Spatial Statistics

Interpreting Calls for Assistance from Fort Worth, Texas Using Spatial Statistics

Problem: The Fort Worth, Texas fire department wants to address four questions within battalion 2’s jurisdiction.  The four questions are:

  1. Do EMS calls have a tendency to cluster?
  2. Do high priority calls have a tendency to cluster?
  3. Are features in the Calls for Service dataset randomly distributed when weighted with priority rankings?
  4. At what distance do densities of calls per block cluster?

Analyses: In order to answer the questions posed by the Fort Worth, TX fire department, we need to use their provided database files along with shape files that contain battalion 2’s jurisdiction.  With this data we will perform our analyses using four different techniques to answer their questions.  These techniques involve using the Nearest Neighbor Tool, the General G Tool, Ripley’s K Function, and Moran’s I.  The four processes are discussed in the following sections.

Analysis 1: Using the EMS Calls data from Feb 2007, we use the Average Nearest Neighbor Tool to determine the distance from each feature to the nearest neighbor.  Our data shows the Nearest Neighbor Index is .915, Observed Mean Distance is 995.29’, Expected Mean Distance is 1087.30’, Z-Score is -1.698, and Confidence level is 90%.  These results demonstrate that there is not a tendency trending towards clustering.  We can therefore reject the null hypothesis that states that the data is randomly distributed.
Picture1

Results 1:
 Map of EMS Call Clustering
whitlow_m_exercise8_1

Analysis 2: Using the Calls for Service data from Feb 2007, we use the General G tool to assess how locations affect clustering.  Our data shows the Maximum Z-Score is 9.861, the Confidence Level is 99%, and the Distance Band of Clustering is 200’.  These results identify the band of 200’ as the point where maximum clustering occurs.   As the band width increases, the z-score decreases.  This means that high priority calls have a higher tendency to cluster at the 200’ width when compared to any width higher than 200’ at 50’ increments.  We can therefore reject the null hypothesis that states that the priority ranking values for the features are randomly distributed.
Picture2

Results 2:
Map of High Priority Call Clustering
whitlow_m_exercise8_2_2

Analysis 3: Using the Calls for service data from Jan 2007, we use the Ripley’s K Function tool to assess the randomness of the features when weighted with priority rankings.  Our K function map peaks at the value of 500 and 900.  Our second graph that uses 99 permutations to account for a larger sample clearly indicates a greater cluster around the value of 900.
Picture3

Results 3:
Map of Calls for Service Weighted by Priority Rankings
whitlow_m_exercise8_3_2

Analysis 4: Using the Calls for service data from Fab 2007, we use the Moran’s I tool to identify the densities of calls per block, and determine where they cluster.  Our data shows the maximum Z-Score is 3.703, with the confidence interval of 99% at the band size of 550.  We can therefore reject the null hypothesis that states that the calls for service are randomly distributed among city blocks.
Picture4

Results 4:
Map of Call Distance Density per Block
whitlow_m_exercise8_4_2

Application and Reflection
Using spatial statistics I can better determine where incidents of trail conflict exist.  Visitor data would have to be collected and entered in order to complete the analysis.  This could be achieved using a spatial join, connecting trail maps with trail conflict incidents.  The results provided through spatial statistics can help park managers assess their trails and trail types in order to minimize reported trail conflict incidents by implementing alternative management protocols.

Problem Description: Using spatial statistics I can better determine where incidents of trail conflict exist.  The results provided through spatial statistics can help park managers assess their trails and trail types in order to minimize reported trail conflict incidents by implementing alternative management protocols.

Data Needed: Visitor data would have to be collected and entered in order to complete the analysis.  These database files would need to be paired up with shape files that contain trail and boundary layers for the park in question.

Analysis Procedures: This process would involve using a spatial join to connect trail maps and trail conflict incidents.  We would need to use the Spatial Autocorrelation (Moran’s I) tool to provide resulting statistics.  The statistics would then need to be interpreted to determine significance level, and where trail conflict needs to be mitigated.