skip to main content

Explore Research Products in the PAR
It may take a few hours for recently added research products to appear in PAR search results.


Title: Interpreting and Stabilizing Machine-Learning Parametrizations of Convection
Abstract Neural networks are a promising technique for parameterizing subgrid-scale physics (e.g., moist atmospheric convection) in coarse-resolution climate models, but their lack of interpretability and reliability prevents widespread adoption. For instance, it is not fully understood why neural network parameterizations often cause dramatic instability when coupled to atmospheric fluid dynamics. This paper introduces tools for interpreting their behavior that are customized to the parameterization task. First, we assess the nonlinear sensitivity of a neural network to lower-tropospheric stability and the midtropospheric moisture, two widely studied controls of moist convection. Second, we couple the linearized response functions of these neural networks to simplified gravity wave dynamics, and analytically diagnose the corresponding phase speeds, growth rates, wavelengths, and spatial structures. To demonstrate their versatility, these techniques are tested on two sets of neural networks, one trained with a superparameterized version of the Community Atmosphere Model (SPCAM) and the second with a near-global cloud-resolving model (GCRM). Even though the SPCAM simulation has a warmer climate than the cloud-resolving model, both neural networks predict stronger heating/drying in moist and unstable environments, which is consistent with observations. Moreover, the spectral analysis can predict that instability occurs when GCMs are coupled to networks that support gravity waves that are unstable and have phase speeds larger than 5 m s −1 . In contrast, standing unstable modes do not cause catastrophic instability. Using these tools, differences between the SPCAM-trained versus GCRM-trained neural networks are analyzed, and strategies to incrementally improve both of their coupled online performance unveiled.  more » « less
Award ID(s):
1835863
PAR ID:
10299894
Author(s) / Creator(s):
; ; ;
Date Published:
Journal Name:
Journal of the Atmospheric Sciences
Volume:
77
Issue:
12
ISSN:
0022-4928
Page Range / eLocation ID:
4357 to 4375
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. ; ( , The Planetary Science Journal)
    Abstract

    In Earth’s current climate, moist convective updraft speeds increase with surface warming. This trend suggests that very vigorous convection might be the norm in extremely hot and humid atmospheres, such as those undergoing a runaway greenhouse transition. However, theoretical and numerical evidence suggests that convection is actually gentle in water-vapor-dominated atmospheres, implying that convective vigor may peak at some intermediate humidity level. Here, we perform small-domain convection-resolving simulations of an Earth-like atmosphere over a wide range of surface temperatures and confirm that there is indeed a peak in convective vigor, which we show occurs nearTs≃ 330 K. We show that a similar peak in convective vigor exists when the relative abundance of water vapor is changed by varying the amount of background (noncondensing) gas at fixedTs, which may have implications for Earth’s climate and atmospheric chemistry during the Hadean and Archean eons. We also show that Titan-like thermodynamics (i.e., a thick nitrogen atmosphere with condensing methane and low gravity) produce a peak in convective vigor atTs≃ 95 K, which is curiously close to the current surface temperature of Titan. Plotted as functions of the saturation-specific humidity at cloud base, metrics of convective vigor from both Earth-like and Titan-like experiments all peak when cloud-base air contains roughly 10% of the condensible gas by mass. Our results point to a potentially common phenomenon in terrestrial atmospheres: that moist convection is most vigorous when the condensible component is between dilute and nondilute abundance.

     
    more » « less
  2. ; ; ; ( , The Planetary Science Journal)
    Abstract

    Storms operated by moist convection and the condensation of CH4or H2S have been observed on Uranus and Neptune. However, the mechanism of cloud formation, thermal structure, and mixing efficiency of ice giant weather layers remains unclear. In this paper, we show that moist convection is limited by heat transport on giant planets, especially on ice giants where planetary heat flux is weak. Latent heat associated with condensation and evaporation can efficiently bring heat across the weather layer through precipitations. This effect was usually neglected in previous studies without a complete hydrological cycle. We first derive analytical theories and show that the upper limit of cloud density is determined by the planetary heat flux and microphysics of clouds but is independent of the atmospheric composition. The eddy diffusivity of moisture depends on the planetary heat fluxes, atmospheric composition, and surface gravity but is not directly related to cloud microphysics. We then conduct convection- and cloud-resolving simulations with SNAP to validate our analytical theory. The simulated cloud density and eddy diffusivity are smaller than the results acquired from the equilibrium cloud condensation model and mixing length theory by several orders of magnitude but consistent with our analytical solutions. Meanwhile, the mass-loading effect of CH4and H2S leads to superadiabatic and stable weather layers. Our simulations produced three cloud layers that are qualitatively similar to recent observations. This study has important implications for cloud formation and eddy mixing in giant planet atmospheres in general and observations for future space missions and ground-based telescopes.

     
    more » « less
  3. ; ; ; ; ; ; ( , Journal of Geophysical Research: Atmospheres)
    Abstract

    Climate models are essential to understand and project climate change, yet long‐standing biases and uncertainties in their projections remain. This is largely associated with the representation of subgrid‐scale processes, particularly clouds and convection. Deep learning can learn these subgrid‐scale processes from computationally expensive storm‐resolving models while retaining many features at a fraction of computational cost. Yet, climate simulations with embedded neural network parameterizations are still challenging and highly depend on the deep learning solution. This is likely associated with spurious non‐physical correlations learned by the neural networks due to the complexity of the physical dynamical system. Here, we show that the combination of causality with deep learning helps removing spurious correlations and optimizing the neural network algorithm. To resolve this, we apply a causal discovery method to unveil causal drivers in the set of input predictors of atmospheric subgrid‐scale processes of a superparameterized climate model in which deep convection is explicitly resolved. The resulting causally‐informed neural networks are coupled to the climate model, hence, replacing the superparameterization and radiation scheme. We show that the climate simulations with causally‐informed neural network parameterizations retain many convection‐related properties and accurately generate the climate of the original high‐resolution climate model, while retaining similar generalization capabilities to unseen climates compared to the non‐causal approach. The combination of causal discovery and deep learning is a new and promising approach that leads to stable and more trustworthy climate simulations and paves the way toward more physically‐based causal deep learning approaches also in other scientific disciplines.

     
    more » « less
  4. ; ; ( , Journal of Advances in Modeling Earth Systems)
    Abstract

    With the recent advances in data science, machine learning has been increasingly applied to convection and cloud parameterizations in global climate models (GCMs). This study extends the work of Han et al. (2020,https://doi.org/10.1029/2020MS002076) and uses an ensemble of 32‐layer deep convolutional residual neural networks, referred to as ResCu‐en, to emulate convection and cloud processes simulated by a superparameterized GCM, SPCAM. ResCu‐en predicts GCM grid‐scale temperature and moisture tendencies, and cloud liquid and ice water contents from moist physics processes. The surface rainfall is derived from the column‐integrated moisture tendency. The prediction uncertainty inherent in deep learning algorithms in emulating the moist physics is reduced by ensemble averaging. Results in 1‐year independent offline validation show that ResCu‐en has high prediction accuracy for all output variables, both in the current climate and in a warmer climate with +4K sea surface temperature. The analysis of different neural net configurations shows that the success to generalize in a warmer climate is attributed to convective memory and the 1‐dimensional convolution layers incorporated into ResCu‐en. We further implement a member of ResCu‐en into CAM5 with real world geography and run the neural‐network‐enabled CAM5 (NCAM) for 5 years without encountering any numerical integration instability. The simulation generally captures the global distribution of the mean precipitation, with a better simulation of precipitation intensity and diurnal cycle. However, there are large biases in temperature and moisture in high latitudes. These results highlight the importance of convective memory and demonstrate the potential for machine learning to enhance climate modeling.

     
    more » « less
  5. ; ; ; ; ; ; ; ; ( , Atmosphereocean)
    Convective parameterization is the long-lasting bottleneck of global climate modelling and one of the most difficult problems in atmospheric sciences. Uncertainty in convective parameterization is the leading cause of the widespread climate sensitivity in IPCC global warming projections. This paper reviews the observations and parameterizations of atmospheric convection with emphasis on the cloud structure, bulk effects, and closure assumption. The representative state-of-the-art convection schemes are presented, including the ECMWF convection scheme, the Grell scheme used in NCEP model and WRF model, the Zhang-MacFarlane scheme used in NCAR and DOE models, and parameterizations of shallow moist convection. The observed convection has self-suppression mechanisms caused by entrainment in convective updrafts, surface cold pool generated by unsaturated convective downdrafts, and warm and dry lower troposphere created by mesoscale downdrafts. The post-convection environment is often characterized by “diamond sounding” suggesting an over-stabilization rather than barely returning to neutral state. Then the pre-convection environment is characterized by slow moistening of lower troposphere triggered by surface moisture convergence and other mechanisms. The over-stabilization and slow moistening make the convection events episodic and decouple the middle/upper troposphere from the boundary layer, making the state-type quasi-equilibrium hypothesis invalid. Right now, unsaturated convective downdrafts and especially mesoscale downdrafts are missing in most convection schemes, while some schemes are using undiluted convective updrafts, all of which favour easily turned-on convection linked to double-ITCZ (inter-tropical convergence zone), overly weak MJO (Madden-Julian Oscillation) and precocious diurnal precipitation maximum. We propose a new strategy for convection scheme development using reanalysis-driven model experiments such as the assimilation runs in weather prediction centres and the decadal prediction runs in climate modelling centres, aided by satellite simulators evaluating key characteristics such as the lifecycle of convective cloud-top distribution and stratiform precipitation fraction. 
    more » « less
  • Citation Formats
  • MLA
  • APA
  • Chicago
  • BibTeX
  • Save / Share this Record
  • Send to Email