Develop a GIS Model to Identify Areas of Greatest Susceptibility to Heat-Related Illness

The GIS model developed for this research identifies both individual census tracts and multiple-tract clusters where susceptibility to heat-related illness is greatest. The model runs in three segments. First, LiDAR and imagery data are used to produce a municipal tree-cover shapefile for the City of Montreal. A LiDAR point cloud classification algorithm classifies point cluster patterns indicative of high vegetation. Figure 3 shows the raw LiDAR point cloud with intensity values. Figure 4 shows the same point cloud after classification. For the City of Montreal, 684 point cloud tiles like the one in Figure 3 were classified.

Figure 3. Unclassified LiDAR point cloud showing intensity values 

Figure 4. Classified LiDAR point cloud with high vegetation in green, buildings in yellow, and ground points in blue. 

A tree canopy footprint layer is derived from the classified point cloud.  Tree-cover is aggregated to the Montreal census tracts, and a choropleth displays percent tree-cover in each tract. The 25% of tracts with least tree-cover are assigned a tree-cover vulnerability index value of 1, for use in determining the highest-priority tracts; all others tracts receive a tree-cover vulnerability value of 0. Tree-cover vulnerability index is stored as a new field in the Montreal Census Tracts attribute table.

Concurrently, census data relating to heat vulnerability factors 1-7 are joined to the Montreal census tracts. An indexing system is applied to all 7 socioeconomic vulnerability variables. For each vulnerability factor, the top 25% of census tracts with the highest vulnerability scores are identified as single-variable priority tracts and assigned a single-vulnerability value of 1; those not within the 25% most vulnerable are assigned a value of 0. Each single-variable vulnerability index is stored as a new field within the Montreal Census Tracts attribute table. This attribute table also stores the tree-cover vulnerability index.

Additionally, a proximity analysis determines the tract-average distance to the nearest publicly accessible cooling center. Point data for 805 cooling centers are joined into a single shape-file. The Euclidean distance tool derives proximity to the nearest cooling center point for 1m cells. Proximity raster cells are converted to points, and spatially joined to the census tract polygons, assigning an average distance value to the tract. The 25% of census tracts with the greatest average distance to cooling centers are assigned a proximity-vulnerability value of 1; all other tracts are assigned a proximity-vulnerability value of 0. The proximity-vulnerability index is joined as a new field to the Montreal Census Tract attribute table.
           

Finally, Equation 1 relates tree-cover and socioeconomic vulnerability factors, and assigns all tracts an indexed combined-vulnerability value.

Final output is a heat vulnerability choropleth, which shows the highest priority tracts based on the indexed multi-variable vulnerability scores. This choropleth map uses a vulnerability scale of 0-1, with the municipal census tract as the spatial unit. A hot-spot analysis using a fixed distance band spatial relationship with Euclidian distances calculates the Getis-Ord Gi* statistics for the final heat vulnerability layer. This analysis indicates where there are high- or low-value spatial clusters, and locates areas of most significant vulnerability.