Author Archives: Doug McNeall

Uncertainty quantification and exascale computing in climate science

I was asked to sit on a discussion panel at this meeting on uncertainty quantification (UQ) with exascale computing. I prepared a short statement (below) on future challenges for UQ at exascale, but I would have made a slightly a longer one (below that) if there was time. A great resource for thinking about the […]

Climate science in 10 minutes

I was recently tasked with explaining the basics of (physical) climate science in 10 minutes, for a general audience of non-scientists. I tweeted a long thread which covered most of the talk here: And the ThreadReader app collected the whole thread for viewing here: https://threadreaderapp.com/thread/1298911570127446016.html

A pairs density ploy of the JULES inputs that pass level 1 constraints.

Visualising input spaces using emulators

In a previous post, I looked at some of the ways we could visualise the input space of climate models, when they are constrained to produce behaviour that looks something like the real world. I used parallel co-ordinates plots and pairs plots to visualise the high (32) dimensional input space of the JULES land surface […]

Visualising weird input spaces

I’ve been working on a fairly large (~500 member) perturbed-parameter ensemble of the land surface model JULES. The model simulates the global historical land surface, and each ensemble member is forced by the same global reanalysis on the HadGEM2 grid scale. Differences in the model output are therefore caused by the different values of the […]

More on surviving social media

I updated this talk [download slides] from last summer (time flies!), offering some personal thoughts on surviving a sometimes-hostile social media environment. [The sound gets better after I decide to hold the tiny mic rather than have it stuck in my shirt.] I was talking to my peers this time, so the mood is somewhat […]

Surviving the climate communications environment

The title was carefully chosen. On Monday, I gave a talk to around 100 engaged and engaging students taking part in the University of Exeter’s Grand Challenge 2017. The students are introduced to a number of global challenges, and expected to work in interdisciplinary groups to come up with solutions. This year, climate change features […]

Sensitivity analysis with R

After last week’s post, I thought it might be useful to have some practical examples of how to do sensitivity analysis (SA) of complex models (like climate models) with an emulator. SA is one of those things that everyone wants to do at some point, and I’ll be able to point people here for code […]

One-at-a-time sensitivity analysis

I use one-at-a-time sensitivity analysis all the time, but it’s not without it’s dangers. It looks like developers are ahead of the statisticians in useful illustrative examples. @ThePracticalDev @testobsessed pic.twitter.com/uKvhzoEPF7 — George Dinwiddie (@gdinwiddie) January 23, 2016 2 unit tests. 0 integration tests pic.twitter.com/V2Z9F4G1sJ — The Practical Dev (@ThePracticalDev) January 14, 2016

Choosing your next design point

You can use the R package DiceOptim to choose the next point to run your expensive simulator. Here’s a gif of function EGO.nsteps() in action, choosing one point at a time, with an initial design of three points.   It doesn’t behave in exactly the way I expected, putting lots of points in that well […]

Gaussian process emulator example

Here’s a little Gaussian process emulator example that I cooked up using the R package DiceKriging. The function is Higdon02, from this useful archive on simulation experiments. I’ve used a constant to initiate the model fit on the smallest data set: km(form=~1, …) rather than a linear term: km(form=~., …) as otherwise you end up […]