What Is the Role and Opportunity of Data Assimilation in the Earth System Digital Twin?

Many initiatives are currently underway to build digital twins of the Earth system, including NASA’s Earth System Digital Twins efforts and the European Union’s Destination Earth program. Digital twins bring together numerical models, artificial intelligence techniques and large amounts of data to build replicas of the Earth system or its components, allowing users to monitor and predict environmental changes. By linking environmental simulations to impact models, digital twins can quickly provide information to policymakers and stakeholders on the effect of proposed intervention strategies on quantities of interest.

In bringing the vast amounts of Earth observation data into digital twins, data assimilation has a fundamental role to play. First, in the building of digital twins, it is important to build good emulators from the best physical models available. Many models of Earth system components, for example land-surface models, rely on a good specification of many model parameters. Constraining these model parameters by data assimilation before emulation can ensure that the emulator is as accurate as possible. Further insights from data assimilation on treating observation errors and biases have the potential to be useful for observationally-based emulators.

Once the digital twin has been built, its use for real-time prediction or near-real-time impact studies requires a continuous update of the model system by assimilating the latest observations. For many proposed uses of digital twins, such as predicting the energy output of wind farms, these observations may be very high resolution in time and space, providing a challenge as to how to extract the most information from the observations in an efficient way. Furthermore, where components of the Earth system are coupled in a digital twin (e.g. linking atmospheric models to hydrology), care must be taken to link the different temporal and spatial information from observations appropriately. Similar challenges are currently being addressed in data assimilation for standard numerical weather prediction. We will highlight where current research in data assimilation could be informative for the design of digital twins and discuss potential avenues for further research.