Pattern Analysis

Analyzing Fire Department Calls for Patterns of Clustering and Other Global Spatial Patterns

I have been tasked with examining the spatial clustering of data for four different scenarios.  The first is to determine if false fire alarms are exhibiting any signs of clustering.  The second is to determine if calls that are highly ranked are clustering.  Third, I used the previous data to determine the maximum z-score, confidence level and distance band for clustering. Finally, the fire department wants to know the density of calls per block and wants to know where the calls begin to cluster.

To aid me in analyzing fire department calls I used ArcGIS Pro version 2.9. The data was provided by NCSU GIS 520 course materials. To solve the four processes I performed different methods.  First, to determine if Fort Worth EMS calls were randomly distributed in 2015 I needed to know the area of the data (Figure 2 – Battalion 2) in order to perform the average nearest neighbor tool which generated a report with significant figures.  With those figures I was able to determine that the calls were clustered by a 90 percent confidence.  Second, in order to determine if the ranking of calls is random or clustered I calculated the distance band from neighbor count to determine the average distance.  I used that average for the high/low cluster tool for incremental distances and created a table of distances, g index and z scores to determine at what distance the z score peaks.  Third, I created two charts; the first I used Ripley’s K function to measure the distances between objects to determine randomness and clustering of ranked EMS calls and generated a graph.  I then ran Ripley’s K function tool again but with 99 permutations and graphed the differences.  I compared the graph’s highest peaks to find the maximum clustering.  Finally, I aggregated the 300 x 300 foot grids by spatial join and changed the join’s properties to show where the join includes patron locations.  I then ran the spatial autocorrelation tool multiple times with varying distances to find the peak z score to determine the clustering of calls.

Figure 1: Workflow Diagram Representing Different Stages Of the Task in Each Column

Identifying how patterns occur and where they occur is important in understanding how geographic patterns behave. The values associate with my data, such as calls or distance, is at first assumed to be random. The p-value associated with my data values identifies the probability that my data is random at a percentage of confidence. The z-score provides information on how far off the data is from the average data sets. These statistical measurements are important to understand the output information ArcGIS Pro displays to understand the data collection and in turn make the best decision from the information displayed.

This task has allowed me to run multiple methods of global spatial pattern processes. Having a well roundedness of multiple tools in the toolbox allows for better choices to be made and how to access the results the user wants.

Figure 2: Map Representing Call Clustering with 90% Confidence

Figure 3: Map Representing Higher Ranked Calls to be Clustering

Figure 4: Map Representing Maximum Z-Score, Confidence Level and Distance Band for Clustering

Figure 5: Map Representing Density of Call per Block and Where Calls Are Clustering

Another example of how some tools from this assignment could be used is if a restaurant wanted to know if customers from the restaurant originate from cluster or random locations. The restaurant also wants to know if the customers that are paying higher prices for their food come from clustered or random locations.  The data to determine this would be addresses of the customers and the amount of money the customers spent at the restaurant both would be collected from the restaurant itself from credit card records.  I would first need the area of the data set in order to use the average nearest neighbor tool then to determine if the customers locations are clustered I would analyze the mean distances and their z-scores.  I would then run the high/low clustering tool with varying distances to determine where the peak location is for the greatest clustering of highest paying customers. I would present the pattern analysis to the restaurant.