To be sustainable and profitable, future landscape decisions need to consider the various services provided by ecosystems (such as timber, food, and water), as well as their responses to climate change. For this reason, there is an ever increasing need to build land use decisions upon an understanding of the integrated water and biogeochemical cycles.
Process-based land surface models are an excellent tool for predicting carbon, water, and surface heat exchanges between the land and atmosphere. However, running simulations using these models can carry a high computational cost, which can limit their use in landscape decision-making. Even advanced supercomputers can take several days to run simulations at a resolution high enough to capture areas the size of individual farms.
The aim of this project was to design and build ‘emulators’ of the UK’s community land surface model, JULES, to help inform future evaluation of ecosystem services related to the carbon and water cycles. An emulator is a statistical model that helps facilitate fast predictions of the output of a computer model. Using emulators to develop fast approximate models is an increasingly common technique when existing approaches entail a high computational cost.
What is JULES?
JULES is the UK’s community land surface model. It is used in the Met Office Unified Model (UM) framework for applications ranging from climate change to weather forecasting. JULES predicts productivity of vegetation, carbon accumulation in vegetation and soils, soil water content, runoff, and emissions of greenhouse gases from vegetation and soils – meaning it can be used for evaluating the impacts of land use change and climate change on several ecosystem services.
JULES is also used for evaluating the global carbon cycle: for example, each year, JULES is used to calculate the net land sink of carbon and emissions due to land use change in the Global Carbon Project. However, in these simulations the spatial resolution (>50km) is inadequate for addressing landscape-scale issues.
Decisions on land use changes should incorporate longer timescale changes and uncertainty quantification, but the computational cost of running JULES at km-scale resolution is high. Building an emulator of JULES allows for more robust statistical calculation of changes in carbon stocks, the water cycle, and their uncertainty, and opens the way for better calibration of JULES.
The project team developed a statistical emulation framework for JULES that provides predictions at a high resolution. The team demonstrated the gains that can be acquired from this approach and have detailed their findings in a paper that is available as a preprint. As JULES is one of the most complex land surface models, this approach can potentially provide substantial gains for all types of land surface model. The project will serve as a building block in a larger project aimed at utilising emulators of more ecosystem services.
Anna Harper (Principal Investigator)
Richard Everson (Co-Investigator)
Peter Challenor (Co-Investigator)
Daniel Williamson (Co-Investigator)
1. Emulation of high-resolution land surface models with sparse Gaussian processes with application to JULES, August 2021.
|Link to the project page on the UKRI Gateway to Research: JULES EMulator of ecosystem services (JEM)|