Speaker: 00:00:02

In this episode, Andy Ashfanden and Doug

Brinkerhoff tell us about their work in

2

00:00:09,822 --> 00:00:14,162

Glaciology and the application of Bayesian

statistics in studying glaciers.

3

00:00:14,162 --> 00:00:18,482

They discuss the use of computer models

and data analysis in understanding glacier

4

00:00:18,482 --> 00:00:23,782

behavior and predicting sea level rise and

a lot of other fascinating topics.

5

00:00:23,782 --> 00:00:28,822

Andy grew up in the Swiss Alps and studied

Earth Sciences with a focus on Atmospheric

6

00:00:28,822 --> 00:00:31,308

and Climate Science and Glaciology.

7

00:00:31,566 --> 00:00:36,866

After his PhD, Andy moved to Fairbanks,

Alaska, and became involved with the

8

00:00:36,866 --> 00:00:42,526

parallel Ice Sheet model, the first open

source and openly developed Ice Sheet

9

00:00:42,526 --> 00:00:43,506

model.

10

00:00:43,766 --> 00:00:49,446

His first PhD student was no other than

Doug Brinkerhoff.

11

00:00:49,446 --> 00:00:54,006

Doug did an MS in computer science at the

University of Montana, focusing on

12

00:00:54,006 --> 00:00:58,386

numerical methods for Ice Sheet modeling,

and then moved to Fairbanks to complete

13

00:00:58,386 --> 00:00:59,806

his PhD with Andy.

14

00:00:59,806 --> 00:01:01,390

Why in Fairbanks?

15

00:01:01,390 --> 00:01:06,550

he became an art invasion after quote,

seeing that uncertainty needs to be

16

00:01:06,550 --> 00:01:09,690

embraced rather than ignored, end quote.

17

00:01:09,690 --> 00:01:13,910

Doug has since moved back to Montana,

becoming faculty in the University of

18

00:01:13,910 --> 00:01:16,050

Montana's computer science department.

19

00:01:16,050 --> 00:01:20,870

Thank you so much to Stephen Lawrence for

inspiring me to do this episode.

20

00:01:20,870 --> 00:01:26,662

This is Learning Vision Statistics,

episode 105, recorded March 7th.

21

00:01:27,566 --> 00:01:49,286

Welcome to Learning Basion Statistics, a

podcast about patient inference, the

22

00:01:49,286 --> 00:01:52,846

methods, the projects, and the people who

make it possible.

23

00:01:52,846 --> 00:01:55,036

I'm your host, Alex Andorra.

24

00:01:55,036 --> 00:01:56,654

You can follow me on Twitter,

25

00:01:56,654 --> 00:02:02,074

Alex underscore and Dora like the country

for any info about the show learnbasedats

26

00:02:02,074 --> 00:02:07,394

.com is left last week show notes becoming

a corporate sponsor unlocking Bayesian

27

00:02:07,394 --> 00:02:11,594

Merch supporting the show on patreon

everything is in there that's

28

00:02:11,594 --> 00:02:16,974

learnbasedats .com if you're interested in

one -on -one mentorship online courses or

29

00:02:16,974 --> 00:02:22,294

statistical consulting feel free to reach

out and book a call at topmate .io slash

30

00:02:22,294 --> 00:02:25,518

Alex underscore and Dora see you around

31

00:02:25,518 --> 00:02:28,422

and best patient wishes to you all.

32

00:02:32,206 --> 00:02:38,106

Andy Ashvanden, Doug Brinkerhoff, welcome

to Learning Asian Statistics.

33

00:02:39,146 --> 00:02:40,886

Thanks for having us.

34

00:02:40,886 --> 00:02:41,466

Thanks, Alex.

35

00:02:41,466 --> 00:02:42,006

Yeah.

36

00:02:42,006 --> 00:02:42,386

Yeah.

37

00:02:42,386 --> 00:02:42,876

Thank you.

38

00:02:42,876 --> 00:02:45,566

Thank you so much for taking the time.

39

00:02:45,566 --> 00:02:48,246

Andy, thank you for putting me in contact

with Doug.

40

00:02:48,246 --> 00:02:52,296

I'm actually happy to have the both of you

on the show today.

41

00:02:52,296 --> 00:02:59,758

I have a lot of questions for you and

yeah, I love that we have an applied.

42

00:02:59,758 --> 00:03:07,298

slide with you Andy and Doug is more on

the stats side of things so that's gonna

43

00:03:07,298 --> 00:03:17,118

be very fun I always love that but before

that yeah let's dug into what you do day

44

00:03:17,118 --> 00:03:21,618

to day how would you guys define the work

you're doing nowadays and how did you end

45

00:03:21,618 --> 00:03:25,468

up working on this maybe let's start with

you Andy

46

00:03:27,758 --> 00:03:33,458

Well, often when people hear the word

glaciologist, they assume I should be

47

00:03:33,458 --> 00:03:36,978

jumping around on the glacier on a daily

basis.

48

00:03:37,458 --> 00:03:39,158

Some of my colleagues do that.

49

00:03:39,158 --> 00:03:46,658

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

job has become a bit more boring in that

50

00:03:46,658 --> 00:03:51,938

sense that most of the time I spend in

front of my computer developing code for

51

00:03:51,938 --> 00:03:56,654

data analysis, data processing, trying to

understand.

52

00:03:56,654 --> 00:03:59,194

what's going on with glaciers.

53

00:03:59,814 --> 00:04:06,774

So it's not as glorious anymore as maybe I

want it to be.

54

00:04:08,334 --> 00:04:11,714

Is there a particular reason for that?

55

00:04:11,714 --> 00:04:19,034

Is it a trend in your film that now more

and more of the work is done with

56

00:04:19,034 --> 00:04:20,094

computers?

57

00:04:20,874 --> 00:04:24,826

I think there is certainly a trend that...

58

00:04:26,126 --> 00:04:30,466

More stuff is being done with computers in

particular, we just have more data

59

00:04:30,466 --> 00:04:34,486

available, you know, starting with the

dawn of the satellite era.

60

00:04:34,486 --> 00:04:43,606

And now with much more dense coverage of

different SAR and optical sensors on

61

00:04:43,606 --> 00:04:44,466

satellites.

62

00:04:44,466 --> 00:04:48,566

So that just has created the need for

doing more computing.

63

00:04:48,566 --> 00:04:52,246

Personally, it just happened.

64

00:04:52,246 --> 00:04:54,190

I did not, you know,

65

00:04:54,190 --> 00:04:59,430

have a master plan going from collecting

field observation on a small glacier to do

66

00:04:59,430 --> 00:05:00,810

large -scale modeling.

67

00:05:00,810 --> 00:05:06,170

It just, my career somehow morphed into

that.

68

00:05:06,170 --> 00:05:07,170

Hmm.

69

00:05:07,450 --> 00:05:08,690

Okay, I see.

70

00:05:08,690 --> 00:05:13,690

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

about that when we start thinking to what

71

00:05:13,690 --> 00:05:15,190

you guys do.

72

00:05:15,270 --> 00:05:21,270

But Doug, yeah, can you tell us what

you're doing nowadays and how you ended up

73

00:05:21,270 --> 00:05:22,342

working on that?

74

00:05:22,638 --> 00:05:24,078

Yeah, sure.

75

00:05:24,998 --> 00:05:31,338

I'm in a computer science department now,

so obviously I spend a lot of time in

76

00:05:31,338 --> 00:05:32,838

front of a computer as well.

77

00:05:32,838 --> 00:05:41,278

But similarly, I got into this notion of

understanding glaciers from a

78

00:05:41,278 --> 00:05:43,058

mountaineering type perspective.

79

00:05:43,058 --> 00:05:47,738

That's what I was interested in and got

into geosciences from there and then took

80

00:05:47,738 --> 00:05:51,086

this sort of roundabout way back to

81

00:05:51,086 --> 00:05:58,766

computers by sort of slowly recognizing

that they were a really helpful tool for

82

00:05:58,766 --> 00:06:02,866

trying to understand what was happening

with these systems.

83

00:06:03,646 --> 00:06:05,206

They definitely are.

84

00:06:05,206 --> 00:06:09,046

I remember that's personally how I ended

up working on stats.

85

00:06:09,846 --> 00:06:18,386

Ironically, I wasn't a big fan of stats

when I was in college.

86

00:06:18,946 --> 00:06:20,366

I loved math.

87

00:06:20,366 --> 00:06:25,506

and algebra and stuff like that but stats

I didn't like that because it was you know

88

00:06:25,506 --> 00:06:28,426

we were doing a lot of pen and paper

computations so I was like I don't

89

00:06:28,426 --> 00:06:33,266

understand like it's just I'm bad at

computing personally so I don't know why

90

00:06:33,266 --> 00:06:38,746

computers don't do that you know and and

then afterwards randomly I I started

91

00:06:38,746 --> 00:06:41,946

working on electoral forecasting and

discovered you could simulate

92

00:06:41,946 --> 00:06:47,150

distributions with the computer and the

computer was doing all the tedious

93

00:06:47,150 --> 00:06:51,069

error -prone and boring work that I used

to not like at all.

94

00:06:51,069 --> 00:06:55,290

And then I could just focus on, okay,

thinking about the model structure and

95

00:06:55,290 --> 00:07:01,270

making sure the model made sense, what we

can say with it, what the model cannot

96

00:07:01,270 --> 00:07:03,570

tell you also, things like that.

97

00:07:03,570 --> 00:07:06,170

That was definitely super interesting.

98

00:07:06,170 --> 00:07:12,350

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

working on stats, ironically.

99

00:07:13,130 --> 00:07:15,730

I had a similar path.

100

00:07:15,730 --> 00:07:16,270

I didn't...

101

00:07:16,270 --> 00:07:24,130

take a stats class until I was in my PhD

and watched Stan or one of these other

102

00:07:24,130 --> 00:07:29,910

MCMC packages work to answer some really

interesting questions that you couldn't do

103

00:07:29,910 --> 00:07:34,430

with the type of stats that people told

you about when you were in high school.

104

00:07:34,790 --> 00:07:39,450

And that became much more intriguing to me

after seeing it applied to ecological

105

00:07:39,450 --> 00:07:43,830

models or election forecasting or any of

these things that you need a computer to

106

00:07:43,830 --> 00:07:45,732

assist with inferences for.

107

00:07:45,998 --> 00:07:53,218

Yeah, for me, taking a stats class as an

undergrad student in the first or second

108

00:07:53,218 --> 00:07:55,558

year, I had the impression that.

109

00:07:57,134 --> 00:08:03,834

the stats department took great pride into

making the class as inaccessible as

110

00:08:03,834 --> 00:08:10,614

possible and just go through like theorems

and proofs and try to avoid like any

111

00:08:10,614 --> 00:08:14,314

connection to the real world, trying to

make it useful for us.

112

00:08:14,314 --> 00:08:21,514

And I also got like really later into it

through Doc mainly, where I thought like,

113

00:08:21,514 --> 00:08:22,904

you know, this kind of makes sense.

114

00:08:22,904 --> 00:08:24,629

That's a good method.

115

00:08:25,005 --> 00:08:27,765

to use to answer a problem I care about.

116

00:08:27,765 --> 00:08:32,845

And before that, we were just giving

hypothetical problems that I had no

117

00:08:32,845 --> 00:08:34,225

connection to.

118

00:08:34,525 --> 00:08:34,645

Yeah.

119

00:08:34,645 --> 00:08:35,785

Yeah.

120

00:08:35,785 --> 00:08:37,155

Yeah, definitely makes sense.

121

00:08:37,155 --> 00:08:39,265

And I resonate with that a lot.

122

00:08:39,785 --> 00:08:44,405

And so today, what are the topics you

focus on?

123

00:08:44,405 --> 00:08:49,245

Are you both working on the same topics or

are you working on slightly different or

124

00:08:49,245 --> 00:08:51,465

completely different topics of your field?

125

00:08:51,465 --> 00:08:54,861

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

I really enjoyed this episode,

126

00:08:54,861 --> 00:08:59,601

I really don't know a lot about classology

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

127

00:08:59,601 --> 00:08:59,981

to be awesome.

128

00:08:59,981 --> 00:09:01,413

I'm going to learn a lot.

129

00:09:03,885 --> 00:09:07,825

Yeah, well, we work together a lot.

130

00:09:08,065 --> 00:09:13,485

We both have our own independent projects,

but I think we work together a lot.

131

00:09:13,485 --> 00:09:18,245

And I would say that you can tell me if

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

132

00:09:18,245 --> 00:09:24,865

would characterize the work that we both

do separately and together as trying to

133

00:09:24,865 --> 00:09:32,205

make glacier evolution forecasts that

actually agree in a meaningful way.

134

00:09:32,205 --> 00:09:36,905

with the observations that exist out there

in the world.

135

00:09:37,205 --> 00:09:41,765

And that sounds sort of like an obvious

thing to do.

136

00:09:41,765 --> 00:09:47,085

Like, yeah, if you have a model of glacier

motion that maybe you use to predict sea

137

00:09:47,085 --> 00:09:51,865

level rise or something like that, like it

ought to agree with the measurements that

138

00:09:51,865 --> 00:09:55,825

people have taken, those people that are

jumping around on the glacier that Andy

139

00:09:55,825 --> 00:09:56,985

mentioned before.

140

00:09:56,985 --> 00:10:01,069

But for a long time, and...

141

00:10:01,069 --> 00:10:04,949

Perhaps now as well, that hasn't been the

case.

142

00:10:04,949 --> 00:10:11,209

And so we're working to make our models

and reality agree as much as possible.

143

00:10:13,169 --> 00:10:14,115

Andy?

144

00:10:15,661 --> 00:10:19,481

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

145

00:10:19,481 --> 00:10:21,317

we do similar things, but...

146

00:10:23,533 --> 00:10:29,393

I see this as a symbiotic relationship

where our independent strengths

147

00:10:31,053 --> 00:10:32,709

taken together, Meg.

148

00:10:34,797 --> 00:10:36,777

I need to rephrase that.

149

00:10:37,777 --> 00:10:44,293

I think the sum of our strength has led to

some...

150

00:10:47,277 --> 00:10:53,977

ways of thinking and breakthroughs that we

may not have done just on our own.

151

00:10:53,977 --> 00:10:57,057

Sorry, that was not a good way of phrasing

it.

152

00:10:57,977 --> 00:11:01,677

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

153

00:11:05,453 --> 00:11:11,633

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

bit more on like model development, on

154

00:11:11,633 --> 00:11:13,983

development of ice flow models.

155

00:11:15,757 --> 00:11:21,977

And as Doc said, we want to make them

agree with observations as good as we can

156

00:11:21,977 --> 00:11:25,277

within observational uncertainties.

157

00:11:25,517 --> 00:11:31,017

And I didn't have the background in

statistics to make that happen, whereas

158

00:11:31,017 --> 00:11:37,517

Doc has both the insight into like how ice

flows and the modeling aspect, but he also

159

00:11:37,517 --> 00:11:41,097

has a much deeper understanding of

statistics in general and patient

160

00:11:41,097 --> 00:11:43,617

statistics in

161

00:11:44,717 --> 00:11:53,057

in particular and we had a lot of

conversations trying to converge on an

162

00:11:53,057 --> 00:11:57,297

approach to make that happen in a

meaningful way.

163

00:11:57,937 --> 00:12:03,517

Because these days if you go and skim

through our literature, almost every third

164

00:12:03,517 --> 00:12:11,357

paper somehow somewhere mentions machine

learning or artificial intelligence or

165

00:12:11,357 --> 00:12:12,257

something.

166

00:12:12,397 --> 00:12:13,557

It's just a buzzword.

167

00:12:13,557 --> 00:12:15,057

It's a big hype.

168

00:12:15,057 --> 00:12:21,417

Most of the time, if you dig deeper, all

you'll find is people do some multilinear

169

00:12:21,417 --> 00:12:24,457

regression and call it machine learning.

170

00:12:25,037 --> 00:12:26,087

That's the best case.

171

00:12:26,087 --> 00:12:31,973

In some cases, I think methods are being

used in places where...

172

00:12:33,453 --> 00:12:37,293

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

demonstrate that this is the right place

173

00:12:37,293 --> 00:12:38,753

to use those methods.

174

00:12:38,753 --> 00:12:45,673

And we are trying to spend time to figure

out where can we use these modern machine

175

00:12:45,673 --> 00:12:51,013

learning methods in a meaningful way that

actually drive science and help us answer

176

00:12:51,013 --> 00:12:52,973

real world questions.

177

00:12:53,453 --> 00:12:55,693

Yeah, yeah, yeah, very good points.

178

00:12:55,693 --> 00:13:01,677

And something I've seen also in my

experience is that, well, the kind of

179

00:13:01,677 --> 00:13:07,397

models and methods you can use is also

determined by the quality and reliability

180

00:13:07,397 --> 00:13:08,757

of your data.

181

00:13:09,477 --> 00:13:17,157

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

give us an idea of what does data look

182

00:13:17,157 --> 00:13:18,136

like in your field?

183

00:13:18,136 --> 00:13:20,797

How big, how reliable are they?

184

00:13:20,797 --> 00:13:25,317

And I think that's going to set us up

nicely to talk about modeling afterwards.

185

00:13:26,437 --> 00:13:27,017

Sure.

186

00:13:27,017 --> 00:13:31,269

So to figure out, you know, how much

187

00:13:31,373 --> 00:13:33,473

a glacier and ice sheet is gonna melt.

188

00:13:33,473 --> 00:13:36,213

There are a few things you need to know.

189

00:13:36,513 --> 00:13:41,733

If you think about it in terms of partial

differential equations, you need initial

190

00:13:41,733 --> 00:13:46,293

conditions and boundary conditions to

solve those equations.

191

00:13:46,713 --> 00:13:57,133

But you also have processes besides those

PDEs that are a surrogate for physics that

192

00:13:57,133 --> 00:13:59,193

we don't understand yet.

193

00:13:59,193 --> 00:14:01,261

So those have parameters.

194

00:14:02,221 --> 00:14:09,031

Often we don't know the values of those

parameters very well.

195

00:14:09,031 --> 00:14:10,373

So we come in with.

196

00:14:13,069 --> 00:14:15,601

a lot of different uncertainties.

197

00:14:18,669 --> 00:14:21,849

Now I forgot what I meant to say.

198

00:14:22,009 --> 00:14:24,049

Sorry, can you repeat the question?

199

00:14:24,049 --> 00:14:31,429

Yeah, I was just asking you how, like what

the typical data look like in your field.

200

00:14:31,429 --> 00:14:32,239

How big are they?

201

00:14:32,239 --> 00:14:34,409

How reliable are they?

202

00:14:34,589 --> 00:14:40,089

And that's usually very important to

understand then what you guys apply as

203

00:14:40,089 --> 00:14:40,741

models.

204

00:14:42,701 --> 00:14:44,035

Yeah, of course.

205

00:14:45,901 --> 00:14:49,701

So one, if you look at the different

conservation equations that we're trying

206

00:14:49,701 --> 00:14:57,261

to solve, conservation of mass, momentum

and energy, for solving conservation of

207

00:14:57,261 --> 00:15:01,361

mass, we need to know the shape of the

glacier, the geometry.

208

00:15:01,921 --> 00:15:06,581

Now, with modern satellites and airplanes,

it's relatively easy to measure the

209

00:15:06,581 --> 00:15:08,629

surface of the ice.

210

00:15:10,221 --> 00:15:15,621

relatively accurately and we can construct

accurate digital elevation models out of

211

00:15:15,621 --> 00:15:16,181

that.

212

00:15:16,181 --> 00:15:23,361

The tricky part is trying to figure out

how thick the ice is, for which we need

213

00:15:23,361 --> 00:15:26,461

grout penetrating radar or seismic

methods.

214

00:15:26,461 --> 00:15:29,541

All of them have large uncertainties.

215

00:15:30,281 --> 00:15:34,741

Doing radar right now cannot be done from

space.

216

00:15:34,841 --> 00:15:37,613

So to figure out the thickness,

217

00:15:37,613 --> 00:15:41,853

at every point in the Greenland ice sheet

or Antarctic ice sheet basically requires

218

00:15:41,853 --> 00:15:44,413

you to fly a plane.

219

00:15:44,833 --> 00:15:48,033

And that's a lot of effort and of course

costs a lot of money.

220

00:15:48,033 --> 00:15:51,293

So you can only do that in targeted areas.

221

00:15:51,293 --> 00:15:56,913

And in the last 10 years, colleagues have

developed methods trying to combine those

222

00:15:56,913 --> 00:16:04,593

observations from our ground penetrating

radar with what we understand how ice

223

00:16:04,593 --> 00:16:07,205

flows, that it, you know,

224

00:16:07,309 --> 00:16:13,769

obeys the laws of physics and conservation

of mass to come up with smarter way to

225

00:16:13,769 --> 00:16:19,063

interpolate your data beyond just doing

creaking.

226

00:16:21,933 --> 00:16:26,233

Now, ice thickness, I'm mentioning that

first because that is the most important

227

00:16:26,233 --> 00:16:26,893

thing.

228

00:16:26,893 --> 00:16:30,433

It defines how the ice flows.

229

00:16:30,433 --> 00:16:32,273

It defines the surface gradient.

230

00:16:32,273 --> 00:16:38,713

And at the end of the day, ice more or

less flows downhill against gravity.

231

00:16:38,713 --> 00:16:42,405

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

232

00:16:44,365 --> 00:16:46,805

you're off to a really bad start.

233

00:16:47,645 --> 00:16:51,865

So reliable ice thickness measurements are

key.

234

00:16:52,965 --> 00:16:57,095

We've made a lot of progress in the last

15 years.

235

00:16:57,095 --> 00:17:03,205

NASA spent approximately $100 million for

a project called Operation IceBridge,

236

00:17:03,205 --> 00:17:06,665

which among other things measured ice

thickness and that just flew over

237

00:17:06,665 --> 00:17:10,705

Greenland every spring for multiple weeks.

238

00:17:11,025 --> 00:17:14,125

And that has given us a much more detailed

picture.

239

00:17:14,125 --> 00:17:19,565

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

how fast it flows.

240

00:17:19,805 --> 00:17:24,725

And you can show that if you use these

newer data set compared to older ice

241

00:17:24,725 --> 00:17:28,785

thickness data sets, that the models are

getting substantially better.

242

00:17:28,885 --> 00:17:34,205

And it also gives us an avenue to test

whenever they add more observations, is

243

00:17:34,205 --> 00:17:37,425

the model getting better or better and

better?

244

00:17:37,445 --> 00:17:41,385

You can go look into individual glaciers

and you may see the model is still

245

00:17:41,385 --> 00:17:42,425

performing poorly.

246

00:17:42,425 --> 00:17:43,853

And,

247

00:17:43,853 --> 00:17:47,113

you may find, well, there is not much data

there.

248

00:17:47,873 --> 00:17:53,753

So hopefully at some point someone goes

out and can fly that glacier.

249

00:17:53,753 --> 00:17:58,753

So this is the main uncertainty that we're

still struggling with despite like 10, 15

250

00:17:58,753 --> 00:18:00,193

years of effort.

251

00:18:00,773 --> 00:18:08,353

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

very important and it's really hard to

252

00:18:08,353 --> 00:18:13,611

quantify the uncertainties is the climate

forcing.

253

00:18:14,829 --> 00:18:20,309

So in order to predict how a glacier flows

and how much it melts, you need to know

254

00:18:20,309 --> 00:18:22,669

how much it snows.

255

00:18:22,809 --> 00:18:26,949

And this is a tough topic.

256

00:18:27,409 --> 00:18:34,209

Both Green and Antarctica are very large,

but they can vary topography over short

257

00:18:34,209 --> 00:18:39,109

scales, which requires high resolution

climate models.

258

00:18:39,549 --> 00:18:43,693

They are expensive, a lot more expensive

to run.

259

00:18:43,693 --> 00:18:46,513

than an ice sheet model these days.

260

00:18:46,513 --> 00:18:51,632

So they can usually do like one simulation

of the past 40 years and that's it.

261

00:18:51,632 --> 00:18:56,971

There is basically no uncertainty

quantification that they do.

262

00:18:58,573 --> 00:19:00,813

up to maybe recently or right now.

263

00:19:00,813 --> 00:19:04,693

I think with machine learning, things may

start to change there too.

264

00:19:05,393 --> 00:19:13,573

So we have products from observations

assimilated into those climate models, but

265

00:19:13,573 --> 00:19:18,913

we often don't know how certain or

uncertain they are because what we have is

266

00:19:18,913 --> 00:19:20,553

spot measurements.

267

00:19:20,553 --> 00:19:24,513

There might be a couple hundred spot

measurements in Greenland or Antarctica

268

00:19:24,513 --> 00:19:27,661

where you can calibrate or validate.

269

00:19:27,661 --> 00:19:28,881

your climate model.

270

00:19:28,881 --> 00:19:32,681

So that's a big uncertainty.

271

00:19:32,821 --> 00:19:34,751

And I've been speaking for a long time.

272

00:19:34,751 --> 00:19:39,221

Maybe Doug wants to chime in and add

something to it.

273

00:19:39,221 --> 00:19:39,771

Yeah.

274

00:19:39,771 --> 00:19:43,841

I mean, sometimes I think when you're

working with these really large couple of

275

00:19:43,841 --> 00:19:50,481

geophysical systems, it can be the line

between model result and data product

276

00:19:50,481 --> 00:19:52,721

becomes a little bit blurred.

277

00:19:53,721 --> 00:19:55,861

So what do we have?

278

00:19:55,861 --> 00:19:56,965

We have...

279

00:19:57,133 --> 00:20:03,193

Direct surface measurements from a variety

of sources maybe over the past 30 years

280

00:20:03,193 --> 00:20:07,953

with varying degrees of spatial

resolution, like Andy said.

281

00:20:07,953 --> 00:20:10,833

It's gotten a lot better in the satellite

era, of course.

282

00:20:10,833 --> 00:20:15,953

We've got these sparse measurements of

thickness that we don't completely

283

00:20:15,953 --> 00:20:19,703

understand the uncertainties for, but

they're pretty accurate.

284

00:20:19,703 --> 00:20:23,213

But they are certainly not everywhere.

285

00:20:24,173 --> 00:20:28,433

with respect to the total area of

glaciated ice on Earth.

286

00:20:28,893 --> 00:20:30,103

What else do we have?

287

00:20:30,103 --> 00:20:35,273

Yeah, we've got a couple snow pit

measurements or shortwave radar that can

288

00:20:35,273 --> 00:20:39,873

measure snow accumulation over a few

places on Earth.

289

00:20:40,593 --> 00:20:46,653

We have optical satellite observations

that can often be leveraged into

290

00:20:46,653 --> 00:20:50,653

understanding the displacement of the

glacier surface.

291

00:20:53,037 --> 00:21:01,957

And there are a couple other somewhat more

esoteric products that we can come up with

292

00:21:01,957 --> 00:21:05,497

hypotheses about how we might use to

constrain glacial ice flow, but we haven't

293

00:21:05,497 --> 00:21:10,717

quite gotten there yet, like the

distribution of dust layers and stuff like

294

00:21:10,717 --> 00:21:13,677

that inside of the ice that you can also

back out from some of these radar

295

00:21:13,677 --> 00:21:14,977

observations.

296

00:21:15,197 --> 00:21:21,645

But taken together, these observations,

the data that we have,

297

00:21:21,645 --> 00:21:26,425

Occupy large amounts of space on a hard

drive in in that sense.

298

00:21:26,425 --> 00:21:33,245

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

individual measurements out there but sort

299

00:21:33,245 --> 00:21:39,045

of relative to the magnitude of the system

that we're looking at and the timeframes

300

00:21:39,045 --> 00:21:44,385

over which we would really like to

Constrain their behavior.

301

00:21:44,385 --> 00:21:47,065

The data is super small.

302

00:21:47,065 --> 00:21:47,855

Okay.

303

00:21:47,855 --> 00:21:48,255

Okay.

304

00:21:48,255 --> 00:21:48,625

I see.

305

00:21:48,625 --> 00:21:51,597

Yeah Thanks guys super

306

00:21:51,597 --> 00:21:56,757

I think super important to set up that

background, that context.

307

00:21:56,917 --> 00:22:02,897

Actually, Doug, you're the patient

statistician of the couple, if I

308

00:22:02,897 --> 00:22:04,577

understood correctly.

309

00:22:06,597 --> 00:22:15,257

Then can you tell us why would patient

statistics be interesting in this context?

310

00:22:16,897 --> 00:22:18,537

Let's start with that.

311

00:22:18,537 --> 00:22:20,973

What would patient statistics?

312

00:22:20,973 --> 00:22:25,033

ring in this context, in this approach of

studying glaciers.

313

00:22:25,113 --> 00:22:26,253

Yeah, sure.

314

00:22:26,253 --> 00:22:27,333

So...

315

00:22:29,741 --> 00:22:31,571

I, yeah, okay.

316

00:22:31,571 --> 00:22:37,721

So I kind of think that most scientific

problems can be cast in a probabilistic

317

00:22:37,721 --> 00:22:38,291

way.

318

00:22:38,291 --> 00:22:44,381

And this is certainly true for

glaciological modeling, where what you

319

00:22:44,381 --> 00:22:49,361

want to do at the end of the day is to

take some assumption that you have about

320

00:22:49,361 --> 00:22:50,821

the way that the world works, right?

321

00:22:50,821 --> 00:22:51,861

A model.

322

00:22:51,861 --> 00:22:56,421

And you want to use that model and you

want to make a prediction about the future

323

00:22:56,421 --> 00:22:59,715

or something that you haven't observed.

324

00:22:59,917 --> 00:23:06,197

But you would also like to ingest all of

the information that you have collected

325

00:23:06,197 --> 00:23:10,497

about the world into that model so that

everything ends up remaining self

326

00:23:10,497 --> 00:23:12,477

-consistent.

327

00:23:12,857 --> 00:23:19,737

And that ends up being a really helpful

paradigm in which to operate for

328

00:23:19,737 --> 00:23:20,797

glaciology.

329

00:23:20,797 --> 00:23:25,057

So typically, you know, the large -scale

goal and what everybody begins their

330

00:23:25,057 --> 00:23:28,377

proposals and papers and stuff with is

like, glaciers are important for

331

00:23:28,377 --> 00:23:29,965

predicting sea level rise.

332

00:23:29,965 --> 00:23:35,405

And to predict sea level rise, what we

need to do is we need to take an ice sheet

333

00:23:35,405 --> 00:23:40,565

model, ice physics model, and project it,

run it into the future, say 200 years or

334

00:23:40,565 --> 00:23:43,565

something like that, and say, well, there

was this much ice to start with, there's

335

00:23:43,565 --> 00:23:49,785

this much ice now, that difference is

gonna turn into sea level rise.

336

00:23:50,125 --> 00:23:53,205

So that's one part of it.

337

00:23:53,485 --> 00:23:59,469

We don't have enough information about how

these systems work to just make

338

00:23:59,469 --> 00:24:01,639

one prediction, right?

339

00:24:01,639 --> 00:24:05,469

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

of places like Andy was saying.

340

00:24:05,889 --> 00:24:10,489

And so the sensible approach to dealing

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

341

00:24:10,489 --> 00:24:17,029

probability distribution over the bed and

let's sample from that probability

342

00:24:17,029 --> 00:24:20,289

distribution and make a whole lot of

different predictions about what sea level

343

00:24:20,289 --> 00:24:25,429

rise is going to be based on all of those

different potential realizations of how

344

00:24:25,429 --> 00:24:27,853

the bed of the glacier might look.

345

00:24:27,853 --> 00:24:30,523

And of course, it's not just the bed

that's uncertain.

346

00:24:30,523 --> 00:24:32,993

There's a bunch of other stuff as well.

347

00:24:33,333 --> 00:24:39,093

And so that's a very Bayesian way of

looking at probability, right?

348

00:24:39,093 --> 00:24:43,353

I mean, you can't hardly escape the

Bayesian paradigm in geophysics, right?

349

00:24:43,353 --> 00:24:47,573

Because we don't have the capacity for

repeat samples.

350

00:24:48,253 --> 00:24:52,733

All we have is just the one data point,

right?

351

00:24:52,733 --> 00:24:54,853

So no replicates here.

352

00:24:54,913 --> 00:24:57,013

No limiting behavior.

353

00:24:57,933 --> 00:25:00,893

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

of ensemble modeling.

354

00:25:00,893 --> 00:25:05,133

That's what we would call that this notion

of randomly sampling from potential model

355

00:25:05,133 --> 00:25:06,433

inputs and running into the future.

356

00:25:06,433 --> 00:25:08,633

That's a super Bayesian idea to begin

with.

357

00:25:08,633 --> 00:25:14,093

And then the other sort of step in this

process is to say, okay, well, I actually

358

00:25:14,093 --> 00:25:18,913

want to constrain what I think the bet is

based on these observations that I have,

359

00:25:18,913 --> 00:25:25,273

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

big pie in the sky view over of what my

360

00:25:25,273 --> 00:25:27,959

bet elevation could be, maybe something.

361

00:25:28,525 --> 00:25:33,725

between 5 ,000 meters above sea level and

10 ,000 meters below.

362

00:25:33,725 --> 00:25:38,225

But then I'm going to take all of these

radar observations that I have and whittle

363

00:25:38,225 --> 00:25:44,535

down the space of possible ways that the

bed could be.

364

00:25:44,535 --> 00:25:51,205

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

posterior inference, right?

365

00:25:51,205 --> 00:25:53,205

Yeah, yeah.

366

00:25:53,205 --> 00:25:54,765

Yeah, for sure.

367

00:25:55,045 --> 00:25:57,645

Thanks to SuperClean.

368

00:25:57,773 --> 00:26:01,253

Maybe a question for the both of you.

369

00:26:01,893 --> 00:26:11,093

Do you have a favorite study or project

where the collaboration between glaciology

370

00:26:11,093 --> 00:26:15,873

and Bayesian stance led to interesting

insights?

371

00:26:16,593 --> 00:26:23,773

And yeah, a study that you particularly

like, whether that's one of yours or a

372

00:26:23,773 --> 00:26:26,327

stunning glaciology from someone else.

373

00:26:30,285 --> 00:26:31,567

What do you think, Andy?

374

00:26:39,597 --> 00:26:40,605

Yeah.

375

00:26:42,669 --> 00:26:49,069

I think as Doug alluded earlier, combining

Bayesian methods with the idea of large

376

00:26:49,069 --> 00:26:56,549

ensembles, thanks to having access to

large high -performance computer systems,

377

00:26:57,289 --> 00:27:04,229

have allowed us for the first time to

investigate the parameter space in a

378

00:27:04,229 --> 00:27:05,039

meaningful way.

379

00:27:05,039 --> 00:27:06,309

Before that,

380

00:27:07,821 --> 00:27:16,101

you would basically hand tune most of what

you did was based on expert judgment.

381

00:27:16,581 --> 00:27:21,801

Like your prior was what you've learned

over the past 10 years, so to speak.

382

00:27:22,721 --> 00:27:24,500

And surprisingly,

383

00:27:26,253 --> 00:27:33,853

Calibration by eyeballing can yield pretty

good results, but it only gives you a

384

00:27:33,853 --> 00:27:39,853

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

any information about the tails.

385

00:27:41,093 --> 00:27:49,833

So, for years, we would publish one study,

a mean of one simulation, maybe a few

386

00:27:49,833 --> 00:27:55,561

simulations, but we didn't look at the

distributions themselves.

387

00:27:55,853 --> 00:28:06,053

and bringing the Bayesian methods into our

field, I think have led to a great deal of

388

00:28:07,917 --> 00:28:14,477

to have led us to discover an

uncomfortable truth that those tails are

389

00:28:14,477 --> 00:28:20,157

really large and they are not normally

distributed.

390

00:28:20,517 --> 00:28:22,053

So ...

391

00:28:24,941 --> 00:28:30,681

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

to understand the tails and understand the

392

00:28:30,681 --> 00:28:38,701

full distribution and not just a mean or a

median or any single point realization of

393

00:28:38,701 --> 00:28:39,429

that.

394

00:28:41,101 --> 00:28:43,961

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

point.

395

00:28:43,961 --> 00:28:51,321

And that reminds me of a study that we

didn't do that I think is really good.

396

00:28:51,401 --> 00:28:58,321

But it merits maybe just explicitly

stating something about glaciological

397

00:28:58,321 --> 00:29:04,401

systems, particularly the ice sheets,

which is that ice flow and in particular

398

00:29:04,401 --> 00:29:06,285

the mechanisms of.

399

00:29:06,285 --> 00:29:11,805

Retreat so the potential for you know

Antarctica or Greenland in some sense to

400

00:29:11,805 --> 00:29:15,645

collapse and not be ice I see anymore to

become ice -free.

401

00:29:15,645 --> 00:29:24,905

That's a super nonlinear process in the

sense that If if say we get the bed wrong

402

00:29:24,905 --> 00:29:34,565

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

were to imagine that the bed is Shallower

403

00:29:34,565 --> 00:29:36,397

than it actually is

404

00:29:36,397 --> 00:29:43,997

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

if the bed is actually shallower than we

405

00:29:43,997 --> 00:29:49,277

think it is, then that doesn't really have

that many implications for sea level

406

00:29:49,277 --> 00:29:49,687

change.

407

00:29:49,687 --> 00:29:54,097

If the things change as normal, if the bed

is, it just melts away.

408

00:29:54,097 --> 00:29:58,257

If the bed is a lot deeper than we think

it is, then all of a sudden you have the

409

00:29:58,257 --> 00:30:03,197

potential for the entire ice sheet to

float and physically disintegrate via

410

00:30:03,197 --> 00:30:03,661

like,

411

00:30:03,661 --> 00:30:07,481

the dramatic sort of calving processes

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

412

00:30:07,481 --> 00:30:12,261

movie Chasing Ice or one of these other

sort of documentaries.

413

00:30:12,641 --> 00:30:20,461

And so the consequences of being wrong are

asymmetric with respect to some of these

414

00:30:20,461 --> 00:30:25,351

unknown factors that govern the system.

415

00:30:25,351 --> 00:30:27,947

And there's a really wonderful paper.

416

00:30:28,205 --> 00:30:33,325

that shows this quite explicitly by a

colleague of ours named Alex Roebol, who

417

00:30:33,325 --> 00:30:39,685

basically just took a simple model of

Antarctica, forced it with sort of

418

00:30:39,685 --> 00:30:47,525

normally distributed melting noise, more

or less, and a bunch of different

419

00:30:47,525 --> 00:30:54,785

scenarios, and showed this really big

systematic bias towards more mass loss on

420

00:30:54,785 --> 00:30:57,421

account of the fundamental

421

00:30:57,421 --> 00:31:03,401

asymmetry in the way that these

glaciological systems respond to errors in

422

00:31:03,401 --> 00:31:04,545

input data.

423

00:31:06,893 --> 00:31:09,793

Yeah, that just sounds very fascinating.

424

00:31:09,793 --> 00:31:12,913

I'm super curious to see one of these

models.

425

00:31:13,213 --> 00:31:20,413

Do you know if there are any open source

packages that, for instance, people

426

00:31:20,413 --> 00:31:28,253

working in your field are using in Python

or in R that kind of wrap the usual models

427

00:31:28,253 --> 00:31:30,113

you guys are working on?

428

00:31:30,113 --> 00:31:35,021

And also, is there any cool data sets that

we can put in the show notes for?

429

00:31:35,021 --> 00:31:37,981

people to look around if they want to.

430

00:31:38,001 --> 00:31:42,761

Any interesting applications that you

think would be interesting, let's put that

431

00:31:42,761 --> 00:31:43,859

in the show notes.

432

00:31:46,637 --> 00:31:51,977

You made some super cool visualizations

for one of those papers a while ago,

433

00:31:51,977 --> 00:31:52,897

didn't you Andy?

434

00:31:52,897 --> 00:31:58,597

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

I'll send you the link.

435

00:31:58,657 --> 00:32:03,917

I think one of our earlier collaborations

where we started exploring the idea of

436

00:32:03,917 --> 00:32:12,117

large ensembles was funded by NASA and

with support from NASA, they helped us

437

00:32:12,117 --> 00:32:13,627

visualizing.

438

00:32:13,709 --> 00:32:18,689

our simulations on their big screens and

narrating it.

439

00:32:18,689 --> 00:32:19,719

I'll send you a link.

440

00:32:19,719 --> 00:32:23,829

That's all open and open source.

441

00:32:24,649 --> 00:32:32,909

With regard to packages, most of those

models that we develop are kind of big

442

00:32:32,909 --> 00:32:33,989

beasts.

443

00:32:34,729 --> 00:32:36,829

It takes a while to learn them.

444

00:32:36,829 --> 00:32:39,205

Right now, there are very few.

445

00:32:41,325 --> 00:32:43,585

wrappers around it in Python.

446

00:32:43,585 --> 00:32:50,485

The model we developed, you can access

stuff through Python, but we're not at the

447

00:32:50,485 --> 00:32:53,605

level to use it as a black box.

448

00:32:53,765 --> 00:32:57,604

Whether you should be able to use it as a

black box is a different question.

449

00:32:57,604 --> 00:33:05,405

But we have a fund a project from the

National Science Foundation that drives us

450

00:33:05,405 --> 00:33:09,517

towards that goal of reducing the barrier

of entry.

451

00:33:09,517 --> 00:33:16,407

and reducing the time to actually do

science by taking steps like this.

452

00:33:16,407 --> 00:33:22,977

So in the next couple of years, our group

and others are working towards a cloud

453

00:33:22,977 --> 00:33:30,057

version of the model that ideally can just

be deployed with the click of a mouse.

454

00:33:30,397 --> 00:33:35,617

And, you know, you, for example, choose

the parameters you are interested in in

455

00:33:35,617 --> 00:33:38,925

your uncertainty quantification.

456

00:33:38,925 --> 00:33:42,305

and the rest is done automatically.

457

00:33:42,305 --> 00:33:47,845

Right now you do need inside knowledge on

HPC systems.

458

00:33:47,985 --> 00:33:50,185

Each HPC system is different.

459

00:33:50,185 --> 00:33:55,885

It can take days or weeks just to get the

model to run because each system has a

460

00:33:55,885 --> 00:33:59,035

different MPI stack, different compilers.

461

00:33:59,035 --> 00:34:01,065

You can run into all sorts of problems.

462

00:34:01,065 --> 00:34:03,125

So that's just one step.

463

00:34:04,505 --> 00:34:08,899

So we are trying to make that easier, but

we are not there yet.

464

00:34:09,517 --> 00:34:17,837

I'll give you an anecdote, which is that

Andy has made a lot of progress utilizing

465

00:34:17,837 --> 00:34:24,257

a very large computational fluid dynamics

code for ice sheet flow called the

466

00:34:24,257 --> 00:34:30,037

parallel ice sheet model, which is

wonderful and super carefully constructed

467

00:34:30,037 --> 00:34:33,937

and really a great piece of software.

468

00:34:33,937 --> 00:34:37,257

But man, I don't have the attention span

to figure out how to learn it.

469

00:34:37,257 --> 00:34:39,213

And so for a lot of the...

470

00:34:39,213 --> 00:34:48,793

A lot of the real Bayesian computation

stuff that we've done, I got tired and

471

00:34:48,793 --> 00:34:55,013

just made Andy run a large ensemble and

then we train a neural network to pretend

472

00:34:55,013 --> 00:34:59,251

to be PISM and we'll sometimes work with

that instead.

473

00:35:01,037 --> 00:35:03,117

Well, that sounds like fun too.

474

00:35:03,117 --> 00:35:04,197

Yeah, and actually...

475

00:35:04,197 --> 00:35:04,957

That's the future.

476

00:35:05,137 --> 00:35:05,937

Yeah.

477

00:35:05,937 --> 00:35:07,757

Yeah, go ahead, Andy.

478

00:35:07,757 --> 00:35:14,697

That's what we're still working on and

what I envision to push a bit further in

479

00:35:14,697 --> 00:35:17,057

the next couple of years as well.

480

00:35:17,377 --> 00:35:18,117

Okay.

481

00:35:18,117 --> 00:35:21,517

Yeah, definitely super, super fascinating.

482

00:35:22,057 --> 00:35:26,857

And yeah, Doug, actually, I wanted to ask

you a bit more about that because you said

483

00:35:26,857 --> 00:35:30,029

you have a background in computer science,

so...

484

00:35:30,029 --> 00:35:35,429

I'm wondering how do we integrate the

Bayesian algorithms into the computational

485

00:35:35,429 --> 00:35:39,409

models that you've talked about for

studying glaciers?

486

00:35:39,429 --> 00:35:42,169

Are you using open source packages?

487

00:35:42,389 --> 00:35:46,489

What does your work look like on that

front?

488

00:35:46,589 --> 00:35:48,109

Yeah, absolutely.

489

00:35:48,509 --> 00:35:56,829

Before I did statistics, I did numerical

methods and I still do a lot of that work.

490

00:35:56,829 --> 00:35:58,797

In particular, I

491

00:35:58,797 --> 00:36:03,917

work in the branch of numerical methods

associated with solving partial

492

00:36:03,917 --> 00:36:11,657

differential equations via the finite

element method, which is, you know,

493

00:36:11,657 --> 00:36:16,017

doesn't really matter how that works, but

there's a really wonderful package for

494

00:36:16,017 --> 00:36:22,445

solving set equations via that method

called FireDrake or

495

00:36:22,445 --> 00:36:26,705

Phoenix, and so it's a really nice open

source Python package that a ton of

496

00:36:26,705 --> 00:36:32,445

scientists are using for all sorts of

different applications in computational

497

00:36:32,445 --> 00:36:33,745

mechanics.

498

00:36:34,105 --> 00:36:39,885

And so I use that for developing sort of

the guts, the dynamical cores, as some

499

00:36:39,885 --> 00:36:42,205

might call them, of these models.

500

00:36:42,805 --> 00:36:51,085

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

allows for a very straightforward

501

00:36:51,085 --> 00:36:57,225

computation of derivatives of the output

of those models with respect to the inputs

502

00:36:57,225 --> 00:37:05,625

of those models, which is super useful for

all sorts of optimization tasks and also

503

00:37:05,625 --> 00:37:13,025

approximation in a Bayesian sense tasks,

MCMC or other approximation methods.

504

00:37:13,025 --> 00:37:18,885

And so my typical workflow now is to take

one of those models and actually wrap it

505

00:37:18,885 --> 00:37:21,399

inside of PyTorch.

506

00:37:21,517 --> 00:37:26,897

which is sort of a general purpose

framework for automatic differentiation

507

00:37:26,897 --> 00:37:29,697

that's popular in the machine learning

community.

508

00:37:29,797 --> 00:37:35,417

And basically what that lets me do is

basically view an ice sheet model as if it

509

00:37:35,417 --> 00:37:38,557

were a function in PyTorch.

510

00:37:38,557 --> 00:37:44,097

And I can put stuff into the model, I can

get stuff out of the model, I can compute

511

00:37:44,097 --> 00:37:49,453

misfits with respect to data between what

the model predicts and what the...

512

00:37:49,453 --> 00:37:57,013

what the data says and basically take

derivatives of that with respect to model

513

00:37:57,013 --> 00:38:00,073

parameters in a very seamless and easy

way.

514

00:38:00,213 --> 00:38:05,793

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

all just mixing and matching various

515

00:38:05,793 --> 00:38:10,437

really awesome open source tools.

516

00:38:12,397 --> 00:38:19,477

Actually, back in the day, when I first

got into this stuff, it was all sort of

517

00:38:19,477 --> 00:38:26,577

making ice sheet model solvers from

scratch in NumPy and then sticking them

518

00:38:26,577 --> 00:38:29,216

into PyMC, which you work on, right?

519

00:38:29,477 --> 00:38:31,197

Yeah, yeah, exactly.

520

00:38:31,237 --> 00:38:33,137

That's why I was also asking.

521

00:38:33,137 --> 00:38:38,657

I was curious if you were using PyMC and

other hood to do that, because it sounds

522

00:38:38,657 --> 00:38:41,933

like it would be an appropriate framework

to...

523

00:38:41,933 --> 00:38:42,733

to use it.

524

00:38:42,733 --> 00:38:44,453

So I was curious.

525

00:38:44,613 --> 00:38:44,713

Yeah.

526

00:38:44,713 --> 00:38:47,023

No, now, well, I would love to.

527

00:38:47,023 --> 00:38:52,393

Nowadays, the problems that we work on

tend to be high dimensional enough that

528

00:38:52,393 --> 00:38:58,363

the MCMC methods generally become very

challenging to work with.

529

00:38:58,363 --> 00:39:04,253

And so we have to do sometimes less good

stuff.

530

00:39:06,513 --> 00:39:11,251

And Andy, how does that look like?

531

00:39:11,469 --> 00:39:14,709

cooperating in these projects, right?

532

00:39:14,709 --> 00:39:15,609

How?

533

00:39:15,609 --> 00:39:18,949

Because you are more on the practical side

of things.

534

00:39:18,949 --> 00:39:24,749

So how do you consume the results of the

model, I'm actually curious.

535

00:39:24,749 --> 00:39:29,669

And because if I understand correctly, you

are intervening before the model, because

536

00:39:29,669 --> 00:39:33,849

I'm guessing you're part of the data

collecting team and you have the domain

537

00:39:33,849 --> 00:39:37,889

knowledge that can be integrated into the

model, if there are priors in the model.

538

00:39:37,889 --> 00:39:41,037

And then afterwards, of course, you're

interpreting...

539

00:39:41,037 --> 00:39:42,837

the results of the model.

540

00:39:43,197 --> 00:39:48,237

But how does that look like to cooperate

with these kind of models and in these

541

00:39:48,237 --> 00:39:49,195

contexts?

542

00:39:50,733 --> 00:39:57,293

Well, the high level view of course is

that when we collaborate, doctors are

543

00:39:57,293 --> 00:40:01,313

thinking and I do the talking or pushing

off the buttons and trying to run the

544

00:40:01,313 --> 00:40:02,453

models.

545

00:40:03,113 --> 00:40:05,313

That would be the simple answer.

546

00:40:07,917 --> 00:40:09,537

A lot of dip.

547

00:40:09,537 --> 00:40:10,959

Workflow.

548

00:40:13,133 --> 00:40:17,029

is still very cumbersome.

549

00:40:18,733 --> 00:40:23,973

So Doug has alluded to the different

methods of collecting data sets, all the

550

00:40:23,973 --> 00:40:29,713

uncertainties associated with them or the

lack of uncertainties with these data

551

00:40:29,713 --> 00:40:30,833

sets.

552

00:40:31,753 --> 00:40:38,153

Things have gotten better, but you can

imagine still each data set, you find it

553

00:40:38,153 --> 00:40:41,313

on a different server with a different way

to access it.

554

00:40:41,313 --> 00:40:43,693

It is probably in a different grid.

555

00:40:43,693 --> 00:40:48,531

It most likely has a different spatial

reference system.

556

00:40:50,285 --> 00:40:57,029

So we are trying to transition from a

state where we spend.

557

00:40:58,605 --> 00:41:07,765

half of our time just trying to come up

with not very robust workflow to get from

558

00:41:07,765 --> 00:41:15,525

the data sets on different servers or

websites to ingesting them into the model

559

00:41:15,525 --> 00:41:19,665

to run the model and then to analyze the

data.

560

00:41:23,181 --> 00:41:30,601

Before we had all that great data, things

were easy and hard at the same time.

561

00:41:30,601 --> 00:41:35,721

All you had were a few data points and you

probably had to write an email to your

562

00:41:35,721 --> 00:41:41,601

colleague asking to get access to the data

point that they may have asked you to be

563

00:41:41,601 --> 00:41:43,961

on your paper in return.

564

00:41:45,181 --> 00:41:50,797

At least now we have traded that for

spending a lot of time trying to find...

565

00:41:50,797 --> 00:41:52,207

figure out those workflows.

566

00:41:52,207 --> 00:41:58,257

And there are lots of initiatives right

now trying to make that workflow easier.

567

00:41:58,617 --> 00:42:00,547

But I don't think we're there.

568

00:42:00,547 --> 00:42:08,297

I still feel like this is sort of half of

my time I'm spending with processing the

569

00:42:08,297 --> 00:42:16,697

data and getting really mad at XRA because

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

570

00:42:16,777 --> 00:42:18,573

It almost always does.

571

00:42:18,573 --> 00:42:24,153

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

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

572

00:42:24,153 --> 00:42:30,033

it's going to be a long afternoon and

sometimes a little bit of yelling too.

573

00:42:30,853 --> 00:42:33,053

I've been there.

574

00:42:33,053 --> 00:42:36,553

I feel like we've had similar afternoons.

575

00:42:38,293 --> 00:42:45,213

But yeah, XRA saves the day most of the

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

576

00:42:45,213 --> 00:42:47,413

to debug for sure.

577

00:42:47,533 --> 00:42:51,793

mainly because there is not a lot of

tutorials on it in my experience.

578

00:42:51,793 --> 00:42:55,893

So you have to figure a lot of these

things on your own.

579

00:42:56,053 --> 00:43:00,913

Yeah, and yeah, I was also curious about

that because on my own also I've been

580

00:43:00,913 --> 00:43:05,733

working with a team of researchers.

581

00:43:05,733 --> 00:43:07,993

So they are marine biologists.

582

00:43:07,993 --> 00:43:09,113

So quite different.

583

00:43:09,113 --> 00:43:16,373

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

water and yeah, basically a study of

584

00:43:16,373 --> 00:43:17,197

trade.

585

00:43:17,197 --> 00:43:22,417

of sharks across the world and that has

been super interesting to work with them

586

00:43:22,417 --> 00:43:28,797

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

the statistical expertise, right?

587

00:43:28,797 --> 00:43:31,717

I have nothing to bring on the shark side

of things.

588

00:43:31,717 --> 00:43:38,977

I've actually learned a lot thanks to them

about sharks and shark trade and things

589

00:43:38,977 --> 00:43:40,737

like that.

590

00:43:41,197 --> 00:43:46,081

And yeah, that to me is also very

interesting because...

591

00:43:46,605 --> 00:43:50,885

the models are getting more and more

intricate.

592

00:43:50,885 --> 00:43:57,185

These are models that now are really hard

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

593

00:43:57,185 --> 00:44:01,245

a statistician already, it's really hard

to come up with that kind of model if

594

00:44:01,245 --> 00:44:04,485

you're really a domain expert.

595

00:44:04,545 --> 00:44:08,145

And at the same time, to develop the

model, you need the domain experts because

596

00:44:08,145 --> 00:44:12,645

otherwise, I could not develop that model

without the domain experts, even though I

597

00:44:12,645 --> 00:44:14,365

know how to code the model.

598

00:44:14,805 --> 00:44:15,813

And...

599

00:44:15,853 --> 00:44:22,793

And I find that also super interesting to

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

600

00:44:22,793 --> 00:44:27,563

also good illustration of what science is,

right?

601

00:44:27,563 --> 00:44:34,753

It's like really the sum is bigger than

each party on its own.

602

00:44:35,833 --> 00:44:41,433

But at the same time, as the statistician,

you know, I'm a bit frustrated because I

603

00:44:41,433 --> 00:44:45,773

know the model, for instance, is not going

to be in the paper, for instance.

604

00:44:45,773 --> 00:44:48,173

The model is going to be the appendix of

the paper.

605

00:44:48,173 --> 00:44:51,513

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

model.

606

00:44:51,513 --> 00:44:54,213

I would definitely focus on that.

607

00:44:54,833 --> 00:45:00,833

But my point is, collaborating with the

domain experts has been also super

608

00:45:00,833 --> 00:45:05,193

interesting because as you were saying,

Andy, there are still some parts of the

609

00:45:05,193 --> 00:45:05,833

workflow.

610

00:45:05,833 --> 00:45:10,013

So on mine, I'm talking about the Bayesian

workflow, which are cleaning, which can

611

00:45:10,013 --> 00:45:13,741

only need to be updated and improved and

working.

612

00:45:13,741 --> 00:45:18,841

like that with people who mainly use the

model and consume it instead of writing it

613

00:45:18,841 --> 00:45:20,961

is super valuable.

614

00:45:21,241 --> 00:45:26,821

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

have stuff to add on that because I'm

615

00:45:26,821 --> 00:45:26,961

listening to you.

616

00:45:26,961 --> 00:45:33,081

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

is going to resonate with anybody that's

617

00:45:33,081 --> 00:45:36,241

trying to work across disciplinary

boundaries, which is, I mean, ultimately

618

00:45:36,241 --> 00:45:39,621

what we need to do across all branches of

science right now, right?

619

00:45:39,621 --> 00:45:42,957

We have all of these amazing statistical

methods and...

620

00:45:42,957 --> 00:45:46,717

numerical methods and also so much

knowledge about the way the structural

621

00:45:46,717 --> 00:45:53,457

assumptions that go into how the world

works and We have to combine those things

622

00:45:53,457 --> 00:46:00,177

to make good progress now, but man if you

if It's very difficult to find a

623

00:46:00,177 --> 00:46:06,877

circumstance in which somebody's really

figured that collaboration out in a in a

624

00:46:06,877 --> 00:46:11,867

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

it's challenging

625

00:46:11,981 --> 00:46:14,741

I agree it's hard.

626

00:46:14,741 --> 00:46:19,781

I've been involved in a bunch of larger

scale projects trying to bring together

627

00:46:19,781 --> 00:46:26,461

data scientists and domain scientists and

it's kind of both parties sort of need to

628

00:46:26,461 --> 00:46:36,541

learn to speak the other parties language

and it especially for the data scientists

629

00:46:36,541 --> 00:46:40,229

it can be a challenge because

630

00:46:40,365 --> 00:46:41,575

you know, let me put it that way.

631

00:46:41,575 --> 00:46:42,735

They have really big hammers.

632

00:46:42,735 --> 00:46:44,785

They have awesome tools.

633

00:46:45,325 --> 00:46:51,265

And we just, you know, in glaciology, we

just started taking baby steps.

634

00:46:51,725 --> 00:46:54,425

So most of these awesome tools we actually

don't need.

635

00:46:54,425 --> 00:47:00,185

We need like what they had in undergrad,

like the most basic neural network or

636

00:47:00,185 --> 00:47:05,845

something like that will already get us

from here to 90%.

637

00:47:05,845 --> 00:47:09,541

So when you collaborate with them, they're

638

00:47:09,901 --> 00:47:11,871

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

639

00:47:11,871 --> 00:47:17,641

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

simple neural network and that will do the

640

00:47:17,641 --> 00:47:18,441

job.

641

00:47:18,441 --> 00:47:26,081

So as Doc said, having being able to

straddle both worlds between the domain

642

00:47:26,081 --> 00:47:32,421

science and the data science is a

challenge and we need more people doing

643

00:47:32,421 --> 00:47:32,711

this.

644

00:47:32,711 --> 00:47:37,001

I think in our field right now, there's

only a handful of people that I would

645

00:47:37,001 --> 00:47:37,887

trust.

646

00:47:37,901 --> 00:47:42,401

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

among them and maybe three or four others.

647

00:47:42,401 --> 00:47:48,421

And I think we need more people who are

capable to, who are bilingual in data

648

00:47:48,421 --> 00:47:52,681

science and in domain science.

649

00:47:53,561 --> 00:48:01,981

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

is that since this is, we're all Bayesian

650

00:48:01,981 --> 00:48:06,893

statistics boosters here, is that Bayes

theorem or maybe more,

651

00:48:06,893 --> 00:48:12,313

more specifically or broadly, the

posterior predictive distribution, if we

652

00:48:12,313 --> 00:48:15,153

can use some technical language for a

second.

653

00:48:15,153 --> 00:48:22,493

It provides an exceptionally useful

blueprint for talking to people across

654

00:48:22,493 --> 00:48:23,913

disciplinary boundaries.

655

00:48:23,913 --> 00:48:29,513

Because I can write this down and I can

say, OK, here are the things, domain

656

00:48:29,513 --> 00:48:31,893

scientists, that I need from you.

657

00:48:31,893 --> 00:48:36,439

I need you to tell me what you want to

predict.

658

00:48:37,453 --> 00:48:42,933

Like in the case of glaciology, that often

ends up being sea level rise or volume

659

00:48:42,933 --> 00:48:43,383

change.

660

00:48:43,383 --> 00:48:45,633

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

661

00:48:45,633 --> 00:48:52,313

I need you to provide to me a set of

structural assumptions that encodes your

662

00:48:52,313 --> 00:48:55,183

best understanding as a domain expert of

how the world works.

663

00:48:55,183 --> 00:48:56,853

That's your numerical model.

664

00:48:56,853 --> 00:48:58,383

It's going to take in some inputs.

665

00:48:58,383 --> 00:49:00,493

It's going to produce some outputs.

666

00:49:00,493 --> 00:49:05,073

I need you to tell me what aspects of that

model you don't feel like you know enough

667

00:49:05,073 --> 00:49:05,955

about.

668

00:49:06,797 --> 00:49:11,357

I need you to tell me what observations

you have available to you.

669

00:49:11,357 --> 00:49:17,497

And then we can put these things all

together in a big flow chart, a graph,

670

00:49:17,577 --> 00:49:18,117

right?

671

00:49:18,117 --> 00:49:23,137

Presumably a directed acyclic graph that

prescribes all of the causal relationships

672

00:49:23,137 --> 00:49:24,437

in the system.

673

00:49:24,437 --> 00:49:29,537

And then once that picture is drawn, me as

a person that understands sort of the

674

00:49:29,537 --> 00:49:33,317

numerical methods, the nuts and bolts of

doing inference and prediction in this

675

00:49:33,317 --> 00:49:34,701

sort of probabilistic framework,

676

00:49:34,701 --> 00:49:39,201

I can take that picture and I can convert

that into code and I can bring to bear the

677

00:49:39,201 --> 00:49:40,841

statistical tools.

678

00:49:41,181 --> 00:49:48,121

So like the Bayesian language of cause and

effect and uncertainty is like a neutral

679

00:49:48,121 --> 00:49:54,901

ground that I think that we can all start

to use to act as a mechanism for

680

00:49:54,901 --> 00:49:58,861

translating the language that we all use

in different fields.

681

00:49:59,401 --> 00:50:04,557

Yeah, learning the Bayes theorem and

whatever is associated with it.

682

00:50:04,557 --> 00:50:13,257

certainly has opened my world quite a bit

in terms of how I think about a problem

683

00:50:13,257 --> 00:50:18,077

and I found it the right way to

encapsulate my thoughts.

684

00:50:18,077 --> 00:50:25,557

And as Doug said, it sort of levels the

playing field that it provides that common

685

00:50:25,557 --> 00:50:32,749

language that the base theorem, I think

it's closely associated with how we

686

00:50:32,749 --> 00:50:36,089

do stuff or think about problems in

geoscience.

687

00:50:36,089 --> 00:50:39,409

And that has started to make things so

much easier.

688

00:50:39,409 --> 00:50:43,209

If you just sit down as Doc said, you

write down the probability of sea level

689

00:50:43,209 --> 00:50:48,369

rise given, and then, you know, you start

with the chain rule, you have your models,

690

00:50:48,369 --> 00:50:52,789

you try to come up with a likelihood

model, you try to come up with priors for

691

00:50:52,789 --> 00:50:54,349

your parameters.

692

00:50:54,569 --> 00:51:01,469

And even as like a non -Basian expert, it

still provides me with a way to think

693

00:51:01,469 --> 00:51:02,597

about it.

694

00:51:02,893 --> 00:51:10,193

and provides me with the tools to talk

about Doc, with Doc and others about the

695

00:51:10,193 --> 00:51:14,291

problems that I have and the goals I want

to achieve.

696

00:51:18,349 --> 00:51:20,289

Yeah, yeah, awesome points.

697

00:51:20,469 --> 00:51:26,509

And definitely agree that, yeah, also

making the effort of making sure we're

698

00:51:26,509 --> 00:51:32,069

talking about the same things and

educating on these concepts is absolutely

699

00:51:32,069 --> 00:51:33,189

crucial.

700

00:51:34,109 --> 00:51:42,669

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

there is a project of yours, and since I

701

00:51:42,669 --> 00:51:47,029

see the time running by, there is

something I really want to ask you about,

702

00:51:47,029 --> 00:51:47,821

and that's...

703

00:51:47,821 --> 00:51:52,221

the Parallel Ice Sheet Model, so PISM.

704

00:51:52,221 --> 00:51:56,921

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

yeah, I'm curious about that.

705

00:51:56,921 --> 00:51:58,831

What does that mean?

706

00:51:58,831 --> 00:52:01,961

What are you doing with this project?

707

00:52:06,413 --> 00:52:12,453

The general ice sheet model or PISM in

short started a little bit before I came

708

00:52:12,453 --> 00:52:14,413

to Alaska as a postdoc.

709

00:52:14,413 --> 00:52:20,713

In fact, few of us may even remember the

time before the first iPhone and PISM

710

00:52:20,713 --> 00:52:28,113

started a year before the, I think the

first iPhone came out and it was the first

711

00:52:28,113 --> 00:52:30,193

open source ice sheet model.

712

00:52:30,193 --> 00:52:35,581

But at the same time, it was the first

openly developed ice sheet model.

713

00:52:35,597 --> 00:52:42,057

Lots of other models have come later and

opened their code after, you know, some,

714

00:52:42,057 --> 00:52:44,337

after they have reached some maturity.

715

00:52:44,337 --> 00:52:51,997

And basically we can go back to commit

number one from 2006 or something like

716

00:52:51,997 --> 00:52:55,037

that and look at the first line that has

been written.

717

00:52:55,037 --> 00:53:01,677

And this is mostly thanks to a

mathematician named Ed Buehler here at the

718

00:53:01,677 --> 00:53:05,493

University of Fairbanks and his, at that

time, grad student.

719

00:53:05,581 --> 00:53:11,061

Chad Brown, who somehow got into ice sheet

modeling, I think similar to Doc, through

720

00:53:11,061 --> 00:53:16,901

mountaineering, going over glaciers,

climbing up on ice and getting fascinated

721

00:53:16,901 --> 00:53:20,881

with ice as a geophysical fluid.

722

00:53:21,361 --> 00:53:27,421

And they started developing a model

slightly differently than it has been

723

00:53:27,421 --> 00:53:32,461

developed in the past by individual

glaciologists without...

724

00:53:32,461 --> 00:53:37,381

often without like a super strong

background in math and numerical analysis.

725

00:53:38,221 --> 00:53:48,301

So PISM started from writing or by writing

validation tests first and then developing

726

00:53:48,301 --> 00:53:52,561

the most appropriate numerical methods to

solve the problem.

727

00:53:53,441 --> 00:53:56,771

And as the name said, the P stands for

parallel.

728

00:53:56,771 --> 00:54:00,141

So it was also one of the first models

that was.

729

00:54:00,141 --> 00:54:07,321

developed from scratch in MPI via PETSI

and could take advantage of larger HP

730

00:54:07,321 --> 00:54:12,521

systems versus at that time when PISM

started, you would run your ice sheet

731

00:54:12,521 --> 00:54:15,021

model on a single core on your laptop.

732

00:54:15,781 --> 00:54:19,201

Since then, the project has grown quite a

bit.

733

00:54:20,021 --> 00:54:24,321

The University of Alaska here is still the

lead developer.

734

00:54:25,501 --> 00:54:29,263

I have full -time software engineer.

735

00:54:29,709 --> 00:54:33,369

who does a lot of the testing code

development, works with users.

736

00:54:33,369 --> 00:54:38,769

We have another team at the Potsdam

Institute for Climate Impact Research in

737

00:54:38,769 --> 00:54:43,009

Potsdam in Germany, who does a lot of the

development as well.

738

00:54:43,009 --> 00:54:51,189

And then there are 30 to 40 -ish users

scattered around the world who either

739

00:54:51,189 --> 00:54:59,053

develop the model or use it purely for

trying to answer scientific questions.

740

00:54:59,053 --> 00:55:08,153

and one of the best compliments we have

ever gotten about our model is, or was

741

00:55:08,153 --> 00:55:15,533

when we found the first publication by

accident of someone who just found the

742

00:55:15,533 --> 00:55:20,553

model online, went on GitHub, downloaded

it, compiled it, figured out how it works

743

00:55:20,553 --> 00:55:24,973

because it is well documented, did some

cool science with it and got it through

744

00:55:24,973 --> 00:55:26,213

peer review.

745

00:55:26,213 --> 00:55:28,813

So they never even had to contact.

746

00:55:28,813 --> 00:55:32,273

the developers to get help to get anything

done.

747

00:55:33,153 --> 00:55:38,953

And for us, that's a big compliment.

748

00:55:38,953 --> 00:55:45,233

There are other models where you kind of

need to take like a one week long course

749

00:55:45,233 --> 00:55:46,993

to even get started.

750

00:55:47,093 --> 00:55:53,053

And we've been trying to maintain that

level of documentation and co

751

00:55:53,053 --> 00:55:57,869

-transparency by keeping a relatively

stable well thought out.

752

00:55:57,869 --> 00:56:01,989

API, something like that.

753

00:56:02,109 --> 00:56:09,269

So through all that backbone development,

it has become one of the leading models to

754

00:56:09,269 --> 00:56:14,629

answer questions revolving around

glaciology and sea level rise.

755

00:56:14,629 --> 00:56:20,909

Of course, again, because it started in

2006, it is starting to age and things

756

00:56:20,909 --> 00:56:25,809

that, for example, Doc mentioned that he's

developing with his fire -direct code

757

00:56:25,809 --> 00:56:27,941

coupled to

758

00:56:28,365 --> 00:56:30,437

um, tight torch.

759

00:56:32,173 --> 00:56:37,893

This is something we cannot yet offer and

it may not be feasible because there's so

760

00:56:37,893 --> 00:56:43,195

much legacy code that we can't handle a

smooth transition.

761

00:56:46,893 --> 00:56:51,213

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

762

00:56:51,213 --> 00:56:53,793

Oh, that's impressive.

763

00:56:54,373 --> 00:56:58,533

And I'm guessing that requires quite a lot

of collaboration with quite a lot of

764

00:56:58,533 --> 00:56:59,193

people.

765

00:56:59,193 --> 00:57:01,673

So well done on that.

766

00:57:01,673 --> 00:57:02,083

Thank you.

767

00:57:02,083 --> 00:57:03,413

That's incredible.

768

00:57:03,673 --> 00:57:04,433

Yeah.

769

00:57:04,433 --> 00:57:13,473

Any links, if there are any links that

people interested in could dig into, feel

770

00:57:13,473 --> 00:57:15,597

free to join that to the show notes.

771

00:57:15,597 --> 00:57:20,337

because I think that's a very interesting

project.

772

00:57:21,797 --> 00:57:29,857

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

preparing for the show that you, and I

773

00:57:29,857 --> 00:57:35,017

think you've talked about that at the

beginning, you work on echo geomorphic

774

00:57:35,017 --> 00:57:36,237

effects.

775

00:57:37,357 --> 00:57:43,577

Can you tell us what this is and what that

means and why that's interesting?

776

00:57:43,577 --> 00:57:44,431

Sure.

777

00:57:44,653 --> 00:57:45,853

Sure, yeah.

778

00:57:45,853 --> 00:57:50,793

I would not say that I am an eco

-geomorphologist by any stretch of the

779

00:57:50,793 --> 00:57:57,273

imagination, but when you work on

glaciology in Alaska, I think we're always

780

00:57:57,273 --> 00:58:03,193

interested in understanding and

communicating the importance of glacial

781

00:58:03,193 --> 00:58:07,093

systems beyond their influence on sea

level rise.

782

00:58:07,093 --> 00:58:13,293

Because it turns out that if you plop a

giant chunk of ice somewhere on the

783

00:58:13,293 --> 00:58:17,133

coastline, it's going to have implications

for what the water chemistry is like and

784

00:58:17,133 --> 00:58:21,913

what the water temperature is like and

what the local climate is like and maybe

785

00:58:21,913 --> 00:58:27,133

more broadly how animals can move around

and a whole bunch of other stuff.

786

00:58:27,133 --> 00:58:31,313

And so one project that I'm super excited

about, we've been working on this for a

787

00:58:31,313 --> 00:58:37,473

couple of years, is to try and understand

the future evolution of a very large

788

00:58:37,473 --> 00:58:40,733

glacier in coastal Alaska called Malaspina

Glacier.

789

00:58:40,733 --> 00:58:42,573

It's very conspicuous.

790

00:58:42,573 --> 00:58:46,253

feature if you ever look at the coastline

of Alaska on Google Earth or something

791

00:58:46,253 --> 00:58:47,633

like that.

792

00:58:47,633 --> 00:58:57,073

And it also happens to sit very close to a

really robust Alaska native community that

793

00:58:57,073 --> 00:59:01,973

uses the forelands of the glacier and the

adjacent areas as hunting and fishing

794

00:59:01,973 --> 00:59:02,933

grounds.

795

00:59:02,933 --> 00:59:08,557

And through the course of our modeling,

and we can say,

796

00:59:08,557 --> 00:59:12,177

this with a fair bit of confidence because

we've done a complete probabilistic

797

00:59:12,177 --> 00:59:18,877

treatment, we can say that it's very

likely that this very large glacier is

798

00:59:18,877 --> 00:59:25,217

more or less going to disappear in the

next certainly century, maybe faster than

799

00:59:25,217 --> 00:59:26,177

that.

800

00:59:26,397 --> 00:59:32,607

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

fjord, Icefield Valley.

801

00:59:32,607 --> 00:59:36,117

The forelands might start to degrade.

802

00:59:36,897 --> 00:59:38,453

And

803

00:59:38,765 --> 00:59:44,965

the whole landscape of that area that

people are using for all sorts of things,

804

00:59:45,065 --> 00:59:51,465

for gathering food and transportation and

a ton of other activities, it's all going

805

00:59:51,465 --> 00:59:52,745

to change a lot.

806

00:59:52,745 --> 00:59:58,195

And so I'm really excited about being able

to utilize some of these modeling tools,

807

00:59:58,195 --> 01:00:03,885

particularly in conjunction with robust

uncertainty quantification frameworks to

808

01:00:03,885 --> 01:00:07,175

provide responsible

809

01:00:07,469 --> 01:00:14,269

defensible predictions about how this

place is going to be different in the

810

01:00:14,269 --> 01:00:16,517

coming years to the people that live

there.

811

01:00:19,597 --> 01:00:20,057

Yeah.

812

01:00:20,057 --> 01:00:21,137

Okay.

813

01:00:21,177 --> 01:00:23,657

That makes more sense now.

814

01:00:24,457 --> 01:00:28,157

And geo -ecomorphitration, that's the

term.

815

01:00:28,477 --> 01:00:29,977

That's pretty impressive.

816

01:00:30,297 --> 01:00:36,237

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

817

01:00:36,237 --> 01:00:43,617

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

of how ecosystems change in response to

818

01:00:43,617 --> 01:00:46,157

changes in the way that the earth shapes.

819

01:00:46,977 --> 01:00:47,257

Yeah.

820

01:00:47,257 --> 01:00:49,677

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

821

01:00:49,677 --> 01:00:52,737

at parties, you know, like Fisher.

822

01:00:54,657 --> 01:00:56,797

Awesome.

823

01:00:56,797 --> 01:00:58,997

Well, thanks a lot, guys.

824

01:00:59,297 --> 01:01:04,907

We're going to start wrapping up because I

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

825

01:01:04,907 --> 01:01:08,757

but of course I still would have lots of

questions.

826

01:01:08,757 --> 01:01:16,497

Maybe, yeah, something I'd like to hear

you both about is potential development,

827

01:01:16,497 --> 01:01:19,365

potential applications of

828

01:01:19,405 --> 01:01:20,935

of what you're doing right now.

829

01:01:20,935 --> 01:01:27,565

Where would you like to see the research

in glaciology and ice sheet modeling going

830

01:01:27,565 --> 01:01:29,105

in the coming years?

831

01:01:29,105 --> 01:01:31,905

What is the most exciting to you?

832

01:01:31,905 --> 01:01:34,369

Maybe Andy first.

833

01:01:37,581 --> 01:01:42,421

Maybe I'll start with the not so exciting

part.

834

01:01:45,165 --> 01:01:51,765

because especially now with those new

methods that we're developing, machine

835

01:01:51,765 --> 01:01:58,265

learning, artificial intelligence and

large data sets, I think there is still a

836

01:01:58,265 --> 01:02:04,045

lot to be done just trying to understand

the data sets we already have with

837

01:02:04,045 --> 01:02:06,385

relatively simple methods.

838

01:02:07,605 --> 01:02:14,085

I say this is not particularly exciting

and it's also harder to get funding to do

839

01:02:14,085 --> 01:02:14,573

that.

840

01:02:14,573 --> 01:02:19,473

funding agencies like to see something

very new, something shiny.

841

01:02:20,853 --> 01:02:25,933

But sometimes you can make a bunch of

progress by just bringing together bits

842

01:02:25,933 --> 01:02:30,593

and pieces that you already have, but you

just never have time for that.

843

01:02:30,593 --> 01:02:36,073

You could develop an algorithm that

describes how a glacier caps off in

844

01:02:36,073 --> 01:02:39,413

Antarctica and you test it and it works

very well there.

845

01:02:39,673 --> 01:02:42,833

But then you have to go on and develop

something new.

846

01:02:43,533 --> 01:02:47,493

you're rarely left with the time to test,

well, would that be a good idea for

847

01:02:47,493 --> 01:02:54,513

Mellaspino Glacier or for a glacier in

anywhere in Alaska or in Greenland as

848

01:02:54,513 --> 01:02:55,513

well?

849

01:02:55,673 --> 01:03:01,533

So if I had some time and some money, this

is where I think I could make a bunch of

850

01:03:01,533 --> 01:03:04,513

progress with relatively little effort.

851

01:03:05,973 --> 01:03:09,053

Maybe Doc wants to start with the shiny

stuff.

852

01:03:10,913 --> 01:03:13,011

Shiny stuff, I don't know.

853

01:03:13,165 --> 01:03:20,345

You know what's always a perpetual source

of inspiration for me is the United States

854

01:03:20,345 --> 01:03:22,665

National Weather Service.

855

01:03:22,665 --> 01:03:29,404

I go on their website and I type in my

town name and I click on a location on a

856

01:03:29,404 --> 01:03:35,045

little map and it shows me a pretty high

accuracy prediction of what the weather is

857

01:03:35,045 --> 01:03:39,725

going to look like where I'm at for the

next like seven days or something like

858

01:03:39,725 --> 01:03:40,585

that.

859

01:03:40,585 --> 01:03:41,997

And I...

860

01:03:41,997 --> 01:03:51,577

It's this innocuous little interface, but

it overlies this incredible system of

861

01:03:51,577 --> 01:03:56,797

computational fluid mechanics combined

with real -time integration of data

862

01:03:56,797 --> 01:03:58,647

products in a probabilistic way.

863

01:03:58,647 --> 01:04:00,737

They're doing ensemble modeling.

864

01:04:01,257 --> 01:04:05,857

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

incredible operational system that has

865

01:04:05,857 --> 01:04:09,897

just a wonderful, useful interface for

people.

866

01:04:09,897 --> 01:04:11,621

And you know...

867

01:04:12,141 --> 01:04:17,761

I think that we are getting maybe to the

point in glaciology with our understanding

868

01:04:17,761 --> 01:04:23,541

of methods and capacities and stuff to

maybe do something like that.

869

01:04:23,541 --> 01:04:31,081

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

real -time forecasting for every little

870

01:04:31,081 --> 01:04:33,129

chunk of glacier ice in the world.

871

01:04:37,773 --> 01:04:39,683

Yeah, that sounds very interesting.

872

01:04:39,683 --> 01:04:41,573

I'm going to look at that page.

873

01:04:41,573 --> 01:04:43,533

Yeah, let's send that to the shuttles.

874

01:04:43,533 --> 01:04:45,083

That sounds very fun.

875

01:04:45,083 --> 01:04:47,913

I know, but that for sure.

876

01:04:47,913 --> 01:04:55,573

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

used application of Bayesian statistics in

877

01:04:55,573 --> 01:04:57,973

geophysics of any of them.

878

01:04:58,133 --> 01:04:59,393

Interesting.

879

01:04:59,673 --> 01:05:06,553

Well, if anybody in the listeners knows

someone working at weather .gov who could

880

01:05:06,553 --> 01:05:07,917

come on the podcast,

881

01:05:07,917 --> 01:05:13,897

to talk about the application of patient

methods at weather .gov.

882

01:05:13,897 --> 01:05:14,857

My door is open.

883

01:05:14,857 --> 01:05:16,977

That would be a great episode.

884

01:05:16,977 --> 01:05:17,437

Yeah.

885

01:05:17,437 --> 01:05:18,497

Absolutely.

886

01:05:18,737 --> 01:05:23,717

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

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

887

01:05:23,717 --> 01:05:28,697

remember, about gravitation waves.

888

01:05:28,697 --> 01:05:31,517

So not gravitational waves, but

gravitation waves.

889

01:05:31,517 --> 01:05:33,317

I didn't know that existed.

890

01:05:33,517 --> 01:05:35,227

That was super interesting.

891

01:05:35,227 --> 01:05:36,877

And I'm going to...

892

01:05:36,877 --> 01:05:43,657

I'm going to link to this episode in the

show notes because that was a very cool

893

01:05:43,657 --> 01:05:48,497

one basically talking about the mass of

really big mountains.

894

01:05:48,497 --> 01:05:55,077

So probably what the mountains you have in

Alaska, Andy and like basically the wave

895

01:05:55,077 --> 01:06:00,977

they create through their gravity, which

is non -negligible in comparison to the

896

01:06:00,977 --> 01:06:04,365

gravity of the earth, which is just pretty

incredible.

897

01:06:04,365 --> 01:06:08,305

and that has impacts on the weather.

898

01:06:08,365 --> 01:06:11,985

So definitely gonna link to that.

899

01:06:12,025 --> 01:06:16,805

Before closing up the show though, I'm

gonna ask you the last two questions I ask

900

01:06:16,805 --> 01:06:18,665

every guest at the end of the show.

901

01:06:18,665 --> 01:06:23,825

First one, if you had unlimited time and

resources, which problem would you try to

902

01:06:23,825 --> 01:06:24,325

solve?

903

01:06:24,325 --> 01:06:29,385

I feel like Andy, you've almost answered

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

904

01:06:29,385 --> 01:06:32,857

Maybe that gives you an opportunity to

answer something else.

905

01:06:34,765 --> 01:06:43,765

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

and I've done modeling of the Antarctic

906

01:06:43,765 --> 01:06:49,085

ice sheet, of the Greenland ice sheet, of

glaciers in the Alps and Scandinavia and

907

01:06:49,085 --> 01:06:50,917

we haven't done much.

908

01:06:52,813 --> 01:06:54,713

with Alaskan glaciers.

909

01:06:54,713 --> 01:07:00,393

Doug was mentioning their projects on

Malaspino glaciers and the surrounding

910

01:07:00,393 --> 01:07:01,633

area.

911

01:07:02,073 --> 01:07:06,513

But because Alaska is so big, the

challenges are equally big.

912

01:07:06,513 --> 01:07:12,653

Understanding the precipitation there,

where you go from sea level up to 5 ,000

913

01:07:12,653 --> 01:07:19,789

meters within a couple tens of kilometers

poses interesting challenges to like,

914

01:07:19,789 --> 01:07:23,589

any modeling or observational approach.

915

01:07:24,149 --> 01:07:30,189

And after living here for that long,

within unlimited resources, I think I

916

01:07:30,189 --> 01:07:33,539

would like to give back to Alaska and

study Alaskan glaciers.

917

01:07:33,539 --> 01:07:39,449

So I would invest in both observational

and modeling capabilities to better

918

01:07:39,449 --> 01:07:44,349

understand how the Arctic here in Alaska

is changing.

919

01:07:47,049 --> 01:07:49,733

That's like, sounds differently like a

920

01:07:49,741 --> 01:07:53,541

a very interesting project.

921

01:07:53,721 --> 01:07:55,481

Doug, what about you?

922

01:07:55,481 --> 01:08:04,201

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

then I suppose I would say what I did

923

01:08:04,201 --> 01:08:12,921

before about this notion of a worldwide,

every glacier forecasting tool that was

924

01:08:12,921 --> 01:08:15,161

widely usable by the general public.

925

01:08:15,161 --> 01:08:17,165

I think I'll stick with that one.

926

01:08:17,165 --> 01:08:22,565

But since my resources are unlimited, I

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

927

01:08:22,565 --> 01:08:28,165

whole bunch of other people to go out and

sort out the whole nuclear fusion thing.

928

01:08:28,165 --> 01:08:31,945

And then there'll be enough electricity to

run my computer.

929

01:08:34,145 --> 01:08:37,425

That sounds like a good thing to do

indeed.

930

01:08:38,365 --> 01:08:43,485

And second question, if you could have

dinner with any great scientific mind,

931

01:08:43,485 --> 01:08:46,733

dead, alive, or fictional, who would it

be?

932

01:08:46,733 --> 01:08:49,333

So Doug, let's start with you.

933

01:08:49,333 --> 01:08:50,473

Sure.

934

01:08:51,893 --> 01:08:54,143

Man, why do we call it Bayesian

statistics?

935

01:08:54,143 --> 01:08:59,113

We should really be calling it Laplacian

statistics, right?

936

01:08:59,113 --> 01:08:59,613

Yeah.

937

01:08:59,613 --> 01:09:07,933

He came up with this notion that we should

view probability as a means for

938

01:09:07,933 --> 01:09:10,393

communicating our knowledge of a process.

939

01:09:10,393 --> 01:09:13,669

And I think that that's the most

940

01:09:14,285 --> 01:09:20,005

Perhaps the most important scientific idea

that nobody ever mentions.

941

01:09:20,005 --> 01:09:21,845

So I'm going to go with Laplace.

942

01:09:21,845 --> 01:09:28,865

I would be really interested to see how he

felt about the application of probability

943

01:09:28,865 --> 01:09:32,365

in that way to these more complicated

systems as well.

944

01:09:33,085 --> 01:09:34,745

I love that.

945

01:09:34,965 --> 01:09:43,225

And not only because that was my personal

answer also in one of the episodes I've

946

01:09:43,225 --> 01:09:44,125

done.

947

01:09:44,845 --> 01:09:45,845

Awesome.

948

01:09:45,885 --> 01:09:47,075

Andy, we'll get to you.

949

01:09:47,075 --> 01:09:51,225

But before that, I found the episode I was

referencing.

950

01:09:51,225 --> 01:09:55,585

So that was episode 64 with Laura

Mansfield.

951

01:09:55,585 --> 01:10:00,025

And we were talking about modeling the

climate and gravity waves.

952

01:10:00,025 --> 01:10:01,645

I think I said gravitational waves.

953

01:10:01,645 --> 01:10:02,185

That was wrong.

954

01:10:02,185 --> 01:10:04,345

That's gravity waves.

955

01:10:05,245 --> 01:10:08,685

Andy, who would you have dinner with?

956

01:10:08,685 --> 01:10:12,165

Well, I feel like I'm pretty blessed.

957

01:10:12,165 --> 01:10:13,861

I think I have...

958

01:10:14,541 --> 01:10:20,321

dinner with great scientific minds on a

regular basis when I have dinner with my

959

01:10:20,321 --> 01:10:23,821

colleagues at scientific conferences.

960

01:10:24,321 --> 01:10:29,797

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

961

01:10:32,013 --> 01:10:34,923

How about I'll meet Aristostinus?

962

01:10:34,923 --> 01:10:36,853

I'm not sure I pronounced that correctly.

963

01:10:36,853 --> 01:10:44,453

He was, I believe, the first one to

estimate the circumference of the earth.

964

01:10:44,473 --> 01:10:49,653

And I think that was like several, couple

hundred years BC.

965

01:10:49,993 --> 01:10:56,693

I'm just curious how people thought about

science in an environment several thousand

966

01:10:56,693 --> 01:10:57,293

years ago.

967

01:10:57,293 --> 01:11:00,677

I would love to chat with someone like far

back who...

968

01:11:01,197 --> 01:11:07,297

came up with like, I think the estimate

that he came up with was maybe within 10 %

969

01:11:07,297 --> 01:11:08,917

or something like that.

970

01:11:08,917 --> 01:11:13,957

And then suddenly like a thousand years

later, people thought yours was flat.

971

01:11:15,469 --> 01:11:19,049

I think that would be an interesting

person to meet.

972

01:11:19,949 --> 01:11:22,589

Yeah, for sure.

973

01:11:23,408 --> 01:11:25,709

Good one.

974

01:11:25,709 --> 01:11:31,629

I think you're the first one to choose

that.

975

01:11:31,629 --> 01:11:34,549

I love it.

976

01:11:35,089 --> 01:11:40,209

What's the most common answer you get for

that question?

977

01:11:41,329 --> 01:11:44,333

Well, that question is...

978

01:11:44,333 --> 01:11:47,213

bit more like the variation is bigger than

the first one.

979

01:11:47,213 --> 01:11:53,033

The first one has a clear winner if I

remember correctly, which is climate

980

01:11:53,033 --> 01:11:54,073

change.

981

01:11:54,553 --> 01:11:58,033

So we have a lot of people who would try

and tackle that.

982

01:11:58,033 --> 01:12:04,133

The second question, I think one of the

most common is Richard Feynman, if I

983

01:12:04,133 --> 01:12:05,113

remember correctly.

984

01:12:05,113 --> 01:12:06,633

I believe so.

985

01:12:07,213 --> 01:12:13,293

Yeah, I think Feynman is the winner, but

it's not...

986

01:12:13,293 --> 01:12:15,553

Pareto distribution.

987

01:12:15,553 --> 01:12:17,273

It's a pretty uniform distribution.

988

01:12:17,273 --> 01:12:19,153

It's not like...

989

01:12:19,153 --> 01:12:19,973

Yeah, I'm curious.

990

01:12:19,973 --> 01:12:23,723

Not a lot of people choose Laplace.

991

01:12:23,723 --> 01:12:25,793

Not a lot of people choose base.

992

01:12:26,353 --> 01:12:30,313

And interestingly, I think nobody chose

base until now.

993

01:12:30,493 --> 01:12:31,023

Yeah.

994

01:12:31,023 --> 01:12:33,193

Not a lot of people have chosen Einstein.

995

01:12:33,193 --> 01:12:37,813

So that's an interesting question because

that kind of goes against prior.

996

01:12:37,813 --> 01:12:39,829

It's hard to guess.

997

01:12:42,477 --> 01:12:43,027

Sorry, Andy.

998

01:12:43,027 --> 01:12:49,517

I would have thought like Einstein or

Newton or Galileo would come up pretty

999

01:12:49,517 --> 01:12:50,757

frequently.

Speaker: 1000

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

No, Galileo, I don't think so.

Speaker: 1001

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

Leonardo da Vinci does come up quite a

lot.

Speaker: 1002

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

But yeah, otherwise, I had Euclid once, of

course.

Speaker: 1003

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

That was a fun one, too.

Speaker: 1004

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,

Speaker: 1005

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

thank you for being so generous.

Speaker: 1006

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

Before we close up though, is there

something I forgot to ask you about and

Speaker: 1007

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

that you would like to mention or talk

about before we close up?

Speaker: 1008

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

I don't think so, not for me.

Speaker: 1009

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

I think it was a pretty comprehensive

journey.

Speaker: 1010

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

Yeah.

Speaker: 1011

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

Great.

Speaker: 1012

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

Believe me, I would still have like, I

could keep you for two hours, but no.

Speaker: 1013

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

Let's be parsimonious.

Speaker: 1014

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

Awesome.

Speaker: 1015

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

Well, again, thank you very much, Andy.

Speaker: 1016

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

Thank you very much, Dag.

Speaker: 1017

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

As usual, those who want to dig deeper,

refer to the show notes because we have.

Speaker: 1018

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

Andy's and Doug's links over there and

also a bit of the work.

Speaker: 1019

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

Speaker: 1020

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

show.

Speaker: 1021

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

Thanks Alex.

Speaker: 1022

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

Thanks Alex.

Speaker: 1023

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

Thanks for having us.

Speaker: 1024

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

This has been another episode of Learning

Bayesian Statistics.

Speaker: 1025

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

Be sure to rate, review, and follow the

show on your favorite podcatcher, and

Speaker: 1026

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

visit learnbaystats .com for more

resources about today's topics, as well as

Speaker: 1027

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

access to more episodes to help you reach

true Bayesian state of mind.

Speaker: 1028

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

That's learnbaystats .com.

Speaker: 1029

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

Our theme music is Good Bayesian by Baba

Brinkman, fit MC Lass and Meghiraam.

Speaker: 1030

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

Check out his awesome work at bababrinkman

.com.

Speaker: 1031

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

I'm your host.

Speaker: 1032

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

Alex Andorra.

Speaker: 1033

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

You can follow me on Twitter at Alex

underscore Andorra, like the country.

Speaker: 1034

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

You can support the show and unlock

exclusive benefits by visiting Patreon

Speaker: 1035

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

.com slash LearnBasedDance.

Speaker: 1036

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

Thank you so much for listening and for

your support.

Speaker: 1037

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

You're truly a good Bayesian.

Speaker: 1038

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

Change your predictions after taking

information.

Speaker: 1039

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

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

amazing, let's adjust those expectations.

Speaker: 1040

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

Let me show you how to be a good Bayesian

Change calculations after taking fresh

Speaker: 1041

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

data in Those predictions that your brain

is making Let's get them on a solid

Speaker: 1042

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

foundation