More examples

Engagement-led adoption, extension and deployment of AI tools

Academic engagement with the public and governmental sectors will continue to be a central component of data science and AI development. These collaborations accelerate access to large datasets from various domains and of different resolutions required to build AI tools and train models that are widely applicable. A crucial yet missing step that many stakeholders, such as in local councils, public health and grassroots community initiatives, need to take before using data for their purposes is localising data for their areas of applications. Currently, UKCP18 Met Office climate projections data exists at multiple temporal-spatial resolutions and Representative Concentration Pathways (RCPs). Additionally, climate projections are not just available for the future, but also past and present. Re-calibration at local levels is required to bias-adjust data in these different periods using on-the-ground weather measurements. This recalibration process is currently performed ad-hoc by climate scientists for specific areas and duration, rather than at scale due to computational requirements for manipulating and processing large datasets and the domain expertise needed to apply sophisticated downscaling techniques. As such, when it has been performed, datasets, and knowledge of the implementation of methods, tend to be restricted to small user groups and shared directly within that network rather than from a single source of truth that all can directly access. This leads to issues with the reproducibility of findings, transparency of methods, and lack of meaningful comparison, or more generally the ability to perform comparisons, of the range of bias adjustment methods used for calibrating projections with local climate data. The CLIM-RECAL project addresses these classic cross-domain data challenges. Developed by DyME researchers, CLIM-RECAL integrates data from a variety of different online and offline places, aligns their scales and standards across domains, and creates understandable and accessible documentation for non-domain expert users. Intending to save a large chunk of research time for anyone working with UKCP18 climate projections data, this has been a rigorous, labour-intensive and extremely necessary undertaking by the researchers in ASG. Taking the learnings from the initial RAMP/DyME work to SPC/ASPICS and now specific applications of DyME simulation models to air quality and extreme heat, the DyME team have evidenced the importance of integrating existing datasets, and are enabling future adoption through CLIM-RECAL. Access to high-quality, directly usable, climate data will allow more people to learn about environmental research, and different stakeholders working in climate change spaces to achieve more.

The Turing has continued to facilitate extensive interdisciplinary exchange, including with government and local authorities, in environmental research. Researchers at the University of Exeter are working with stakeholders within local councils to build an app to help councils plan and mitigate the impact of climate change. The Local Climate Adaptation Tool (LCAT), though initially developed for Cornwall, has now extended to include other local government partners across the UK. Based on a qualitative evidence review, LCAT builds health impact pathways and displays relevant data of interest to stakeholders. The LCAT tool developers work with DyME-CHH to source the appropriate climate data for their analysis. Future iterations of the app will include outputs from DyME-CHH and the related Turing project to provide appropriately adjusted climate data at high granularity, as well as estimates of future population demographics. These integrations from DyME-CHH to the app were identified in collaboration with the Exeter group while working on different projects, which would have been missed if not for collaborative meetings and events. Furthermore, DyME-CHH outputs can now leverage the extensive network built by LCAT to disseminate and ensure findings, which are integrated into management decisions. The projected demographic output DyME-CHH was of particular interest to collaborators and will provide a new application of SPC beyond its initial application in health & environment, amassing software engineering expertise. This is also an opportunity to co-produce documentation that is understandable to stakeholders in social, environmental and health research, ensuring the work is disseminated to its maximum potential.

Another example is EnergyFlex developed under the ASG project for simulating energy efficiency opportunities for households. Energy flexibility is key to delivering a reliable, sustainable energy system, reducing pressure on energy production systems to continue using non-renewable energy sources (Energy Efficient Cities initiative). EnergyFlex is an agent-based microsimulation approach that models the energy performance of the housing stock by estimating when people in an urban area are likely to be using energy. Working with local authorities, EnergyFlex also generates synthetic housing stocks and provides a way to explore inequalities in energy efficiency, helping target households in more vulnerable socio-economic circumstances in need of retrofit. To address the challenges in residential decarbonisation in the UK, supported by the RAM team, a workshop with stakeholders from government and industry was convened to identify action plans, in addition to the possible avenues for using microsimulation models such as EnergyFlex. Since the workshop, several steps have been taken that include documentation, dissemination and engagement with stakeholders in the energy and housing domains to explore the use of the tool for strategic planning and policy-making. This project has involved a close collaboration with the Data & Analytics Facility for National Infrastructure (DAFNI), through which modules of code have been uploaded as standalone executable versions of the model which can be run in the cloud by users without any coding requirement. EnergyFlex models will also be pooled by DyME-CHH, which has already integrated the SPC and QUANT models.

Although it’s clear that mathematical modelling and simulation methods are pushing environmental research forward, researchers often lack the high throughput resources required to train and run those computationally intensive models. RADDISH, an ASG-funded project, with collaborators from University College London, provides a fast implementation of state-of-art data assimilation code targeted at High-Performance Computing (HPC). RADDISH, Real-time Advanced Data Assimilation for Digital Simulation of numerical twins on HPC, was initially developed to cope with the timescales, data streams, and complex nature of modelling in different scenarios for weather forecasting. Recognising the lack of tools and capacity to efficiently run simulation models, RADDISH provides an effective solution for the computational implementation of modelling and simulation in HPC. The tool has been rigorously optimised for forecasting tsunamis and modelling the evacuation of people and vessels in coastal areas (Multlevel Bayesian Quadrature Preprint).

Recent collaborations with the government and public stemming from the COVID-19 pandemic response have provided new opportunities to work across different fields. These modes of working are also helping find applications for existing technology and interdisciplinary approaches to address urgent societal issues related to climate change. Microsimulation techniques, as investigated in the Environment and Sustainability theme, improve our understanding of the impact of environmental policies and other interventions on a population through long-term projection. As already in practice, AI tools and app-based solutions will continue to assist local governments, individuals and communities in identifying ways to eliminate or minimise human-induced heat, air pollution and activities that add to the greenhouse gas emissions and damage our planetary ecosystem.