Bushfire reporting has undergone a radical shift with the advent of social media. While "000" calls were once the norm, people have turned heavily to social media to post information about bushfires, and as a consequence are falling foul to the bystander effect; believing someone else has already reported the incident. Fire agencies need to increasingly rely on social media to enhance their situational awareness. By monitoring social media they can identify unreported incidents, assess public response, and gain insight into fire behaviour. But with billions of social media posts, how do agencies sift through to find meaningful public information?
Taking a human-centred service design approach, our solution improves past processes and reduces cognitive load. We created a machine learning (ML) based on social media keywords and hashtags; through a process of user research, prototyping, and continuous improvement. By providing an interface for updating the ML model parameters, we enable agencies to continuously adapt to social media trends. Computer vision and real-time inference technologies have been implemented to find pertinent imagery of fire location and behaviour, automatically attaching this data to relevant incidents. Athena’s Social Intelligence Dashboard is a world-first emergency services workflow providing highly relevant public information to firefighters.
The Black Summer bushfires of 2019-2020 killed 33 people, displaced or killed an estimated 3 billion animals, destroyed 3,000 homes and 16 million hectares of land, causing almost 10% of Australia's GDP in damages. Athena's Social Intelligence Dashboard helps prevent such catastrophic events from happening again. We streamlined a 2-hour, multi-system workflow into a single collaborative platform,now taking just 5-15 minutes. We’ve already alerted authorities to 2 unreported fires and provided ongoing intelligence for 43 more. Early detection and better understanding of fire behaviour, enhances response times, minimises fire spread, and reduces damages to infrastructure and the environment.
Significant effort has been placed to ensure the Dashboard is malleable and able to be viewed in multiple contexts. We achieved this by visualising social media spatially, on a global feed and tailored incident dashboards. The feed aggregates posts from various social media platforms into a single view, using ML to identify relevant content through location inference, keyword matching, and image recognition. The feed can be customised to each user, allowing them to focus on specific incidents or areas they are responsible for. All location-based posts are displayed geospatially, providing context with other data like surface temperature, fire spread predictions, people movement, and aerial imagery. Posts are then validated as intelligence and shared with the appropriate users involved in managing the related area. To ensure scalability and adaptability, we added a global filters setting page, allowing fire agencies to continuously update the parameters of the ML model within the platform. The Dashboard is underpinned by image recognition, machine learning, and computer vision technology. Social media posts featuring fire or smoke are identified through this solution. The model learns from a firefighters' interactions to become smarter over time; consequently, the more our users engage with the Dashboard, the smarter it becomes.