For exercise 3, I finally left the world of Arc and took my data to R.

### Question/Objective:

Test the radar backscatter between different-facing aspects and determine if aspect influences radar signal.

### Tool used:

I used R to run the statistical analysis, but still needed to extract the data using Arc.

### What I did:

I extracted the data from ArcGIS by creating 1000 random points througout my scene and running the “extract data to points” tool to pull out the backscatter and aspect at each point. Then I exported the table of point data to Excel to clean up any missing values that might throw off R. There were 4 values in total that were missing either the backscatter or aspect value. Then I saved this table as a .csv and brought it into R.

Once in R, the data needed a little bit of cleaning up. Since the default aspects from Arc range from -1 (flat) and 0 (north) to 360 (still north, all the way around the compass), using just the numeric value to classify the points would not work well. I used the following lines of code to change the aspects to factors:

NEWDATA$ASPECT <- cut(DATA$ASPECT,breaks=c(-1,0,22.5,67.5,112.5,157.5,202.5,247.5,292.5,337.5,360),labels = c(“FLAT”,”N”,”NE”,”E”,”SE”,”S”,”SW”,”W”,”NW”,”No”),include.lowest = TRUE)

levels(NEWDATA$ASPECT) <-c(“FLAT”,”N”,”NE”,”E”,”SE”,”S”,”SW”,”W”,”NW”,”N”)

The second line combines the two “north” aspects into one (i.e. 0-22.5 degrees and 337-360 degrees).

### Results:

The results were promising. I expected to find no difference between aspects, and the scatterplots showed that, at least upon visual inspection, the aspects showed no difference between each other.

A boxplot also showed little difference in the average Gamma0 values (the backscatter).

Next step was to run a statistical analysis. Since the data were not normally distributed, I had to do something other than an ANOVA, so after some internet research I went with a Kruskal-Wallis test. This tests for difference in population distributions (i.e. if they are identical or not) when the distributions are not normal. The code is simple enough:

kruskal.test(GAMMA~ASPECT, data=DATA)

I also conducted an ANOVA on the log-transformed Gamma0 data, which appeared more normal than before. None of the results were significant from any test, meaning that the aspect does not have a statistically significant effect on the backscatter, which is the result I was hoping for.

### Critique of the Method:

This method was useful for my dataset. I had a rather large dataset to work with in Arc, and didn’t find the statistical tools that I needed in Arc either. I am comfortable in R, and prefer it for larger datasets since it doesn’t have the bells and whistles that Arc does.