Module 1 of GIS 5100 [Applications in GIS] was very interesting and insightful; it was also a great introduction into one of the many ways GIS systems can be used to improve the world in which we dwell. The module focused on using different hotspot analysis methods to map crime in two cities; after the information was mapped, the data could by further analyzed to see which mapping technique would be more effective at preventing the same type of crimes from taking place in the future.
The first half of the module was based on datasets from Washington D.C. for 2018. First, we created a thematic, choropleth map [see GIS 5007, Module 5 for more information on choropleth mapping] of burglaries throughout 2018 [per 1000 homes] that occurred in each census tract that lies within Washington D.C. [above, left]. While this map does provide insight into which census tracts had more burglaries, it can often be visually misleading to the user, due to the varying sizes of the tracts themselves.
Next, we mapped all the assaults with dangerous weapon incidents that occurred throughout Washington D.C. in 2018. For this map, we used the Kernel Density Estimation model; this type of model places a user-defined grid over the city and searched each grid cell for how many assaults occurred within a user-defined radius. Each incident is then given a weight that is calculated according to its proximity from the grid cell's centroid. Finally, the density of each cell is mapped, creating a map that shows the assault hotspots of 2018 [above, right].
On the second half of the lab, we analyzed datasets for the city of Chicago, IL to determine which method would give the most accurate predictions of future occurrences of a specific crime [homicide]. To do this, we created three maps of all homicide hotspots that occurred in 2017 throughout the city using three different mapping techniques: Grid Overlay, Kernel Density Estimation, and Local Moran's I Spatial Autocorrelation [see below].
For the Grid Overlay map, a grid was placed over the entire city and a spatial join function was run to obtain a number count of homicides within each grid cell. Once this count was determined, the top 20% of grid cells with the highest homicide counts were mapped. The Kernel Density Estimation technique was described earlier in the post, and the areas having greater than 3x the mathematical mean were mapped as hotspots. Finally, Local Moran's I is a method that determines where homicide cases are random or where they are clustered. This intensive formula is beyond the scope of this discussion, but ESRI's documentation on this Spatial Autocorrelation method can be found here. The formula determined where the clusters [hotspots] exist, and they were mapped in the righthand frame on the map displayed above.
The remainer of the lab focused on analyzing the statistics produced by each mapping technique. To accomplish this, the total area of each hot spot mapping method was calculated [square miles]. Next, all homicides that occurred throughout Chicago, IL in 2018 were overlayed on each map and a spatial join function was executed to determine the number of 2018 homicides that took place within the 2017 hotspots. Then, the percentages of 2018 homicides within each 2017 hotspot were calculated. Finally, the density was calculated by taking the number of 2018 homicides that occurred within 2017 hotspots and dividing that number by the total area of the 2017 crime hotspots [this was done for all three maps].
After a careful analysis of all these statistics, I would recommend the Kernel Density Estimation map as the most effective means to efficiently allocate resources, thereby reducing the number of homicides in the future. The reasons for this choice are as follows:
The spatial area of the KDE map is not the biggest or the smallest; too small of an area would place all resources closely together, causing an increase in response times to areas not within those confined spaces. Conversely, too large of an area, and all resources are spread out too thinly, also causing increased response times. Secondly, the percentage of 2018 homicides happening in 2017 crime hotspots was remarkably high compared to the Grid Overlay map, and only marginally smaller than Local Moran's I. The reason I chose KDE over Local Moran's I was due to the fact that Local Moran's I had a 38% increase in land area over KDE, which would force a sparser allocation of resources. Thirdly, the density of Grid Overlay was higher than that of KDE [Local Moran's I homicide was substantially lower], but this was based on a smaller geographical area and a lower count of 2018 homicides taking place within 2017 hotspots. A careful reasoning of this statistic concluded with KDE being the better option.
This lab was very interesting and intriguing; an excellent example of how GIS is being used on a professional level every day. I look forward to advancing further in this class to see how GIS systems can be utilized in other real-world scenarios.
No comments:
Post a Comment