More examples

Multi-level interventions for mitigation and adaptation

Approaches for climate response will involve adaptation and mitigation options for securing not only our food production capacity but also other aspects of the environment to protect our existing carbon stocks and reduce carbon emissions. By taking a data-centric approach, it is possible to integrate long-term time-series data with causal knowledge of physical and socio-economic influences. Multi-disciplinary integration has allowed AI researchers to build better models for predicting future climate conditions for urban farming, material development, peatland preservation and assessing interventions for climate action.

A striking example of adaptation from the ASG repertoire is the Growing Underground Farm project led in collaboration with researchers in the University of Cambridge, the British Geological Survey and Imperial College London. As wireless sensor technology, large cloud databases, and computer processing power become more available, digital twins are becoming attractive for integrating agriculture into built environments. ‘Growing Underground Farm’ is utilising digital twin for building a template for space-efficient local farms ensuring the production of greens all year round, while ensuring minimal environmental impact. The world’s first underground farm has been established 33 metres below the pavements of London’s Clapham High Street in London. A former World War II air raid shelter, proposed but never joined the tunnels to the London Underground system post-war, is now stacked with racks of fresh green leaves and microgreens thriving under banks of LED lights. CROP, the digital twin of this farm, represents the reality of the environment through real-time streams of data and provides feedback for optimal management of the ‘twinned’ object. This includes three crucial elements. First, data from an extensive and robust monitoring system that tracks the observable environmental conditions in the underground farm is supported by data curation to ensure the quality and tractability of data. Second, analysing observable data along with information collected by farm operators helps identify key parameters of the farm environment and thereby crop yield. Third, using data modelling techniques to identify critical trends and changes, forecast potential future operational scenarios, and provide feedback on the influence of recent events on the farm environment. CROP’s cloud-based web application provides a 3D model of the farm with links to get live data streams from sensors within the farm. A database stores incoming data from the sensors and the associated models providing access to visualisations of current and historic conditions in the farm and crop growth data. Subsequent versions of CROP included the implementation of a temperature forecasting model, a scenario evaluation tool and the incorporation of more detailed crop monitoring with yield data, giving insight into improving farm efficiency and crop yield in a built environment (Ward et al., 2020).

In addition to building synthetic environments, knowledge from natural and living systems can be applied in creating adaptation options that improve our ability to engineer new materials that meet advanced functional and sustainability requirements. Molecular simulation investigation inspires biomimicry approaches that transfer knowledge from biological systems in human designs to create bio-inspired energy-efficient materials that don’t add to greenhouse gas emissions. AI can speed up the search for alternative or new materials by sifting through millions of potential molecules or biological states and identifying candidates for lab-based testing and development. Researchers within the ‘molecular structure from images under physical constraints’ project have built a deep learning-based framework, affinity-VAE, where VAE stands for Variational AutoEncoder, a class of CNN used to reduce the dimensionality of 3D volume data while encoding it into a low-dimensional latent representation. Affinity-VAE uses Convolutional Neural Networks (CNN) to determine the probability of a given class member in high throughput imaging data for reconstructing the molecular structure. This approach can automatically cluster and classify objects in multidimensional (2D and 3D) image data, including simulated biological electron cryo-tomography (cryo-ET) data based on their similarity. Affinity-VAE shows potential for the discovery of new candidates in experimental data that can contribute to designing biomimetic materials with environmental-friendly properties and functions. (Mirecka et al., 2022).

Mitigation responses need to focus on high-risk natural resources such as peatlands that store more carbon than all other vegetation types in the world combined. Peatlands are vulnerable to rapid climate change and disturbances such as wildfire and drainage, with the risk that their huge carbon stocks could be released into the atmosphere, further accelerating global warming. To counteract this threat, government, environmental organisations and industry are investing in climate actions to protect undisturbed peatlands and to restore those that have been damaged by human activities. An ASG project on understanding the risk and uncertainty in peatland carbon emissions is drawing information on various qualities of peatlands and their carbon emissions to scope a high-level Bayesian network. In collaboration with UKCEH and the Queens Mary University of London, the research team works to sketch out government policy, specific management interventions and natural mechanisms of resilience that are linked to peatland carbon emissions via causal relationships. This causal knowledge is built on numerical modelling, a scientific understanding of peatland functioning and expert advice on management activities and socio-economic factors. The project is developing a framework to integrate different types of information on peatlands: quantitative data on greenhouse gas fluxes and environmental drivers, such as temperature and precipitation; semi-quantitative understanding of controls and interactions, such as feedback between peat formation and water loss; and qualitative information on socio-economic factors, such as land-use policy and site-level management actions. Part of this research assimilates long-term records of carbon emissions from two Scottish peatlands into a process-based model enabling the identification of key environmental drivers and quantifying uncertainties on peatland carbon emissions. The other part of the work develops a causal framework that links carbon emissions predicted by the process-based model to the broader ecological and hydrological functioning of peatlands and human interventions. Predictive models developed in this project can accurately predict future peatland carbon emissions and the benefits of specific policies and interventions that drive policy and management of peatlands. <citation needed>

Another nature-based mitigation project comes from the Data Science for Climate Resilience in East Africa project carried out in collaboration with the International Small Group and Tree Planting Program (TIST) and the University of Exeter. The project uses the cloud-based Google Earth’s catalogue of satellite imagery and geospatial datasets to identify indicators of plant productivity and drought. The specific satellite data for this project is taken from LandSat-7 which gives a low-resolution but long-term survey of the areas of interest, primarily East Africa and Sentinel which gives high-resolution images for directly studying individual farms. By combining these indices with the detailed organisational data from TIST, models are being generated to infer tree cover and study landscape-scale changes occurring due to TIST and Kenyan Northern Rangelands Trust conservation practices over their histories of community-led tree planting activities. The satellite data analysis combined with forest cover change datasets shows that community-led activities are observable as increasing greening trends with clear secondary effects on the landscape, such as overspilling or an increase in forest cover near TIST groves. The high temporal resolution of the satellite data is allowing the development of novel tools for assessing the impacts of tree planting and conservation farming, using indicators of plant productivity, vegetation structure and drought indices. These computational methods potentially provide cheap and effective ways to monitor progress in tree growth, resilience, and conservation as well as enhance field surveys and land management. TIST regularly seeds an area by bringing existing members to speak with locals about the benefits of the program that have a significant social and environmental impact. Improved and transparent metrics for assessing the success of their activities have a direct value to tree planting and conservation organisations, supporting their growth and making the benefits of participation clear to members and funders. Addressing some primary concerns regarding drought and soil erosion, the project increases an understanding of the impact of tree planting programmes and engages members within the growing organisation sustainably. An understanding of the social factors at play: for example how information spreads, how the diversity or gender make-up of the engaging groups affects their success, and how the organisation grows, giving invaluable lessons for the future of similar schemes in different countries (Buxton et al., 2021).

AI-assisted mitigation and adaptation approaches from a variety of solutions can be transferred to other ecosystems, building a better understanding of interactions between environmental, social, biological and economic processes and extending the potential to transform predictions of climate change and intervention options.