Evaluate the Strengths, Weaknesses, and Accuracy of the Model
The model’s tree coverage calculation is compared to a canopy cover shapefile produced by the Montreal Infrastructure, Roads and Transportation Service for 2015, and assessed for accuracy. The two shapefiles are overlaid and compared for major differences in area coverage. The official canopy cover shapefile is derived from 2015 LiDAR data, likely the same used in this analysis, making it a very strong tool for accuracy evaluation. The comparison of the canopy cover created through this analysis and the pre-existing canopy cover shows if the methods used in this analysis are accurate or if there exists uncertainty in the model. The total model accuracy is found using a ground truthing method, a common practice in verifying remote sensing data (He et al., 2013). Census tracts will be selected from multiple areas with socioeconomic characteristics of both high heat-vulnerability and low-vulnerability. Using satellite imagery and Google Earth Street View, a visual comparison will be drawn between the modelled raster and the reference areas.
The LiDAR analysis is not without limitations. Tree canopy cover can be difficult to measure, due in part to the natural variance of tree canopies. Trees near buildings may be exluded due to misclassification errors. While tree canopy cover is being used as a metric for cooling, other geographic features that affect temperature are not being taken into effect such as shade caused by tall buildings. Additionally, lower point density in the LiDAR dataset results in less accurate tree canopy cover estimations.
The model's primary strength is that it is easily applied to other areas for similar studies. All data is from open sources that are regularly available to city planners and the MCE weighting method is comprehensive and reproducible. Using an index scoring method in the MCE also makes the model easy to interpret. The use of a 25% threshold for identifying the most at-risk areas is easily understood even by those who have limited background with spatial analysis tools. Another model strength is that it produces visual representation of areas most at risk using a large variety of criteria, greatly aiding in decision-making for planners and for when they are presenting projects to the public. The final map produced by the model should clearly display the different areas of high and minimal risk.
The scale at which the model produces results is a weakness and an area that will require greater development in the future. As the socioeconomic data included in this model are available only at the census tract level it is not possible to make inferences regarding the need for increased tree cover at very fine levels, such as which specific city blocks are most vulnerable to heat illness from a socioeconomic standpoint. The data available are also from different time periods which may lessen the strength of the results. Finally, this analysis requires LiDAR data of the study area, which is not always a resource available to smaller municipalities due to budgetary constraints.