Earth System Modeling 2.0: A Blueprint for Models That Learn From Observations and Targeted High-Resolution Simulations (1709.00037v3)

Abstract: Climate projections continue to be marred by large uncertainties, which originate in processes that need to be parameterized, such as clouds, convection, and ecosystems. But rapid progress is now within reach. New computational tools and methods from data assimilation and machine learning make it possible to integrate global observations and local high-resolution simulations in an Earth system model (ESM) that systematically learns from both. Here we propose a blueprint for such an ESM. We outline how parameterization schemes can learn from global observations and targeted high-resolution simulations, for example, of clouds and convection, through matching low-order statistics between ESMs, observations, and high-resolution simulations. We illustrate learning algorithms for ESMs with a simple dynamical system that shares characteristics of the climate system; and we discuss the opportunities the proposed framework presents and the challenges that remain to realize it.

• The paper proposes a novel framework that synergizes global satellite observations and targeted high-resolution simulations to refine model parameterizations.
• The paper demonstrates that ensemble Kalman inversion and MCMC methods effectively reduce uncertainties in climate projections by optimizing model parameters.
• The paper shows that integrating observational data with learning algorithms enhances model adaptability and predictive accuracy for future climate conditions.

Overview of Earth System Modeling 2.0: A Blueprint for Models That Learn From Observations and Targeted High-Resolution Simulations

This paper proposes a novel framework for improving Earth System Models (ESMs) through the integration of data assimilation and machine learning techniques, aimed at learning from both global observations and targeted high-resolution simulations. The authors outline a method for ESMs to systematically learn from observational data—both from global-scale satellite observations and localized high-resolution simulations—thereby improving parameterization schemes, reducing uncertainties in climate projections, and enhancing predictive capabilities.

Key Proposals

1. Utilization of Global Observations: The paper highlights the potential of leveraging extensive global datasets, particularly from satellite observations, which provide comprehensive information about atmospheric and Earth system variables, such as temperature, humidity, and cloud cover. These datasets can be used to refine parameterization schemes by matching low-order statistics between ESMs and observations.
2. Incorporation of High-Resolution Simulations: Recognizing computational constraints that limit the global implementation of high-resolution models, the authors propose targeted use of high-fidelity local simulations, such as nested simulations within selected grid columns of ESMs. These simulations can aid in understanding subgrid-scale processes that are otherwise averaged out in coarser global models.
3. Learning Algorithms: The paper explores the potential of ensemble Kalman inversion and Markov chain Monte Carlo (MCMC) methods as viable options for the learning algorithms needed to optimize ESM parameters. These algorithms allow for the assimilation of diverse data sources, accommodating both computable parameters derived from simulations and non-computable parameters reliant on empirical data.

Numerical Results and Implications

The authors present numerical results using the Lorenz-96 model as a test case for parameter learning algorithms. The findings show that both ensemble Kalman inversion and MCMC methods can effectively identify model parameters, offering insights into the balance between computational cost and the precision of parameter estimates. These methodologies emphasize the importance of understanding covariance structures and employing time-averaged statistics, which can reduce dependency on initial conditions and enhance the robustness of model predictions.

Impact and Future Directions

The proposed framework has significant implications for advancing climate modeling by enhancing the accuracy and reliability of ESMs. The integration of global observations and high-resolution simulations not only aids in reducing biases but also improves the ability of models to exploit emergent constraints—statistical relationships that tie together observable variability in current climates with responses to future climate change.

In future developments, refining learning algorithms to operate in an online setting, where data is continuously assimilated during model runs, could further enhance model adaptability. Additionally, sophisticated strategies for targeting high-resolution simulations could optimize information gain, improving model efficiency.

In conclusion, Earth System Modeling 2.0, as envisioned by the authors, represents a sophisticated approach to climate modeling, aligning with contemporary computational capabilities and observational data availability. By embedding learning processes within ESMs, this framework offers a path forward in addressing long-standing challenges in climate prediction, paving the way for more credible climate projections and informed policy decisions.