Forest Fire Detection System Based on Convolutional Neural Network Using MODIS Satellite Data in Indonesia

Forest and land fires are a severe problem in Indonesia, with widespread impacts on the environment, public health, and economy. Early detection is crucial in fire prevention and mitigation efforts to minimize losses. This study proposes a convolutional neural network (CNN)-based forest fire detection system utilizing moderate resolution imaging spectroradiometer (MODIS) satellite data. MODIS imagery is used because of its wide spatial coverage, high temporal resolution, and consistent availability across Indonesia. The research methodology includes extracting hotspot data from MODIS imagery, preprocessing the data to improve their quality, and labelling images based on actual fire events. Furthermore, a CNN is used to classify images into fire and non-fire categories. The CNN model is trained on the preprocessed data, including historical data, and validated using ground truth from official reports on forest fires in Indonesia. The results demonstrate that our CNN-based detection system achieves high accuracy >90% and superior hotspot sensitivity compared with conventional methods. The proposed system improves the effectiveness of early warnings, supports decision making by authorities, and contributes to sustainable forest fire mitigation efforts in Indonesia. Future work will aim to achieve high fire hotspot detection accuracy.