Ecosystems, human lives and air quality, as well as the integrity and stability of ecosystems continue to be threatened by wildfires, requiring the need for robust and real-time detection systems. Conventional airborne patrols or surface sensors are restrained by coverage limitations, reaction delays, maintenance problems and exorbitant costs, which are not suitable for real-time wildfire detection. The recent technologies from satellite-based remote sensing, with the help of deep learning models, offer promising solutions. This paper is a comprehensive literature review of previous works on wildfire detection methods that utilize satellite images and advanced machine learning methods such as CNNs, RNNs and the fusion of them. It further investigates emerging technologies like multi- or hyper-spectral images, anomaly analysis using infrared radiation, spatio-temporal data fusion methods, as well as the use of unmanned aerial vehicles (UAVs) to complement satellite data. The application of the fusion of geostationary with polar-orbiting satellite data to enhance temporal sampling as well as location accuracy is also discussed. The use of ensemble learning techniques as well as AI-based anomaly discovery methods is also discussed as being central in enhancing robustness as well as minimizing false alarms. It finally outlines important challenges like the interference caused by cloud coverage, the domain limitations in spatial measures, real-time processing bottlenecks in the data, as well as generalizability to geographic areas. It concludes with research directions, prescribing scalable, interpretable, as well as multi-modal AI schemes towards operational wildfire monitoring and management.