Digital twins of physical and human systems informed by real-time data, are becoming ubiquitous across weather forecasting, disaster preparedness, and urban planning, but researchers lack the tools to run these models efficiently, limiting progress. This project is adapting methods originally developed in weather forecasting to cope with the timescales, data streams, and complex nature of modelling in different scenarios to enable digital twinning to become a reality across a wide range of fields. These tools could change how we model evacuation for tsunamis, traffic flow in smart cities, or protect communities against flash flooding, and more.
Explaining the science
This project goes by the acronym RADDISH (Real-time Advanced Data assimilation for Digital Simulation of numerical twins on HPC).
How can we include streams of observations to modify modelling of events in real time? For instance, buoys and satellites can measure tsunami waves at some sparse locations at some times: how do we improve warnings of inundation based upon tsunami computer models at the coast according to such data? Another example is crowding of public places where computer models of people’s movements can make use of data collected by cameras to improve their evacuation scenarios in real time. The latest statistical approaches take into account the current state of the model, and feed the measurements into the model to modify its behaviour. These methods need to be computationally efficient and accurate and must scale up in order to be used in very realistic settings.
- Adapt a suite of existing agent-based models (that are being developed through existing Turing and aligned projects) so that they can leverage state-of-art GPU computer clusters and data assimilation (DA) techniques. Initially, simple (toy) agent-based models of people’s behaviour in a neighbourhood will be developed for initial DA implementation, before moving on to more complicated models that simulate realistic urban environments (i.e. more akin to digital twins).
- Couple the agent-based models with a physical tsunami model to simulate the impact of a physical emergency on the flow of individuals (the latter, e.g., reacting to the emergency by evacuating away from the tsunami).
- Develop new implementations of state-of-art DA code (ranging from high-dimensional particle filters to well-studied ensemble Kalman filters, or emulators, as appropriate) targeted at high performance computing to allow data to be assimilated into the models in real-time. New DA machinery that has been designed for human-environmental systems and operates in real-time will be invaluable for scientists who are attempting to simulate human-environmental systems.
The team will also be running stakeholder workshops throughout to ensure that the tools developed reflect the needs and uses of the target user community.
Our models are of people in cities (e.g. migrations, evacuations) and tsunamis. They are used for purely scientific purposes (i.e. investigating hazards and individual responses to them) and for studies of real areas in order to assess resilience of infrastructure and to design evacuation policies.
Projects applying the tools developed by RADDISH to real-world scenarios such as evacuations for tsunamis in India and Indonesia have already been funded.
From a wider perspective, the tools developed by this project could also be applied to modelling of polices resources during riots, flood warning systems in the UK, and modelling of smart cities.
RADDISH team members presented at the workshop on “Data Assimilation and Uncertainty Quantification at the exascale”
RADDISH team members join the Met Office Academic Partnership
Find out more
To find out more about the project, visit the GitHub project page.
Please contact the project organisers via GitHub.