Ed Gerber

Ed Gerber

Professor at the Courant Institute

Using machine learning to improve climate prediction!

less than 1 minute read

Published:

Zac Espinosa’s study to replace a physics-based gravity wave parameterization with a neural network based emulator was just accepted in Geophysical Research Letters!

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.

You May Also Enjoy

Stressed out jet streams!

1 minute read

Published:

Xingjian (Ken) Yan, a precocious undergraduate (now bound for a PhD at MIT) working with Lei Wang and I just submitted a paper exploring the utility of the traffic jam theory of blocking onset for perdiction to Geophysical Research Letters. Ken defined and explored “flux exceedance events”, meteorological situations where the jet stream gets overloaded with storm activity. Nakamura and Huang suggested that this overloaded jet situation creates a pile up storm activity – a traffic jam – leading to blocking events. Ken found that the climatological structure of exceedance events is remarkably similar to that of blocks, but that they appear to be distinct phenomenon: an overloaded jet stream is unfortunately not a reliable harbinger of an atmospheric block.

Protection without poison

1 minute read

Published:

Aaron Match developed a new theory for why the ozone layer is up in the stratosphere, reaching a maximum 26 km above the surface in paper submitted to the journal of Atmospheric Chemistry and Physics. In 1880, Walter Hartley deduced that ozone must be absorbing UV-B and UV-C radiation from the sun, but since it isn’t present at the surface (except in polluted air), this ozone must be somewhere “up there”. The stratosphere wasn’t discovered for a couple more decades, but it turns out that ozone that shields us from this harmful UV radiation is largely between 16 and 40 km above us. It’s a good thing its there: ozone is toxic. If it were uniformly distributed through the atmosphere, the concentration at the surface would be 8 times the EPA safety limit. Why is ozone safely up in the stratosphere, where it protects us without poisoning us? Advanced chemistry climate models can accurately predict the distribution of ozone, but in turns out our text book understanding of the ozone layer were incomplete and gave the wrong explanation for why it reaches a maximum in the stratosphere at 26 km.

Using Explainable AI and Transfer Learning to understand and predict the maintenance of Atlantic blocking with limited observational data

1 minute read

Published:

Huan Zhang submitted our paper on how to use machine learning to predict the persistence of blocking events to the new Journal of Geophysical Research Machine Learning and Computation, published by the AGU. Blocking events are persistent high pressure systems that “block” the flow of the jet stream. They are associated with extreme weather, as they shift the direction of storms, and, in summer, create heat domes that drive heat waves. A key element of a block is its persistence. Huan developed a convolutional neural network to predict whether an nascent blocking anomaly would persist, or fade away, and then interrogated the network to understand why it worked, and how it could be trained with the short observational record.

So, you have an imperfect data-driven parameterization for climate modeling. How can you make it better?

1 minute read

Published:

Minah Yang submitted a paper on how to combat data imbalance in regression problems to JAMES. The goal is to improve the performance of data-driven paramterizations, particularly for profiles that are rare, but important. This is a data imbalance problem: we need ensure that the parameterization works well on input-output pairs that are seldom seen in training. Minah proposes a technique based on histogram equalization, visualized below with help from Cece, Minah’s faithful companion! The idea is to oversample or reweight these rare cases during training, to ensure the method learns from them.