Sunday, December 10, 2017

Special Topics in GIS Lab 15

Dasymetric mapping is 

"a technique in which attribute data that is organized by a large or arbitrary area unit is more accurately distributed within that unit by the overlay of geographic boundaries that exclude, restrict, or confine the attribute in question. For example, a population attribute organized by census tract might be more accurately distributed by the overlay of water bodies, vacant land, and other land-use boundaries within which it is reasonable to infer that people do not live."


For the lab this week we focused on dasymetric mapping and I found this definition from esri helpful so I wanted to add it here. This week really focused a lot on basic arcGIS skills with an in depth understanding of how populations are distributed. For the lab we used a raster product of population density or imperviousness to get more accurate estimates for populations in certain areas.

This image is a layout of the census tracts used for the population clipped and erased to remove hydrology features. It is overlayed on the raster of imperviousness to show the population densities. This product along with many statistical editing allowed us to calculate more accurate population estimates.

Tuesday, December 5, 2017

Special Topics in GIS Lab 14

This week in the lab we focused on the topic of Gerrymandering which is the representation of of a population through skewed congressional districts. We looked at metrics that could be used to measure the skewness of the districts. The two metrics used were compactness and community. Compactness measured how spread out and unnatural the district's shape was while community measured how many counties the district overlapped either completely or partially. For compactness the equation 400pi(area)/(perimeter)^2  can be used to get a value. This equation produces a result from 0-100 where 100 is the best compactness (a circle). The lowest compactness or 0 would be the worst case where the perimeter is very large and the area very small. For community the measurements were derived from running an intersect to see how many counties were divided into sections by the districts spatially and how many counties were included in each district total. Then the total divided counties were divided by the total counties within the districts to display a percentage of counties divided. Below are some examples of worst compactness and community scores. The first demonstrates low compactness and the second low community.

Wednesday, November 29, 2017

Special Topics in GIS Lab 13

This week in the lab we worked with scale and resolution. The lab focused on differences between line and polygon features in the first half and raster DEMs in the second portion described below.

To compare the two DEMs I focused on slope, aspect, and overall elevation. First the slopes of each DEM were mapped and the averages compared. Then for further analysis the slope maps were combined to display the slope difference between the two DEMs. Second the aspect of each DEM was mapped and compared to analyze the slope direction and general direction the terrain was facing. This analysis gave more of a horizontal representation of the differences rather than strictly elevation differences. Finally the overall elevation was compared for the two DEMs. The standard statistics were calculated and compared to determine the data represented.

The results clearly show that there is a difference between the SRTM and LiDAR data. Both of the DEMs were analyzed at 90m resolution. The slope of the separate DEMs seemed to follow the terrain feature of the area which was a drainage. The LiDAR model had a steeper average slope which revealed itself in the higher elevations of the terrain feature. The slope difference between the models is also shown in the visual. The areas where the slope is the steepest between the valleys and ridges tended to have a larger difference while relatively flat areas had a smaller difference.
The aspect maps of the DEMs at 90m resolution were fairly similar. The aspect analysis was meant to look at the models horizontal of directional facing representations rather than elevation. The results show that even though he models a very close match there are still some areas that have different aspects. This shows that the slopes differed between the certain areas enough to cause the software to display a different facing direction for the certain area. The differences are slight and nowhere near opposite direction, but if an accurate representation is vital then consideration should be taken in which model to use for a study.

Finally the elevation of the two models differs. In the table the results show that the LiDAR model has a lower average elevation along with a higher maximum and lower minimum elevation. To me these results show the LiDAR model as more complete. In the elevation comparison map the areas of largest elevation difference follow the trend in the slope difference map. They are in the specific areas of the terrain feature where the valley is transitioning to a ridge and has a steeper slope. The one interesting result is that the large differences in elevations are only seen on the southern side of the drainage in areas with steep slopes. The norther side of the terrain feature shows smaller differences between the two elevation models. This could be due to the location of the sensor platforms when the separate model’s data was collected.

The creation of the LiDAR and SRTM both involve air/space borne sensors. LiDAR can also be collected with ground based sensors, but larger areas tend to be collected on using airborne sensors. These two sensors are both collecting elevation data, but one is tens to hundreds of miles in the air, while the other is less than a mile above the ground. The fact that the SRTM data is at 90m is impressive considering how high up in space the sensor is located. Overall I would have to say that the LiDAR model is a more accurate representation of the drainage feature in all perameters due to the fact that the sensor is collecting data far closer to the source, the model was derived from a higher resolution product, and the LiDAR sensor was probably built for high resolution large scale ground mapping.

Wednesday, November 22, 2017

Special Topics in GIS Lab 12

This week in the lab we focused on Graphically Weighted Regression or (GWR). This type of regression analysis uses the same analysis as the OLS, but takes into account a spacial aspect. For the lab this week we used a county area with locations and types of different crimes. The goal was to select a particular dependent variable and figure out which independent variables are able to describe the dependent variable the best. Using techniques learned from previous labs I created a correlation matrix and exploratory regression in ArcGIS to determine the most descriptive regression was completed using three distinct independent variables. I then used these variables to create an OLS and GWR analysis for comparison. After reviewing the results it was clear that the GWR was more than 10% more accurate in describing the dependent variable than the OLS. This is due to the GWR using the spatial aspect rather than analyzing as a whole.

Monday, November 13, 2017

Special Topics in GIS Lab 11

Using regression analysis in ArcGIS this week was eye opening. Prior to this week the analysis was all done in Excel and displayed in spreadsheets. This week we were able to pull the information into ArcGIS and perform the same analysis in a way that produced a visual representation as well as a summary of statistical reports that help the user analyse the testing. First you need the data for the analysis to be performed on. Then you simply use the Ordinary Least Squares tool in ArcGIS to set your data set, dependent, and independent variable(s) and let the program do the work. The results are provided in an easy to read report that gives all the information you need to decide if the results fit your analytical needs. This week in the lab we worked on interpreting the results. The focus was on the many values that the program calculates to analyse the regression. The tool that you run in ArcGIS even incorporated instructions int he results for reading the reported variables and determining if the result is what you are looking for.

The output of the OLS tool produces a residual map that visually shows the difference the actual data is from an expected value on the regression line. This visual aid can help us determine the accuracy and bias that could be present in the results. For instance if there is a pattern in the results or areas where the residuals are similar in a certain area it could show a bias or non-random correlation. This information is good to have because it could clue you into the fact that you need to use another variable to run the regression analysis.

Finally at the end of the lab (of course) it was revealed that rather than running the tool many times on different independent variables you can simply run another tool that includes all of the variables you wish to analyse and reports the most accurate result possible. The capability of technology today is impressive and makes me wonder what the engineers and analysts will have the software calculating for us next!

Monday, November 6, 2017

Special Topics in GIS Lab 10

The lab this week focused on regression and how to analyze data to determine predicted values. The focus this week was in Microsoft Excel rather than ArcGIS. The analysis was formula intensive, but allowed for a better understanding of how to use Excel to gather the data parameters you want for analysis. The Excel program also has a great tool that can be used to accomplish the majority of the regression analysis in one easy step.
This example sums up the processes we used in Excel this week. The goal is to determine if the data points you collect or are given have any correlation and if so how well are the values predictors for each other. In our example two stations reported rainfall annually for a given number of years. One station failed to report for a span of years and it is your job to predict the missing amounts to complete the dataset. In order to do that you perform a regression analysis on the rest of the data to find the slope and intercept value of a "trendline" for the data. This line equation can then be used to enter your x-values from one station and find our y-values for the missing station numbers. The values will not be perfect, but they will be predictions in line with the rest of the data. 

Tuesday, October 31, 2017

Special Topics in GIS Lab 9

This week in the lab we learned about accuracy statistics when using Digital Elevation Models (DEMs). In order to accomplish the analysis process you need a DEM to analyse and test elevation points that cover the DEM extent. This analysis is only for the vertical elevation accuracy. We focused on horizontal accuracy in the first part of the semester. For the vertical accuracy statistics we plotted the test points against the DEM then derived the elevation values of the DEM pixels at the test point locations. By comparing the two values we were able to calculate RMSE including the 95th and 68th percentile statistics. The image below shows the result of the vertical accuracy testing. The letters a through e are the different types of land cover represented in the test  points. These were provided as comparison factors for the accuracy statistics. The overall accuracy is listed at the end under All.