I have recently started a new position as a Lecturer at the University of Auckland where I will continue researching regional climate change and extreme event predictability with a touch of machine learning. Anyone interested in working on these problems Down Under-er please reach out!
This research was part of my postdoctoral fellowship with Stanford Data Science and I am eternally grateful for the funding and freedom to dedicate to this project.
implying that ocean variability provides additional predictability of regional warming.
This paper demonstrates that multi-year extreme event prediction can be tackled through targeted methodologies that identify extreme-event covariates that are more predictable than the extremes themselves
Then, we train simple machine learning models to predict the onset of these summertime warming jumps in climate models, and verify on observations. We show skill in predicting warming jumps, independent of the warming signal...
We first show that abrupt jumps in regional average summertime temperatures correspond to a significantly heightened likelihood of experiencing a three-day heat event over the same period.
Our new preprint proposes a framework for predicting summertime temperature jumps on 1-5 year timescales.
eartharxiv.org/repository/v...
Our new paper shows how recent prescribed (Rx) burns in the western US impacted later wildfires. We find that Rx fires reduced wildfire severity + net smoke emissions, even when factoring in smoke from Rx fires. But, we find that these Rx fires were less effective in the wildland-urban interface.
Future TC risk is uncertain — but here’s the twist: what drives that uncertainty changes with your risk model setup.
We unpacked that. Full paper here: www.science.org/doi/10.1126/...
@adamsobel.bsky.social @scamargo.bsky.social @nblmndl.bsky.social and other wonderful co-authors 🙏!
This study also raises a fun and interesting question – since the neural networks predict observations better than they predict the climate models they were trained on, does this mean that machine learning models are also suffering from our perennial friend – the signal-to-noise paradox? 5/6
We also find that the pattern learned by the neural network is significantly correlated with historic temperature variability over North America – implying that predictions of SSTs in the North Pacific can be used to predict multi-year regional temperature variability over North America. 4/6
When we apply the neural net to observations (out-of-sample!) we find that the observations are as well predicted, if not BETTER predicted than the climate model data used to train the neural (what?!) 3/6
We train a neural net on climate model data to predict SST variability in the North Pacific ocean on 1-5 year timescales and then pick apart the pattern best learned by the neural net in the climate model data. 2/6
New paper by me and Noah Diffenbaugh on identifying predictable Pacific variability in the observational record 1/6 agupubs.onlinelibrary.wiley.com/doi/10.1029/...
So proud of @amaz.bsky.social who defended her PhD today!!!! Dr Allie!!! So well deserved!! She absolutely killed it ❤️
