Proudly sponsored by PyMC Labs, the Bayesian Consultancy. Book a call, or get in touch!

• My Intuitive Bayes Online Courses
• 1:1 Mentorship with me

In this episode, Andy Aschwanden and Doug Brinkerhoff tell us about their work in glaciology and the application of Bayesian statistics in studying glaciers. They discuss the use of computer models and data analysis in understanding glacier behavior and predicting sea level rise, and a lot of other fascinating topics.

Andy grew up in the Swiss Alps, and studied Earth Sciences, with a focus on atmospheric and climate science and glaciology. After his PhD, Andy moved to Fairbanks, Alaska, and became involved with the Parallel Ice Sheet Model, the first open-source and openly-developed ice sheet model.

His first PhD student was no other than… Doug Brinkerhoff! Doug did an MS in computer science at the University of Montana, focusing on numerical methods for ice sheet modeling, and then moved to Fairbanks to complete his PhD. While in Fairbanks, he became an ardent Bayesian after “seeing that uncertainty needs to be embraced rather than ignored”. Doug has since moved back to Montana, becoming faculty in the University of Montana’s computer science department.

Our theme music is « Good Bayesian », by Baba Brinkman (feat MC Lars and Mega Ran). Check out his awesome work at https://bababrinkman.com/ !

Thank you to my Patrons for making this episode possible!

Yusuke Saito, Avi Bryant, Ero Carrera, Giuliano Cruz, Tim Gasser, James Wade, Tradd Salvo, William Benton, James Ahloy, Robin Taylor,, Chad Scherrer, Zwelithini Tunyiswa, Bertrand Wilden, James Thompson, Stephen Oates, Gian Luca Di Tanna, Jack Wells, Matthew Maldonado, Ian Costley, Ally Salim, Larry Gill, Ian Moran, Paul Oreto, Colin Caprani, Colin Carroll, Nathaniel Burbank, Michael Osthege, Rémi Louf, Clive Edelsten, Henri Wallen, Hugo Botha, Vinh Nguyen, Marcin Elantkowski, Adam C. Smith, Will Kurt, Andrew Moskowitz, Hector Munoz, Marco Gorelli, Simon Kessell, Bradley Rode, Patrick Kelley, Rick Anderson, Casper de Bruin, Philippe Labonde, Michael Hankin, Cameron Smith, Tomáš Frýda, Ryan Wesslen, Andreas Netti, Riley King, Yoshiyuki Hamajima, Sven De Maeyer, Michael DeCrescenzo, Fergal M, Mason Yahr, Naoya Kanai, Steven Rowland, Aubrey Clayton, Jeannine Sue, Omri Har Shemesh, Scott Anthony Robson, Robert Yolken, Or Duek, Pavel Dusek, Paul Cox, Andreas Kröpelin, Raphaël R, Nicolas Rode, Gabriel Stechschulte, Arkady, Kurt TeKolste, Gergely Juhasz, Marcus Nölke, Maggi Mackintosh, Grant Pezzolesi, Avram Aelony, Joshua Meehl, Javier Sabio, Kristian Higgins, Alex Jones, Gregorio Aguilar, Matt Rosinski, Bart Trudeau, Luis Fonseca, Dante Gates, Matt Niccolls, Maksim Kuznecov, Michael Thomas, Luke Gorrie, Cory Kiser, Julio, Edvin Saveljev, Frederick Ayala, Jeffrey Powell, Gal Kampel, Adan Romero and Will Geary.

Visit https://www.patreon.com/learnbayesstats to unlock exclusive Bayesian swag ;)

Takeaways:

- Computer models and data analysis play a crucial role in understanding glacier behavior and predicting sea level rise.

- Reliable data, especially on ice thickness and climate forcing, are essential for accurate modeling.

- The collaboration between glaciology and Bayesian statistics has led to breakthroughs in understanding glacier evolution forecasts.

-There is a need for open-source packages and tools to make glaciological models more accessible. Glaciology and ice sheet modeling are complex fields that require collaboration between domain experts and data scientists.

- The use of Bayesian statistics in glaciology allows for a probabilistic framework to understand and communicate uncertainty in predictions.

- Real-time forecasting of glacier behavior is an exciting area of research that could provide valuable information for communities living near glaciers.

-There is a need for further research in understanding existing data sets and developing simpler methods to analyze them.

- The future of glaciology research lies in studying Alaskan glaciers and understanding the challenges posed by the changing Arctic environment.

Chapters:

00:00 Introduction and Background

08:54 The Role of Statistics in Glaciology

31:46 Open-Source Packages and Tools

52:06 The Power of Bayesian Statistics in Glaciology

01:06:34 Understanding Existing Data Sets and Developing Simpler Methods

Links from the show:

• Andy’s website: https://glaciers.gi.alaska.edu/people/aschwanden
• Doug’s website: https://dbrinkerhoff.org/
• Andy on GitHub: https://github.com/aaschwanden 
• Doug on GitHub: https://github.com/douglas-brinkerhoff/
• Andy on Twitter: https://twitter.com/glacierandy?lang=fr
• Andy on Google Scholar: https://scholar.google.com/citations?user=CuvsLvMAAAAJ&hl=en
• Doug on Google Scholar: https://scholar.google.com/citations?user=FqU6ON8AAAAJ&hl=en
• LBS #64, Modeling the Climate & Gravity Waves, with Laura Mansfield: https://learnbayesstats.com/episode/64-modeling-climate-gravity-waves-laura-mansfield/
• Parallel Ice Sheet Model: www.pism.io
• PISM on GitHub: https://github.com/pism/pism
• Greenland View of Three Simulated Greenland Ice Sheet Response Scenarios: https://svs.gsfc.nasa.gov/4727/

Transcript

This is an automatic transcript and may therefore contain errors. Please get in touch if you're willing to correct them.

In this episode, Andy Ashfanden and Doug

Brinkerhoff tell us about their work in

2

Glaciology and the application of Bayesian

statistics in studying glaciers.

3

They discuss the use of computer models

and data analysis in understanding glacier

4

behavior and predicting sea level rise and

a lot of other fascinating topics.

5

Andy grew up in the Swiss Alps and studied

Earth Sciences with a focus on Atmospheric

6

and Climate Science and Glaciology.

7

After his PhD, Andy moved to Fairbanks,

Alaska, and became involved with the

8

parallel Ice Sheet model, the first open

source and openly developed Ice Sheet

9

model.

10

His first PhD student was no other than

Doug Brinkerhoff.

11

Doug did an MS in computer science at the

University of Montana, focusing on

12

numerical methods for Ice Sheet modeling,

and then moved to Fairbanks to complete

13

his PhD with Andy.

14

Why in Fairbanks?

15

he became an art invasion after quote,

seeing that uncertainty needs to be

16

embraced rather than ignored, end quote.

17

Doug has since moved back to Montana,

becoming faculty in the University of

18

Montana's computer science department.

19

Thank you so much to Stephen Lawrence for

inspiring me to do this episode.

20

This is Learning Vision Statistics,

episode 105, recorded March 7th.

21

Welcome to Learning Basion Statistics, a

podcast about patient inference, the

22

methods, the projects, and the people who

make it possible.

23

I'm your host, Alex Andorra.

24

You can follow me on Twitter,

25

Alex underscore and Dora like the country

for any info about the show learnbasedats

26

.com is left last week show notes becoming

a corporate sponsor unlocking Bayesian

27

Merch supporting the show on patreon

everything is in there that's

28

learnbasedats .com if you're interested in

one -on -one mentorship online courses or

29

statistical consulting feel free to reach

out and book a call at topmate .io slash

30

Alex underscore and Dora see you around

31

and best patient wishes to you all.

32

Andy Ashvanden, Doug Brinkerhoff, welcome

to Learning Asian Statistics.

33

Thanks for having us.

34

Thanks, Alex.

35

Yeah.

36

Yeah.

37

Thank you.

38

Thank you so much for taking the time.

39

Andy, thank you for putting me in contact

with Doug.

40

I'm actually happy to have the both of you

on the show today.

41

I have a lot of questions for you and

yeah, I love that we have an applied.

42

slide with you Andy and Doug is more on

the stats side of things so that's gonna

43

be very fun I always love that but before

that yeah let's dug into what you do day

44

to day how would you guys define the work

you're doing nowadays and how did you end

45

up working on this maybe let's start with

you Andy

46

Well, often when people hear the word

glaciologist, they assume I should be

47

jumping around on the glacier on a daily

basis.

48

Some of my colleagues do that.

49

I've done it for years, but these days my

job has become a bit more boring in that

50

sense that most of the time I spend in

front of my computer developing code for

51

data analysis, data processing, trying to

understand.

52

what's going on with glaciers.

53

So it's not as glorious anymore as maybe I

want it to be.

54

Is there a particular reason for that?

55

Is it a trend in your film that now more

and more of the work is done with

56

computers?

57

I think there is certainly a trend that...

58

More stuff is being done with computers in

particular, we just have more data

59

available, you know, starting with the

dawn of the satellite era.

60

And now with much more dense coverage of

different SAR and optical sensors on

61

satellites.

62

So that just has created the need for

doing more computing.

63

Personally, it just happened.

64

I did not, you know,

65

have a master plan going from collecting

field observation on a small glacier to do

66

large -scale modeling.

67

It just, my career somehow morphed into

that.

68

Hmm.

69

Okay, I see.

70

And well, I'm guessing we'll talk more

about that when we start thinking to what

71

you guys do.

72

But Doug, yeah, can you tell us what

you're doing nowadays and how you ended up

73

working on that?

74

Yeah, sure.

75

I'm in a computer science department now,

so obviously I spend a lot of time in

76

front of a computer as well.

77

But similarly, I got into this notion of

understanding glaciers from a

78

mountaineering type perspective.

79

That's what I was interested in and got

into geosciences from there and then took

80

this sort of roundabout way back to

81

computers by sort of slowly recognizing

that they were a really helpful tool for

82

trying to understand what was happening

with these systems.

83

They definitely are.

84

I remember that's personally how I ended

up working on stats.

85

Ironically, I wasn't a big fan of stats

when I was in college.

86

I loved math.

87

and algebra and stuff like that but stats

I didn't like that because it was you know

88

we were doing a lot of pen and paper

computations so I was like I don't

89

understand like it's just I'm bad at

computing personally so I don't know why

90

computers don't do that you know and and

then afterwards randomly I I started

91

working on electoral forecasting and

discovered you could simulate

92

distributions with the computer and the

computer was doing all the tedious

93

error -prone and boring work that I used

to not like at all.

94

And then I could just focus on, okay,

thinking about the model structure and

95

making sure the model made sense, what we

can say with it, what the model cannot

96

tell you also, things like that.

97

That was definitely super interesting.

98

So yeah, like that's also how I ended up

working on stats, ironically.

99

I had a similar path.

100

I didn't...

101

take a stats class until I was in my PhD

and watched Stan or one of these other

102

MCMC packages work to answer some really

interesting questions that you couldn't do

103

with the type of stats that people told

you about when you were in high school.

104

And that became much more intriguing to me

after seeing it applied to ecological

105

models or election forecasting or any of

these things that you need a computer to

106

assist with inferences for.

107

Yeah, for me, taking a stats class as an

undergrad student in the first or second

108

year, I had the impression that.

109

the stats department took great pride into

making the class as inaccessible as

110

possible and just go through like theorems

and proofs and try to avoid like any

111

connection to the real world, trying to

make it useful for us.

112

And I also got like really later into it

through Doc mainly, where I thought like,

113

you know, this kind of makes sense.

114

That's a good method.

115

to use to answer a problem I care about.

116

And before that, we were just giving

hypothetical problems that I had no

117

connection to.

118

Yeah.

119

Yeah.

120

Yeah, definitely makes sense.

121

And I resonate with that a lot.

122

And so today, what are the topics you

focus on?

123

Are you both working on the same topics or

are you working on slightly different or

124

completely different topics of your field?

125

Because I have to say, and that's also why

I really enjoyed this episode,

126

I really don't know a lot about classology

and what you guys are doing, so it's going

127

to be awesome.

128

I'm going to learn a lot.

129

Yeah, well, we work together a lot.

130

We both have our own independent projects,

but I think we work together a lot.

131

And I would say that you can tell me if

you don't agree with this, Andy, but I

132

would characterize the work that we both

do separately and together as trying to

133

make glacier evolution forecasts that

actually agree in a meaningful way.

134

with the observations that exist out there

in the world.

135

And that sounds sort of like an obvious

thing to do.

136

Like, yeah, if you have a model of glacier

motion that maybe you use to predict sea

137

level rise or something like that, like it

ought to agree with the measurements that

138

people have taken, those people that are

jumping around on the glacier that Andy

139

mentioned before.

140

But for a long time, and...

141

Perhaps now as well, that hasn't been the

case.

142

And so we're working to make our models

and reality agree as much as possible.

143

Andy?

144

I agree with Doc and I see it as a...

145

we do similar things, but...

146

I see this as a symbiotic relationship

where our independent strengths

147

taken together, Meg.

148

I need to rephrase that.

149

I think the sum of our strength has led to

some...

150

ways of thinking and breakthroughs that we

may not have done just on our own.

151

Sorry, that was not a good way of phrasing

it.

152

So while I'm coming a bit more...

153

In the past 10 years, I've been focusing a

bit more on like model development, on

154

development of ice flow models.

155

And as Doc said, we want to make them

agree with observations as good as we can

156

within observational uncertainties.

157

And I didn't have the background in

statistics to make that happen, whereas

158

Doc has both the insight into like how ice

flows and the modeling aspect, but he also

159

has a much deeper understanding of

statistics in general and patient

160

statistics in

161

in particular and we had a lot of

conversations trying to converge on an

162

approach to make that happen in a

meaningful way.

163

Because these days if you go and skim

through our literature, almost every third

164

paper somehow somewhere mentions machine

learning or artificial intelligence or

165

something.

166

It's just a buzzword.

167

It's a big hype.

168

Most of the time, if you dig deeper, all

you'll find is people do some multilinear

169

regression and call it machine learning.

170

That's the best case.

171

In some cases, I think methods are being

used in places where...

172

they haven't, we haven't been able to

demonstrate that this is the right place

173

to use those methods.

174

And we are trying to spend time to figure

out where can we use these modern machine

175

learning methods in a meaningful way that

actually drive science and help us answer

176

real world questions.

177

Yeah, yeah, yeah, very good points.

178

And something I've seen also in my

experience is that, well, the kind of

179

models and methods you can use is also

determined by the quality and reliability

180

of your data.

181

So I'm actually curious, Andy, if you can

give us an idea of what does data look

182

like in your field?

183

How big, how reliable are they?

184

And I think that's going to set us up

nicely to talk about modeling afterwards.

185

Sure.

186

So to figure out, you know, how much

187

a glacier and ice sheet is gonna melt.

188

There are a few things you need to know.

189

If you think about it in terms of partial

differential equations, you need initial

190

conditions and boundary conditions to

solve those equations.

191

But you also have processes besides those

PDEs that are a surrogate for physics that

192

we don't understand yet.

193

So those have parameters.

194

Often we don't know the values of those

parameters very well.

195

So we come in with.

196

a lot of different uncertainties.

197

Now I forgot what I meant to say.

198

Sorry, can you repeat the question?

199

Yeah, I was just asking you how, like what

the typical data look like in your field.

200

How big are they?

201

How reliable are they?

202

And that's usually very important to

understand then what you guys apply as

203

models.

204

Yeah, of course.

205

So one, if you look at the different

conservation equations that we're trying

206

to solve, conservation of mass, momentum

and energy, for solving conservation of

207

mass, we need to know the shape of the

glacier, the geometry.

208

Now, with modern satellites and airplanes,

it's relatively easy to measure the

209

surface of the ice.

210

relatively accurately and we can construct

accurate digital elevation models out of

211

that.

212

The tricky part is trying to figure out

how thick the ice is, for which we need

213

grout penetrating radar or seismic

methods.

214

All of them have large uncertainties.

215

Doing radar right now cannot be done from

space.

216

So to figure out the thickness,

217

at every point in the Greenland ice sheet

or Antarctic ice sheet basically requires

218

you to fly a plane.

219

And that's a lot of effort and of course

costs a lot of money.

220

So you can only do that in targeted areas.

221

And in the last 10 years, colleagues have

developed methods trying to combine those

222

observations from our ground penetrating

radar with what we understand how ice

223

flows, that it, you know,

224

obeys the laws of physics and conservation

of mass to come up with smarter way to

225

interpolate your data beyond just doing

creaking.

226

Now, ice thickness, I'm mentioning that

first because that is the most important

227

thing.

228

It defines how the ice flows.

229

It defines the surface gradient.

230

And at the end of the day, ice more or

less flows downhill against gravity.

231

So if you don't know how thick the ice is,

232

you're off to a really bad start.

233

So reliable ice thickness measurements are

key.

234

We've made a lot of progress in the last

15 years.

235

NASA spent approximately $100 million for

a project called Operation IceBridge,

236

which among other things measured ice

thickness and that just flew over

237

Greenland every spring for multiple weeks.

238

And that has given us a much more detailed

picture.

239

of where the ice is, how thick it is, and

how fast it flows.

240

And you can show that if you use these

newer data set compared to older ice

241

thickness data sets, that the models are

getting substantially better.

242

And it also gives us an avenue to test

whenever they add more observations, is

243

the model getting better or better and

better?

244

You can go look into individual glaciers

and you may see the model is still

245

performing poorly.

246

And,

247

you may find, well, there is not much data

there.

248

So hopefully at some point someone goes

out and can fly that glacier.

249

So this is the main uncertainty that we're

still struggling with despite like 10, 15

250

years of effort.

251

Now, the second one where it's, that is

very important and it's really hard to

252

quantify the uncertainties is the climate

forcing.

253

So in order to predict how a glacier flows

and how much it melts, you need to know

254

how much it snows.

255

And this is a tough topic.

256

Both Green and Antarctica are very large,

but they can vary topography over short

257

scales, which requires high resolution

climate models.

258

They are expensive, a lot more expensive

to run.

259

than an ice sheet model these days.

260

So they can usually do like one simulation

of the past 40 years and that's it.

261

There is basically no uncertainty

quantification that they do.

262

up to maybe recently or right now.

263

I think with machine learning, things may

start to change there too.

264

So we have products from observations

assimilated into those climate models, but

265

we often don't know how certain or

uncertain they are because what we have is

266

spot measurements.

267

There might be a couple hundred spot

measurements in Greenland or Antarctica

268

where you can calibrate or validate.

269

your climate model.

270

So that's a big uncertainty.

271

And I've been speaking for a long time.

272

Maybe Doug wants to chime in and add

something to it.

273

Yeah.

274

I mean, sometimes I think when you're

working with these really large couple of

275

geophysical systems, it can be the line

between model result and data product

276

becomes a little bit blurred.

277

So what do we have?

278

We have...

279

Direct surface measurements from a variety

of sources maybe over the past 30 years

280

with varying degrees of spatial

resolution, like Andy said.

281

It's gotten a lot better in the satellite

era, of course.

282

We've got these sparse measurements of

thickness that we don't completely

283

understand the uncertainties for, but

they're pretty accurate.

284

But they are certainly not everywhere.

285

with respect to the total area of

glaciated ice on Earth.

286

What else do we have?

287

Yeah, we've got a couple snow pit

measurements or shortwave radar that can

288

measure snow accumulation over a few

places on Earth.

289

We have optical satellite observations

that can often be leveraged into

290

understanding the displacement of the

glacier surface.

291

And there are a couple other somewhat more

esoteric products that we can come up with

292

hypotheses about how we might use to

constrain glacial ice flow, but we haven't

293

quite gotten there yet, like the

distribution of dust layers and stuff like

294

that inside of the ice that you can also

back out from some of these radar

295

observations.

296

But taken together, these observations,

the data that we have,

297

Occupy large amounts of space on a hard

drive in in that sense.

298

They're big like like there's a a ton of

individual measurements out there but sort

299

of relative to the magnitude of the system

that we're looking at and the timeframes

300

over which we would really like to

Constrain their behavior.

301

The data is super small.

302

Okay.

303

Okay.

304

I see.

305

Yeah Thanks guys super

306

I think super important to set up that

background, that context.

307

Actually, Doug, you're the patient

statistician of the couple, if I

308

understood correctly.

309

Then can you tell us why would patient

statistics be interesting in this context?

310

Let's start with that.

311

What would patient statistics?

312

ring in this context, in this approach of

studying glaciers.

313

Yeah, sure.

314

So...

315

I, yeah, okay.

316

So I kind of think that most scientific

problems can be cast in a probabilistic

317

way.

318

And this is certainly true for

glaciological modeling, where what you

319

want to do at the end of the day is to

take some assumption that you have about

320

the way that the world works, right?

321

A model.

322

And you want to use that model and you

want to make a prediction about the future

323

or something that you haven't observed.

324

But you would also like to ingest all of

the information that you have collected

325

about the world into that model so that

everything ends up remaining self

326

-consistent.

327

And that ends up being a really helpful

paradigm in which to operate for

328

glaciology.

329

So typically, you know, the large -scale

goal and what everybody begins their

330

proposals and papers and stuff with is

like, glaciers are important for

331

predicting sea level rise.

332

And to predict sea level rise, what we

need to do is we need to take an ice sheet

333

model, ice physics model, and project it,

run it into the future, say 200 years or

334

something like that, and say, well, there

was this much ice to start with, there's

335

this much ice now, that difference is

gonna turn into sea level rise.

336

So that's one part of it.

337

We don't have enough information about how

these systems work to just make

338

one prediction, right?

339

Like we don't know the bed in a whole lot

of places like Andy was saying.

340

And so the sensible approach to dealing

with that is to say, well, let's put a

341

probability distribution over the bed and

let's sample from that probability

342

distribution and make a whole lot of

different predictions about what sea level

343

rise is going to be based on all of those

different potential realizations of how

344

the bed of the glacier might look.

345

And of course, it's not just the bed

that's uncertain.

346

There's a bunch of other stuff as well.

347

And so that's a very Bayesian way of

looking at probability, right?

348

I mean, you can't hardly escape the

Bayesian paradigm in geophysics, right?

349

Because we don't have the capacity for

repeat samples.

350

All we have is just the one data point,

right?

351

So no replicates here.

352

No limiting behavior.

353

And so, you know, there's just this notion

of ensemble modeling.

354

That's what we would call that this notion

of randomly sampling from potential model

355

inputs and running into the future.

356

That's a super Bayesian idea to begin

with.

357

And then the other sort of step in this

process is to say, okay, well, I actually

358

want to constrain what I think the bet is

based on these observations that I have,

359

which is to say, I'm going to start with a

big pie in the sky view over of what my

360

bet elevation could be, maybe something.

361

between 5 ,000 meters above sea level and

10 ,000 meters below.

362

But then I'm going to take all of these

radar observations that I have and whittle

363

down the space of possible ways that the

bed could be.

364

And that's, I mean, that is nothing if not

posterior inference, right?

365

Yeah, yeah.

366

Yeah, for sure.

367

Thanks to SuperClean.

368

Maybe a question for the both of you.

369

Do you have a favorite study or project

where the collaboration between glaciology

370

and Bayesian stance led to interesting

insights?

371

And yeah, a study that you particularly

like, whether that's one of yours or a

372

stunning glaciology from someone else.

373

What do you think, Andy?

374

Yeah.

375

I think as Doug alluded earlier, combining

Bayesian methods with the idea of large

376

ensembles, thanks to having access to

large high -performance computer systems,

377

have allowed us for the first time to

investigate the parameter space in a

378

meaningful way.

379

Before that,

380

you would basically hand tune most of what

you did was based on expert judgment.

381

Like your prior was what you've learned

over the past 10 years, so to speak.

382

And surprisingly,

383

Calibration by eyeballing can yield pretty

good results, but it only gives you a

384

median or a mean, and it doesn't give you

any information about the tails.

385

So, for years, we would publish one study,

a mean of one simulation, maybe a few

386

simulations, but we didn't look at the

distributions themselves.

387

and bringing the Bayesian methods into our

field, I think have led to a great deal of

388

to have led us to discover an

uncomfortable truth that those tails are

389

really large and they are not normally

distributed.

390

So ...

391

It's we've realized it's really important

to understand the tails and understand the

392

full distribution and not just a mean or a

median or any single point realization of

393

that.

394

So yeah, okay, so that's a really good

point.

395

And that reminds me of a study that we

didn't do that I think is really good.

396

But it merits maybe just explicitly

stating something about glaciological

397

systems, particularly the ice sheets,

which is that ice flow and in particular

398

the mechanisms of.

399

Retreat so the potential for you know

Antarctica or Greenland in some sense to

400

collapse and not be ice I see anymore to

become ice -free.

401

That's a super nonlinear process in the

sense that If if say we get the bed wrong

402

and it's too shallow if we if we if we

were to imagine that the bed is Shallower

403

than it actually is

404

then maybe, or I'll rephrase that and say,

if the bed is actually shallower than we

405

think it is, then that doesn't really have

that many implications for sea level

406

change.

407

If the things change as normal, if the bed

is, it just melts away.

408

If the bed is a lot deeper than we think

it is, then all of a sudden you have the

409

potential for the entire ice sheet to

float and physically disintegrate via

410

like,

411

the dramatic sort of calving processes

that maybe you've seen if you've seen the

412

movie Chasing Ice or one of these other

sort of documentaries.

413

And so the consequences of being wrong are

asymmetric with respect to some of these

414

unknown factors that govern the system.

415

And there's a really wonderful paper.

416

that shows this quite explicitly by a

colleague of ours named Alex Roebol, who

417

basically just took a simple model of

Antarctica, forced it with sort of

418

normally distributed melting noise, more

or less, and a bunch of different

419

scenarios, and showed this really big

systematic bias towards more mass loss on

420

account of the fundamental

421

asymmetry in the way that these

glaciological systems respond to errors in

422

input data.

423

Yeah, that just sounds very fascinating.

424

I'm super curious to see one of these

models.

425

Do you know if there are any open source

packages that, for instance, people

426

working in your field are using in Python

or in R that kind of wrap the usual models

427

you guys are working on?

428

And also, is there any cool data sets that

we can put in the show notes for?

429

people to look around if they want to.

430

Any interesting applications that you

think would be interesting, let's put that

431

in the show notes.

432

You made some super cool visualizations

for one of those papers a while ago,

433

didn't you Andy?

434

Well, I can't take credit for that, but

I'll send you the link.

435

I think one of our earlier collaborations

where we started exploring the idea of

436

large ensembles was funded by NASA and

with support from NASA, they helped us

437

visualizing.

438

our simulations on their big screens and

narrating it.

439

I'll send you a link.

440

That's all open and open source.

441

With regard to packages, most of those

models that we develop are kind of big

442

beasts.

443

It takes a while to learn them.

444

Right now, there are very few.

445

wrappers around it in Python.

446

The model we developed, you can access

stuff through Python, but we're not at the

447

level to use it as a black box.

448

Whether you should be able to use it as a

black box is a different question.

449

But we have a fund a project from the

National Science Foundation that drives us

450

towards that goal of reducing the barrier

of entry.

451

and reducing the time to actually do

science by taking steps like this.

452

So in the next couple of years, our group

and others are working towards a cloud

453

version of the model that ideally can just

be deployed with the click of a mouse.

454

And, you know, you, for example, choose

the parameters you are interested in in

455

your uncertainty quantification.

456

and the rest is done automatically.

457

Right now you do need inside knowledge on

HPC systems.

458

Each HPC system is different.

459

It can take days or weeks just to get the

model to run because each system has a

460

different MPI stack, different compilers.

461

You can run into all sorts of problems.

462

So that's just one step.

463

So we are trying to make that easier, but

we are not there yet.

464

I'll give you an anecdote, which is that

Andy has made a lot of progress utilizing

465

a very large computational fluid dynamics

code for ice sheet flow called the

466

parallel ice sheet model, which is

wonderful and super carefully constructed

467

and really a great piece of software.

468

But man, I don't have the attention span

to figure out how to learn it.

469

And so for a lot of the...

470

A lot of the real Bayesian computation

stuff that we've done, I got tired and

471

just made Andy run a large ensemble and

then we train a neural network to pretend

472

to be PISM and we'll sometimes work with

that instead.

473

Well, that sounds like fun too.

474

Yeah, and actually...

475

That's the future.

476

Yeah.

477

Yeah, go ahead, Andy.

478

That's what we're still working on and

what I envision to push a bit further in

479

the next couple of years as well.

480

Okay.

481

Yeah, definitely super, super fascinating.

482

And yeah, Doug, actually, I wanted to ask

you a bit more about that because you said

483

you have a background in computer science,

so...

484

I'm wondering how do we integrate the

Bayesian algorithms into the computational

485

models that you've talked about for

studying glaciers?

486

Are you using open source packages?

487

What does your work look like on that

front?

488

Yeah, absolutely.

489

Before I did statistics, I did numerical

methods and I still do a lot of that work.

490

In particular, I

491

work in the branch of numerical methods

associated with solving partial

492

differential equations via the finite

element method, which is, you know,

493

doesn't really matter how that works, but

there's a really wonderful package for

494

solving set equations via that method

called FireDrake or

495

Phoenix, and so it's a really nice open

source Python package that a ton of

496

scientists are using for all sorts of

different applications in computational

497

mechanics.

498

And so I use that for developing sort of

the guts, the dynamical cores, as some

499

might call them, of these models.

500

And it's a nice tool in the sense that it

allows for a very straightforward

501

computation of derivatives of the output

of those models with respect to the inputs

502

of those models, which is super useful for

all sorts of optimization tasks and also

503

approximation in a Bayesian sense tasks,

MCMC or other approximation methods.

504

And so my typical workflow now is to take

one of those models and actually wrap it

505

inside of PyTorch.

506

which is sort of a general purpose

framework for automatic differentiation

507

that's popular in the machine learning

community.

508

And basically what that lets me do is

basically view an ice sheet model as if it

509

were a function in PyTorch.

510

And I can put stuff into the model, I can

get stuff out of the model, I can compute

511

misfits with respect to data between what

the model predicts and what the...

512

what the data says and basically take

derivatives of that with respect to model

513

parameters in a very seamless and easy

way.

514

And there's a, I mean, I don't know, it's

all just mixing and matching various

515

really awesome open source tools.

516

Actually, back in the day, when I first

got into this stuff, it was all sort of

517

making ice sheet model solvers from

scratch in NumPy and then sticking them

518

into PyMC, which you work on, right?

519

Yeah, yeah, exactly.

520

That's why I was also asking.

521

I was curious if you were using PyMC and

other hood to do that, because it sounds

522

like it would be an appropriate framework

to...

523

to use it.

524

So I was curious.

525

Yeah.

526

No, now, well, I would love to.

527

Nowadays, the problems that we work on

tend to be high dimensional enough that

528

the MCMC methods generally become very

challenging to work with.

529

And so we have to do sometimes less good

stuff.

530

And Andy, how does that look like?

531

cooperating in these projects, right?

532

How?

533

Because you are more on the practical side

of things.

534

So how do you consume the results of the

model, I'm actually curious.

535

And because if I understand correctly, you

are intervening before the model, because

536

I'm guessing you're part of the data

collecting team and you have the domain

537

knowledge that can be integrated into the

model, if there are priors in the model.

538

And then afterwards, of course, you're

interpreting...

539

the results of the model.

540

But how does that look like to cooperate

with these kind of models and in these

541

contexts?

542

Well, the high level view of course is

that when we collaborate, doctors are

543

thinking and I do the talking or pushing

off the buttons and trying to run the

544

models.

545

That would be the simple answer.

546

A lot of dip.

547

Workflow.

548

is still very cumbersome.

549

So Doug has alluded to the different

methods of collecting data sets, all the

550

uncertainties associated with them or the

lack of uncertainties with these data

551

sets.

552

Things have gotten better, but you can

imagine still each data set, you find it

553

on a different server with a different way

to access it.

554

It is probably in a different grid.

555

It most likely has a different spatial

reference system.

556

So we are trying to transition from a

state where we spend.

557

half of our time just trying to come up

with not very robust workflow to get from

558

the data sets on different servers or

websites to ingesting them into the model

559

to run the model and then to analyze the

data.

560

Before we had all that great data, things

were easy and hard at the same time.

561

All you had were a few data points and you

probably had to write an email to your

562

colleague asking to get access to the data

point that they may have asked you to be

563

on your paper in return.

564

At least now we have traded that for

spending a lot of time trying to find...

565

figure out those workflows.

566

And there are lots of initiatives right

now trying to make that workflow easier.

567

But I don't think we're there.

568

I still feel like this is sort of half of

my time I'm spending with processing the

569

data and getting really mad at XRA because

it doesn't quite do what I want it to do.

570

It almost always does.

571

what I want it to do and it's amazing and

if it doesn't do what I want it to do then

572

it's going to be a long afternoon and

sometimes a little bit of yelling too.

573

I've been there.

574

I feel like we've had similar afternoons.

575

But yeah, XRA saves the day most of the

time but when it doesn't, yeah, it's hard

576

to debug for sure.

577

mainly because there is not a lot of

tutorials on it in my experience.

578

So you have to figure a lot of these

things on your own.

579

Yeah, and yeah, I was also curious about

that because on my own also I've been

580

working with a team of researchers.

581

So they are marine biologists.

582

So quite different.

583

It's got to do with water too, but liquid

water and yeah, basically a study of

584

trade.

585

of sharks across the world and that has

been super interesting to work with them

586

because of course I'm here, I'm there for

the statistical expertise, right?

587

I have nothing to bring on the shark side

of things.

588

I've actually learned a lot thanks to them

about sharks and shark trade and things

589

like that.

590

And yeah, that to me is also very

interesting because...

591

the models are getting more and more

intricate.

592

These are models that now are really hard

and I'm like, damn, if you're not kind of

593

a statistician already, it's really hard

to come up with that kind of model if

594

you're really a domain expert.

595

And at the same time, to develop the

model, you need the domain experts because

596

otherwise, I could not develop that model

without the domain experts, even though I

597

know how to code the model.

598

And...

599

And I find that also super interesting to

see that in a way because it's like, it's

600

also good illustration of what science is,

right?

601

It's like really the sum is bigger than

each party on its own.

602

But at the same time, as the statistician,

you know, I'm a bit frustrated because I

603

know the model, for instance, is not going

to be in the paper, for instance.

604

The model is going to be the appendix of

the paper.

605

I'm like, oh my God, but it's a beautiful

model.

606

I would definitely focus on that.

607

But my point is, collaborating with the

domain experts has been also super

608

interesting because as you were saying,

Andy, there are still some parts of the

609

workflow.

610

So on mine, I'm talking about the Bayesian

workflow, which are cleaning, which can

611

only need to be updated and improved and

working.

612

like that with people who mainly use the

model and consume it instead of writing it

613

is super valuable.

614

So yeah, I don't know, Doug, maybe if you

have stuff to add on that because I'm

615

listening to you.

616

Yeah, I mean, what you're saying, I think,

is going to resonate with anybody that's

617

trying to work across disciplinary

boundaries, which is, I mean, ultimately

618

what we need to do across all branches of

science right now, right?

619

We have all of these amazing statistical

methods and...

620

numerical methods and also so much

knowledge about the way the structural

621

assumptions that go into how the world

works and We have to combine those things

622

to make good progress now, but man if you

if It's very difficult to find a

623

circumstance in which somebody's really

figured that collaboration out in a in a

624

in a problem -free way, it's Yeah, it's

it's challenging

625

I agree it's hard.

626

I've been involved in a bunch of larger

scale projects trying to bring together

627

data scientists and domain scientists and

it's kind of both parties sort of need to

628

learn to speak the other parties language

and it especially for the data scientists

629

it can be a challenge because

630

you know, let me put it that way.

631

They have really big hammers.

632

They have awesome tools.

633

And we just, you know, in glaciology, we

just started taking baby steps.

634

So most of these awesome tools we actually

don't need.

635

We need like what they had in undergrad,

like the most basic neural network or

636

something like that will already get us

from here to 90%.

637

So when you collaborate with them, they're

638

I can't blame them, I would get bored too.

639

But it's like, no, no, we just need like a

simple neural network and that will do the

640

job.

641

So as Doc said, having being able to

straddle both worlds between the domain

642

science and the data science is a

challenge and we need more people doing

643

this.

644

I think in our field right now, there's

only a handful of people that I would

645

trust.

646

that they're able to do that, Doc is one

among them and maybe three or four others.

647

And I think we need more people who are

capable to, who are bilingual in data

648

science and in domain science.

649

But the one, so the thing I'll say I guess

is that since this is, we're all Bayesian

650

statistics boosters here, is that Bayes

theorem or maybe more,

651

more specifically or broadly, the

posterior predictive distribution, if we

652

can use some technical language for a

second.

653

It provides an exceptionally useful

blueprint for talking to people across

654

disciplinary boundaries.

655

Because I can write this down and I can

say, OK, here are the things, domain

656

scientists, that I need from you.

657

I need you to tell me what you want to

predict.

658

Like in the case of glaciology, that often

ends up being sea level rise or volume

659

change.

660

And it's like, OK, I can work with that.

661

I need you to provide to me a set of

structural assumptions that encodes your

662

best understanding as a domain expert of

how the world works.

663

That's your numerical model.

664

It's going to take in some inputs.

665

It's going to produce some outputs.

666

I need you to tell me what aspects of that

model you don't feel like you know enough

667

about.

668

I need you to tell me what observations

you have available to you.

669

And then we can put these things all

together in a big flow chart, a graph,

670

right?

671

Presumably a directed acyclic graph that

prescribes all of the causal relationships

672

in the system.

673

And then once that picture is drawn, me as

a person that understands sort of the

674

numerical methods, the nuts and bolts of

doing inference and prediction in this

675

sort of probabilistic framework,

676

I can take that picture and I can convert

that into code and I can bring to bear the

677

statistical tools.

678

So like the Bayesian language of cause and

effect and uncertainty is like a neutral

679

ground that I think that we can all start

to use to act as a mechanism for

680

translating the language that we all use

in different fields.

681

Yeah, learning the Bayes theorem and

whatever is associated with it.

682

certainly has opened my world quite a bit

in terms of how I think about a problem

683

and I found it the right way to

encapsulate my thoughts.

684

And as Doug said, it sort of levels the

playing field that it provides that common

685

language that the base theorem, I think

it's closely associated with how we

686

do stuff or think about problems in

geoscience.

687

And that has started to make things so

much easier.

688

If you just sit down as Doc said, you

write down the probability of sea level

689

rise given, and then, you know, you start

with the chain rule, you have your models,

690

you try to come up with a likelihood

model, you try to come up with priors for

691

your parameters.

692

And even as like a non -Basian expert, it

still provides me with a way to think

693

about it.

694

and provides me with the tools to talk

about Doc, with Doc and others about the

695

problems that I have and the goals I want

to achieve.

696

Yeah, yeah, awesome points.

697

And definitely agree that, yeah, also

making the effort of making sure we're

698

talking about the same things and

educating on these concepts is absolutely

699

crucial.

700

And, well, Andy, so to shift gears a bit,

there is a project of yours, and since I

701

see the time running by, there is

something I really want to ask you about,

702

and that's...

703

the Parallel Ice Sheet Model, so PISM.

704

I don't think we've mentioned it yet, and

yeah, I'm curious about that.

705

What does that mean?

706

What are you doing with this project?

707

The general ice sheet model or PISM in

short started a little bit before I came

708

to Alaska as a postdoc.

709

In fact, few of us may even remember the

time before the first iPhone and PISM

710

started a year before the, I think the

first iPhone came out and it was the first

711

open source ice sheet model.

712

But at the same time, it was the first

openly developed ice sheet model.

713

Lots of other models have come later and

opened their code after, you know, some,

714

after they have reached some maturity.

715

And basically we can go back to commit

716

that and look at the first line that has

been written.

717

And this is mostly thanks to a

mathematician named Ed Buehler here at the

718

University of Fairbanks and his, at that

time, grad student.

719

Chad Brown, who somehow got into ice sheet

modeling, I think similar to Doc, through

720

mountaineering, going over glaciers,

climbing up on ice and getting fascinated

721

with ice as a geophysical fluid.

722

And they started developing a model

slightly differently than it has been

723

developed in the past by individual

glaciologists without...

724

often without like a super strong

background in math and numerical analysis.

725

So PISM started from writing or by writing

validation tests first and then developing

726

the most appropriate numerical methods to

solve the problem.

727

And as the name said, the P stands for

parallel.

728

So it was also one of the first models

that was.

729

developed from scratch in MPI via PETSI

and could take advantage of larger HP

730

systems versus at that time when PISM

started, you would run your ice sheet

731

model on a single core on your laptop.

732

Since then, the project has grown quite a

bit.

733

The University of Alaska here is still the

lead developer.

734

I have full -time software engineer.

735

who does a lot of the testing code

development, works with users.

736

We have another team at the Potsdam

Institute for Climate Impact Research in

737

Potsdam in Germany, who does a lot of the

development as well.

738

And then there are 30 to 40 -ish users

scattered around the world who either

739

develop the model or use it purely for

trying to answer scientific questions.

740

and one of the best compliments we have

ever gotten about our model is, or was

741

when we found the first publication by

accident of someone who just found the

742

model online, went on GitHub, downloaded

it, compiled it, figured out how it works

743

because it is well documented, did some

cool science with it and got it through

744

peer review.

745

So they never even had to contact.

746

the developers to get help to get anything

done.

747

And for us, that's a big compliment.

748

There are other models where you kind of

need to take like a one week long course

749

to even get started.

750

And we've been trying to maintain that

level of documentation and co

751

-transparency by keeping a relatively

stable well thought out.

752

API, something like that.

753

So through all that backbone development,

it has become one of the leading models to

754

answer questions revolving around

glaciology and sea level rise.

755

Of course, again, because it started in

756

that, for example, Doc mentioned that he's

developing with his fire -direct code

757

coupled to

758

um, tight torch.

759

This is something we cannot yet offer and

it may not be feasible because there's so

760

much legacy code that we can't handle a

smooth transition.

761

Yeah, I didn't know that project was that.

762

Oh, that's impressive.

763

And I'm guessing that requires quite a lot

of collaboration with quite a lot of

764

people.

765

So well done on that.

766

Thank you.

767

That's incredible.

768

Yeah.

769

Any links, if there are any links that

people interested in could dig into, feel

770

free to join that to the show notes.

771

because I think that's a very interesting

project.

772

Doug, I'm also curious, I think I've seen

preparing for the show that you, and I

773

think you've talked about that at the

beginning, you work on echo geomorphic

774

effects.

775

Can you tell us what this is and what that

means and why that's interesting?

776

Sure.

777

Sure, yeah.

778

I would not say that I am an eco

-geomorphologist by any stretch of the

779

imagination, but when you work on

glaciology in Alaska, I think we're always

780

interested in understanding and

communicating the importance of glacial

781

systems beyond their influence on sea

level rise.

782

Because it turns out that if you plop a

giant chunk of ice somewhere on the

783

coastline, it's going to have implications

for what the water chemistry is like and

784

what the water temperature is like and

what the local climate is like and maybe

785

more broadly how animals can move around

and a whole bunch of other stuff.

786

And so one project that I'm super excited

about, we've been working on this for a

787

couple of years, is to try and understand

the future evolution of a very large

788

glacier in coastal Alaska called Malaspina

Glacier.

789

It's very conspicuous.

790

feature if you ever look at the coastline

of Alaska on Google Earth or something

791

like that.

792

And it also happens to sit very close to a

really robust Alaska native community that

793

uses the forelands of the glacier and the

adjacent areas as hunting and fishing

794

grounds.

795

And through the course of our modeling,

and we can say,

796

this with a fair bit of confidence because

we've done a complete probabilistic

797

treatment, we can say that it's very

likely that this very large glacier is

798

more or less going to disappear in the

next certainly century, maybe faster than

799

that.

800

And when that happens, it'll open up a new

fjord, Icefield Valley.

801

The forelands might start to degrade.

802

And

803

the whole landscape of that area that

people are using for all sorts of things,

804

for gathering food and transportation and

a ton of other activities, it's all going

805

to change a lot.

806

And so I'm really excited about being able

to utilize some of these modeling tools,

807

particularly in conjunction with robust

uncertainty quantification frameworks to

808

provide responsible

809

defensible predictions about how this

place is going to be different in the

810

coming years to the people that live

there.

811

Yeah.

812

Okay.

813

That makes more sense now.

814

And geo -ecomorphitration, that's the

term.

815

That's pretty impressive.

816

Geo -geomorphology, I guess that's...

817

I guess you'd say that that'd be the study

of how ecosystems change in response to

818

changes in the way that the earth shapes.

819

Yeah.

820

That's what you want to do to say...

821

at parties, you know, like Fisher.

822

Awesome.

823

Well, thanks a lot, guys.

824

We're going to start wrapping up because I

don't want to take too much of your time,

825

but of course I still would have lots of

questions.

826

Maybe, yeah, something I'd like to hear

you both about is potential development,

827

potential applications of

828

of what you're doing right now.

829

Where would you like to see the research

in glaciology and ice sheet modeling going

830

in the coming years?

831

What is the most exciting to you?

832

Maybe Andy first.

833

Maybe I'll start with the not so exciting

part.

834

because especially now with those new

methods that we're developing, machine

835

learning, artificial intelligence and

large data sets, I think there is still a

836

lot to be done just trying to understand

the data sets we already have with

837

relatively simple methods.

838

I say this is not particularly exciting

and it's also harder to get funding to do

839

that.

840

funding agencies like to see something

very new, something shiny.

841

But sometimes you can make a bunch of

progress by just bringing together bits

842

and pieces that you already have, but you

just never have time for that.

843

You could develop an algorithm that

describes how a glacier caps off in

844

Antarctica and you test it and it works

very well there.

845

But then you have to go on and develop

something new.

846

you're rarely left with the time to test,

well, would that be a good idea for

847

Mellaspino Glacier or for a glacier in

anywhere in Alaska or in Greenland as

848

well?

849

So if I had some time and some money, this

is where I think I could make a bunch of

850

progress with relatively little effort.

851

Maybe Doc wants to start with the shiny

stuff.

852

Shiny stuff, I don't know.

853

You know what's always a perpetual source

of inspiration for me is the United States

854

National Weather Service.

855

I go on their website and I type in my

town name and I click on a location on a

856

little map and it shows me a pretty high

accuracy prediction of what the weather is

857

going to look like where I'm at for the

next like seven days or something like

858

that.

859

And I...

860

It's this innocuous little interface, but

it overlies this incredible system of

861

computational fluid mechanics combined

with real -time integration of data

862

products in a probabilistic way.

863

They're doing ensemble modeling.

864

There's so much to it, and it's this

incredible operational system that has

865

just a wonderful, useful interface for

people.

866

And you know...

867

I think that we are getting maybe to the

point in glaciology with our understanding

868

of methods and capacities and stuff to

maybe do something like that.

869

And that's what I'm most excited about is

real -time forecasting for every little

870

chunk of glacier ice in the world.

871

Yeah, that sounds very interesting.

872

I'm going to look at that page.

873

Yeah, let's send that to the shuttles.

874

That sounds very fun.

875

I know, but that for sure.

876

Weather .gov, I bet it's the most widely

used application of Bayesian statistics in

877

geophysics of any of them.

878

Interesting.

879

Well, if anybody in the listeners knows

someone working at weather .gov who could

880

come on the podcast,

881

to talk about the application of patient

methods at weather .gov.

882

My door is open.

883

That would be a great episode.

884

Yeah.

885

Absolutely.

886

I've done a somewhat, I mean, a related

episode a few months or years ago, I don't

887

remember, about gravitation waves.

888

So not gravitational waves, but

gravitation waves.

889

I didn't know that existed.

890

That was super interesting.

891

And I'm going to...

892

I'm going to link to this episode in the

show notes because that was a very cool

893

one basically talking about the mass of

really big mountains.

894

So probably what the mountains you have in

Alaska, Andy and like basically the wave

895

they create through their gravity, which

is non -negligible in comparison to the

896

gravity of the earth, which is just pretty

incredible.

897

and that has impacts on the weather.

898

So definitely gonna link to that.

899

Before closing up the show though, I'm

gonna ask you the last two questions I ask

900

every guest at the end of the show.

901

First one, if you had unlimited time and

resources, which problem would you try to

902

solve?

903

I feel like Andy, you've almost answered

that, but I'm still gonna ask you again.

904

Maybe that gives you an opportunity to

answer something else.

905

Yes, I've came to Alaska over 15 years ago

and I've done modeling of the Antarctic

906

ice sheet, of the Greenland ice sheet, of

glaciers in the Alps and Scandinavia and

907

we haven't done much.

908

with Alaskan glaciers.

909

Doug was mentioning their projects on

Malaspino glaciers and the surrounding

910

area.

911

But because Alaska is so big, the

challenges are equally big.

912

Understanding the precipitation there,

where you go from sea level up to 5 ,000

913

meters within a couple tens of kilometers

poses interesting challenges to like,

914

any modeling or observational approach.

915

And after living here for that long,

within unlimited resources, I think I

916

would like to give back to Alaska and

study Alaskan glaciers.

917

So I would invest in both observational

and modeling capabilities to better

918

understand how the Arctic here in Alaska

is changing.

919

That's like, sounds differently like a

920

a very interesting project.

921

Doug, what about you?

922

Well, yeah, if I'm limited to glaciology,

then I suppose I would say what I did

923

before about this notion of a worldwide,

every glacier forecasting tool that was

924

widely usable by the general public.

925

I think I'll stick with that one.

926

But since my resources are unlimited, I

guess while I'm doing that, I will pay a

927

whole bunch of other people to go out and

sort out the whole nuclear fusion thing.

928

And then there'll be enough electricity to

run my computer.

929

That sounds like a good thing to do

indeed.

930

And second question, if you could have

dinner with any great scientific mind,

931

dead, alive, or fictional, who would it

be?

932

So Doug, let's start with you.

933

Sure.

934

Man, why do we call it Bayesian

statistics?

935

We should really be calling it Laplacian

statistics, right?

936

Yeah.

937

He came up with this notion that we should

view probability as a means for

938

communicating our knowledge of a process.

939

And I think that that's the most

940

Perhaps the most important scientific idea

that nobody ever mentions.

941

So I'm going to go with Laplace.

942

I would be really interested to see how he

felt about the application of probability

943

in that way to these more complicated

systems as well.

944

I love that.

945

And not only because that was my personal

answer also in one of the episodes I've

946

done.

947

Awesome.

948

Andy, we'll get to you.

949

But before that, I found the episode I was

referencing.

950

So that was episode 64 with Laura

Mansfield.

951

And we were talking about modeling the

climate and gravity waves.

952

I think I said gravitational waves.

953

That was wrong.

954

That's gravity waves.

955

Andy, who would you have dinner with?

956

Well, I feel like I'm pretty blessed.

957

I think I have...

958

dinner with great scientific minds on a

regular basis when I have dinner with my

959

colleagues at scientific conferences.

960

But if I just pick one person, let's...

961

How about I'll meet Aristostinus?

962

I'm not sure I pronounced that correctly.

963

He was, I believe, the first one to

estimate the circumference of the earth.

964

And I think that was like several, couple

hundred years BC.

965

I'm just curious how people thought about

science in an environment several thousand

966

years ago.

967

I would love to chat with someone like far

back who...

968

came up with like, I think the estimate

that he came up with was maybe within 10 %

969

or something like that.

970

And then suddenly like a thousand years

later, people thought yours was flat.

971

I think that would be an interesting

person to meet.

972

Yeah, for sure.

973

Good one.

974

I think you're the first one to choose

that.

975

I love it.

976

What's the most common answer you get for

that question?

977

Well, that question is...

978

bit more like the variation is bigger than

the first one.

979

The first one has a clear winner if I

remember correctly, which is climate

980

change.

981

So we have a lot of people who would try

and tackle that.

982

The second question, I think one of the

most common is Richard Feynman, if I

983

remember correctly.

984

I believe so.

985

Yeah, I think Feynman is the winner, but

it's not...

986

Pareto distribution.

987

It's a pretty uniform distribution.

988

It's not like...

989

Yeah, I'm curious.

990

Not a lot of people choose Laplace.

991

Not a lot of people choose base.

992

And interestingly, I think nobody chose

base until now.

993

Yeah.

994

Not a lot of people have chosen Einstein.

995

So that's an interesting question because

that kind of goes against prior.

996

It's hard to guess.

997

Sorry, Andy.

998

I would have thought like Einstein or

Newton or Galileo would come up pretty

999

frequently.

01:12:51,597 --> 01:12:55,657

No, Galileo, I don't think so.

01:12:55,657 --> 01:12:58,877

Leonardo da Vinci does come up quite a

lot.

01:12:59,937 --> 01:13:05,677

But yeah, otherwise, I had Euclid once, of

course.

01:13:05,677 --> 01:13:08,269

That was a fun one, too.

01:13:08,269 --> 01:13:12,929

Awesome guys, well I think we can call it

a show, I've taken enough of your time,

01:13:12,929 --> 01:13:14,849

thank you for being so generous.

01:13:14,849 --> 01:13:21,749

Before we close up though, is there

something I forgot to ask you about and

01:13:21,749 --> 01:13:25,797

that you would like to mention or talk

about before we close up?

01:13:28,365 --> 01:13:30,125

I don't think so, not for me.

01:13:30,125 --> 01:13:33,845

I think it was a pretty comprehensive

journey.

01:13:33,845 --> 01:13:34,965

Yeah.

01:13:34,965 --> 01:13:36,085

Great.

01:13:36,225 --> 01:13:40,485

Believe me, I would still have like, I

could keep you for two hours, but no.

01:13:40,665 --> 01:13:43,045

Let's be parsimonious.

01:13:43,405 --> 01:13:43,885

Awesome.

01:13:43,885 --> 01:13:46,725

Well, again, thank you very much, Andy.

01:13:46,725 --> 01:13:47,955

Thank you very much, Dag.

01:13:47,955 --> 01:13:54,381

As usual, those who want to dig deeper,

refer to the show notes because we have.

01:13:54,381 --> 01:13:59,341

Andy's and Doug's links over there and

also a bit of the work.

01:13:59,701 --> 01:14:04,921

And on that note, thanks again, Andy and

Doug for taking the time and being on this

01:14:04,921 --> 01:14:05,961

show.

01:14:06,341 --> 01:14:06,751

Thanks Alex.

01:14:06,751 --> 01:14:07,361

Thanks Alex.

01:14:07,361 --> 01:14:09,063

Thanks for having us.

01:14:13,517 --> 01:14:17,217

This has been another episode of Learning

Bayesian Statistics.

01:14:17,217 --> 01:14:22,157

Be sure to rate, review, and follow the

show on your favorite podcatcher, and

01:14:22,157 --> 01:14:27,077

visit learnbaystats .com for more

resources about today's topics, as well as

01:14:27,077 --> 01:14:31,817

access to more episodes to help you reach

true Bayesian state of mind.

01:14:31,817 --> 01:14:33,767

That's learnbaystats .com.

01:14:33,767 --> 01:14:38,587

Our theme music is Good Bayesian by Baba

Brinkman, fit MC Lass and Meghiraam.

01:14:38,587 --> 01:14:41,747

Check out his awesome work at bababrinkman

.com.

01:14:41,747 --> 01:14:42,925

I'm your host.

01:14:42,925 --> 01:14:43,905

Alex Andorra.

01:14:43,905 --> 01:14:48,165

You can follow me on Twitter at Alex

underscore Andorra, like the country.

01:14:48,165 --> 01:14:53,225

You can support the show and unlock

exclusive benefits by visiting Patreon

01:14:53,225 --> 01:14:55,405

.com slash LearnBasedDance.

01:14:55,405 --> 01:14:57,865

Thank you so much for listening and for

your support.

01:14:57,865 --> 01:15:00,095

You're truly a good Bayesian.

01:15:00,095 --> 01:15:03,585

Change your predictions after taking

information.

01:15:03,585 --> 01:15:10,221

And if you're thinking I'll be less than

amazing, let's adjust those expectations.

01:15:10,221 --> 01:15:15,601

Let me show you how to be a good Bayesian

Change calculations after taking fresh

01:15:15,601 --> 01:15:21,661

data in Those predictions that your brain

is making Let's get them on a solid

01:15:21,661 --> 01:15:23,501

foundation