GEOG 566






         Advanced spatial statistics and GIScience

June 5, 2017

My Spatial Problem Revised

Filed under: 2017,Final Project @ 1:03 pm

1) The Question I asked:

Does site aspect play a significant role in radar backscatter after processing?

2) Description of the Dataset

The backscatter dataset comes from the Sentinel-1 satellite constellation, which is equipped with C-band radar and has global coverage. The return time for a given area is 6-12 days, with a ground resolution on the order of 10 m (after pre-processing steps have been completed). I was looking at a single point in time which covered much of Western Oregon. I am using plots of land owned by Starker Forests Inc., which can be seen in Figure 1 below. Although the information on forest cover was not provided, using the plots significantly reduced the size of the area of interest (for ease of use in ArcMap), as well as ensured that at the least the areas would be even-aged, monoculture stands of Douglas fir.

Figure 1: Area of Interest with values extracted at points (green).

3) Hypotheses

The null hypothesis I was working with was that there is no significant difference between aspects in average backscatter. This is assumed to be true once the preprocessing steps have been conducted, but my project was to investigate if that is a safe assumption.

4) Approach

The first analysis I used was to test for autocorrelation. This was a frustrating process in ArcMap due to the clunky nature of that program, however once I exported my data from Arc and brought it into R the process became much smoother. The autocorrelation step was an investigative move to test if the data points I would be using were a representative sample.

The next analysis was a comparison of aspect and backscatter values, which involved another grueling battle with Arc which ended with a semi-automated workflow that exported data for later use in R.

The final analysis was to look closer at the (non)relationship between aspect and backscatter by using various statistical tools in R (namely the bartlett test and the Kruskal-Wallis test).

5) Results

I produced a few maps, but the important results came in the form of graphs and statistical values. The most important was the finding that backscatter did not correlate at all with aspect via a Kruskal-Wallis test (figure 5). Combined with a bartlett test on homogeneity of variances, I found that the values of gamma-naught (backscatter) did not vary between the different aspects. These two tests show that there is no evidence that backscatter changes with aspect (all else being equal).

Figure 2: Variogram of the datapoints used in Excercise 1

Figure 3: Scatterplot of the backscatter values according to aspect (clockwise from top-left: NW, N, NE, E, SE, S, SW, W, center: Flat).

Figure 4: Boxplot showing the means and variation of backscatter (y axis) and aspect (x axis).

Figure 5: Results of Kruskal-Wallis test on both Gamma~Aspect and log(Gamma)~Aspect. Both tests resulted in high p-values (p > 0.6), which indicates the values of gamma-naught (backscatter) do not come from different distributions. If they did, that would mean that aspect plays a role in the distribution of backscatter.

6) Significance

The significance of this project is fairly limited to users of Sentinel-1 data who are using the same pre-processing as me. It is mainly a validation of the radiometric and geometric calibration process (an important step nonetheless).

7) What did I learn about programs?

I solidified my disdain for ArcGIS as a tool for statistical analysis on large datasets. I now see Arc as a gateway software that you can pass data through, pull out what you need based on geographic means (or with model builder), and then export the data to use in R or some other program.

I learned how useful R can be in geostatistical analysis, and was exposed to a few new software packages which I will be using in my thesis work.

8) What did I learn about statistics?

I learned a few new statistical tools such as variograms and the Kruskal-Wallis test. I didn’t learn enough to explain them very well to anyone else, but I did learn enough to know what the results from each mean in relation to my data.

9) Response to critique and comments

The responses and comments I received were very helpful. David and Marja both had thoughtful insight on my tutorial 2, and brought up approaches that I did not think to do. Lody and Audrey were also helpful during the first phases (tutorial 1) since they had similar projects to mine and a similar background so we could help guide each other.

The response from Dr. Jones during class was the most helpful though. I would frequently come to a dead end in my processing or my approach, but Dr. Jones always had an idea to change the direction toward an ultimate goal by taking smaller steps. Her comments on Canvas were not as helpful in giving direction.

Print Friendly, PDF & Email


No Comments

No comments yet.

RSS feed for comments on this post.

Sorry, the comment form is closed at this time.

© 2019 GEOG 566   Powered by WordPress MU    Hosted by blogs.oregonstate.edu