skip to main content


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
NSF-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. 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. Abstract

    This study investigates the emergence of hurricane‐like vortices in idealized simulations of rotating moist convection. A Boussinesq atmosphere with simplified thermodynamics for phase transitions is forced by prescribing the temperature and humidity at the upper and lower boundaries. The governing equations are solved numerically using a variable‐density incompressible Navier‐Stokes solver with adaptive mesh refinement to explore the behavior of moist convection under a broad range of conditions. In the absence of rotation, convection aggregates into active patches separated by large unsaturated regions. Rotation modulates this statistical equilibrium state so that the self‐aggregated convection organizes hurricane‐like vortices. The warm and saturated air converges to the center of the vortices, and the latent heat released through the upwelling, forms the warm core structure. These hurricane‐like vortices share characteristics similar to tropical cyclones in the earth's atmosphere. The hurricane‐like vortices occur under conditionally unstable conditions where the potential energy given at the boundaries is large enough, corresponding to a moderate rate of rotation. This regime shares many similar characteristics to the tropical atmosphere indicating that the formation of intense meso‐scale vortices is a general characteristic of rotating moist convection. The model used here does not include any interactions with radiation, wind‐evaporation feedback, or cloud microphysics, indicating that, while these processes may be relevant for tropical cyclogenesis in the Earth atmosphere, they are not its primary cause. Instead, our results confirm that the formation and maintenance of hurricane‐like vortices involve a combination of atmospheric dynamics under the presence of rotation and of phase transitions.

     
    more » « less
  3. Abstract

    A framework is introduced to investigate the indirect effect of aerosol loading on tropical deep convection using three-dimensional limited-domain idealized cloud-system-resolving model simulations coupled with large-scale dynamics over fixed sea surface temperature. The large-scale circulation is parameterized using the spectral weak temperature gradient (WTG) approximation that utilizes the dominant balance between adiabatic cooling and diabatic heating in the tropics. The aerosol loading effect is examined by varying the number of cloud condensation nuclei (CCN) available to form cloud droplets in the two-moment bulk microphysics scheme over a wide range of environments from 30 to 5000 cm−3. The radiative heating is held at a constant prescribed rate in order to isolate the microphysical effects. Analyses are performed over the period after equilibrium is achieved between convection and the large-scale environment. Mean precipitation is found to decrease modestly and monotonically when the aerosol number concentration increases as convection gets weaker, despite the increase in cloud liquid water in the warm-rain region and ice crystals aloft. This reduction is traced down to the reduction in surface enthalpy fluxes as an energy source to the atmospheric column induced by the coupling of the large-scale motion, though the gross moist stability remains constant. Increasing CCN concentration leads to 1) a cooler free troposphere because of a reduction in the diabatic heating and 2) a warmer boundary layer because of suppressed evaporative cooling. This dipole temperature structure is associated with anomalously descending large-scale vertical motion above the boundary layer and ascending motion at lower levels. Sensitivity tests suggest that changes in convection and mean precipitation are unlikely to be caused by the impact of aerosols on cloud droplets and microphysical properties but rather by accounting for the feedback from convective adjustment with the large-scale dynamics. Furthermore, a simple scaling argument is derived based on the vertically integrated moist static energy budget, which enables estimation of changes in precipitation given known changes in surfaces enthalpy fluxes and the constant gross moist stability. The impact on cloud hydrometeors and microphysical properties is also examined, and it is consistent with the macrophysical picture.

     
    more » « less
  4. 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. Abstract

    To explore the interactions among column processes in the Community Atmosphere Model (CAM), the single‐column version of CAM (SCAM) is integrated for 1000 days in radiative‐convective equilibrium (RCE) with tropical values of boundary conditions, spanning a parameter or configuration space of model physics versions (v5 vs. v6), vertical resolution (standard and 60 levels), sea surface temperature (SST), and some interpretation‐driven experiments. The simulated time‐mean climate is reasonable, near observations and RCE of a cyclic cloud‐resolving model. Updraft detrainment in the deep convection scheme produces distinctive grid‐scale structures in humidity and cloud, which also interact with radiative transfer processes. These grid artifacts average out in multi‐column RCE results reported elsewhere, illustrating the nuts‐and‐bolts interpretability that SCAM adds to the hierarchy of model configurations. Multi‐day oscillations of precipitation arise from descent of warm convection‐capping layers starting near the tropopause, eventually reset by a burst of convective deepening. Experiments reveal how these oscillations depend critically on an internal parameter that controls the number of neutral buoyancy levels allowed for determining cloud top and computing dilute convective available potential energy in the deep convection scheme, and merely modified a little by disabling cloud‐base radiation (heating of cloud base). This strong dependence of transient behavior in 1D on this parameter will be tested in the second part of this work, in which SCAM is coupled to a parameterized dynamics of two‐dimensional, linearized gravity wave, and in the 3D simulations in future study.

     
    more » « less