ADD-TREES:
AI-elevated Decision-support via Digital Twins for Restoring and Enhancing Ecosystem Services
Applications:
ADD-TREES is a pioneering Artificial Intelligence (AI) research project providing innovative technologies that will aid crucial decisions about land use changes, with a focus on initiatives to create new woodlands and forests. It aims to feed directly into the UK Government Environment Act 2021 target of planting half a million hectares of new trees to increase carbon capture and other greenhouse gases as part of the UK’s strategy to achieve Net Zero by 2050.
Through four interlinked work packages, researchers will develop user-bespoke digital twins and co-created decision support tools directly with and for policy makers and large landholders. Our engaged users include Defra, MoD, National Trust, Network Rail, Forestry England and the Woodland Trust.
Methods:
Developing bespoke and adaptive methods for automatic emulation based on deep Gaussian Processes for the non-stationary, spatio-temporal and non-Gaussian process-based model outputs critical for landscape decision problems faced by our users. These techniques must be deployed in supercomputing environments for training and enable users to obtain fast predictions on their own laptops.
Embed modern data fusion techniques into the Joint UK Land Environment Simulator (JULES) to enable climate and soil forcing data (temperature, rainfall, humidity, wind speed, atmospheric pressure and carbon) to be downscaled to arbitrary scales, so that local user-bespoke JULES simulations can inform decision makers of the carbon consequences of planting different species on their land under climate change, and to combine the data fusion enhanced JULES with pest, disease and drought mortality modelling in order to embed user-specific local tree-stock risk quantification, accounting for climate extremes, into decision support tools.
Embed UQ, in particular automatic calibration/data assimilation into agricultural models and farmer behaviour modelling. We will build the capacity to synthesise satellite data regularly and automatically on crop distribution and government data on rural payments to automatically inform parametrisations for farm revenue and costs, and farmer behaviours and risk attitudes in a changing climate.
To deliver an explorable space of target-compatible polices and planting for users to interact with, enabling them to search for policies that are palatable to a decision maker when considering their goals and preferences.
Bring together the AI-enhanced modelling and the automatic emulation for those models to co-design and generate a user-bespoke AI-enhanced decision support tools on an open-source R-Shiny interactive app. Tools shaped and scaled to the exact decision needs of any user.