Source: Water Resources Research
Earth is a dynamic system composed of interconnected components, including the atmosphere, hydrosphere, lithosphere, and biosphere. Because of the inherent complexity of this system, computer models with dozens to hundreds of parameters are required to unravel the interactions that occur between these realms and to explore potential future change.
As Earth system models become more sophisticated, there is a growing need for increasingly efficient methods of conducting sensitivity analyses; such tools are used to evaluate the degree to which each input factor affects the model output.
But in a new study, Gupta and Razavi argue that the conventional performance metric–based methods currently used to evaluate model sensitivity are based upon flawed reasoning. The team revisited the theoretical basis of sensitivity analysis for dynamical Earth system models; after reviewing the fundamental principles, they developed a novel theoretical approach that offers an accurate assessment of parameter importance for both individual time steps and cumulative model runs.
The team then demonstrated the potential of this new approach with a case study that analyzes the sensitivity of a conceptual hydrological model response to changes in its 10 parameters when applied to the Oldman Basin in the Canadian Rockies. Their results indicate that the conventional performance metric–based and new, metric-free approaches differ substantially when ranking which parameters strongly influence model output.
These findings highlight the importance of readdressing the basis for sensitivity analyses to improve our understanding of the Earth system, as well as the models being used to represent it. Because this new approach is computationally efficient, the authors argue that it is broadly applicable, with historical conditions as well as predictive scenarios. (Water Resources Research, https://doi.org/10.1029/2018WR022668, 2018)
—Terri Cook, Freelance Writer
Cook, T. (2019), Reframing sensitivity analysis in Earth system models, Eos, 100, https://doi.org/10.1029/2019EO116217. Published on 21 February 2019.
Text © 2019. The authors. CC BY-NC-ND 3.0
Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.