During data preprocessing, explanatory datasets that did not have measurements for the entirety of the study area were assigned values of zero in place of NoData. This was required to complete the study overall as points with NoData values cannot be used in a regression analysis. It is recognized that assigned zero to cells lacking measurement records may misrepresent the input data and therefore increase overall model uncertainty. Had this step not been completed however, the vast majority of data points would have been deemed invalid for any type of regression. Given the high volume and density of data, the multivariate nature of the regression, and the steps to achieve randomization, it is suspected that biases from assuming NoData to zero are minimal. Nonetheless, it is necessary to recognize the aforementioned assumptions as a cause of elevated uncertainty.  

Additionally, this study originally intended on using the full 200 thousand point fire dataset however running the GWR models proved unfeasible with the available computer hardware specifications. All the steps outlined in the methodology section were repeated using 100 and 50 thousand point dataset, however the GWR analysis would often resulting in memory errors. These processing issues are well documented in Li and Fotherinham (2020).