Navigating Uncertainty in Climate Futures: Using New Matilda R Package to Explore Uncertainty in Climate Projections
Matilda v1.0 gives scientists a powerful analytical workflow to investigate uncertainty propagation in climate change projections from reduced complexity climate models (RCMs).
Erik Mclean on Pexels.com
One primary scientific goal is to understand the interactions between humans and the Earth system. We can use mathematical models to help predict future global changes based on decisions humans make. Some models are deterministic and provide a singular outcome. This type of modeling overlooks a key factor: uncertainty. Scientists tackle this problem by developing probabilistic models. Probabilistic models consider many possible outcomes based on inherent uncertainty of model parameters and can be achieved using Bayesian statistical methodologies. Embracing model uncertainty offers a clearer vision of possible futures for the Earth system. This approach helps to provide informed analysis to understand the effects of global change.
This R package gives scientists a powerful analytical workflow to investigate uncertainty propagation in climate change projections from reduced complexity climate models (RCMs). By propagating parameter uncertainty and applying likelihood-based model weighting techniques to analyze the realism of RCM ensemble members, researchers can explore diverse scenarios of Earth's future changes within a robust Bayesian framework. This not only enhances our understanding of Earth system dynamics but also enables more nuanced analyses of climate interactions. Ultimately, this R package can be used to analyze how various human-Earth system interactions can lead to global climate change.
Matilda R package provides the user with a turn-key mechanism for incorporating parameter uncertainty into climate model projections using the Hector RCM. The package uses prior knowledge from the literature to define parametric probability distribution functions (PDFs) for six key model parameters that drive Hector’s carbon cycle. Matilda explores the uncertainty space of model parameters and creates a perturbed parameter ensemble (PPE) by iteratively running Hector with sampled parameter sets. The resulting PPE members can be weighted by computing member likelihood against historical observations, or ‘criteria’. Using a Bayesian framework, the semi-informed PDFs of ensemble members are updated with likelihood weights to produce a posterior PDF. The posterior ensemble can then be used to define PDFs of future climate change metrics informed by historic observations as well as to compute probabilistic results for future climate outcomes. The package offers significant flexibility in selecting weighting criteria, including historical temperature, atmospheric CO2 concentrations, and ocean carbon uptake, or users can create custom weighting criteria. The open-source design makes Matilda accessible to the broad R user community and aims to limit user-programming responsibility.