ADD-TREES: AI-elevated Decision-support via Digital Twins for Restoring and Enhancing Ecosystem Services

ADD-TREES：通过数字孪生提供人工智能提升的决策支持，以恢复和增强生态系统服务

• 批准号: EP/Y005597/1
• 负责人: Daniel Williamson
• 金额: $ 212.72万
• 依托单位: UNIVERSITY OF EXETER
• 依托单位国家: 英国
• 项目类别: Research Grant
• 财政年份: 2023
• 资助国家: 英国
• 起止时间: 2023 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=EP%2FY005597%2F1
• 关键词: ADD; TREES; AI; elevated; Decision

The UK's ambitions to achieve Net Zero by 2050 depend critically on large-scale greenhouse gas removal (GGR) that can offset emissions from difficult-to-decarbonise sectors. Capturing carbon in growing trees represents the only GGR technology that can be scaled up immediately and at relatively low cost to meet that requirement. As such, in the Environment Act (2021) the UK government committed to ambitious and legally-binding targets for the rapid expansion of UK woodland. Over the next few years, significant decisions must be made regarding where to plant half a million hectares of trees, decisions that will shape the UK countryside for generations to come. Deciding where to plant trees, which species to plant and when to plant them is complicated. How much GGR a particular woodland expansion strategy realises depends on a myriad of factors including how planting impacts on soil carbon stocks, how different tree species respond to spatially-varying environmental conditions under a changing climate and the vulnerability of planted trees to pests and disease. To complicate things further, in most cases new woodlands will be established on farmland. So planting comes at the cost of lost food production. That is important not only to landowners who are unlikely to consider planting trees unless compensated for lost farm income but also to policy makers who may have concerns over UK food security. Moreover, land use underpins a variety of important ecosystem services. Decisions over where to plant trees has significant implications for, amongst other things, flood mitigation, water quality, pollination, biodiversity and human health. The capacity to unravel that complexity and inform decision making, exists in the sophisticated science and socio-economic models developed by the academic community. Those models can simulate tree growth and GGR across the UK under climate change. They can estimate farm income changes from tree-planting and predict uptake of policy packages incentivising such land use change. They can even identify the impacts of tree planting on the flows of a whole array of ecosystems services. Unfortunately, these state-of-the-art models may take days to run and require expertise and specialist software that is simply not available to the diverse collection of policy makers and land managers engaged in tree-planting decisions. The central objective of this project is to bridge that gap, leveraging AI technologies to provide bespoke, AI-generated decision support tools that synthesise and present the information contained within state-of-the-art models in ways that can properly inform policy and planting decisions. Delivering this vision requires the embedding of existing AI technologies into the models themselves, allowing those models to be automatically scaled to the spatial and temporal resolution that best suits some particular decision problem. In addition, AI methods will be used to automatically build and link fast-running emulators of those scientific models. Powering decision-support tools with this AI-generated, fast-running modelling capacity will provide users with unprecedented capabilities to explore in real time tree-planting decisions and their numerous consequences. We will deliver co-designed tools to our project partners at Defra, National Trust, Forestry England, the Ministry of Defence, the National Forest Company, Network Rail and Woodland Trust. Moreover, our AI methods will ensure that this modelling technology is accessible to all landowners and policy makers engaged in tree-planting decisions. Configured through simple interfaces, the AI will assemble bespoke decision-support tools shaped and scaled to the exact decision needs of any user. Through this project the knowledge embedded in the science community's latest modelling and data will be transferred into the hands of the users that will shape the UK's Net Zero contribution from trees.

英国到2050年实现净零排放的雄心在很大程度上取决于大规模的温室气体清除（GGR），这可以抵消难以脱碳的行业的排放。在树木生长过程中捕获碳是唯一可以立即扩大规模、成本相对较低以满足这一要求的GGR技术。因此，在《环境法》（2021年）中，英国政府承诺为英国林地的快速扩张制定雄心勃勃的、具有法律约束力的目标。在接下来的几年里，必须做出重大决定，在哪里种植50万公顷的树木，这些决定将塑造英国农村的后代。决定在哪里种树、种什么树、什么时候种树是很复杂的。一种特定的林地扩张战略实现多少GGR取决于无数因素，包括种植对土壤碳储量的影响，不同树种如何应对气候变化下空间变化的环境条件，以及种植树木对病虫害的脆弱性。更复杂的是，在大多数情况下，新的林地将建立在农田之上。所以种植是以损失粮食产量为代价的。这不仅对土地所有者很重要，因为他们不太可能考虑种树，除非农业收入损失得到补偿，而且对可能担心英国食品安全的政策制定者也很重要。此外，土地利用支撑着各种重要的生态系统服务。在何处种树的决定，除其他外，对减轻洪水、水质、授粉、生物多样性和人类健康具有重大影响。揭开这种复杂性并为决策提供信息的能力存在于学术界开发的复杂科学和社会经济模型中。这些模型可以模拟气候变化下英国各地的树木生长和GGR。它们可以估计植树带来的农业收入变化，并预测鼓励这种土地利用变化的一揽子政策的实施情况。他们甚至可以确定植树对一系列生态系统服务流动的影响。不幸的是，这些最先进的模型可能需要几天的时间来运行，并且需要专业知识和专业软件，而参与植树决策的各种政策制定者和土地管理者根本无法获得这些知识和软件。该项目的核心目标是弥合这一差距，利用人工智能技术提供定制的、人工智能生成的决策支持工具，这些工具可以综合和呈现最先进模型中包含的信息，从而为政策和种植决策提供适当的信息。实现这一愿景需要将现有的人工智能技术嵌入到模型本身，允许这些模型自动缩放到最适合某些特定决策问题的空间和时间分辨率。此外，人工智能方法将用于自动构建和链接这些科学模型的快速运行模拟器。通过这种人工智能生成的快速运行建模能力，为决策支持工具提供动力，将为用户提供前所未有的能力，以实时探索植树决策及其众多后果。我们将向Defra、国民信托、英格兰林业、国防部、国家森林公司、网络铁路和林地信托的项目合作伙伴提供共同设计的工具。此外，我们的人工智能方法将确保所有参与植树决策的土地所有者和政策制定者都可以使用这种建模技术。通过简单的界面配置，人工智能将组装定制的决策支持工具，形成并扩展到任何用户的确切决策需求。通过这个项目，嵌入在科学界最新模型和数据中的知识将被转移到用户手中，这将塑造英国树木的净零贡献。

项目成果

Toward machine-assisted tuning avoiding the underestimation of uncertainty in climate change projections.

• DOI: 10.1126/sciadv.adf2758
• 发表时间: 2023-07-21
• 期刊: SCIENCE ADVANCES
• 影响因子: 13.6
• 作者: Hourdin, Frederic;Ferster, Brady;Deshayes, Julie;Mignot, Juliette;Musat, Ionela;Williamson, Daniel
• 通讯作者: Williamson, Daniel