Introduction

Recent developments in fast and scalable inverse modeling and data assimilation methods in hydrology

Recent developments in fast and scalable inverse modeling and data assimilation methods in hydrology

CP-2021-12
Recent developments in fast and scalable inverse modeling and data assimilation methods in hydrology

Ghorbanidehno, Hojat, Amalia Kokkinaki, Jonghyun Lee, and Eric Darve

Journal of Hydrology 591, 125266, https://doi.org/10.1016/j.jhydrol.2020.125266 (2020)

The last twenty years have brought significant advances in hydrology and hydrogeology, especially in the area of data availability and predictive modeling capabilities. Remote sensing, imaging technology and in situ continuous monitoring devices have allowed the collection of large and complex datasets. At the same time, the computational power, efficiency and parallelization capabilities of numerical models have been constantly increasing, making possible simulations at larger scales and finer resolutions. Harnessing these new capabilities can result in better characterization of complex heterogeneous systems through inverse modeling, and better accuracy in predicting the behavior of dynamic systems, through data assimilation. Keeping up with developments in data collection and simulation capabilities, significant research has been conducted on developing computationally efficient inverse modeling and data assimilation algorithms. These new algorithms need to be able to handle expensive linear algebra computations, excessive data storage needs, and large numbers of repeated numerical simulations. The algorithms that have been developed aim to achieve a reasonable trade-off between computational cost and estimation accuracy. Two of the most common techniques used to handle this trade-off are ensemble-based methods and eigenspectrum-based compression of covariances. Both of these approaches can be combined with fast linear algebra techniques, further improving their computational efficiency. This focused review discusses computationally efficient inverse modeling and data assimilation methods developed in the past two decades in the context of their predecessor algorithms, outlining their similarities and differences and the applications they have been used for. By presenting a conceptual map of approaches that have been used to reduce inverse modeling and data assimilation costs, we hope to increase the visibility and understanding of these computational tools, as they are becoming more available to the broader scientific community and a bigger part of data-driven decision making in hydrological and hydrogeological applications.