Capability – Machine Learning and Artificial Intelligence
Efficient AI-based Wildfire forecasting and model parameter identification
Real-time forecasting of wildfire dynamics, which has raised increasing attention recently in fire safety science world-widely, is extremely challenging due to the complexities of the physical models and the geographical features. Running physics-based simulations for large-scale wildfires can be computationally difficult, if not infeasible. We propose a novel algorithm scheme, which combines reduced-order modelling (ROM), recurrent neural networks (RNN), data assimilation (DA) and error covariance tuning for real-time forecasting/monitoring of the burned area.
Real-time Modelling of Socio-physical Wildfire Dynamics
We aim to make use of the large amount of real-time social media data published online during events, by combining this with satellite & geophysical data to inform disaster models and create more socialized, integrated forecasts which inform and coordinate various wildfire management stakeholders over the course of a wildfire event. We treat connected individuals during events to be a network of noisy remote multimodal sensors with variable reliability and location, which respond to physical stimuli. The extent of this multimodality is facilitated by the comprehensiveness of Natural Language Processing methods implemented on real-time streams of posts from related users, creating a socialized abstraction of a traditional sensing network. This affords the modelling of social dynamics alongside physical wildfire activity, using machine learning methods to model and predict socio-physical wildfire metrics.
In brief, our capacities/activities related to machine learning and artificial intelligence can be summarized as:
- Building fast surrogate models for high-dimensional dynamical systems (e.g., simulation of fire propagation)
- Reduced-order-modelling/feature extractions for large physical fields
- Parameter identification for complex fire predictive models
- Real-time monitoring and modelling of online social channels for disaster detection and classification.
- Natural language processing of large amounts of textual social data and information extrapolation to predict wildfire activity
Who to Contact
If you have general queries about our capabilities and research in this area, please get in touch with our key contact point below.
Dr Sibo Cheng (ICL) Email: email@example.com
People working in this area
Dr Rossella Arcucci (lead) (Department of Earth Science and Engineering, ICL) – Lecturer in Data Science and Machine Learning – read more
Dr Sibo Cheng (ICL)- PDRA – Machine learning for wildfire forecasting accuracy
Jake Lever (DSI, ICL) – PhD student – Machine Learning and Data Assimilation for Accurate Wildfire Predictions and Estimations
The Centre is directed by a six-strong Leadership Team of a Director and Associate Directors, covering a wide range of areas of expertise that are crucial for the materialisation of our Centre: