Northern+Wiscosin

Lab 7

Downlable Landsat 5 scenes for this area with cloud cover less than 20%:
 * Date ||= Cloudy Cover ||
 * 2010/09/01 ||= 0% ||
 * 2008/07/09 ||= 3% ||
 * 2008/05/22 ||= 0% ||
 * 2008/03/19 ||= 0% ||
 * 2008/03/03 ||= 0% ||
 * 2007/04/18 ||= 0% ||
 * 2006/06/02 ||= 0% ||
 * 2006/05/17 ||= 10% ||
 * 2005/08/02 ||= 0% ||
 * 2000/09/05 ||= 0% ||
 * 1999/07/17 ||= 0% ||
 * 1999/05/30 ||= 0% ||
 * 1998/05/27 ||= 10% ||
 * 1998/04/09 ||= 0% ||
 * 1998/03/24 ||= 0% ||
 * 1997/06/09 ||= 0% ||
 * 1996/08/09 ||= 10% ||
 * 1991/02/17 ||= 10% ||
 * 1989/07/21 ||= 10% ||
 * 1987/06/14 ||= 0% ||
 * 1985/06/24 ||= 0% ||

Based on the cloud condition in these available secenes, I will choose the secenes of 1987/06/14 and 1997/06/09 for the following analysis.

Image for 1997/06/09 before atmospheric correction (RGB: Band 432)

Image for 1997/06/09 after atmospheric correction (RGB: Band 432)

Spetral profile for a vegetation pixel before and after the atmospheric correction

Before:

After:



Image for 1987/06/14 before atmospheric correction (RGB: Band 432)

Image for 1987/06/14 afte atmospheric correction (RGB: Band 432)

Spetral profile for the same vegetation pixel before and after the atmospheric correction

Beofore
 * [[image:plot1987_before.jpg]]

after:


 * Lab 9: Spatial Filter**

The original subset image (RGB432)


 * Median Filter 3 x 3** (RGB432)**:**

Median Filter 9 x 9 (RGB432):

Compared to the median filter with the kernel size of 3x3, the 9x9 median filter give a much more smoothed image on which the edge information is completely suppressed.

**Sobel Edge detector (RGB432)**:



The edge detector sharpens the contrast of the edge of the ground objects.


 * Principle Component Analysis**

True color original image:
 * Pc1:**

The first PCA component contains most of the information of the oringinal image, except urban area.

Pc2:

The second PCA component mainly contains the information of water.

Pc3:

**Pc4:**

It is interesting to find that the urban information is in the 5th PCA component.

**Pc5:**

**Pc6:**


 * Statistics:**


 * Eigenvalue Plot:**



Band 1 7 3167 296.224488 185.631984 1 1055621.718951 Band 2 44 3234 599.924633 230.748766 2 620694.708192 Band 3 30 3373 462.774332 363.124643 3 15867.081482 Band 4 -184 6810 3671.787197 804.823689 4 8486.785285 Band 5 -36 5183 1907.319012 596.949900 5 1273.684820 Band 6 -15 5584 967.928355 692.210084 6 864.908240
 * Basic Stats Min Max Mean Stdev Num Eigenvalue**

Band 1 34459.233586 41036.961630 66215.344624 -47402.857513 96104.469458 122430.377488 Band 2 41036.961630 53244.992842 81302.861630 -32152.362779 126949.654775 152240.642307 Band 3 66215.344624 81302.861630 131859.506645 -97669.674992 191115.706763 243616.092952 Band 4 -47402.857513 -32152.362779 -97669.674992 647741.170091 11863.745870 -140691.975391 Band 5 96104.469458 126949.654775 191115.706763 11863.745870 356349.183243 382570.367418 Band 6 122430.377488 152240.642307 243616.092952 -140691.975391 382570.367418 479154.800563
 * Covariance Band 1 Band 2 Band 3 Band 4 Band 5 Band 6**

Band 1 1.000000 0.958039 0.982314 -0.317286 0.867267 0.952793 Band 2 0.958039 1.000000 0.970310 -0.173130 0.921625 0.953133 Band 3 0.982314 0.970310 1.000000 -0.334198 0.881663 0.969198 Band 4 -0.317286 -0.173130 -0.334198 1.000000 0.024694 -0.252540 Band 5 0.867267 0.921625 0.881663 0.024694 1.000000 0.925839 Band 6 0.952793 0.953133 0.969198 -0.252540 0.925839 1.000000
 * Correlation Band 1 Band 2 Band 3 Band 4 Band 5 Band 6**

Band 1 -0.173433 -0.210649 -0.345675 0.331967 -0.511454 -0.658992 Band 2 -0.022279 -0.071790 -0.039164 -0.924822 -0.336849 -0.155091 Band 3 -0.203290 -0.173816 -0.339841 -0.182847 0.781324 -0.411181 Band 4 -0.324928 -0.470127 -0.543690 0.004407 -0.105958 0.605440 Band 5 0.576795 -0.774563 0.243915 0.025843 0.056681 -0.063127 Band 6 -0.699877 -0.315111 0.639054 0.019539 0.006422 -0.045438
 * Eigenvector Band 1 Band 2 Band 3 Band 4 Band 5 Band 6**

According to the eigenvalues given in the basic statistics, the percentage of variability explained by each PCA component is respectively: 61.99%, 36.45%, 0.93%, 0.5%, 0.07% and 0.05%. The first three PCA component can explain about 99.38% of the variability of the original image.

According to the correlation matrix, the correlation efficient is highest for the visible blue band and the visible red band, which is 0.982314, the least correlation should be the correlation coefficient with the least absolute value and it is the correlation coefficient for band 2 and band 4


 * True color image of the inverse PCA**

There is a better contrast for the true color image created by the inverse PCA than the original true color image. This could be caused by leaving out the last three PCA component which mainly contain the noise of the image and the inverse PCA could be used as an noise filter procedure.


 * Spear Change detection**


 * Two color Multi-view for band 4**

. If the objects are shown in cyan on the two color multiview image, it indicates that the energy in the near infared band is reflected more strongly and this could be related to the increase of vegetation. In contrast the objects shown as red on the two color multiview image is related to vegetation removal.

**//Image transform-Principle component analysis//** 

**//Both the third and the fourth principle components contain the land cover change information. But compared to the third. ////principle component, the fourth principle component also contains the area without change, in other words, there is a greater contrast between the changed and unchanged part of the original image in the third principle component than that on the fourth principle component. And the objects shown as bright indicate //** **//the expansion of urban area while the dark objects represent the increase of vegetation //**

//Image subsractive//

Change in NDVI 

Red Blue Ratio

Man made




 * Lab11&12**

Training and validation pixels collected for the 1987 image


 * ROI separability for image of 1987:**

1987 Classified image:


 * Legend**


 * Water**:Blue
 * Forest**:Green
 * Agriculture**:Sea Green
 * Fallow Agriculture:**Yellow
 * Bare soil**:Magenta


 * Confusion matrix for 1987 classified image**


 * Training and validation pixels collected for the 1997 image**


 * ROI separability for image of 1997:**
 * Confusion matrix for 1997 classified image**


 * 1997 Classified image:**

**Legend**

**Water**:Red **Forest**:Green **Agriculture**:Blue **Fallow Agriculture:**Yellow **Bare soil**:Magenta

**Question 1: Describe which class has the lowest user accuracy. Is that accuracy still acceptable?** On both of the classified images of 1987 and 1997, agriculture has the lowest user accuracy. According to the confusion matrix, commission errors are the largest for agriculture on both of the two classified images and the commission error is mainly caused by the difficulty in distinguishing agriculture from forest and bare soil. On one hand, vegetation cover for training pixels collected at some agriculture fields is as high as that of forest, which tend to lead the maximum likehood classifier to classify some forest pixels as agriculture. On the other hand, the purity of some agriculture training pixels is influenced by the soil background, which makes it have similar spectral characteristics to that of bare soil. As the major area of the subset image is forest, so the low user accuracy for agriculture is still acceptable. **Question 2: Compare the two classified images (time 1 and time 2) and state what you find. How do they compare?**  From the classified images, it is interesting to find that the major land cover inter-conversions are among the classes of agriculture, fallow and bare soil. There is a 36% decrease in agriculture while a 32% and 57% in bare soil and fallow agriculture. As there is only four days difference for the two images in terms of Julian dates, it is surprisingly to find the changes in the schedules of the agricultural activity. I think this may be sign of an adaption to the possible climate change in this area, which mean although the dates for the two images are close, people in this area gradually change their farming schedule which leads to conversion of agriculture to fallow or bare and vice versa. **Question 3: Which classification is more accurate?** The classification for the 1997 image is more accurate in terms of the overall accuracy (96.06% compared to the 95.11% for 1987 image) and the kappa coefficient (0.9470 compared to the 0.9356 for 1987 image). The improved classification accuracy can be attributed to more training pixels collected in the classification for the 1997 image (8012 training pixels collected for image of 1997 and 6002 training pixels collected for image of 1987). **Question 4: What is the largest changing class in km2 (note you can change the units under Options) and %? Look at the Class Changes.** The largest changing class in km2 is agriculture which shows a decrease of 32.15 km2. The class with the largest change in terms of percentage is bare soil which shows a 55.2% increase. The class change in km2 and % is respectively shown in figure 1 and 2. ** Question 5: **** Compare your results with the “change detection lab” from report 2. Discuss how the changed classes are visible in your change detection. ** The dominant changes found in the spear change detection lab are the inter-conversion among the three land cover classes of agriculture, fallow agriculture and bare soil. Assume both of the images area shown as band 4, band 3 and band 2 respectively for red, green and blue channels, the agriculture is shown as bright red with smooth surface and regular shaped land patches (most of the agriculture patches are rectangular) and some bare soil classes is shown as dark green patches with regular or irregular shapes while the fallow agriculture have almost the same features as that of agriculture except that they are always shown in bright cyan. If the bare soil class, keep the rectangular shape, we are pretty sure it is converted from agriculture fields and it is in irregular shape and in a location that is previously occupied by forest, it is possible that the bare soil came from a clear cut of the forest. The inter-conversion between agriculture and fallow agriculture is easy to locate, they are always kept in the same shape and only differ in the color.