive image was NDVI values for 2010 – NDVI values for 2009. A similar pattern is observed with this NDVI image. Most of the image is brighter, especially in the western portion of the image while a darker in the eastern portion. The dark eastern portion corresponds to the cloudy haze that dominated the 2009 image. The third image that was processed was the principle component image for band 4. This image appeared to me to show the most change. The north west portion of the image shows a dark tint which indicates change and corresponds with the multiview of band 4. The area of the river is very bright which corresponds with the observation from the multiview band as well. It appears the area around the widest portion of the river if shifting and the urbanized area around the river is changing as well. The fact that the river appears to contain more water in 2010 (more vegetation seems to be popping out of the river in 2009 than in 2010 which would lead one to think that there is less water in the river) is in interesting point. More rain would generally lead to healthier vegetation and therefore higher NDVI values in 2010 but combined with the MODIS data, the higher temperatures in 2010 seem to limit the health of the vegetation which could help explain the higher vegetation reflectance in 2009.

=Lab 11 and 12 Supervised Classification models= ==


 * 2009 Image**==

Input File: 2009 ROI Name: (Jeffries-Matusita, Transformed Divergence) Terrace Farming [Green] 500 points: Water [Blue] 531 points: (1.99999248 2.00000000) Urban [Yellow] 679 points: (1.99999993 1.99999999) Forest [Cyan] 632 points: (1.99863624 2.00000000) Urban Agriculture [Magenta] 583 points: (1.99999219 1.99999766) Cloud [Red] 469 points: (1.99999991 2.00000000) Water [Blue] 531 points: Terrace Farming [Green] 500 points: (1.99999248 2.00000000) Urban [Yellow] 679 points: (1.99421037 1.99979242) Forest [Cyan] 632 points: (2.00000000 2.00000000) Urban Agriculture [Magenta] 583 points: (1.99981331 2.00000000) Cloud [Red] 469 points: (1.99999995 2.00000000) Urban [Yellow] 679 points: Terrace Farming [Green] 500 points: (1.99999993 1.99999999) Water [Blue] 531 points: (1.99421037 1.99979242) Forest [Cyan] 632 points: (2.00000000 2.00000000) Urban Agriculture [Magenta] 583 points: (1.97266830 1.99068399) Cloud [Red] 469 points: (1.99999898 2.00000000) Forest [Cyan] 632 points: Terrace Farming [Green] 500 points: (1.99863624 2.00000000) Water [Blue] 531 points: (2.00000000 2.00000000) Urban [Yellow] 679 points: (2.00000000 2.00000000) Urban Agriculture [Magenta] 583 points: (2.00000000 2.00000000) Cloud [Red] 469 points: (1.99999994 2.00000000) Urban Agriculture [Magenta] 583 points: Terrace Farming [Green] 500 points: (1.99999219 1.99999766) Water [Blue] 531 points: (1.99981331 2.00000000) Urban [Yellow] 679 points: (1.97266830 1.99068399) Forest [Cyan] 632 points: (2.00000000 2.00000000) Cloud [Red] 469 points: (1.99997690 2.00000000) Cloud [Red] 469 points: Terrace Farming [Green] 500 points: (1.99999991 2.00000000) Water [Blue] 531 points: (1.99999995 2.00000000) Urban [Yellow] 679 points: (1.99999898 2.00000000) Forest [Cyan] 632 points: (1.99999994 2.00000000) Urban Agriculture [Magenta] 583 points: (1.99997690 2.00000000) Pair Separation (least to most); Urban [Yellow] 679 points and Urban Agriculture [Magenta] 583 points - 1.97266830 Water [Blue] 531 points and Urban [Yellow] 679 points - 1.99421037 Terrace Farming [Green] 500 points and Forest [Cyan] 632 points - 1.99863624 Water [Blue] 531 points and Urban Agriculture [Magenta] 583 points - 1.99981331 Urban Agriculture [Magenta] 583 points and Cloud [Red] 469 points - 1.99997690 Terrace Farming [Green] 500 points and Urban Agriculture [Magenta] 583 points - 1.99999219 Terrace Farming [Green] 500 points and Water [Blue] 531 points - 1.99999248 Urban [Yellow] 679 points and Cloud [Red] 469 points - 1.99999898 Terrace Farming [Green] 500 points and Cloud [Red] 469 points - 1.99999991 Terrace Farming [Green] 500 points and Urban [Yellow] 679 points - 1.99999993 Forest [Cyan] 632 points and Cloud [Red] 469 points - 1.99999994 Water [Blue] 531 points and Cloud [Red] 469 points - 1.99999995 Urban [Yellow] 679 points and Forest [Cyan] 632 points - 2.00000000 Forest [Cyan] 632 points and Urban Agriculture [Magenta] 583 points - 2.00000000 Water [Blue] 531 points and Forest [Cyan] 632 points - 2.00000000



Confusion Matrix: [2009] (1500x1500x1) Overall Accuracy = (3004/3156) 95.1838% Kappa Coefficient = 0.9421 Ground Truth (Percent) Class Terrace Farmi Water Urban ForestUrban Agricul Unclassified 0.00 0.00 0.00 0.00 0.00 Terrace Farmi 99.25 0.00 0.00 16.58 0.00 Water [Blue] 0.00 100.00 0.19 0.00 0.00 Urban [Yellow 0.00 0.00 90.70 0.00 1.39 Forest [Cyan] 0.00 0.00 0.00 83.42 0.00 Urban Agricul 0.75 0.00 9.11 0.00 98.61 Cloud [Red] 4 0.00 0.00 0.00 0.00 0.00 Total 100.00 100.00 100.00 100.00 100.00 Ground Truth (Percent) Class Cloud Total Unclassified 0.00 0.00 Terrace Farmi 0.00 15.46 Water [Blue] 0.00 17.33 Urban [Yellow 0.00 15.37 Forest [Cyan] 0.00 14.83 Urban Agricul 0.00 17.40 Cloud [Red] 4 100.00 19.61 Total 100.00 100.00 Class Commission Omission Commission Omission (Percent) (Percent) (Pixels) (Pixels) Terrace Farmi 19.06 0.75 93/488 3/398 Water [Blue] 0.18 0.00 1/547 0/546 Urban [Yellow 1.44 9.30 7/485 49/527 Forest [Cyan] 0.00 16.58 0/468 93/561 Urban Agricul 9.29 1.39 51/549 7/505 Cloud [Red] 4 0.00 0.00 0/619 0/619 Class Prod. Acc. User Acc. Prod. Acc. User Acc. (Percent) (Percent) (Pixels) (Pixels) Terrace Farmi 99.25 80.94 395/398 395/488 Water [Blue] 100.00 99.82 546/546 546/547 Urban [Yellow 90.70 98.56 478/527 478/485 Forest [Cyan] 83.42 100.00 468/561 468/468 Urban Agricul 98.61 90.71 498/505 498/549 Cloud [Red] 4 100.00 100.00 619/619 619/619

2010 Image
Input File: 2010 ROI Name: (Jeffries-Matusita, Transformed Divergence) Cloud [Red] 251 points: Terrace Farming [Green] 332 points: (1.99859164 2.00000000) Water [Blue] 804 points: (2.00000000 2.00000000) Urban [Yellow] 663 points: (1.99958883 2.00000000) Forest [Cyan] 1251 points: (1.99846951 2.00000000) Urban Agriculture [Magenta] 465 points: (1.99113245 2.00000000) Terrace Farming [Green] 332 points: Cloud [Red] 251 points: (1.99859164 2.00000000) Water [Blue] 804 points: (1.99999982 2.00000000) Urban [Yellow] 663 points: (1.98680501 1.99497280) Forest [Cyan] 1251 points: (1.99660342 2.00000000) Urban Agriculture [Magenta] 465 points: (1.99390123 1.99991154) Water [Blue] 804 points: Cloud [Red] 251 points: (2.00000000 2.00000000) Terrace Farming [Green] 332 points: (1.99999982 2.00000000) Urban [Yellow] 663 points: (1.99940126 1.99999488) Forest [Cyan] 1251 points: (2.00000000 2.00000000) Urban Agriculture [Magenta] 465 points: (2.00000000 2.00000000) Urban [Yellow] 663 points: Cloud [Red] 251 points: (1.99958883 2.00000000) Terrace Farming [Green] 332 points: (1.98680501 1.99497280) Water [Blue] 804 points: (1.99940126 1.99999488) Forest [Cyan] 1251 points: (1.99268381 1.99999813) Urban Agriculture [Magenta] 465 points: (1.99742510 1.99988289) Forest [Cyan] 1251 points: Cloud [Red] 251 points: (1.99846951 2.00000000) Terrace Farming [Green] 332 points: (1.99660342 2.00000000) Water [Blue] 804 points: (2.00000000 2.00000000) Urban [Yellow] 663 points: (1.99268381 1.99999813) Urban Agriculture [Magenta] 465 points: (1.89396997 1.98761815) Urban Agriculture [Magenta] 465 points: Cloud [Red] 251 points: (1.99113245 2.00000000) Terrace Farming [Green] 332 points: (1.99390123 1.99991154) Water [Blue] 804 points: (2.00000000 2.00000000) Urban [Yellow] 663 points: (1.99742510 1.99988289) Forest [Cyan] 1251 points: (1.89396997 1.98761815) Pair Separation (least to most); Forest [Cyan] 1251 points and Urban Agriculture [Magenta] 465 points - 1.89396997 Terrace Farming [Green] 332 points and Urban [Yellow] 663 points - 1.98680501 Cloud [Red] 251 points and Urban Agriculture [Magenta] 465 points - 1.99113245 Urban [Yellow] 663 points and Forest [Cyan] 1251 points - 1.99268381 Terrace Farming [Green] 332 points and Urban Agriculture [Magenta] 465 points - 1.99390123 Terrace Farming [Green] 332 points and Forest [Cyan] 1251 points - 1.99660342 Urban [Yellow] 663 points and Urban Agriculture [Magenta] 465 points - 1.99742510 Cloud [Red] 251 points and Forest [Cyan] 1251 points - 1.99846951 Cloud [Red] 251 points and Terrace Farming [Green] 332 points - 1.99859164 Water [Blue] 804 points and Urban [Yellow] 663 points - 1.99940126 Cloud [Red] 251 points and Urban [Yellow] 663 points - 1.99958883 Terrace Farming [Green] 332 points and Water [Blue] 804 points - 1.99999982 Water [Blue] 804 points and Urban Agriculture [Magenta] 465 points - 2.00000000 Water [Blue] 804 points and Forest [Cyan] 1251 points - 2.00000000 Cloud [Red] 251 points and Water [Blue] 804 points - 2.00000000



Confusion Matrix: [Memory4] (1500x1500x1)

Overall Accuracy = (3227/3280) 98.3841% Kappa Coefficient = 0.9782

Ground Truth (Percent) Class CloudTerrace Farmi Water Urban Forest Unclassified 0.00 0.00 0.00 0.00 0.00 Cloud [Red] 3 100.00 0.00 0.00 0.46 0.00 Terrace Farmi 0.00 100.00 0.12 4.29 0.00 Water [Blue] 0.00 0.00 97.39 0.00 0.00 Urban [Yellow 0.00 0.00 2.49 95.10 0.00 Forest [Cyan] 0.00 0.00 0.00 0.15 100.00 Total 100.00 100.00 100.00 100.00 100.00

Ground Truth (Percent) Class Total Unclassified 0.00 Cloud [Red] 3 7.74 Terrace Farmi 10.79 Water [Blue] 23.87 Urban [Yellow 19.54 Forest [Cyan] 38.05 Total 100.00

Class Commission Omission Commission Omission (Percent) (Percent) (Pixels) (Pixels) Cloud [Red] 3 1.18 0.00 3/254 0/251 Terrace Farmi 8.19 0.00 29/354 0/325 Water [Blue] 0.00 2.61 0/783 21/804 Urban [Yellow 3.12 4.90 20/641 32/653 Forest [Cyan] 0.08 0.00 1/1248 0/1247

Class Prod. Acc. User Acc. Prod. Acc. User Acc. (Percent) (Percent) (Pixels) (Pixels) Cloud [Red] 3 100.00 98.82 251/251 251/254 Terrace Farmi 100.00 91.81 325/325 325/354 Water [Blue] 97.39 100.00 783/804 783/783 Urban [Yellow 95.10 96.88 621/653 621/641 Forest [Cyan] 100.00 99.92 1247/1247 1247/1248

Question 1: Describe what class has the lowers user accuracy? Is that accuracy still acceptable?

For the 2009 image my lowest user accuracy classification is terrace farming with a user accuracy of 80.94%. This is an acceptable accuracy; however, on the image it doesn’t necessarily match what appears to be terrace farming. For the 2010 image, the lowest user accuracy is 91.81% which is again under terrace farming. This accuracy is an acceptable value.

Question 2: Compare the two classified images and state what you find. How do they compare.

There are several things that are interesting about my two classification images. My 2009 image (although appearing to be cloudier) has less cloud area than my 2010 image classified even though the 4,3,2 band image shows less cloud cover in the 2010 images. The 2010 image has a white haze that overlay the mountainous region towards the northwest of the image that ENVI classification will put into clouds. One thing that was interesting about the classification of the urban agriculture environment is that the haze of the 2009 image actually aided the accuracy of the classification. Because of this urban haze, the urban agriculture has vastly different pixel characteristics than the similar forest region in the area. So the 2009 image is much more accurate for containing the urban agriculture region to the urban area. Another interesting trait about the 2009 image was the amount of forest that was characterized as terrace farming. The terrace farming classification was based upon the exposed dirt that would be left over after harvesting would occur. The 2010 image occurs two weeks before the anniversary date of the 2009 image which means that the 2010 image is a couple of weeks earlier in the growing season. This could account for the large amount of the terrace farming that appears to be missing from the 2010 image. I believe that the two week time period is when the crops would be harvested and cut down creating more barren terrace farming land visible compared to the un-harvested terrace farming land that is vegetated during the 2010 image. The 2010 image appears to have a more accurate classification of the amount of forest; however, because of the absence of this urban haze in the 2010 image, the urban agriculture and the forest classification are somewhat similar which creates incorrectly classified urban agriculture penetrating deep into the mountains.

Question 3: Which Classification is more accurate? The classification that is the most accurate for the urban portion of the region is the 2009 image which was helped by the urban haze. The 2010 image had a better classification of the forest and terrace farming area in the mountainous region. Although it incorrectly classified the urban agriculture in the mountains, the amount of terrace farming is much more accurate.

Question 4: What is the largest changing class in km2 and %. Based on percentages, the cloud classification increased the most with a 273.378%, but it only had a change of 55.25 km2. Based solely on area, the largest change would be in the forest classification which changed by 630.96 km2 with 218.037% pixel change. A significant change that appears not related to misclassification is the amount of water in the river and with decreases of 19.365% of pixels and 10.77 km2.

Question 5: Compare your results with the change detection lab from report 2. Discuss how the changed classes are visible in your detection. In the second report, one thing I noticed or thought I had noticed was the apparent widening of the river but according to the classification analysis, the river is not widening, but rather decreasing. Because of the restraints of having to get 6 or more classifications, the best place to classify the image was in the region with an urban presence. But, because of the cloud and haze amount in this region, this created great errors in the classification. Concurrently, the area that seemed to have the most change appears to be in the region opposite of where I took the classification so much of the change detection didn’t correlate very much.