In 2023, UTS and the NSW SES were awarded a $4.9M grant to accelerate the development of Network Sensing Technology and create an innovative and revolutionary platform to improve situational awareness for the NSW SES Command Centre, to enable improved and intelligent disaster response.
Through this project we proved our leading Network Sensing technology works in real time and is highly accurate. We also created a powerful Digital Twin for visualising water levels on the ground, with inbuilt highly accurate predictive intelligence with the ability to integrate real time, network sensing data. This Digital Twin provides street-level detail of flood water—information that the SES could not previously access.
Our groundbreaking Network Sensing and Digital Twin technology was showcased on 16 June LIVE on site at the UTS Rowers’ Club on the Parramatta River, demonstrating the applicability of the technology to the world outside the laboratory.
In close collaboration with the NSW SES, over a period of 18 months, the team conducted extensive field trials over hundreds of hours to validate the accuracy of the network sensing technology we had been pioneering in the lab and in the field for 7 years prior. The Digital Twin was also tailored to the needs of the NSW SES, who lack real-time, street-level data for flood monitoring and management.
Digital Twin and Predictive Intelligence
A Digital Twin (DT) is a digital copy of a physical object, process or system which helps to simulate real situations and their outcomes.
As part of the Smart Flood and Storm Intelligence Project with the NSW SES, the team at UTS developed a unique and powerful DT as part of the Flood and Storm Intelligence project to provide real-time water level and flood monitoring and prediction.
This AI-hydrodynamic hybrid framework integrates fine-scale hydrodynamic-based simulations with advanced deep learning techniques to achieve rapid, high-resolution, high-accuracy flood prediction and visualisation for next-level detail and accuracy.
The challenge with current flood monitoring and prediction tools
Currently, the government flood watch service bases their predictions on a conventional approach. These methods are limited by their slow response times.
They typically take several hours to complete a single flood event’s simulation, and can only provide high-level predictions.
People know the water is going to come, but they don’t know when, where exactly and how high. Properties and lives are lost because of this lack of both geographic and time information.
Why is this Flood and Storm Intelligence Digital Twin different?
This AI-empowered Flood Digital Twin employs advanced AI techniques and interprets large volumes of data for real-time flood prediction at high resolution and high accuracy.
The DT processes data from a wide variety of sources:
- River gauges
- BoM Rainfall forecasts
- 5G Sensing for initial ground conditions
- Digital Elevation Model data, DEM to understand geographic and structural variables
It then predicts water level maps, and finally displays them on an interactive GIS map.
- developed a hybrid approach, that learns from existing hydrodynamic expert knowledge (the classical model) with deep learning transformers (the AI part).
Our success
- The Smart Flood and Storm Intelligence DT developed with the NSW SES mapped the Wagga Wagga region along the Murrumbidgee River, an area of about 1,000 sq. km
- The calibrated Flood used previous flood study results and river gauge rating curves.
- Large volumes of simulation data were used to develop and train powerful AI models, which learned complex patterns to predict water levels at 1.3 billion grid points at a resolution of 5m x 5m, in seconds—thousands of times faster than most existing solutions.
- This sequential data is loaded to the front end and rendered on a map, in a matter of minutes (2.5 minutes).
- We successfully delivered this Flood Digital Twin.
- Our Flood DT is a proven game-changer for flood prediction in real time and at high resolution—down to street level. Both flood extent and water levels are predicted with over 90% accuracy.