Deep learning to represent subgrid processes in climate models

By Stephan Rasp, Michael S. Pritchard, Pierre Gentine. 2018

892 citations recorded by the source corpus.

Significance Current climate models are too coarse to resolve many of the atmosphere’s most important processes. Traditionally, these subgrid processes are heuristically approximated in so-called parameterizations. However, imperfections in these parameterizations, especially for clouds, have impeded progress toward more accurate climate predictions for decades. Cloud-resolving models alleviate many of the gravest issues of their coarse counterparts but will remain too computationally demanding for climate change predictions for the foreseeable future. Here we use deep learning to leverage the power of short-term cloud-resolving simulations for climate modeling. Our data-driven model is fast and accurate, thereby showing the potential of machine-learning–based approaches to climate model development. The representation of nonlinear subgrid processes, especially clouds, has been a major source of uncertainty in climate models for decades. Cloud-resolving models better represent many of these processes and can now be run globally but only for short-term simulations of at most a few years because of computational limitations. Here we demonstrate that deep learning can be used to capture many advantages of cloud-resolving modeling at a fraction of the computational cost. We train a deep neural network to represent all atmospheric subgrid processes in a climate model by learning from a multiscale model in which convection is treated explicitly. The trained neural network then replaces the traditional subgrid parameterizations in a global general circulation model in which it freely interacts with the resolved dynamics and the surface-flux scheme. The prognostic multiyear simulations are stable and closely reproduce not only the mean climate of the cloud-resolving simulation but also key aspects of variability, including precipitation extremes and the equatorial wave spectrum. Furthermore, the neural network approximately conserves energy despite not being explicitly instructed to. Finally, we show that the neural network parameterization generalizes to new surface forcing patterns but struggles to cope with temperatures far outside its training manifold. Our results show the feasibility of using deep learning for climate model parameterization. In a broader context, we anticipate that data-driven Earth system model development could play a key role in reducing climate prediction uncertainty in the coming decade.

Subfields: Atmospheric Science, Global and Planetary Change

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Atmospheric Science

In Global and Planetary Change

Citations in the corpus

Stratospheric Aerosol Injection as a Deep Reinforcement Learning Problem (2019, background): “Recently, the advent of deep learning has led to a number of successful emulation attempts of full GCMs used for weather prediction (Düben, 2018), as well as for sub-grid scale processes (Brenowitz, 2018; Rasp, 2018), including precipitation (O’Gorman, 2018).”
Fifty Years of Research on the Madden‐Julian Oscillation: Recent Progress, Challenges, and Perspectives (2020, methodology): “This approach involves fitting a statistical model to the output of relatively expensive physical models (e.g., CRMs) that more faithfully represent the subgrid processes (Brenowitz & Bretherton, 2018; Gentine et al., 2018; O'Gorman & Dwyer, 2018; Rasp et al., 2018; Schneider et al., 2017).”
Visualizing the Consequences of Climate Change Using Cycle-Consistent Adversarial Networks (2019, background): “…applicable across climate science and ML problems, which requires deep collaboration for synergistic advancements in both disciplines (Monteleoni et al., 2013; Joppa, 2017; Racah et al., 2017; Schneider et al., 2017; Gil et al., 2018; Hwang et al., 2018; Karpatne et al., 2018; Rasp et al., 2018).”
Neural general circulation models for weather and climate (2023, background): “9–22.”
Challenges and design choices for global weather and climate models based on machine learning (2018, positive): “processes using techniques from machine learning was investigated in more detail, with promising results using both NNs (Brenowitz and Bretherton, 2018; Gentine et al., 2018; Rasp et al., 2018) and random forest decision trees (O’Gorman and Dwyer, 2018).”
Integrating Scientific Knowledge with Machine Learning for Engineering and Environmental Systems (2020, background): “For example, cloud-resolving models (CRM) need to run at sub-kilometer horizontal resolution to be able to effectively represent boundary-layer eddies and low clouds, which are crucial for the modeling of Earth’s energy balance and the cloud–radiation feedback [230].”
Confronting the Challenge of Modeling Cloud and Precipitation Microphysics (2020, methodology): “Another approach that has gained recent traction within the atmospheric science community has been to develop parameterizations wholly via machine learning (e.g., Brenowitz & Bretherton, 2018; Gentine et al., 2018; O’Gorman & Dwyer, 2018; Rasp et al., 2018).”
Tackling Climate Change with Machine Learning (2019, background): “ul if they are closely integrated into existing scientiﬁc models. This has been emphasized, again and again, by authors who have laid future paths for artiﬁcial 39 intelligence within climate science [477, 484, 485, 495, 500, 514]. New models need to leverage existing knowledge to make good predicti”
Improving Data‐Driven Global Weather Prediction Using Deep Convolutional Neural Networks on a Cubed Sphere (2020, background): “Another machine-learning effort has focused on the improvement of physics parameterizations in GCMs for both weather forecasting and climate prediction [4, 32].”
Physics-informed machine learning: case studies for weather and climate modelling (2021, background): “…modelling: (i) emulating complex physical processes that are either poorly understood or not sufﬁciently well represented by existing models [46,91–95]; (ii) downscaling coarse data to produce high-ﬁdelity high-resolution data [37,89,96,97]; and (iii) forecasting the spatio-temporal dynamics…”
Quantifying Progress Across Different CMIP Phases With the ESMValTool (2020, background): “Stevens, 2019) or completely novel approaches to parameterize clouds and convection in climate models such as, for instance, machine learning based cloud parametrizations (Gentine et al., 2018; Rasp et al., 2018)”
Finite basis physics-informed neural networks (FBPINNs): a scalable domain decomposition approach for solving differential equations (2021, background): “For example, many of today’s applications require the study of multi-physics, multi-scale systems, which can be very challenging to incorporate and model accurately, often requiring the use of adaptive schemes and subgrid parameterisations [Nochetto et al., 2009, Rasp et al., 2018].”
WeatherBench: A Benchmark Data Set for Data‐Driven Weather Forecasting (2020, methodology): “Plain neural networks are very bad at extrapolating to climates beyond what they have seen in the training data set (Rasp et al., 2018).”