Dave’s plain language summary does a nice job of explaining the big picture. Parameterizations are reduced-complexity models that estimate the effects of physical processes smaller than what can be resolved by the grid of a weather or climate model. While necessary for realistic simulations, they are a source of uncertainty in climate projections. Recently, machine learning has been used to augment or replace conventional parameterizations of atmospheric gravity waves, a type of motion by which disturbances near the Earth’s surface can affect the wind higher up. We compare several machine learning approaches to the gravity wave parameterization problem. In particular, we test neural networks against random and boosted forests, which are built around flowchart-like models called regression trees. We find that boosted forests, though not widely used for climate model parameterization, are especially successful, scoring as well as or better than neural networks on various performance metrics. We then provide proof-of-concept of a novel method to retrain the boosted forest so that it uses its input data more in line with the physics of the system, and show that this technique improves the forest’s behavior when used together with an atmospheric model.
]]>This is a tour-de-force study, where Qiang emulated orographic, frontal, and convective gravity wave parameterizations in WACCM, which has perhaps the most sophisticated treatment of gravity waves on the market! He first had to overcome set imbalance to make training possible, and then probed uncertainty in the emulators, showing that many of the current approaches provided equivalent information. Finally, he showed that while the machine learning based emulators couldn’t generalize out-of-set to climate perturbations, with just a bit of data from the new climate, transfer learning can fix a scheme. This suggests that a small amount of data from high resolution simulations in warm climates might be sufficient for the purposes of data-driven parameterizations trained on today’s climate.
]]>This was a true team effort, involving Dave Connelly (our regression tree expert), Zihan Shao (our support vector regression guru), and Minah Yang (encoder-decoder extraordinaire) at NYU, with support from Steve Hardimann at the UK MetOffice (our dilated convolutional neural network tsar).
]]>It was an amazing experience. With the miracles of modern technology, er, Zoom, Slack, and email, research did happen (stay tuned)! Papers were published (and they did make it to the publications page)! In most ways, it wasn’t that much of a rest period (though I did have time to go running and hiking in many amazing places), and I did take a break from my website.
I’m back now. Well, have been for 3 months, but moving with a family was quite an operation. Ugh, there a still a few boxes in my office that need to get back to the apartment, but the website is back in business!
]]>At the AMS Conference on Atmospheric and Oceanic Fluid Dynamics, June 13-17:
Marguerite Brown will present A Theory for the Scaling of Moist Quasi-Geostrophic Turbulence
Dave Connelly will present Online Performance and Calibration of Data-Driven Gravity Wave Parameterizations
Aaron Match will present The Decade the Quasi-Biennial Oscillation Faltered: Testing Classical Understanding of the QBO against Weakening and Disruptions
Minah Yang will present Sampling Strategies for Training Machine Learning Emulators of Gravity Wave Parameterizations
]]>At the SIAM Conference on Mathematics of Planet Earth (MPE22), July 13-15, 2022
The ozone layer absorbs ultraviolet light otherwise harmful to life. Due to the Montreal Protocol, the ozone layer is generally recovering from depletion. In response to global warming, ozone is predicted to increase in the upper stratosphere but decrease in the tropical lower stratosphere. These decreases have previously been attributed to strengthening of stratospheric upwelling or to the deepening of the troposphere under global warming. We draw on elements of these prevailing explanations to quantify a new mechanism, which explains roughly half of the decrease: deepening of the troposphere converts stratospheric air into tropospheric air, the low ozone anomalies from which are then transported upwards into the tropical lower stratosphere by the background upwelling.
]]>The effect of moist processes, especially the impact of latent heating associated with condensation, on the size and strength of mid-latitude storms is not well understood. Such insight is particularly needed in the context of global warming, as we expect moisture to play a more important role in a warmer world. In this study, we provide intuition into how including condensation can result in mid-latitude storms that grow faster and have features on both larger and smaller scales than their dry counterparts. We provide a framework for quantifying these changes and verify it for the special case where it is raining everywhere. These findings can be extended to the more realistic situation where it is only raining locally.
]]>Atmospheric gravity waves (GWs) or “buoyancy waves” are generated by perturbations in a stably-stratified environment. They mediate momentum transport between the lower and middle atmospheres and play a leading-order role in driving middle atmospheric circulation. Due to computational constraints, global climate models (GCMs) “parameterize” or estimate the effect of GWs on the large-scale flow. Current climate predictions are sensitive to uncertainties in these representations. Here, we examine whether machine learning, given limited data, can be used for gravity wave parameterization (GWP) in climate prediction. This approach represents an appealing technique to build data-driven GWPs that can reduce existing uncertainties by incorporating observations.
]]>We use our idealized model of the atmosphere to understand the mechanism by which the stratospheric ozone loss (or recovery) impacts the troposphere. It is well known that the formation of the ozone hole caused a poleward shift of the tropospheric jet stream. We show that planetary scale stationary waves modulate the timing and amplitude of the jet stream shift, ultimately weakening its response to stratospheric ozone perturbations.
]]>What’s new? We’ve coupled the neural network based parameterization to the atmospheric model, and probed it’s ability to capture the response of the original gravity wave parameterization to global warming! As you can see in the picture above, the neural network captures the weakening of the QBO and the reduction of its period, as simulated by the original physics based parameterization. The weakening of the QBO in response to enhanced greenhouse gas forcing is observed across all models, and appears to be happening already in our atmosphere. The change in its period, however, is not robust across models and hence remains an open question. Our ultimate goal is to build machine learning based parameterizations constrained by data, so we can say something more definitive about the response of the QBO (and the atmosphere more generally) to global warming.
For the details, please see the paper: A Deep Learning Parameterization of Gravity Wave Drag Coupled to an Atmospheric Global Climate Model, just revised for Geophysical Research Letters. It was led by Zachary Espinosa, a graduate student working with Aditi Sheshadri at Stanford University, in collaboration with Gerald Cain and Kevin DallaSanta.
]]>