<p>Recent climate studies increasingly indicate that the impacts of climate change have been underestimated, with the coming decades expected to be characterised by natural disasters of growing scale, frequency, and severity. Larger and more severe events tend to develop more rapidly and over wider areas, often overwhelming existing social and institutional mitigation systems. This underscores the need for monitoring and modelling approaches capable of supporting real-time disaster detection while also capturing social dynamics relevant to emergency management and decision-making. During disaster events, social media users generate large volumes of heterogeneous and noisy data that convey subjective yet often highly localised information about evolving conditions. Disaster-related posts frequently include images and videos that encode additional visual evidence of hazard occurrence and severity. Social media platforms can therefore be viewed as digital sensing environments in which physical events are represented through user-generated content that can be mapped back onto the physical landscape. This paper contributes to the development of Human Sensor Networks (HSNs) for wildfire monitoring by integrating social media data with satellite-based observations. Events are represented in the digital environment through both textual and visual content shared online, which serve as signal readings for modelling wildfire activity in near real time. Machine learning methods are applied to extract wildfire-relevant information from social media imagery, forming a lightweight and flexible detection pipeline. To improve robustness and reduce uncertainty, predictions derived from human-sensor data are validated using Earth observation satellite imagery. The paper presents vision-based models for smoke and wildfire detection in both ground-level social media images and satellite-based contexts, demonstrating the value of multi-source data integration for accurate and timely wildfire monitoring.</p>