Wildfires represent a significant natural disaster with the potential to inflict widespread damage on both ecosystems and property. In recent years, there has been a growing interest in leveraging deep learning (DL) techniques for predicting the spread of wildfires (WS). However, existing studies have predominantly employed combined features with uniform weighting, overlooking the varying temporal resolutions they can offer (hourly, daily, and constant). As such, this study proposes a hybrid multi-temporal convolutional neural network (CNN) model called FirePred to fill this knowledge gap. In particular, 177 wildfire events were utilized along with related environmental variables between the years 2002 and 2018 in British Columbia, Canada. In pursuit of optimizing the model's performance, an exhaustive exploration of parameter configurations and settings was conducted. This involved assessing diverse combinations of loss functions, padding sizes, batch sizes, and thresholds. Notably, this rigorous analysis yielded an exceptional F1-score of 94% utilizing the most effective parameter set. In addition, to examine the versatility of our proposed model, we conducted an assessment using a dataset encompassing 10 instances of wildfires that transpired in Alaska between 2016 and 2019, as well as a wildfire occurrence in Nova Scotia during 2023. The findings revealed that the performance of the model can be influenced by regional parameters. Finally, the implementation of an uncertainty protocol discovered that the edges of the wildfire contribute the most to the uncertainty.