Will Gregory

🚨 Happy to share that our 'GPSat' paper on scalable interpolation of satellite altimetry data, is now published in @NatureComms ! We hope that this can be used to fill-in observational gaps in (high res) altimetry data @M2LInES @Princeton More to come! doi.org/10.1038/s41467…

Hey academic twitter. Like many others, I'm moving over to LinkedIn for science content - add me as a connection if you're already on there! I'll mainly post about sea ice, climate and machine learning :) linkedin.com/in/will-gregor…

New paper by @willjgregory_ and colleagues presents a high resolution reconstruction of #SeaIce fields. @Princeton, @UCL, @M2LInES, @CPOM_news nature.com/articles/s4146…

Great thread about recent algorithmic developments in sea ice remote sensing! Exciting times lay ahead!

The code is openly available at github.com/CPOMUCL/GPSat

This was a great collaborative effort with @esa_cryosat @ES_UCL , @ai_ucl, and @turinginst, featuring: Ronald MacEachern, So Takao, Isobel Lawrence, @carmenjulianab, Marc Deisenroth, and @micheltsamados

🚨 Happy to share that our 'GPSat' paper on scalable interpolation of satellite altimetry data, is now published in @NatureComms ! We hope that this can be used to fill-in observational gaps in (high res) altimetry data @M2LInES @Princeton More to come! doi.org/10.1038/s41467…

This is a great community effort and has an important take-home message: Sea ice prediction systems do have skill on seasonal timescales, and provide most utility in extreme sea ice years.

Happy to share the final version of our new review paper (led by @_the_iceman), which is out in BAMS (#OpenAccess): doi.org/10.1175/BAMS-D… This community-driven paper explores the utility of #Arctic sea ice prediction through a collection of statistical and dynamical models.

I am very glad to invite you to the second #SuperIce webinar on Sea Ice and Data assimilation/machine learning on June 24th at 14:00 CEST. We will have the pleasure to have two presentations:

still waiting on those 40 hospitals first, buds

The work i’ll be talking about: doi.org/10.1029/2023MS… doi.org/10.1029/2023GL…

This morning at #EGU24 i’ll be talking about our recent @M2LInES work on improving climate model simulations of sea ice using data assimilation and machine learning. Come check it out in Room 1.34 at 10:45! @EGU_CR

Want to 🍶 Overcome computational bottlenecks? 📊 Improve estimates of ocean topography & sea ice thickness? 👉"Scalable interpolation of satellite altimetry data with probabilistic machine learning"👈 @willjgregory_ @micheltsamados @carmenjulianab 👉researchsquare.com/article/rs-420…

Check out our new #GPSat Python programming library for efficient interpolation of satellite altimetry (e.g. @esa_cryosat @NASA_ICESat2) data using scalable Gaussian Process (GP) Huge credit to all the team and @willjgregory_ for leading us on this one x.com/willjgregory_/…

🚨 Very excited to share this pre-print! In this work we developed an open-source Python library, GPSat, for investigating the scaling potential of local Gaussian Process (GP) models for optimal interpolation of satellite altimetry data 🧵 doi.org/10.21203/rs.3.…

Huge shout out to the co-authors of this work: Ronald MacEachern, So Takao, Isobel Lawrence, Carmen Nab (@carmenjulianab), Marc Deisenroth and Michel Tsamados (@micheltsamados) @M2LInES

We really hope that GPSat can be a community tool which people can use and contribute to. If you're interested, it's openly available on Github (github.com/CPOMUCL/GPSat). There's also an example notebook which can be run interactively in Google Colab. Please help us find bugs! 🪲

We find that we can achieve a ~8x computational speedup using a dedicated GP library (GPflow), as opposed to using a standard NumPy implementation. Utilizing GPUs then give us another ~5x speedup