Introducing Astrape — Volume 3/3
Model Space Exploration + Uncertainty Quantification
In the previous parts of this series, we introduced Astrape, the world’s fastest and highest-resolution AEM inversion software, and demonstrated its remarkable processing capabilities enabled by GPU-based acceleration.
Now, in this third and final installment, we delve into how Astrape’s advanced computational power allows for sophisticated model space exploration and uncertainty quantification, transforming the approach to geophysical inversions and resource evaluation in mineral exploration.
Methodical Hyperparameter Search
One of the primary challenges in geophysical inversion is the selection of hyperparameters, which can significantly influence the outcome of the inversion process. These hyperparameters can include initial models and regularization parameters. Traditional methods often rely on manually selecting and varying these parameters, which introduces bias and limits the exploration of potential model solutions. Astrape addresses this limitation by enabling systematic hyperparameter search, allowing for a more comprehensive and unbiased exploration of the model space. Astrape allows users to define search spaces for hyperparameters, utilizing various search techniques such as logarithmic, uniform, and grid searches. Moreover, it supports state-of-the-art search and scheduling algorithms like Population-Based Training, early stopping based on comparisons with other runs, and Bayesian Optimization (BayesOpt). This flexibility empowers users to efficiently manage and organize multiple inversion experiments, systematically exploring the impact of different hyperparameter configurations.
Quantify Model Uncertainty
Inversion problems in geophysics are inherently underconstrained, meaning that multiple models can fit the observed data equally well. For example, a 1 km x 1 km section only has upwards of 40,000 data points, but the model has over 750,000 parameters to fit. Traditional inversion techniques typically yield a single model that fits the data well, but this model represents only one of many possible solutions. Such an approach provides a limited view of the subsurface and fails to capture the uncertainty associated with the inversion process.
Astrape takes a different approach by leveraging the models that reach the global minimum during the inversion process as starting points for further exploration. By varying regularization parameters or modifying the model directly, Astrape can push the inversion process into different regions of the model space. This method allows for the generation of a large and diverse sequence of models, all of which are consistent with the observed data.
This ensemble forms the basis for empirical uncertainty quantification. By analyzing variations among these models, Astrape provides a robust estimation of the uncertainties associated with the inversion results. This is particularly valuable in resource evaluation, where understanding the uncertainty in the model can guide critical decisions such as where to drill.
Resource Evaluation: Informed Decision-Making in Mineral Exploration
As the demand for critical minerals continues to rise, accurate resource evaluation becomes ever more important. By quantifying the uncertainty of the conductivity and volume of geologic targets, Astrape provides essential information that underpins economic decision-making in mineral exploration. The ability to quantify uncertainty allows exploration teams to better estimate the potential of a given area and to optimize drilling strategies. By identifying areas of higher uncertainty, exploration efforts can be focused on reducing these uncertainties, leading to more accurate resource estimates and more efficient exploration programs.
A New Era of Geophysical Inversion
Astrape represents a significant leap forward in geophysical inversion technology, offering unprecedented speed, scalability, and sophistication. By enabling systematic hyperparameter search and comprehensive uncertainty quantification, Astrape allows geophysicists to iterate faster and provides a more complete picture of the subsurface, empowering exploration teams to make better-informed decisions. As the field of geophysics continues to evolve, tools like Astrape will be instrumental in meeting the growing demands for critical minerals and advancing our understanding of the Earth’s subsurface.
This concludes our series on Astrape, showcasing its journey from rapid inversion processing to advanced model space exploration and uncertainty quantification. As we move forward, the insights and technologies developed with Astrape will continue to push the boundaries of what is possible in modern geophysics.
References
- Xiaolong Wei and Jiajia Sun, (2021), “Uncertainty analysis of 3D potential-field deterministic inversion using mixed Lp norms,” GEOPHYSICS 86: G133-G158. https://doi.org/10.1190/geo2020-0672.1
