Igniting Precision: Amplifying Wildfire Prediction in Diverse Regions via Teacher-Student Model Fusion

Accurate wildfire prediction in diverse and geographically dispersed areas is crucial for effective wildfire management. However, the limited availability of labeled data in data-challenged regions, along with the unique characteristics of these areas, poses challenges for training robust prediction models. This study investigates the performance of a convolutional neural network (CNN) on datasets comprising Landsat images from Canada and Alaska. Through principal component analysis (PCA), the study uncovers distinct differences in data distribution between the two regions. It is observed that the reduced data size of the Alaskan dataset, along with its distinct data distribution, leads to a decrease in the CNN's accuracy to 75% compared to an impressive 98% achieved on the Canadian dataset. To address this limitation, we propose a teacher-student model approach, transferring knowledge from a CNN trained on the larger Canadian dataset. The results demonstrate a significant accuracy improvement to 88.96% on the Alaskan dataset. Our findings highlight the effectiveness of the teacherstudent model in mitigating data scarcity challenges, enhancing wildfire prediction capabilities in regions with limited training data. This research contributes to improved wildfire monitoring and prevention strategies in challenging geographical locations.