Objective 4 - Strengths and Weaknesses
We assess the strengths and weaknesses of the wildfire risk model by reviewing our approaches in model design, as well as the types of consolidated data incorporated into our model. No model will predict future fire incidents and risks with 100% accuracy. Nevertheless, identifying these strengths and weakness can aid future research in model design and data needs to create more suitable models.
The most considerable strength of the model overall is its ability to incorporate fire susceptibility and vulnerability together to create the risk model. This is a useful metric, as it will help to show areas that are most important for increased fire management or additional safety precautions (Thompson et al., 2011). This risk model creates information that is more useful for fire managers than fire susceptibility or vulnerability models can produce alone through the consideration of affected human populations (Martinez et al., 2009; Thompson et al., 2011).
The model contains various weaknesses, many of which can be divided into issues involving the data that is available and uncertainties involving the model itself. Much of the data is at differing spatial resolutions, and the finer data, therefore, needed to be coarsened to match with the coarsest layer. As a result, much of the analysis ended up being only as accurate as the coarsest data layer. This resulted in a risk model that is coarser overall than most of the factors are on their own. The factors/variables in the model were assigned weights based on a pairwise comparison. This comparison method assigns weights based on logic, and therefore, contains a large amount of subjectivity. There were attempts to minimize this subjectivity by weighting factors based on literature, but the overall net result still contains room for error.
Susceptibility model accuracy assessment:
The overall accuracy of the susceptibility model for identifying areas where wildfires are most likely to occur can be measured using a quantitative process. This process will determine whether “high susceptibility” areas identified in the susceptibility model actually experienced high occurrences of past forest fires. This will be done by comparing the susceptibility model output to the actual wildfire occurrence data for the 2015 fire season, a data layer that has been published by Natural Resources Canada (Table 11: Data Needs Section). The output of the susceptibility model will be ranked based on relative fire susceptibility of the area (Table 9). These parameters for susceptibility classification will be used to reclassify the susceptibility model output layer. The data will be already placed on a scale of 0-100 based on the standardization that was completed in the susceptability analysis (Objective 2).
Table 9. Ranking of wildfire susceptibility based on susceptibility model output values
From there, the susceptibility layer will be reclassified to include cells with high wildfire susceptibility valued as 1, and all other cells valued as 0 (using “Reclassify Tool” in ArcGIS). This will produce an output map that consists solely of highly susceptible areas. A similar process will be done to produce the 2015 fire event layer, except this time setting all cells that represent fire events in 2015 valued as 2, and all other cells valued as 0 (using “Reclassify Tool” in ArcGIS). This area will be compared to the actual fire events that occurred in 2015 using the “Raster Calculator Tool” in ArcGIS (Eq 5).
Equation 5. Areas with high susceptibility and fire events combined.
Fire Susceptibility vs. Fire Model Layer= (2015 Fire Events layer) + (High susceptibility)
The output layer will have four different values which represent the four different consequences of the model’s predictions (Table 10).
Table 10. Meaning of output values from Equation 5.
From here, the count of cells properly predicted as high susceptibility will be compared to the number of total fire occurrence cells (Eq.6). This will produce a final accuracy value (as a percentage of properly identified cells) for the susceptibility MCE model.
Equation 6. Final accuracy rating of susceptibility model.
Accuracy Rating (%) = Fire occurrence and High Susceptibility ranking/ (Fire occurrence without a high susceptibility ranking + Fire occurrence and High Susceptibility ranking)* 100%