Today, the rapid growth of artificial intelligence (AI) algorithms has enabled many practical applications. For example, generative AI can create text, images, and code from natural-language instructions. Consider recommendation algorithms that analyse user behaviour to suggest products, movies, or music. In particular, computer vision algorithms have become very powerful in identifying objects and faces in images, supporting applications such as autonomous vehicles and medical image analysis.
The development of computer algorithms has long been closely intertwined with advances in geophysical research. A classic example is parallel computing for seismic exploration, where large-scale numerical calculations are used to process seismic measurements and construct detailed images of the subsurface. As geophysical datasets have grown in size and complexity, this has driven the development of increasingly efficient algorithms for signal processing, wave propagation, inversion, and high- performance computing. In turn, advances in computer science have made it possible to tackle geophysical problems at increasingly larger spatial and temporal scales.
Another important connection is parameter estimation and inverse modelling. Geophysicists routinely need to infer properties of the Earth’s interior, such as seismic velocities, densities, or material parameters, from indirect observations. This has stimulated the development of sophisticated optimization algorithms, statistical methods, Bayesian inference, and machine-learning approaches. More broadly, many innovations in computational mathematics have emerged from the need to solve demanding geophysical problems, illustrating a longstanding relationship in which advances in computing enable new scientific discoveries, while challenging geophysical applications motivate the development of new algorithms.
Where I see a crucial benefit of AI is in Digital Twins of geophysical environments. These environments include the atmosphere, cryosphere, biosphere, and the subsurface solid Earth. These environments are all composed of multi-parameter domains with limited or indirect observations coupled by physical and mathematical relationships. There is always a certain amount of uncertainty in scientific research. For a Digital Twin to be useful, it must provide reliable information to support decisions about future development, economic growth, public engagement, hazard prediction, and process monitoring. Here, the development of AI techniques can help in several ways.
The Solid Earth is a complex system with many interrelated processes. Classical models of the structure and composition of the Earth’s crust and mantle tend to focus on one or two parameters, e.g. crustal structure and density, electrical conductivity and water content, or temperature and pressure. Increasingly, it becomes clear that we miss the complex and significant interplay among these physical parameters, in which composition plays a critical role. More data acquisition infrastructures are being deployed, both on the ground and in space, delivering tonnes of information about these parameters. Only by studying Earth as a holistic object and combining all this information can we advance our understanding of our home planet. This complexity demands specialised AI inference algorithms to determine and map unknown physical relationships and fully explore the deep Earth.
Classical uncertainty approaches struggle to map the uncertainty of a correlated, multi-parameter environment. Performing brute-force or Monte Carlo exploration of such a space becomes very costly when the number of parameters grows beyond 10. Here, sophisticated inference techniques can help us explore the uncertainty space of the physical model more consistently and faster. This would help us better judge the model’s quality and validity, and identify new regions of the parameter space that need exploration. In particular, correlations between parameters and their spatial, temporal, and spectral nature can deepen our understanding of our home planet.
The purpose of these Digital Twins is to be able to scan through numerous scenarios, where the models are initialised with real-world conditions and predict what a possible future would look like. They can predict when certain large events will happen or the probability of certain outcomes. With this information, informed decisions can be made. This means that these models need to be fast, resemble reality as closely as possible, and show the uncertainty propagation of their estimates. Fast, precise calculations are needed, and AI techniques can help. Smartly connecting learnable networks to complex physical interrelationships requires sophisticated algorithms that are transparent, precise, and verifiable. Only then will Digital Twins be of use to society in addressing geoscientific problems.
Satellite infrastructure provides a tremendous amount of new data each day. Orbiting satellites are constantly monitoring geo-related processes. This time-varying signal must be incorporated into complex models and Digital Twins of the geosphere. A seemingly rigid and static mathematical model needs to be constantly updated. It needs to adapt to this new data and learn how to cope with it effectively. A true Digital Twin should be updatable with the newest information. I imagine that this can only be achieved by connecting real-time data streams to the model. Real-time means that the algorithms behind the model need to be able to assess the data for correctness, and updating the model also needs to be done in real time. Batch updating will be too slow and could even be too computationally inefficient. A Digital Twin will be constantly learning, something that neural networks are designed for. I am convinced that in the coming decades, geophysics will move towards more probabilistic modelling while retaining deterministic models for testing and verification.
Our observational platforms become smarter, our models become more complex, and AI algorithms become necessary. This allows us to understand more complex processes in our environment. By learning about these processes, we can make better, more informed decisions that help us cooperate with the world around us, rather than pursue single-objective strategies that have proven disastrous. We should utilise these tools to improve our understanding of the world and our place in it.
License: CC BY-NC-ND 4.0
This work is licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International.
What does this mean?
Share – Copy and redistribute the material in any medium or format.
The licensor cannot revoke these freedoms as long as you follow the license terms.




Academic articles from a similar field of interest
Open Access Government produces compelling and informative news, publications, eBooks, and academic research articles for the public and private sector looking at health, diseases & conditions, workplace, research & innovation, digital transformation, government policy, environment, agriculture, energy, transport and more.
As a Crossref Sponsored Member we are able to connect your content with a global network of online scholarly research, currently over 20,000 other organizational members from 160 countries. Crossref drive metadata exchange and support nearly 2 billion monthly API queries, facilitating global research communication.
© Adjacent Digital Politics Ltd