Urban complexity and digital twins: leveraging machine learning for hyper-local heat stress forecasting

Climate change is increasing the number, severity, and frequency of extreme weather events, such as heatwaves, which kill thousands of people worldwide and strain healthcare systems, particularly in cities. Those responsible for managing the societal consequences of these events require easy and uniform access to the best data, models, and decision support tools in order to mitigate impacts on affected populations and enable appropriate responses. To achieve this, spatially and temporally accurate information on the occurrence, location, and duration of heatwaves is required.

A digital replica of the Earth system, a foundational component to a digital twin (DT), can provide such information by producing bespoke cutting-edge numerical simulations of upcoming heatwaves and combining several components of the value chain – from observations to modelling, to answering strategic and socioeconomic questions – in a single workflow. To better understand how heatwaves may affect a city across different neighbourhoods, however, km-resolution DT forecasts must be zoomed in at the metre scale as events unfold. This necessitates taking into account urban complexity, which can be achieved by combining operational weather forecasts with quality-controlled station observations and land-use data. By using such an approach, machine learning (ML) has shown to enhance forecasts in urban areas to much higher resolutions than standard operational forecasts. Yet, the use of ML to forecast downscaling has so far involved offline, city-specific applications with limited scalability and testing in real-world settings.

The US-EU AI for the Public Good partnership seeks to address these limitations. As part of the partnership, ML solutions are being implemented to deliver hyper-local data from the European Commission's Destination Earth (DestinE) initiative and operationally generate forecast maps of heat in terms of thermophysiological stress. This presentation will provide an overview, plans, and current status of the partnership.