Cluster Analysis

As an element of the FireMapper project, the fire history polygons of the Santa Monica Mountains study area were evaluated to determine their extent of spatial clustering. The clustering analysis was intended to determine if the fire locations were clustered, and also whether or not they were associated with street segments.  


First a subset of fire history polygons were selected from the California State database of wildland fires from 1887 to the present.  From that dataset, a point shapefile of the polygon centroids was developed. The point shapefile was then edited to eliminate all wildland fires over 1000 acres. From the remaining records, the point distances from the nearest street location was determined using the GEO Wizard toolset. 

With the above dataset, three methods of spatial clustering analysis were applied. First, the Average Nearest Neighbor method was used to determine if clustering was present within the dataset. (Figure 1) It was determined that there was significant clustering at greater than the .01 level, with a Z score of -3.36.  Second, the Morans I Spatial Autocorrelation tool was employed to evaluate the clustering of the fire polygons relative to the street segments within the study area. (Figure 2) This analysis determined that the locations were highly clustered, with a Z score of 5.18 and with less than a 1% likelihood of this pattern being a random occurrence. Finally, the dataset was rendered using the Getis-Ord Gi, Hot Spot Analysis tool within the ArcMap Spatial Statistics toolbox. (Figure 3)


Both methods of analysis show high levels of clustering. At the most general level the Average Nearest Neighbor Distance method determined that the locations are highly clustered, with a Z score of -3.36 and a P-value of less than .01. When the clustering was evaluated relative to the clustering associated with street segments, the clustering yielded a Z score of 5.18 and a  P-value of less than .01. The rendering of the cluster analysis, clearly illustrates the clustering pattern, with high levels of clustering from the foothills of the San Gabriels through Griffith Park, and continuing across the Santa Monica Mountains. The cluster analysis clearly shows that the location of wildland fires, in the study area, are highly clustered and are highly associated with the nearest street segment.

Figure: 1.

Average Nearest Neighbor Distance output

Figure: 2.

Spatial Autocorrelation (Morgan’s I) output

Figure: 3.

Hot Spot Analysis (Getis-Ord Gi) output