Advancing E3SM Land Model Calibration Using AI-Based Uncertainty Quantification
Calibration and uncertainty quantification (UQ) of the E3SM Land Model (ELM) at each site are essential for advancing our predictive understanding of ecosystem responses to climate change. However, traditional UQ methods are often computationally intensive, limiting their practical use for site-specific model calibration. Our research tackles this challenge by introducing a novel UQ method that utilizes generative artificial intelligence (AI) techniques. This approach has successfully quantified parameter posterior uncertainty, achieving a 30-fold improvement in computational efficiency compared to traditional Markov Chain Monte Carlo (MCMC) sampling.
Our innovative method significantly lowers the computational costs associated with model calibration and UQ, thereby making site-specific parameter estimation globally feasible. This advancement enhances our ability to accurately calibrate the ELM to align with observational data, enabling more precise simulations of ecosystems and consequently improving our predictive understanding of the impacts of climate change on the ecosystem.
Land surface models are essential for simulating environmental processes and aiding climate-resilient decision-making. Traditionally, calibrating these models has been costly and time-consuming. To address this, we developed a new method called diffusion-based uncertainty quantification (DBUQ), which is faster and less memory-intensive than previous methods. Using supervised learning, DBUQ quickly generates samples to accurately approximate parameter posterior distributions. We tested this method on the Energy Exascale Earth System Model land model at the Missouri Ozark AmeriFlux forest site, finding that it can produce results similar to traditional methods but 30 times faster. This efficiency suggests that DBUQ could revolutionize the calibration of land surface models globally, enhancing our predictive understanding of climate impacts on ecosystems.