Contact:

James (DaD. Carroll
jdcarro2@ncsu.edu

+1-919-559-2894

More info:

Refer to the links below to learn more about the use of different ArcMAP tools and resources.

Links:

- Average Nearest Neighbor Tool

- High/Low Clustering (Getis-Ord General G) Tool

- Multi-Distance Spatial Cluster Analysis Tool

- Spatial AutoCorrelation Tool

 

Spatial Statistics


The Problem

This exercise introduced the capabilities of ArcGIS for analyzing spatial patterns. The project mission is to develop a professional GIS analysis of the call data for the City of Fort Worth, Texas for the Fort Worth Fire Department to determine whether the EMS calls for Battalion 2 tend to cluster. If there are clustering patterns, then they need the data analysis as a basis for making decisions about stationing EMS Intensive Care Units (EICU) at locations near the hot spots. We are required to provide our analysis and recommendations which support either clustering patterns, or a null hypothesis which will require no location changes.

Strategies

TThis GIS analysis will be done using ArcGIS 10.2.2 ArcCatalog and ArcMap with the specific focus on the use of Spatial Statistics Tools for Analyzing Patterns. The data was supplied through the download of the "analyzing_patterns_data" zip file. The data to be used are two geodatabases; Census which contains road data, and the City of Fort Worth which contains all Fire Department and Call data. The data files have the associated tables and grids necessary for the analysis. To verify the validity of any one processing analysis four different GIS Tools will be used to create a full analysis to support the final result and recommendation. A final map file will be produced for future geoprocessing analysis for the City of Fort Worth Planning Department.

Methods

Part One:
1. Open the "Tutorial 8-1.mxd" file in ArcMap.
2. The First Step of Data Analysis needs to determine if a Average Nearest Neighbor Tool analysis will provide a result documenting a tendency of clustering locations before proceeding further. The index calculation results needs to show that the average distance from a feature to all its neighbors when compared to the average distance for random distribution supports a conclusion of non-random location clustering. The square footage area attribute of Battalion 2 will be used in combination with the calls for the month of February 2007 for this analysis. The resultant z-score with associated data and confidence level will be recorded.
3. Determine the correct data files and layers, features, and attributes to use for analysis. Turn on the "Incident_Feb07" layer.
4. Create a definition query to limit the data where the field incl_type is between 700 and 745 inclusively with incident types 7131, 7351, and 7352 are included.
6. Open the Average Nearest Neighbor Tool and enter the following parameters: Input Feature Class: Incident_Feb07, Area: 520175356 square feet, Check the Generate Report Box.
7. Open the results window and record the following values: Observed Mean Distance, Expected Mean Distance, Nearest Neighbor Ratio, and Z-score. Then open the HTML Report File to review the graphic display and record the high confidence level result. Record the results.
8. Create a presentation map and save the map file.


Part Two:
1. Open the "Tutorial 8-2.mxd" file in ArcMap.
2. Turn on the on the "Calls For Service-Jan07" layer.
3. The Second Step of Data Analysis will be to determine if a Getis-Ord General G Tool analysis will identify clustering of high or low values associated with features over the entire study area. It will determine whether the values are clustered and the distance over which these values are considered to be significant. This analysis will be run multiple times using a the recommended 10 Distance Bands, and different distances, specifically the recommended 200-1200 feet in 200 feet increments, to see if location affects the clustering calculations. In addition, the geographic influence within the features can be determined to provide greater accuracy. The resultant z-score, G index, and confidence level will be recorded for each distance.
4. Determine the correct data files and layers, features, and attributes to use for analysis.
5. Zoom into an area in the Data Frame that exhibits visual grouping.
6 . Open the Calculate Distance Band form Neighbor Count Tool and enter the following parameters: Input Feature: Calls For Service-Jan 07, Neighbors: 7. Open the results box and note the minimum, average, and maximum distance. The average distance to find seven neighbors is around 1,000 feet.
7. Run the High/Low Clustering (Getis-Ord General G) Tool using the parameters of: Input Feature Class: Calls For Service-Feb 07, Input Field: FEE, Generate Report Box, Conceptualization of Spatial Relationships: Fixed Distance Band, Distance Band: 200. Open the HTML Report Filef or review and record the General G index and the Z-score for 200 feet.
8. Run the tool 5 more times and increase the band distance by 200 feet for each. Review and record the results.
9. Create a presentation map and save the file.


Part Three:
1. Open the "Tutorial 8-3.mxd" file in ArcMap.
2. Turn on the on the "Calls For Service-Jan07" layer.
3. The Third Step of Data Analysis will be to determine if a Ripleys K Function analysis will identify clustering values based on the distance between all features and graph the relationships. For analysis I will use the recommended 10 distance bands with a beginning distance of 200 feet and a distance increment of 100 feet, also 99 permutations to increase the confidence in the resulting calculations, a Weight Field, and Minimum Enclosing Rectangle. A new floating point field, Distance, will be added to the graph, and then a new graph will be calculated to provide a more accurate data comparison to determine if there are any specific peaks in the graph indicating clustering. The Distance attribute will be analyzed and recorded.
4. Determine the correct data files and layers, features, and attributes to use for analysis.
5 . Open the Multi-Distance Spatial Cluster Analysis (Ripleys K Function) Tool and enter the following parameters: Input Feature Class: Calls For Service-Jan 07, Number of Distance Bands: 10, Compute Confidence Envelope: 99, Check the Display Results Graphically box, Weight Field: FEE, Beginning Distance: 200, Distance Increment: 100. Boundary Connection Method: None, and Study Area Method: Minimum Enclosing Rectangle.
Open the results window to review the results.
6. A table named KFunctionJan07ConfEnv has also been created in the Table of Contents. Open the table and create a new floating point field named Difference and calculate the following value: [ObserveK]-[HiConfEnv]. Create a Graph with the type: Vertical Line, Y Field: Difference, and X Label Field: ExpectedK. Review the graph, add it to the Layout, then resize and move it into the Title Bar.
7. Create a presentation map and save the file.


Part Four:
1. Open the "Tutorial 8-4.mxd" file in ArcMap.
2. Turn on the on the "Calls For Service-Feb07" layer.
3. The Fourth Step of the Data Analysis will be to determine if a Morans 1 analysis will identify whether there is a tendency of the pattern of attribute values as related to feature locations
show a tendency to be clustered, random, or dispersed. A polygon based on a 200 foot grid will be joined with the call feature locations to make a data layer containing the output grid features and count of call locations as an attribute value. The Spatial Autocorrelation Tool will be run for this new layer using the Calls Count, the recommended 250-650 feet distance in increments of 50 feet, and Zones of Indifference. The resultant z-score with associated data and confidence level will be recorded for each distance for analysis.
All four processes will be analyzed separately and in combination to confirm the decision to accept a null hypothesis which supports a random pattern, or reject a null hypothesis which confirms a clustered pattern.
4. Determine the correct data files and layers, features, and attributes to use for analysis.
5. Turn on the "200 ft Grid" layer for the "Calls For Service-Feb07".
6. Create a spatial join data file by using the Join Data Tool to join the 200 ft Grid and "Calls for Service-Feb07" layers.
7. Remove all cells with zero value by a Definition Query to allow only COUNT values that exceed zero.
8. Open the Spatial AutoCorrelation Tool and run the tool multiple time by entering the following parameters: Input Feature Class: Calls for Service-Feb07, Input Field: Count_, Check the Generate Report box, Conceptualization of Spatial Relationships: Zone of Indifference, Distance Band: (from 250 ft to 600 ft in increments of 50 ft).
9. Open the results window and record the z-score for each distance band.
10. Create a presentation map and save the file.


Part Five:
Based on the results of Part Four when compared to the first three analyses, what is the decision? To reject or accept the Null Hypothesis?
What will the report and recommendations be to the Fort Worth Fire Department about future station locations?

Discussion

The most complex aspect of this exercise was repeatedly inputting the correct parameters when using the various spatial statistics tools, and then understanding the results. The K Function graph was a little difficult to understand because there were no real peaks, but the Difference attributes helped to explain the data better because it was closely related to the results of the other Tool analyze. The results for Exercise 8-4 showed the highest z-score of 3.703 at 550 Feet with a 99% positive clustering pattern (not random). The p-value is a very low 0.000213, which indicates it is very unlikely the spatial pattern is random. The data from the Getis-Ord General G analysis indicated a spatial clustering within the 500-600 Feet range with a 99% confidence level, and Ripleys K Function analysis also indicated a spatial clustering within the 500-600 Feet range. A null hypothesis should be rejected based on the data results.

Evaluation

The results for Exercise 8-4 showed the highest z-score of 3.703 at 550 Feet with a 99% positive clustering pattern (not random). The p-value is a very low 0.000213, which indicates it is very unlikely that the spatial pattern is random. The data from the Getis-Ord General G analysis indicated a spatial clustering within the 500-600 Feet range with a 99% confidence level, and Ripleys K Function analysis also indicated a spatial clustering within the 500-600 Feet range. A null hypothesis should be rejected based on the data results.