1
00:00:07,760 --> 00:00:10,640
Speaker 1: When I moved to southern California, I felt this immediate,

2
00:00:10,760 --> 00:00:13,680
Speaker 1: immense relief, not just because I was free of the

3
00:00:13,720 --> 00:00:17,400
Speaker 1: tyranny of outside clothing, but because I was released from

4
00:00:17,400 --> 00:00:20,119
Speaker 1: the anxiety of not knowing if the weather was going

5
00:00:20,200 --> 00:00:23,400
Speaker 1: to ruin my plans. Were you planning an outdoor birthday

6
00:00:23,400 --> 00:00:26,520
Speaker 1: party for your toddler, No need to make backup plans

7
00:00:26,600 --> 00:00:28,920
Speaker 1: just in case it rains. Do you need to drive

8
00:00:28,920 --> 00:00:31,800
Speaker 1: a few hours away, No problem. You don't have to

9
00:00:31,800 --> 00:00:35,800
Speaker 1: worry that a snowstorm might make the roads impassable because

10
00:00:35,960 --> 00:00:38,680
Speaker 1: I could predict the weather myself since it was the

11
00:00:38,680 --> 00:00:41,519
Speaker 1: same every single day. But not all of us are

12
00:00:41,640 --> 00:00:44,120
Speaker 1: lucky enough to live in such calm climbs, so it's

13
00:00:44,159 --> 00:00:47,519
Speaker 1: still very important that we try to anticipate storms so

14
00:00:47,560 --> 00:00:50,639
Speaker 1: that they're less fortunate among us can be prepared. It's

15
00:00:50,680 --> 00:00:54,600
Speaker 1: not often described as important physics, but predicting the weather

16
00:00:54,720 --> 00:00:58,280
Speaker 1: is one of physics' great success stories. John Martin, professor

17
00:00:58,320 --> 00:01:01,720
Speaker 1: of atmospheric oceanic sciences, told me that weather predictions are

18
00:01:01,800 --> 00:01:05,520
Speaker 1: quote the most unheralded scientific advance of the second half

19
00:01:05,560 --> 00:01:07,960
Speaker 1: of the twentieth century. If you keep score every day,

20
00:01:08,000 --> 00:01:10,520
Speaker 1: I can't believe how well we predict the weather three

21
00:01:10,520 --> 00:01:13,720
Speaker 1: to five days in advance. In thirty years, we've gone

22
00:01:13,760 --> 00:01:16,320
Speaker 1: from predictions from one to two days to now five

23
00:01:16,360 --> 00:01:21,399
Speaker 1: to seven days. We have made unbelievable progress. So how

24
00:01:21,440 --> 00:01:24,720
Speaker 1: does that all work? What is the physics underlying the weather,

25
00:01:24,760 --> 00:01:27,080
Speaker 1: why has it gotten better? And what can we expect

26
00:01:27,120 --> 00:01:29,800
Speaker 1: into the future. I talked to Professor Martin and my

27
00:01:29,840 --> 00:01:33,120
Speaker 1: good friend Professor Jane Baldwin here at UC Irvine about

28
00:01:33,160 --> 00:01:35,840
Speaker 1: how the weather all works. So we'll dig into all

29
00:01:35,880 --> 00:01:38,720
Speaker 1: of that in today's episode, dedicated to all of y'all

30
00:01:38,760 --> 00:01:42,960
Speaker 1: who still experience regular weather. Welcome to Daniel and Kelly's

31
00:01:43,040 --> 00:01:44,960
Speaker 1: extraordinarily sunny universe.

32
00:01:58,160 --> 00:02:02,320
Speaker 2: Hello Kelly Leadersmith. I studied fites and space and I

33
00:02:02,400 --> 00:02:03,840
Speaker 2: love rainy days.

34
00:02:05,720 --> 00:02:05,920
Speaker 3: Hi.

35
00:02:06,000 --> 00:02:08,840
Speaker 1: I'm Daniel. I'm a particle physicist, and I can predict

36
00:02:08,840 --> 00:02:11,400
Speaker 1: the weather in California for the next one hundred years

37
00:02:11,400 --> 00:02:12,600
Speaker 1: with my eyes closed.

38
00:02:13,080 --> 00:02:16,160
Speaker 2: How boring, how massively dull.

39
00:02:16,280 --> 00:02:20,120
Speaker 1: How wonderfully, delightfully, predictably, reliably boring.

40
00:02:20,480 --> 00:02:23,560
Speaker 2: Oh, you know, one of my favorite weather moments, I

41
00:02:23,600 --> 00:02:26,640
Speaker 2: have to admit, was a southern California morning. So I

42
00:02:26,760 --> 00:02:29,160
Speaker 2: was a visiting scholar at the University of California, Santa

43
00:02:29,160 --> 00:02:31,520
Speaker 2: Barbara for a little while, and I had an office

44
00:02:31,600 --> 00:02:34,200
Speaker 2: that was like right out on the ocean. Was amazing.

45
00:02:34,200 --> 00:02:36,560
Speaker 2: And when I was driving in one day, there was

46
00:02:36,760 --> 00:02:39,240
Speaker 2: just a little bit of water on the ground and

47
00:02:39,440 --> 00:02:42,000
Speaker 2: the car tires were kicking up a little bit of

48
00:02:42,080 --> 00:02:45,880
Speaker 2: a spray, and there were literally rainbows following all of

49
00:02:45,960 --> 00:02:48,840
Speaker 2: the cars into school. And then I got out of

50
00:02:48,880 --> 00:02:50,840
Speaker 2: the car and the rain had stopped, and there was

51
00:02:50,880 --> 00:02:53,600
Speaker 2: a rainbow over the ocean and there were hummingbirds and

52
00:02:53,639 --> 00:02:56,400
Speaker 2: it was like a Disney movie scene. I expected like

53
00:02:56,639 --> 00:02:58,360
Speaker 2: a bunny to hop out and be like, can I

54
00:02:58,360 --> 00:03:02,120
Speaker 2: help you with anything? Anyway, it was. It was kind

55
00:03:02,120 --> 00:03:03,480
Speaker 2: of magical. I'll give you that.

56
00:03:03,600 --> 00:03:06,040
Speaker 1: California is heaven. Yes, what happens when you die in

57
00:03:06,080 --> 00:03:08,560
Speaker 1: Virginia is you end up in California.

58
00:03:08,639 --> 00:03:12,800
Speaker 2: Do you know that not all California as southern California.

59
00:03:12,080 --> 00:03:13,559
Speaker 1: I mean all of real California.

60
00:03:13,720 --> 00:03:16,360
Speaker 2: Oh, I see, because northern California's got some weather.

61
00:03:18,880 --> 00:03:21,120
Speaker 1: You're absolutely right. In fact, I heard Katrina say something

62
00:03:21,120 --> 00:03:24,600
Speaker 1: really insightful the other day. You know, she's from northern California.

63
00:03:24,600 --> 00:03:27,239
Speaker 1: But now we've lived in southern California for quite a while,

64
00:03:27,600 --> 00:03:29,880
Speaker 1: and she said to somebody that she's now a complete

65
00:03:29,919 --> 00:03:33,520
Speaker 1: Californian because she's lived in both northern and southern California.

66
00:03:33,600 --> 00:03:35,920
Speaker 1: And I was like, Oh, that's cool. She's like accepted

67
00:03:35,960 --> 00:03:40,960
Speaker 1: southern California, which is hard for Northern California's I'm aware, yes,

68
00:03:41,480 --> 00:03:44,240
Speaker 1: not everything is Southern California unfortunately.

69
00:03:45,120 --> 00:03:48,760
Speaker 2: Oh I really like the variability Virginia weather is amazing

70
00:03:48,760 --> 00:03:50,360
Speaker 2: for me. But so my question for you is, what

71
00:03:50,440 --> 00:03:53,520
Speaker 2: is the worst weather situation that you've experienced?

72
00:03:53,800 --> 00:03:57,320
Speaker 1: Great question. I was on the East coast, of course,

73
00:03:57,440 --> 00:04:00,440
Speaker 1: you're doing a college tour with my son, and we

74
00:04:00,440 --> 00:04:03,560
Speaker 1: were in Massachusetts. I think we were visiting Amherst or

75
00:04:03,560 --> 00:04:06,160
Speaker 1: maybe it was Williams, I don't remember. And there was

76
00:04:06,200 --> 00:04:09,200
Speaker 1: some freak tornado which tore up a bunch of trees

77
00:04:09,280 --> 00:04:12,040
Speaker 1: and knocked down a bunch of power lines and there

78
00:04:12,160 --> 00:04:16,440
Speaker 1: was no power in the whole town for like almost

79
00:04:16,440 --> 00:04:19,360
Speaker 1: half a day. It was crazy and the winds were

80
00:04:19,360 --> 00:04:22,560
Speaker 1: insane and it felt a little scary, like we saw

81
00:04:22,680 --> 00:04:25,120
Speaker 1: like huge branches flying by the window.

82
00:04:26,120 --> 00:04:27,440
Speaker 2: Yeah, yep.

83
00:04:28,040 --> 00:04:29,840
Speaker 1: And he didn't end up going to school there.

84
00:04:31,520 --> 00:04:34,080
Speaker 2: Yeah, I get that. I get that. So we lived

85
00:04:34,080 --> 00:04:38,919
Speaker 2: in Alabama, Tuscaloosa, and we moved there pretty soon after

86
00:04:38,960 --> 00:04:42,640
Speaker 2: that giant tornado that like made the news, and you

87
00:04:42,680 --> 00:04:45,039
Speaker 2: could see the path of the tornado because like, you know,

88
00:04:45,040 --> 00:04:47,000
Speaker 2: you'd be driving through an area with lots of like

89
00:04:47,000 --> 00:04:49,680
Speaker 2: you know, Starbucks, Panera, lots of stores or whatever, and

90
00:04:49,720 --> 00:04:52,599
Speaker 2: then suddenly there would be a like air an opening

91
00:04:52,600 --> 00:04:55,000
Speaker 2: in between all of the stores with nothing, and like

92
00:04:55,040 --> 00:04:58,200
Speaker 2: the tornado had just gone through there and just absolutely

93
00:04:58,240 --> 00:05:00,680
Speaker 2: picked up and thrown everything that was in there, and

94
00:05:00,720 --> 00:05:03,040
Speaker 2: even after they cleaned it out, there were still, you know,

95
00:05:03,120 --> 00:05:05,400
Speaker 2: you could tell where the tornado had gone. And we

96
00:05:05,400 --> 00:05:08,680
Speaker 2: were also in Houston during some pretty bad storms and

97
00:05:08,720 --> 00:05:11,360
Speaker 2: we had the kids and our dog and our cats

98
00:05:11,400 --> 00:05:13,919
Speaker 2: in a little hallway in the interior of the house

99
00:05:14,200 --> 00:05:16,800
Speaker 2: and my in laws were visiting, and my mother in

100
00:05:16,880 --> 00:05:19,400
Speaker 2: law was so sweet. She like looked around and she

101
00:05:19,400 --> 00:05:21,000
Speaker 2: she was trying to see, you know, who could get

102
00:05:21,080 --> 00:05:24,000
Speaker 2: hurt and how, And she gave her glasses to Zach

103
00:05:24,080 --> 00:05:26,159
Speaker 2: in case there was any like flying glass and she

104
00:05:26,279 --> 00:05:29,600
Speaker 2: just insisted that he have her glasses. And I was

105
00:05:29,760 --> 00:05:32,040
Speaker 2: like in that moment, I was like, gosh, you are

106
00:05:32,080 --> 00:05:35,120
Speaker 2: the sweetest person in the whole world, Like you are

107
00:05:35,160 --> 00:05:37,240
Speaker 2: thinking about the tiny little things you could do to

108
00:05:37,279 --> 00:05:39,960
Speaker 2: help the people around you, and anyway, she's she's the best.

109
00:05:40,080 --> 00:05:43,480
Speaker 1: Yeah, but we've all been caught in surprise weather, right.

110
00:05:43,640 --> 00:05:46,920
Speaker 1: I remember going backpacking in Arkansas one time and being

111
00:05:46,920 --> 00:05:49,279
Speaker 1: caught in a snowstorm and the temperatures dropped into the

112
00:05:49,320 --> 00:05:51,560
Speaker 1: teens and we weren't one hundred percent sure we were

113
00:05:51,560 --> 00:05:54,360
Speaker 1: going to make it. Oh, and everybody's been, like, you know,

114
00:05:54,440 --> 00:05:56,760
Speaker 1: caught in a snowstorm or a rainstorm or in a

115
00:05:56,839 --> 00:06:00,320
Speaker 1: heat wave. Right, And these things are exciting, they can

116
00:06:00,360 --> 00:06:02,800
Speaker 1: be dramatic, but they can also be very dangerous, right.

117
00:06:02,880 --> 00:06:03,080
Speaker 2: Yeah.

118
00:06:03,080 --> 00:06:06,280
Speaker 1: People die in these crazy weather storms, and so it's

119
00:06:06,360 --> 00:06:08,680
Speaker 1: valuable to be able to know in advance what's going

120
00:06:08,760 --> 00:06:11,000
Speaker 1: to happen, not just so you can plain your picnics,

121
00:06:11,000 --> 00:06:15,839
Speaker 1: but also you can survive the increasingly dramatic weather that

122
00:06:15,880 --> 00:06:17,760
Speaker 1: we're all facing as the planet warms.

123
00:06:17,960 --> 00:06:20,640
Speaker 2: Yeah, that's right, more severe weather is becoming more common.

124
00:06:20,680 --> 00:06:22,920
Speaker 2: And so today we're going to talk about how good

125
00:06:22,960 --> 00:06:25,200
Speaker 2: we are at making predictions and how we go about

126
00:06:25,200 --> 00:06:27,000
Speaker 2: making those predictions exactly.

127
00:06:27,120 --> 00:06:29,760
Speaker 1: And I wanted to pull back the curtain on like

128
00:06:29,800 --> 00:06:32,120
Speaker 1: the science of this, how does this actually happen, What

129
00:06:32,160 --> 00:06:35,120
Speaker 1: are we doing, why is it hard? What are the challenges?

130
00:06:35,440 --> 00:06:38,120
Speaker 1: What improvements might we be seeing in the next five

131
00:06:38,240 --> 00:06:41,279
Speaker 1: or ten years. What problems are just fundamentally impossible and

132
00:06:41,360 --> 00:06:44,279
Speaker 1: might never be solved. And so today we're going to

133
00:06:44,279 --> 00:06:46,520
Speaker 1: dig into science of all that. But before we explain

134
00:06:46,560 --> 00:06:49,040
Speaker 1: to you how the experts do it, I was wondering

135
00:06:49,080 --> 00:06:52,000
Speaker 1: what everybody knew about how weather predictions happen. How do

136
00:06:52,080 --> 00:06:55,080
Speaker 1: those numbers end up on your phone? So I went

137
00:06:55,120 --> 00:06:57,640
Speaker 1: out there to ask our listeners what they knew about

138
00:06:57,680 --> 00:07:00,640
Speaker 1: how we predict the weather. If you would like to

139
00:07:00,640 --> 00:07:02,800
Speaker 1: answer these kind of questions for a future episode, don't

140
00:07:02,800 --> 00:07:06,200
Speaker 1: be shy, right to us two questions at Danielankelly dot org.

141
00:07:06,320 --> 00:07:09,880
Speaker 1: We will send you fun questions every week in your inbox.

142
00:07:10,360 --> 00:07:12,240
Speaker 1: In the meantime, think about it for a minute. What

143
00:07:12,280 --> 00:07:15,400
Speaker 1: do you know about how we predict the weather? Here's

144
00:07:15,440 --> 00:07:20,920
Speaker 1: what our listeners had to say. Sophisticated computer models, which

145
00:07:21,200 --> 00:07:25,320
Speaker 1: with an understanding if case theory, allows us to understand

146
00:07:25,360 --> 00:07:26,320
Speaker 1: the limitations.

147
00:07:27,000 --> 00:07:30,000
Speaker 4: Predicting the weather is like quantum particles.

148
00:07:30,640 --> 00:07:34,440
Speaker 1: There are many probabilities, but it is not known until

149
00:07:34,560 --> 00:07:39,280
Speaker 1: it is observed. Meteorologists they look at the current weather,

150
00:07:39,400 --> 00:07:41,760
Speaker 1: and they try to predict it by looking at the

151
00:07:42,280 --> 00:07:43,520
Speaker 1: moving clouds and all of.

152
00:07:43,480 --> 00:07:52,360
Speaker 3: That, by measuring with velocity and atmospheric pressure and maybe

153
00:07:52,560 --> 00:07:57,120
Speaker 3: modeling these that in supercomputers.

154
00:07:57,320 --> 00:07:59,720
Speaker 1: When a cow lies down in the field, it's going

155
00:07:59,800 --> 00:08:03,000
Speaker 1: to and when my knee aches, it's gonna snow.

156
00:08:03,360 --> 00:08:06,080
Speaker 4: Running multiple models.

157
00:08:05,880 --> 00:08:11,320
Speaker 3: Big computers, really really big computers.

158
00:08:10,840 --> 00:08:14,040
Speaker 4: Feed dad data, two complicated models that run on very

159
00:08:14,040 --> 00:08:15,200
Speaker 4: parfa spoken.

160
00:08:14,880 --> 00:08:21,280
Speaker 1: Pere i'd say, with surface measurements, satellite information and sophisticated

161
00:08:21,840 --> 00:08:25,280
Speaker 1: models and perhaps even artificial.

162
00:08:24,760 --> 00:08:30,920
Speaker 4: Intelligence observations taken by ships, planes, ground stations, satellites combined

163
00:08:30,960 --> 00:08:34,840
Speaker 4: with models built by really really smart people that run

164
00:08:34,880 --> 00:08:38,360
Speaker 4: on some of the fastest computers that humans have ever built.

165
00:08:38,440 --> 00:08:41,360
Speaker 1: There are sophisticated bottles that use a wide range of

166
00:08:41,400 --> 00:08:44,200
Speaker 1: observational and predictive inputs.

167
00:08:43,840 --> 00:08:46,920
Speaker 4: By observing weather patterns and the types of whether those

168
00:08:46,960 --> 00:08:48,040
Speaker 4: patterns tend to bring.

169
00:08:48,320 --> 00:08:51,360
Speaker 2: So I don't know if there's actually like scientific evidence

170
00:08:51,480 --> 00:08:54,160
Speaker 2: that sometimes knees will ache if like a stormfront is

171
00:08:54,200 --> 00:08:56,160
Speaker 2: coming through. But I have to admit that there's a

172
00:08:56,200 --> 00:08:57,719
Speaker 2: part of me that really hopes that if I get

173
00:08:57,800 --> 00:09:00,720
Speaker 2: arthright is when I'm older, I do have the ability

174
00:09:00,760 --> 00:09:02,880
Speaker 2: to tell when the weather's come in because I'll feel

175
00:09:02,880 --> 00:09:06,160
Speaker 2: like I'm really intimately connected to my environment. Oh, the

176
00:09:06,200 --> 00:09:09,120
Speaker 2: knees acted up again. Storms come and get the goats

177
00:09:09,120 --> 00:09:09,640
Speaker 2: in the barn.

178
00:09:10,240 --> 00:09:13,400
Speaker 1: I think that really shows your fundamental optimistic nature, Kelly,

179
00:09:13,440 --> 00:09:16,040
Speaker 1: because you're like, oh, I forget arthritis. There'll be a

180
00:09:16,080 --> 00:09:18,079
Speaker 1: silver lining. I can predict the weather.

181
00:09:18,600 --> 00:09:21,120
Speaker 2: You know, life is easier when you try to see

182
00:09:21,120 --> 00:09:21,880
Speaker 2: the silver lining.

183
00:09:22,640 --> 00:09:23,480
Speaker 1: That's wonderful.

184
00:09:23,600 --> 00:09:26,800
Speaker 2: But our audience had great answers, and they were, you know,

185
00:09:26,840 --> 00:09:28,839
Speaker 2: a lot of them said, you know exactly the right thing,

186
00:09:28,840 --> 00:09:31,720
Speaker 2: which is you've got to have data. Those are the

187
00:09:31,760 --> 00:09:33,960
Speaker 2: observations and you feed them into computers.

188
00:09:34,200 --> 00:09:37,440
Speaker 1: Yeah, essentially, and that's the big picture, not just of

189
00:09:37,480 --> 00:09:40,920
Speaker 1: weather prediction but any kind of prediction. There are two

190
00:09:41,360 --> 00:09:44,920
Speaker 1: fundamental ingredients to how you make a prediction. There's the

191
00:09:44,960 --> 00:09:48,240
Speaker 1: models and then there's the data. So let's take those

192
00:09:48,280 --> 00:09:51,080
Speaker 1: each in turn. When we say the models, we mean

193
00:09:51,280 --> 00:09:54,280
Speaker 1: like we're running a computer simulation or you're calculating things

194
00:09:54,320 --> 00:09:57,840
Speaker 1: on paper. Fundamentally, this is encoding the rules of the

195
00:09:57,880 --> 00:10:01,560
Speaker 1: system what the future can be given what the past was.

196
00:10:02,040 --> 00:10:04,079
Speaker 1: And this doesn't have to be some really complicated thing

197
00:10:04,280 --> 00:10:06,800
Speaker 1: like the weather over ist endbull. Think about a much

198
00:10:06,840 --> 00:10:09,959
Speaker 1: simpler situation, like you're tossing a ball in your backyard.

199
00:10:10,400 --> 00:10:12,560
Speaker 1: You want to know where does it go? Well, the

200
00:10:12,600 --> 00:10:15,200
Speaker 1: laws of physics predict the future, right, this is the model.

201
00:10:15,240 --> 00:10:17,800
Speaker 1: These are the rules that tell you how the past

202
00:10:18,120 --> 00:10:21,120
Speaker 1: becomes the future. Right. And in this case it's simple.

203
00:10:21,160 --> 00:10:24,120
Speaker 1: It's a parabola. It flies through the air. Things to

204
00:10:24,200 --> 00:10:26,680
Speaker 1: keep in mind here, though, is that a model like

205
00:10:26,760 --> 00:10:30,480
Speaker 1: this is always approximate. If I use f eicals MA

206
00:10:30,760 --> 00:10:33,240
Speaker 1: and I just account for gravity, ignore air resistance. When

207
00:10:33,280 --> 00:10:35,680
Speaker 1: I'm describing the ball, I'm going to get a quick answer,

208
00:10:35,720 --> 00:10:37,719
Speaker 1: and it's gonna be pretty good. It's not going to

209
00:10:37,760 --> 00:10:41,840
Speaker 1: be exactly bang on correct. It can't account for everything,

210
00:10:41,880 --> 00:10:44,160
Speaker 1: all the little wind gusts and the air resistance and

211
00:10:44,200 --> 00:10:46,400
Speaker 1: the slight change in humidity and maybe the spin on

212
00:10:46,440 --> 00:10:50,400
Speaker 1: the ball. My model ignores some details, and that's crucial. Right.

213
00:10:50,440 --> 00:10:53,360
Speaker 1: If I included every single particle in the backyard, I

214
00:10:53,400 --> 00:10:55,839
Speaker 1: would never get a calculation. So in order to make

215
00:10:55,880 --> 00:10:58,200
Speaker 1: this tractable, I got to simplify the problem. I got

216
00:10:58,200 --> 00:11:00,680
Speaker 1: to pull out the things that are important and ignore

217
00:11:00,679 --> 00:11:03,679
Speaker 1: the things I think are unimportant. Because I don't think

218
00:11:03,720 --> 00:11:05,680
Speaker 1: they're going to make a big enough difference in the answer.

219
00:11:06,040 --> 00:11:08,240
Speaker 1: And this is where the juice is. This is what

220
00:11:08,320 --> 00:11:12,120
Speaker 1: physics is, right. Physics is taking the universe and simplifying

221
00:11:12,160 --> 00:11:15,320
Speaker 1: it into a model that represents the bits you're excited about,

222
00:11:15,360 --> 00:11:17,480
Speaker 1: the bits you think are interesting and irrelevant, and then

223
00:11:17,840 --> 00:11:20,320
Speaker 1: you use those rules and manipulate it. That's your model

224
00:11:20,440 --> 00:11:22,920
Speaker 1: of the universe, and the model gives you an answer,

225
00:11:23,160 --> 00:11:25,400
Speaker 1: and hopefully, if the model is close enough to your

226
00:11:25,440 --> 00:11:27,880
Speaker 1: description of the universe, the answer you get from the

227
00:11:27,920 --> 00:11:30,840
Speaker 1: model is similar to the answer in the actual universe.

228
00:11:31,200 --> 00:11:33,360
Speaker 2: So one thing I think that's amazing is that something

229
00:11:33,360 --> 00:11:35,240
Speaker 2: as simple as throwing a ball up in the air

230
00:11:35,480 --> 00:11:37,800
Speaker 2: and then seeing where it lands is something we can't

231
00:11:37,840 --> 00:11:40,600
Speaker 2: completely model because there's so many complicating things. And now

232
00:11:40,640 --> 00:11:44,119
Speaker 2: you're talking about weather, which is so much more complicated

233
00:11:44,120 --> 00:11:46,480
Speaker 2: and requires so many more inputs. And of course you

234
00:11:46,520 --> 00:11:48,760
Speaker 2: can update your model. So you know, if you threw

235
00:11:48,800 --> 00:11:50,160
Speaker 2: the ball in the air and you were like, you

236
00:11:50,160 --> 00:11:52,679
Speaker 2: know what, it's a windy day, I absolutely need to

237
00:11:52,679 --> 00:11:55,920
Speaker 2: add wind. Now you've learned something, you add wind, and

238
00:11:55,960 --> 00:11:58,160
Speaker 2: so you know, it's an iterative process where you keep

239
00:11:58,200 --> 00:12:01,040
Speaker 2: trying to say what is in important and do I

240
00:12:01,120 --> 00:12:03,480
Speaker 2: need to include it? And does it make my predictions better?

241
00:12:03,640 --> 00:12:07,360
Speaker 2: But I also will note that you put predicting weather

242
00:12:08,000 --> 00:12:11,800
Speaker 2: under the physics umbrella. You think you guys get to

243
00:12:11,800 --> 00:12:13,000
Speaker 2: claim weather predictions.

244
00:12:13,600 --> 00:12:15,880
Speaker 1: I mean, we're not using economics to predict the weather.

245
00:12:16,520 --> 00:12:19,280
Speaker 1: What else is in the running for taking credit for

246
00:12:19,320 --> 00:12:21,880
Speaker 1: predicting the weather? Is it chemistry?

247
00:12:22,320 --> 00:12:24,880
Speaker 2: I feel like that also is some ecology, you know,

248
00:12:25,080 --> 00:12:26,760
Speaker 2: like because you're tracking.

249
00:12:26,520 --> 00:12:27,960
Speaker 1: Like cowfarts or something.

250
00:12:28,200 --> 00:12:29,679
Speaker 2: No, no, you like, you know, a.

251
00:12:29,679 --> 00:12:33,080
Speaker 1: Fart player role. Actually, so do you think.

252
00:12:33,000 --> 00:12:36,720
Speaker 2: That Noah has cow farts in their weather prediction models?

253
00:12:38,200 --> 00:12:42,319
Speaker 1: I think the climate models do include bovine methane emissions. Yes,

254
00:12:42,640 --> 00:12:45,880
Speaker 1: so not the daily predictions, but the bigger trends. Yes,

255
00:12:45,960 --> 00:12:48,760
Speaker 1: cow farts do help determine the future of our planet.

256
00:12:48,960 --> 00:12:49,360
Speaker 2: Amazing.

257
00:12:49,360 --> 00:12:50,920
Speaker 1: I want to go back to the point you made earlier.

258
00:12:51,000 --> 00:12:54,360
Speaker 1: You're exactly right that we're always approximating, and not just

259
00:12:54,440 --> 00:12:58,199
Speaker 1: when we're doing the weather, not just when we're tossing balls, always,

260
00:12:58,720 --> 00:13:03,160
Speaker 1: every single time, every model is approximation. There's this famous phrase.

261
00:13:03,160 --> 00:13:05,480
Speaker 1: I a memember who said it, like all models are wrong,

262
00:13:05,640 --> 00:13:08,360
Speaker 1: some of them are useful, even our description of like

263
00:13:08,559 --> 00:13:11,240
Speaker 1: the fundamental particles in the universe. As far as we know,

264
00:13:11,480 --> 00:13:15,920
Speaker 1: these are approximations. Every bit of science we have has

265
00:13:16,040 --> 00:13:19,720
Speaker 1: boundaries of where it's relevant because there are approximations made

266
00:13:20,080 --> 00:13:23,920
Speaker 1: when we construct those models everything, literally everything. We have

267
00:13:24,000 --> 00:13:27,680
Speaker 1: no piece of science that isn't an approximation of the universe.

268
00:13:28,280 --> 00:13:31,120
Speaker 1: Maybe one day we have a theory of everything, and

269
00:13:31,160 --> 00:13:34,600
Speaker 1: it's beautiful and we can do exact calculations on very

270
00:13:34,720 --> 00:13:38,440
Speaker 1: very simple situations. But we're not there. We may never

271
00:13:38,520 --> 00:13:40,920
Speaker 1: be there, and even if we are there, it will

272
00:13:40,960 --> 00:13:43,920
Speaker 1: be totally impractical for anything useful. Like you couldn't use

273
00:13:44,400 --> 00:13:47,360
Speaker 1: string theory to predict the path of a hurricane because

274
00:13:47,640 --> 00:13:51,040
Speaker 1: the complexity would be insane, Right, how many strings are

275
00:13:51,040 --> 00:13:53,640
Speaker 1: you modeling? The amount of computation required to do it

276
00:13:53,720 --> 00:13:57,480
Speaker 1: exactly would be impossible. So it's always an approximation. It's

277
00:13:57,520 --> 00:14:00,920
Speaker 1: just a question of which approximations. That's where the science

278
00:14:00,960 --> 00:14:03,240
Speaker 1: comes in, like which ones are important. Having a nose

279
00:14:03,840 --> 00:14:06,720
Speaker 1: for what to approximate and what not to approximate, that's

280
00:14:06,720 --> 00:14:09,320
Speaker 1: what helps some scientists make more progress than others.

281
00:14:09,480 --> 00:14:11,440
Speaker 2: Yeah, And I think another thing to just sort of

282
00:14:11,480 --> 00:14:14,800
Speaker 2: note is that because this is a human endeavor. Sometimes

283
00:14:14,920 --> 00:14:17,959
Speaker 2: you're limited by what you can afford to get data

284
00:14:18,000 --> 00:14:20,200
Speaker 2: on you know, like maybe you do want to know

285
00:14:20,640 --> 00:14:23,360
Speaker 2: how much cows are farting, but in order to get

286
00:14:23,360 --> 00:14:26,440
Speaker 2: that data, you would need seventy billion dollars so that

287
00:14:27,040 --> 00:14:30,320
Speaker 2: farmers could attach sensors to the rear end of every cow.

288
00:14:30,400 --> 00:14:33,400
Speaker 2: And so, like, you know, sometimes you know there's data

289
00:14:33,440 --> 00:14:34,920
Speaker 2: you want, but you can't get it because there's not

290
00:14:35,040 --> 00:14:37,080
Speaker 2: enough money or it's not possible. Maybe one day you

291
00:14:37,080 --> 00:14:39,000
Speaker 2: can get it, Maybe those sensors will become cheap.

292
00:14:39,160 --> 00:14:42,080
Speaker 1: Is seventy billion dollars your like a fantastical number for

293
00:14:42,120 --> 00:14:44,360
Speaker 1: some like absurd amount of money for a science experiment?

294
00:14:44,920 --> 00:14:47,640
Speaker 2: Yeah, I guess that's wow. Yeah what is yours? I

295
00:14:47,640 --> 00:14:49,280
Speaker 2: guess you're a physicist, so it's going to be like.

296
00:14:49,360 --> 00:14:52,680
Speaker 1: Well, that's embarrassing because our next project is one hundred

297
00:14:52,720 --> 00:14:56,600
Speaker 1: billion dollars, So we're like already above the Kelly threshold

298
00:14:56,640 --> 00:14:58,440
Speaker 1: for like absurd amounts of money.

299
00:14:58,600 --> 00:15:01,400
Speaker 2: But wait, like, okay, but that's not like your personal project.

300
00:15:01,440 --> 00:15:04,560
Speaker 2: That's like LHC or like a new particle collider or something.

301
00:15:04,640 --> 00:15:07,240
Speaker 1: Right, Yeah, the new next particle collider budget is about

302
00:15:07,280 --> 00:15:10,440
Speaker 1: one hundred billion, Yes exactly, so more than a planet

303
00:15:10,480 --> 00:15:12,200
Speaker 1: wide cow farts sensor network.

304
00:15:12,360 --> 00:15:15,280
Speaker 2: Well, you guys better make some really important discoveries with

305
00:15:15,320 --> 00:15:17,880
Speaker 2: that money. Otherwise I'm disappointed because I want to know

306
00:15:17,920 --> 00:15:19,120
Speaker 2: what's happening with the cow farts.

307
00:15:20,280 --> 00:15:22,400
Speaker 1: But you bring up another point, which is the data.

308
00:15:22,560 --> 00:15:25,160
Speaker 1: So models are useful. There are a system that tell

309
00:15:25,240 --> 00:15:28,640
Speaker 1: us how the past becomes the future, but you also

310
00:15:28,680 --> 00:15:31,640
Speaker 1: need some data so you know which past you had. Right,

311
00:15:32,080 --> 00:15:36,680
Speaker 1: models describe essentially any possible universe. The rules determine which

312
00:15:36,840 --> 00:15:39,760
Speaker 1: set of universes we might live in, but the data

313
00:15:39,800 --> 00:15:42,760
Speaker 1: constrain it. It tells us which past we had. So

314
00:15:42,800 --> 00:15:45,080
Speaker 1: the rules tell you how the past becomes the future,

315
00:15:45,280 --> 00:15:47,680
Speaker 1: but you need to know which past we were in

316
00:15:47,720 --> 00:15:50,640
Speaker 1: so we know which future will have. So in our

317
00:15:50,680 --> 00:15:53,320
Speaker 1: ball tossing analogy, there's lots of different ways that could

318
00:15:53,320 --> 00:15:55,440
Speaker 1: toss a ball. I Coatalla said high or low, or

319
00:15:55,440 --> 00:15:58,520
Speaker 1: fast or slow or east or west. The rules connect

320
00:15:58,600 --> 00:16:02,360
Speaker 1: the initial conditions that data the past to the future.

321
00:16:02,560 --> 00:16:05,040
Speaker 1: But you need to know where did I throw the ball.

322
00:16:05,160 --> 00:16:07,720
Speaker 1: So if I'm writing a simulation of that ball toss,

323
00:16:07,800 --> 00:16:09,360
Speaker 1: I got to encode in the laws of physics. But

324
00:16:09,400 --> 00:16:11,320
Speaker 1: then I need a data point. I need to say

325
00:16:11,480 --> 00:16:13,560
Speaker 1: the ball was here and it was moving in this

326
00:16:13,640 --> 00:16:16,920
Speaker 1: direction at this velocity. Then I can predict the future.

327
00:16:17,280 --> 00:16:21,160
Speaker 1: Without that, it's useless. Right, So you need these two components.

328
00:16:21,160 --> 00:16:24,000
Speaker 1: You need the models, plus you need the data, and

329
00:16:24,040 --> 00:16:27,520
Speaker 1: then you need more data. Say I'm predicting the ball toss,

330
00:16:27,680 --> 00:16:29,560
Speaker 1: I want to check in halfway and say, hey, it's

331
00:16:29,600 --> 00:16:32,960
Speaker 1: my model correct, doesn't need an adjustment. A way to

332
00:16:33,000 --> 00:16:35,920
Speaker 1: improve your modeling is to shorten the prediction time, to

333
00:16:35,920 --> 00:16:38,000
Speaker 1: say I'm not going to predict the whole path. I'm

334
00:16:38,000 --> 00:16:39,440
Speaker 1: going to predict the second and then I'm going to

335
00:16:39,480 --> 00:16:41,400
Speaker 1: take a measurement and if it's off, I'm going to

336
00:16:41,440 --> 00:16:43,920
Speaker 1: correct it so that if my model has veered off

337
00:16:43,920 --> 00:16:47,600
Speaker 1: from reality, it doesn't get further off. And so the

338
00:16:47,600 --> 00:16:49,640
Speaker 1: more data you have, the better your model is going

339
00:16:49,680 --> 00:16:52,280
Speaker 1: to be. So you need these two elements dancing together,

340
00:16:52,560 --> 00:16:54,000
Speaker 1: the models and the data.

341
00:16:54,160 --> 00:16:56,600
Speaker 2: Yeah, and I checked my weather app today and the

342
00:16:56,640 --> 00:16:59,600
Speaker 2: prediction for tomorrow was changed, And so I'm guessing we

343
00:16:59,640 --> 00:17:03,280
Speaker 2: do this same thing with weather Way update. So I

344
00:17:03,280 --> 00:17:05,200
Speaker 2: think we should talk in a second about what kinds

345
00:17:05,200 --> 00:17:08,720
Speaker 2: of data we collect to help us inform models. But

346
00:17:08,840 --> 00:17:11,080
Speaker 2: I guess my first question is we've talked about models

347
00:17:11,119 --> 00:17:13,879
Speaker 2: in general. How long have we been trying to model weather?

348
00:17:14,760 --> 00:17:15,840
Speaker 2: Aristotle problems, So.

349
00:17:18,400 --> 00:17:20,800
Speaker 1: People have had some crazy ideas about the weather for

350
00:17:20,880 --> 00:17:24,560
Speaker 1: thousands of years. The first real weather models were conceived

351
00:17:24,600 --> 00:17:28,159
Speaker 1: of in the nineteen twenties. And remember we didn't have

352
00:17:28,200 --> 00:17:31,080
Speaker 1: computers really until the fifties or so, so this was

353
00:17:31,119 --> 00:17:34,480
Speaker 1: like a conception and somebody did a proof of principal prediction.

354
00:17:34,800 --> 00:17:37,760
Speaker 1: They tried to predict the weather six hours later. They

355
00:17:37,800 --> 00:17:39,920
Speaker 1: took a bunch of measurements and said, let's try to

356
00:17:39,960 --> 00:17:43,320
Speaker 1: do some calculations. We have an early model. That calculation

357
00:17:43,480 --> 00:17:44,760
Speaker 1: took six weeks.

358
00:17:45,200 --> 00:17:46,240
Speaker 2: So not helpful.

359
00:17:46,600 --> 00:17:50,000
Speaker 1: Not helpful exactly, but they did it and it wasn't terrible,

360
00:17:50,080 --> 00:17:51,919
Speaker 1: and they sort of proved like, hey, you know, if

361
00:17:51,960 --> 00:17:54,679
Speaker 1: you could do this calculation more quickly, then maybe you

362
00:17:54,680 --> 00:17:57,479
Speaker 1: could even know the weather in advance. Oh my gosh,

363
00:17:57,600 --> 00:18:00,840
Speaker 1: what an idea. Right, Yeah, it was until the nineteen

364
00:18:00,920 --> 00:18:03,399
Speaker 1: fifties that we had the first computing models to do

365
00:18:03,440 --> 00:18:07,359
Speaker 1: these calculations. So we can make predictions in time shorter

366
00:18:07,440 --> 00:18:10,080
Speaker 1: than the prediction period. You could have enough data and

367
00:18:10,200 --> 00:18:13,120
Speaker 1: run your model and get an answer before the universe

368
00:18:13,200 --> 00:18:16,520
Speaker 1: revealed it, right, that's that's a prediction instead of a

369
00:18:16,560 --> 00:18:17,280
Speaker 1: post addiction.

370
00:18:18,320 --> 00:18:18,800
Speaker 2: That's better.

371
00:18:18,840 --> 00:18:21,439
Speaker 1: So we've been doing this for decades and the last

372
00:18:21,560 --> 00:18:24,119
Speaker 1: you know, seventy years or so have been improving the

373
00:18:24,160 --> 00:18:26,160
Speaker 1: models and improving the data.

374
00:18:26,200 --> 00:18:28,680
Speaker 2: Man, it's exciting to think that we, you know, we're

375
00:18:28,720 --> 00:18:33,280
Speaker 2: going from slide rules to make these predictions to massive supercomputers.

376
00:18:33,880 --> 00:18:36,680
Speaker 2: I'm appreciating my weather apps a bit more.

377
00:18:37,960 --> 00:18:40,679
Speaker 1: And also, like, six weeks sounds ridiculous. I don't know

378
00:18:40,680 --> 00:18:42,320
Speaker 1: that I could do that in six weeks. Oh, it's

379
00:18:42,320 --> 00:18:45,800
Speaker 1: an amazing calculation. And think about like not just the ideas,

380
00:18:45,800 --> 00:18:49,040
Speaker 1: but all the grunt work doing those calculations and the

381
00:18:49,119 --> 00:18:51,679
Speaker 1: human error that's possible. Like, it's amazing they did it

382
00:18:51,680 --> 00:18:54,399
Speaker 1: in six weeks, you know, So don't laugh at that.

383
00:18:54,600 --> 00:18:57,840
Speaker 2: Absolutely, So we've been doing this since the nineteen fifties.

384
00:18:57,920 --> 00:19:00,439
Speaker 2: Let's talk about what kind of data we're collect to

385
00:19:00,520 --> 00:19:02,960
Speaker 2: inform these models when we get back from the break.

386
00:19:22,760 --> 00:19:24,840
Speaker 2: All right, and we are back, So now we're going

387
00:19:24,880 --> 00:19:27,280
Speaker 2: to talk about the kinds of data that we use

388
00:19:27,400 --> 00:19:31,200
Speaker 2: to make weather predictions. And I'm gonna bet it involves satellites.

389
00:19:33,160 --> 00:19:36,640
Speaker 1: Always going with space first, right, yep, yep. It does

390
00:19:36,680 --> 00:19:41,359
Speaker 1: involve satellites, but there's an amazing, incredible variety of data

391
00:19:41,400 --> 00:19:44,160
Speaker 1: sources we have to understand the weather. And yet still

392
00:19:44,160 --> 00:19:47,520
Speaker 1: it's not nearly enough. Right as you'll hear, our weather

393
00:19:47,560 --> 00:19:50,399
Speaker 1: prediction would be so much better if we had more data.

394
00:19:50,480 --> 00:19:53,280
Speaker 1: We're really limited by the data. But we have lots

395
00:19:53,280 --> 00:19:56,040
Speaker 1: of different kinds. We have weather stations on the surface

396
00:19:56,160 --> 00:19:58,520
Speaker 1: and so a lot of these are called like automatic

397
00:19:58,520 --> 00:20:01,560
Speaker 1: weather stations that are scattered across the country. They're just

398
00:20:01,560 --> 00:20:04,800
Speaker 1: basically a bunch of sensors with a battery and like

399
00:20:04,880 --> 00:20:08,040
Speaker 1: either a wind turbine or a solar panel to get power,

400
00:20:08,520 --> 00:20:11,960
Speaker 1: and they measure things like temperature and pressure and wind

401
00:20:12,040 --> 00:20:16,359
Speaker 1: speed and precipitation, just the raw measurements you need to know,

402
00:20:16,400 --> 00:20:19,320
Speaker 1: like what's going on out there, what is the state

403
00:20:19,520 --> 00:20:22,480
Speaker 1: of the weather right now, because again, if you want

404
00:20:22,520 --> 00:20:24,560
Speaker 1: to predict the future weather, you've got to know what's

405
00:20:24,560 --> 00:20:25,919
Speaker 1: going on right now.

406
00:20:26,119 --> 00:20:28,359
Speaker 2: So is this like a citizen science thing where like

407
00:20:28,480 --> 00:20:31,040
Speaker 2: I could purchase one of these weather stations and hook

408
00:20:31,080 --> 00:20:33,400
Speaker 2: it into what's happening at like the national level.

409
00:20:33,600 --> 00:20:35,879
Speaker 1: Yes and no. So there are a few sort of

410
00:20:35,880 --> 00:20:39,200
Speaker 1: official stations. There's a bunch of different networks. The highest

411
00:20:39,280 --> 00:20:42,760
Speaker 1: quality ones. There's like ten thousand of these scattered around

412
00:20:42,760 --> 00:20:45,880
Speaker 1: the earth, and they're operated by weather services and government agencies.

413
00:20:46,520 --> 00:20:49,320
Speaker 1: But there's a bigger network of like quarter million of

414
00:20:49,359 --> 00:20:52,080
Speaker 1: these things. Some of these are personal weather stations that yeah,

415
00:20:52,240 --> 00:20:55,760
Speaker 1: people just build and publish the data. And there's an

416
00:20:55,800 --> 00:21:02,200
Speaker 1: amazing network it's called COCO ras COOCOHS Community Collaborative Rain,

417
00:21:02,400 --> 00:21:05,480
Speaker 1: Hail and snow Wow. If you can just like build

418
00:21:05,520 --> 00:21:09,000
Speaker 1: your own device and add it to the network and contribute,

419
00:21:09,040 --> 00:21:12,160
Speaker 1: and I think that's super awesome because it's definitely limited

420
00:21:12,480 --> 00:21:16,200
Speaker 1: by the data we have. One problem is that these

421
00:21:16,240 --> 00:21:18,200
Speaker 1: things tend to be where the people are, Like we

422
00:21:18,280 --> 00:21:20,960
Speaker 1: have a few, you know, top of Mount Washington or whatever,

423
00:21:21,240 --> 00:21:23,400
Speaker 1: but mostly these things are put up by people where

424
00:21:23,400 --> 00:21:26,840
Speaker 1: people are near, and so like there's lots in India,

425
00:21:26,920 --> 00:21:30,879
Speaker 1: for example, but very few across Siberia, And often the

426
00:21:30,880 --> 00:21:33,879
Speaker 1: best ones are at places like airports. Airports really need

427
00:21:33,920 --> 00:21:36,679
Speaker 1: to know whether so they have excellent weather stations. But

428
00:21:36,840 --> 00:21:39,760
Speaker 1: like the weather at LaGuardia is not the same as

429
00:21:39,760 --> 00:21:43,440
Speaker 1: the weather in Manhattan, and so often the airport weather

430
00:21:43,480 --> 00:21:46,400
Speaker 1: stations are very very precise and used heavily in the models,

431
00:21:46,840 --> 00:21:48,760
Speaker 1: but they're not giving you the measurements exactly where you

432
00:21:48,800 --> 00:21:49,399
Speaker 1: want them to be.

433
00:21:49,640 --> 00:21:53,240
Speaker 2: Okay, So is that a problem for just the people

434
00:21:53,240 --> 00:21:56,840
Speaker 2: who are in areas where there's not enough weather detectors

435
00:21:57,359 --> 00:21:59,480
Speaker 2: or is that a problem for all of us, because

436
00:21:59,520 --> 00:22:03,119
Speaker 2: what's happened in Siberia is important to what's happening in India.

437
00:22:03,240 --> 00:22:06,520
Speaker 1: Yeah, what happens in Siberia doesn't stay in Siberia. Unfortunately.

438
00:22:08,960 --> 00:22:12,840
Speaker 1: It contributes to uncertainty and error across the model. And

439
00:22:12,880 --> 00:22:14,600
Speaker 1: the Earth is one big system, which is why you

440
00:22:14,600 --> 00:22:16,359
Speaker 1: can't just be like, I'm only going to predict the

441
00:22:16,400 --> 00:22:18,800
Speaker 1: weather Manhattan. I only need to think about Manhattan. You

442
00:22:18,880 --> 00:22:20,800
Speaker 1: need to model the whole planet in order to get

443
00:22:20,800 --> 00:22:23,720
Speaker 1: the weather in Manhattan. So yeah, absolutely, And that's why

444
00:22:23,760 --> 00:22:25,960
Speaker 1: we have lots of different kinds of sensors, not just

445
00:22:26,040 --> 00:22:30,080
Speaker 1: these automatic weather stations. We also have things like weather radar,

446
00:22:30,160 --> 00:22:32,359
Speaker 1: and you might have seen these on your local weather channel.

447
00:22:32,400 --> 00:22:35,399
Speaker 1: Like let's look at the Doppler, this measure of precipitation.

448
00:22:35,520 --> 00:22:39,760
Speaker 1: It also measures the velocity of those rain drops. And

449
00:22:39,800 --> 00:22:41,600
Speaker 1: this is a really cool story because it comes out

450
00:22:41,600 --> 00:22:44,399
Speaker 1: of World War Two. It's another example of like reusing

451
00:22:44,480 --> 00:22:47,719
Speaker 1: military technology and infrastructure after World War two to do

452
00:22:47,800 --> 00:22:48,400
Speaker 1: some science.

453
00:22:48,600 --> 00:22:54,440
Speaker 2: Thank you war Oh boy, hot take pull it back.

454
00:22:55,119 --> 00:22:58,480
Speaker 1: Well, there you are again finding the silver lining. Tens

455
00:22:58,480 --> 00:23:02,040
Speaker 1: of millions of people died, but we have better weather predictions.

456
00:23:02,520 --> 00:23:05,280
Speaker 1: So the way radar works is that it sends these

457
00:23:05,600 --> 00:23:09,159
Speaker 1: pulses of microwave radiation. The wavelengths are like one to

458
00:23:09,200 --> 00:23:12,239
Speaker 1: ten centimeters, that's the microwave region. And it sends a

459
00:23:12,240 --> 00:23:14,679
Speaker 1: pulse for like a microsecond, and then it listens for

460
00:23:14,760 --> 00:23:18,680
Speaker 1: return signals. So like it sends this pulse and rain

461
00:23:18,800 --> 00:23:22,280
Speaker 1: drops will reflect, so it gets the signal back and

462
00:23:22,320 --> 00:23:24,439
Speaker 1: it listens for like a few milliseconds, and then it

463
00:23:24,440 --> 00:23:27,439
Speaker 1: sends another pulse, and so it can tell where the

464
00:23:27,440 --> 00:23:30,520
Speaker 1: clouds are, and it can tell the velocity of those

465
00:23:30,520 --> 00:23:34,560
Speaker 1: clouds by the change in frequency. This is the Doppler effect, right,

466
00:23:34,560 --> 00:23:36,920
Speaker 1: And this is exactly the same effect as like stars

467
00:23:36,960 --> 00:23:39,200
Speaker 1: are moving away from you, so their light is red

468
00:23:39,240 --> 00:23:42,440
Speaker 1: shifted when the radar pulse comes back. If the frequency

469
00:23:42,480 --> 00:23:46,200
Speaker 1: is shifted, you can tell which direction that rain drop

470
00:23:46,359 --> 00:23:46,879
Speaker 1: is moving.

471
00:23:47,080 --> 00:23:50,119
Speaker 2: So that sounds complicated because like there's not just one

472
00:23:50,520 --> 00:23:52,720
Speaker 2: rain drop out there, there's a bunch and so I

473
00:23:52,720 --> 00:23:56,440
Speaker 2: can imagine like your pulse getting lost as it bounces

474
00:23:56,480 --> 00:23:58,920
Speaker 2: off of multiple rain drops and doesn't make it back

475
00:23:58,920 --> 00:24:01,000
Speaker 2: to you. What am I miss saying? This sounds hard?

476
00:24:01,119 --> 00:24:03,879
Speaker 1: No, it is hard, But you're not detecting individual rain drops.

477
00:24:03,880 --> 00:24:07,159
Speaker 1: You're detecting clouds mostly like which direction is this cloud going?

478
00:24:07,760 --> 00:24:10,399
Speaker 1: And you know, initially this was a problem because in

479
00:24:10,440 --> 00:24:12,680
Speaker 1: World War Two, radar operators were trying to use radar

480
00:24:12,720 --> 00:24:15,760
Speaker 1: to discover like enemy planes, and they noticed, like man,

481
00:24:15,880 --> 00:24:18,600
Speaker 1: clouds are getting in the way. And then other folks

482
00:24:18,680 --> 00:24:21,280
Speaker 1: were like, oh wait, you can use radar to see clouds. Awesome,

483
00:24:21,600 --> 00:24:25,960
Speaker 1: and so and so. Then after World War Two they

484
00:24:26,000 --> 00:24:28,359
Speaker 1: started using this to measure the velocity of clouds and

485
00:24:28,400 --> 00:24:31,360
Speaker 1: to see them. And there's this moment in like nineteen

486
00:24:31,480 --> 00:24:34,800
Speaker 1: sixty one when Hurricane Carlo was approaching the coast of

487
00:24:34,800 --> 00:24:37,320
Speaker 1: Texas and Dan Rather went down there to a weather

488
00:24:37,359 --> 00:24:40,400
Speaker 1: station and they were using radar to see the clouds

489
00:24:40,440 --> 00:24:42,480
Speaker 1: and to see their direction, and he had them drawn

490
00:24:42,520 --> 00:24:45,320
Speaker 1: like the coast of Texas over this image of the

491
00:24:45,400 --> 00:24:48,280
Speaker 1: hurricane that showed everybody like, wow, this is a massive

492
00:24:48,359 --> 00:24:51,840
Speaker 1: hurricane moving fast towards the shore and probably save thousands

493
00:24:51,880 --> 00:24:56,240
Speaker 1: of lives because he publicized this like incoming storm much

494
00:24:56,280 --> 00:24:59,040
Speaker 1: more rapidly than we could otherwise without this kind of technology.

495
00:24:59,320 --> 00:24:59,439
Speaker 3: Wo.

496
00:25:00,240 --> 00:25:01,920
Speaker 1: Yeah, this weather radar is really helpful.

497
00:25:02,040 --> 00:25:03,840
Speaker 2: Do you think it still has the same effector or

498
00:25:03,840 --> 00:25:05,639
Speaker 2: do you think people are just kind of like, oh,

499
00:25:05,680 --> 00:25:08,600
Speaker 2: there's hurricanes, I've seen them before. They get big, and

500
00:25:08,640 --> 00:25:09,640
Speaker 2: they don't always leave.

501
00:25:09,760 --> 00:25:12,119
Speaker 1: People don't always leave. There's always somebody who's going to

502
00:25:12,240 --> 00:25:14,919
Speaker 1: ride out the storm, right, Yeah. And I don't know

503
00:25:14,960 --> 00:25:17,000
Speaker 1: with the psychology there, but at least now we can

504
00:25:17,040 --> 00:25:20,440
Speaker 1: inform people further in advance and let them know where

505
00:25:20,440 --> 00:25:22,879
Speaker 1: these things are likely to go. But there's still always uncertainty,

506
00:25:23,160 --> 00:25:24,920
Speaker 1: and we'll talk about that in a minute. You don't

507
00:25:24,960 --> 00:25:27,520
Speaker 1: just have one weather prediction. You have an ensemble. You

508
00:25:27,520 --> 00:25:30,919
Speaker 1: have an envelope of predictions because you don't have perfect

509
00:25:31,040 --> 00:25:33,439
Speaker 1: data and you don't have a perfect model, and so

510
00:25:33,600 --> 00:25:36,000
Speaker 1: often what you do is you vary your data a

511
00:25:36,000 --> 00:25:38,399
Speaker 1: little bit within the uncertainties and run the model again,

512
00:25:38,520 --> 00:25:40,280
Speaker 1: and then you get a different prediction. And I'll give

513
00:25:40,280 --> 00:25:42,760
Speaker 1: you a sense of the spread of the possible outcomes.

514
00:25:43,240 --> 00:25:45,360
Speaker 1: So you might see when there's like a hurricane approaching

515
00:25:45,400 --> 00:25:48,520
Speaker 1: the coast of Florida. They have a bunch of possible trajectories.

516
00:25:48,560 --> 00:25:50,840
Speaker 1: Those are all like different runs of the weather model,

517
00:25:51,080 --> 00:25:54,560
Speaker 1: assuming different initial conditions. Because we have uncertainty, we don't

518
00:25:54,600 --> 00:25:55,840
Speaker 1: have perfect data.

519
00:25:55,920 --> 00:26:00,200
Speaker 2: I personally really enjoy learning about the uncertainty in life

520
00:26:00,200 --> 00:26:02,359
Speaker 2: in general. And whenever I look at those I have

521
00:26:02,440 --> 00:26:06,320
Speaker 2: this weird feeling of security, like, yeah, like they figured

522
00:26:06,320 --> 00:26:08,359
Speaker 2: it out and they know what the errors are. We're good,

523
00:26:08,600 --> 00:26:11,600
Speaker 2: we know what to avoid. Maybe that's maybe that's a

524
00:26:11,600 --> 00:26:14,000
Speaker 2: little bit giving it a little too much credit, but

525
00:26:14,000 --> 00:26:14,760
Speaker 2: it's still amazing.

526
00:26:14,880 --> 00:26:18,520
Speaker 1: And another really important source of uncertainty in our models

527
00:26:18,680 --> 00:26:21,600
Speaker 1: is what's happening in the ocean, like how hot is it,

528
00:26:21,680 --> 00:26:24,080
Speaker 1: how cold is it, how things circulating, all this kind

529
00:26:24,080 --> 00:26:27,240
Speaker 1: of stuff, and so we need data about the ocean.

530
00:26:27,240 --> 00:26:29,040
Speaker 1: But not a lot of people live in the ocean,

531
00:26:29,080 --> 00:26:31,679
Speaker 1: so we don't have like these automatic weather stations, but

532
00:26:31,720 --> 00:26:34,720
Speaker 1: we do have buoy's. These are like floating weather stations,

533
00:26:35,280 --> 00:26:38,480
Speaker 1: and around the world there's a couple of thousand of these,

534
00:26:38,560 --> 00:26:42,480
Speaker 1: depending on the type, that have these like temperature sensors

535
00:26:42,560 --> 00:26:46,080
Speaker 1: on the surface. But we also have this hilarious data

536
00:26:46,359 --> 00:26:50,040
Speaker 1: from what's going on deeper in the ocean that historically

537
00:26:50,080 --> 00:26:54,880
Speaker 1: has come from people on ships taking a bucket, dropping

538
00:26:54,920 --> 00:26:57,760
Speaker 1: it into the ocean, pulling it up, and then measuring

539
00:26:57,760 --> 00:27:00,439
Speaker 1: the temperature of the water. And it's like really that

540
00:27:00,560 --> 00:27:03,320
Speaker 1: lo fi. But for many years that's all we had.

541
00:27:03,400 --> 00:27:06,280
Speaker 1: We had like no other way reliably to know how

542
00:27:06,320 --> 00:27:09,440
Speaker 1: cold is it in the ocean. And this is an

543
00:27:09,440 --> 00:27:12,200
Speaker 1: example of like it's not just data. You need to

544
00:27:12,200 --> 00:27:15,080
Speaker 1: take data and interpret it and clean it and correct it.

545
00:27:15,520 --> 00:27:17,280
Speaker 1: And I spoke to an expert here, you see, I

546
00:27:17,440 --> 00:27:19,600
Speaker 1: Jane Baldwin, who told me that like you had to

547
00:27:19,640 --> 00:27:21,720
Speaker 1: correct for like how long the bucket was out of

548
00:27:21,720 --> 00:27:24,639
Speaker 1: the water before they dunked the thermometer in it, and

549
00:27:24,840 --> 00:27:27,400
Speaker 1: how Japanese ships and US ships used a different bucket

550
00:27:27,680 --> 00:27:29,800
Speaker 1: and it had different effects, and like you got to

551
00:27:29,880 --> 00:27:32,040
Speaker 1: really know, you got to be an expert and how

552
00:27:32,080 --> 00:27:33,879
Speaker 1: this data was taken and what it really means.

553
00:27:34,080 --> 00:27:35,560
Speaker 2: Yeah, So for a while I was doing some water

554
00:27:35,640 --> 00:27:38,240
Speaker 2: quality work and we had this like tube and you

555
00:27:38,280 --> 00:27:40,720
Speaker 2: would put the tube underwater and then you'd sort of

556
00:27:40,720 --> 00:27:44,080
Speaker 2: press a button and like caps would pop into place

557
00:27:44,119 --> 00:27:46,160
Speaker 2: on both sides of the tube, and then you could

558
00:27:46,240 --> 00:27:47,640
Speaker 2: lift it up and so you could get a water

559
00:27:47,680 --> 00:27:51,000
Speaker 2: sample from specifically different depths, and it was it was

560
00:27:51,000 --> 00:27:52,480
Speaker 2: always kind of fun to use that device.

561
00:27:52,720 --> 00:27:55,560
Speaker 1: Yeah, and you might think, like that's ridiculous, what a

562
00:27:55,840 --> 00:27:58,320
Speaker 1: silly system, And it's a little bit silly, but if

563
00:27:58,359 --> 00:28:01,680
Speaker 1: it's the only data you have, it's better than no data. Yeah, right,

564
00:28:01,920 --> 00:28:04,920
Speaker 1: as long as you understand the uncertainties in it. And

565
00:28:05,040 --> 00:28:07,600
Speaker 1: my friend Jane was telling me that misunderstanding this data

566
00:28:07,840 --> 00:28:11,040
Speaker 1: might be a cause for some weird pauses and global

567
00:28:11,080 --> 00:28:13,680
Speaker 1: warming trends, that it could just be like a misinterpretation

568
00:28:13,880 --> 00:28:17,920
Speaker 1: of this ship bucket dunk data.

569
00:28:16,960 --> 00:28:19,159
Speaker 2: I know, we're so moch.

570
00:28:20,560 --> 00:28:24,000
Speaker 1: These days. We have these cool robotic floats that like

571
00:28:24,040 --> 00:28:26,119
Speaker 1: float on the surface of the ocean and then dive

572
00:28:26,200 --> 00:28:29,639
Speaker 1: down up to two thousand meters measure things down in

573
00:28:29,720 --> 00:28:31,480
Speaker 1: the ocean, and then come back up and beam it

574
00:28:31,680 --> 00:28:35,120
Speaker 1: to satellites or whatever. So we're getting better obviously, yes,

575
00:28:35,320 --> 00:28:39,280
Speaker 1: But you know what's really valuable is longitudinal data. Like

576
00:28:39,760 --> 00:28:41,800
Speaker 1: you want data as far back as you can so

577
00:28:41,800 --> 00:28:44,640
Speaker 1: you can understand bigger trends. So you can't just like say, oh,

578
00:28:44,680 --> 00:28:47,280
Speaker 1: that ship bucket dunk data is ridiculous, let's ignore it.

579
00:28:47,280 --> 00:28:49,680
Speaker 1: It's the only data you have for like thirty years

580
00:28:49,840 --> 00:28:52,440
Speaker 1: and so trends in that data do tell you something

581
00:28:52,720 --> 00:28:53,240
Speaker 1: very cool.

582
00:28:53,600 --> 00:28:56,360
Speaker 2: Okay, so now we've gone down deep, how do we

583
00:28:56,400 --> 00:28:58,320
Speaker 2: get data from up high? Yeah?

584
00:28:58,360 --> 00:29:01,160
Speaker 1: Because the weather's not just at the surface, right, And

585
00:29:01,320 --> 00:29:03,840
Speaker 1: the weather folks call the surface the two meter level

586
00:29:03,840 --> 00:29:06,400
Speaker 1: because they want to measure the temperature not on the

587
00:29:06,480 --> 00:29:09,200
Speaker 1: ground literally, but like two meters up like where your

588
00:29:09,200 --> 00:29:11,720
Speaker 1: head is, essentially. But they also need to know what's

589
00:29:11,760 --> 00:29:14,000
Speaker 1: going on even further, so we take measurements in the

590
00:29:14,080 --> 00:29:17,959
Speaker 1: upper atmosphere. We use weather balloons, and these are literally

591
00:29:17,960 --> 00:29:20,360
Speaker 1: what you imagine. You put like a bunch of helium

592
00:29:20,440 --> 00:29:23,120
Speaker 1: in a balloon and you put a weather station on

593
00:29:23,160 --> 00:29:26,720
Speaker 1: it that commissure altitude, pressure, temperature, humidity, wind speed, et cetera.

594
00:29:27,360 --> 00:29:30,360
Speaker 1: And you just let it go and it rises because

595
00:29:30,360 --> 00:29:33,760
Speaker 1: helium rises, and as it goes up, the balloon expands

596
00:29:33,800 --> 00:29:36,080
Speaker 1: because the pressure in the upper atmosphere is less, and

597
00:29:36,160 --> 00:29:39,120
Speaker 1: eventually it pops and then the thing comes back down.

598
00:29:39,200 --> 00:29:42,600
Speaker 1: So these are like one time uses, right, and they

599
00:29:42,600 --> 00:29:44,320
Speaker 1: can go up like twenty kilometers.

600
00:29:44,680 --> 00:29:47,800
Speaker 2: When I visited Saint Catherine's University in Minnesota to give

601
00:29:47,840 --> 00:29:50,080
Speaker 2: a talk, they had a special day where they launched

602
00:29:50,120 --> 00:29:52,960
Speaker 2: a weather balloon, like you know for my visit and

603
00:29:53,600 --> 00:29:55,959
Speaker 2: you know, did a demonstration for all the students and

604
00:29:56,040 --> 00:29:57,960
Speaker 2: it was the coolest thing ever.

605
00:29:57,920 --> 00:30:01,680
Speaker 1: Super cool. Right, These are amazing experiments. And I know

606
00:30:01,760 --> 00:30:05,120
Speaker 1: people who do physics experiments on balloons where they like

607
00:30:05,160 --> 00:30:07,000
Speaker 1: go the Antarctic and they let up a balloon and

608
00:30:07,000 --> 00:30:09,120
Speaker 1: it floats in the atmosphere for like up to a

609
00:30:09,200 --> 00:30:12,000
Speaker 1: month or something, and like, wow, that's really brave work

610
00:30:12,000 --> 00:30:14,520
Speaker 1: because you spent like four years building this instrument and

611
00:30:14,520 --> 00:30:16,560
Speaker 1: then you're putting it on a balloon and to the

612
00:30:16,600 --> 00:30:20,440
Speaker 1: atmosphere and sometimes it's just gone, like it just disappears

613
00:30:20,520 --> 00:30:23,880
Speaker 1: and you lose your whole thesis. And this seems like

614
00:30:23,960 --> 00:30:26,440
Speaker 1: kind of bespoke, right, and it is. There's like a

615
00:30:26,480 --> 00:30:29,440
Speaker 1: couple hundred launches per day in the United States, but

616
00:30:29,480 --> 00:30:32,200
Speaker 1: it's not reliable. It's not like the place you've visited.

617
00:30:32,360 --> 00:30:34,680
Speaker 1: They do exactly the same balloon launch every single day

618
00:30:34,720 --> 00:30:37,120
Speaker 1: at the same time, right, which is the most useful

619
00:30:37,160 --> 00:30:40,360
Speaker 1: thing for a weather model. It's like reliable data and

620
00:30:40,400 --> 00:30:42,360
Speaker 1: we don't have a lot of them. But again, this

621
00:30:42,520 --> 00:30:44,920
Speaker 1: helps you probe the upper atmosphere. We don't have many

622
00:30:44,920 --> 00:30:47,760
Speaker 1: ways to measure the temperature in the upper atmosphere. This

623
00:30:47,800 --> 00:30:49,160
Speaker 1: is a really powerful one.

624
00:30:49,320 --> 00:30:50,640
Speaker 2: Do we also use planes.

625
00:30:51,040 --> 00:30:53,960
Speaker 1: We do use planes because every airplane you've been on

626
00:30:54,480 --> 00:30:57,760
Speaker 1: has really valuable information about weather because it samples from

627
00:30:57,840 --> 00:31:01,000
Speaker 1: the two meter level up to like thirty thousand feet.

628
00:31:01,160 --> 00:31:04,520
Speaker 1: An aircraft, of course have sensors to measure wind speed

629
00:31:04,520 --> 00:31:06,720
Speaker 1: and temperature and all this kind of stuff. So every

630
00:31:06,720 --> 00:31:10,200
Speaker 1: commercial airplane has these sensors, collects this data and then

631
00:31:10,400 --> 00:31:14,200
Speaker 1: sells it to the government. Noah buys this data because

632
00:31:14,240 --> 00:31:16,640
Speaker 1: there's so many flights, Like look at a map of

633
00:31:16,680 --> 00:31:19,320
Speaker 1: airplane flights for a single day in the United States.

634
00:31:19,360 --> 00:31:22,920
Speaker 1: There are so many flights they crisscross the country, and

635
00:31:22,960 --> 00:31:25,880
Speaker 1: it's incredibly valuable data. And this is usually very high

636
00:31:26,000 --> 00:31:28,880
Speaker 1: quality data because it's very important for these planes to

637
00:31:29,000 --> 00:31:29,840
Speaker 1: understand the weather.

638
00:31:30,160 --> 00:31:32,360
Speaker 2: I had no idea Noah was getting access to all

639
00:31:32,400 --> 00:31:33,960
Speaker 2: of that data. That's super cool.

640
00:31:34,080 --> 00:31:37,120
Speaker 1: It's super cool. Basically, any way you can imagine to

641
00:31:37,280 --> 00:31:41,520
Speaker 1: learn the state of the weather somewhere on Earth, somebody's

642
00:31:41,560 --> 00:31:43,160
Speaker 1: doing it, because the more data we have, the better

643
00:31:43,200 --> 00:31:45,840
Speaker 1: these models get. But then of course we can go

644
00:31:46,000 --> 00:31:48,800
Speaker 1: all the way up to space, right because there are

645
00:31:48,880 --> 00:31:51,560
Speaker 1: places where there are no automatic weather stations and there

646
00:31:51,600 --> 00:31:53,800
Speaker 1: are no buoys and there are no airplane flights, yet

647
00:31:53,800 --> 00:31:57,800
Speaker 1: they still contribute to the weather prediction in Kansas or

648
00:31:57,840 --> 00:32:01,720
Speaker 1: in Mexico City or whatever. So we have satellites, and

649
00:32:01,760 --> 00:32:04,240
Speaker 1: since about nineteen seventy nine we've had weather satellites. We

650
00:32:04,280 --> 00:32:07,120
Speaker 1: of course had satellites earlier than that, but none devoted

651
00:32:07,200 --> 00:32:11,480
Speaker 1: to like gathering weather data, and the primarily cover things

652
00:32:11,520 --> 00:32:14,080
Speaker 1: like storm systems and cloud patterns. They can tell you

653
00:32:14,120 --> 00:32:16,400
Speaker 1: where the snow is. They can also tell you like

654
00:32:16,440 --> 00:32:19,400
Speaker 1: where wildfires are, which is an important part of the weather.

655
00:32:20,320 --> 00:32:21,240
Speaker 2: Yeah, and so they.

656
00:32:21,120 --> 00:32:24,040
Speaker 1: Can't directly measure like what is the temperature in Houston

657
00:32:24,120 --> 00:32:27,440
Speaker 1: right now, but they can make indirect measurements like, for example,

658
00:32:27,760 --> 00:32:32,080
Speaker 1: they can measure the amount of infrared radiation from the surface,

659
00:32:32,520 --> 00:32:35,400
Speaker 1: and that is connected to the temperature, but it's actually

660
00:32:35,400 --> 00:32:38,520
Speaker 1: connected to the temperature of the surface, not the two

661
00:32:38,600 --> 00:32:41,800
Speaker 1: meter level. Right, So, like how hot is the blacktop

662
00:32:41,840 --> 00:32:44,120
Speaker 1: in Houston right now? That's what your satellite is telling you,

663
00:32:44,400 --> 00:32:46,520
Speaker 1: and you have to use that to infer how hot

664
00:32:46,600 --> 00:32:49,800
Speaker 1: is it two meters above the blacktop in Houston, which

665
00:32:49,840 --> 00:32:51,160
Speaker 1: is what you actually want to know.

666
00:32:51,600 --> 00:32:53,160
Speaker 2: That sounds hard, it's hard.

667
00:32:53,320 --> 00:32:56,280
Speaker 1: Yeah, exactly. And so we also don't have a lot

668
00:32:56,320 --> 00:32:59,640
Speaker 1: of satellites because they're expensive. There's something like twenty satellites

669
00:32:59,680 --> 00:33:03,880
Speaker 1: are between geostationary and polar orbits. Eight of them are

670
00:33:03,920 --> 00:33:06,600
Speaker 1: operated by Noah. But there's a bunch out there. But

671
00:33:06,640 --> 00:33:08,760
Speaker 1: my friend the climate scientist says that we might be

672
00:33:08,880 --> 00:33:11,320
Speaker 1: on the verge of having a lot more data because

673
00:33:11,960 --> 00:33:14,200
Speaker 1: launching stuff in a space is cheaper, and now we

674
00:33:14,240 --> 00:33:18,200
Speaker 1: can do like small satellites, CubeSats. These might give us

675
00:33:18,240 --> 00:33:21,680
Speaker 1: more data, not as high quality as like the dedicated

676
00:33:22,040 --> 00:33:26,160
Speaker 1: you know, super nerd designed billion dollar satellites. On the

677
00:33:26,240 --> 00:33:28,400
Speaker 1: other hand, we don't know what the future holds for

678
00:33:28,480 --> 00:33:32,640
Speaker 1: like supporting and operating these satellites. This requires money to

679
00:33:32,800 --> 00:33:35,800
Speaker 1: fund these things and have people interpreting these things. We

680
00:33:35,880 --> 00:33:38,200
Speaker 1: don't know how long the government is going to continue

681
00:33:38,240 --> 00:33:40,560
Speaker 1: to support it. They could just like unfund this stuff

682
00:33:40,640 --> 00:33:44,280
Speaker 1: or turn off weather stations. And you know it's more

683
00:33:44,320 --> 00:33:45,920
Speaker 1: than just like, oh, we turned it off for a year.

684
00:33:46,480 --> 00:33:50,960
Speaker 1: Having continuous records is super important for these models for

685
00:33:51,080 --> 00:33:53,560
Speaker 1: predicting the immediate weather, but also for the long term

686
00:33:53,600 --> 00:33:56,480
Speaker 1: climate models, which are essentially an average of the weather,

687
00:33:56,920 --> 00:33:59,440
Speaker 1: and so even turning it off briefly, could be very

688
00:33:59,480 --> 00:34:02,320
Speaker 1: damaging for our abilities to do long term predictions.

689
00:34:02,680 --> 00:34:04,760
Speaker 2: And I'm kind of blown away by the fact that

690
00:34:04,800 --> 00:34:07,080
Speaker 2: we only have twenty satellites. I guess I had assumed,

691
00:34:07,080 --> 00:34:09,400
Speaker 2: since you know, there's like five thousand satellites up there

692
00:34:09,480 --> 00:34:11,360
Speaker 2: or something. I guess most of them are dedicated to

693
00:34:11,440 --> 00:34:14,720
Speaker 2: like beaming cat videos to us from anywhere in the world.

694
00:34:14,800 --> 00:34:18,880
Speaker 2: But like weather seems so important, you know, for farmers,

695
00:34:19,000 --> 00:34:21,680
Speaker 2: for like people who are traveling, just like for everything.

696
00:34:21,800 --> 00:34:24,440
Speaker 1: Yeah, that's true, but the satellites don't give you a

697
00:34:24,480 --> 00:34:27,480
Speaker 1: direct measurement of what you're most interested in. They're essentially

698
00:34:27,520 --> 00:34:30,320
Speaker 1: like really good for filling in the gaps or places

699
00:34:30,360 --> 00:34:32,640
Speaker 1: where you have no other measurements. So yeah, it would

700
00:34:32,680 --> 00:34:35,160
Speaker 1: be great, but they're also super expensive, so you'll hear

701
00:34:35,160 --> 00:34:37,040
Speaker 1: at the end, I asked one of the climate scientists

702
00:34:37,080 --> 00:34:38,800
Speaker 1: I spoke to, like, if you had a billion dollars,

703
00:34:38,840 --> 00:34:41,480
Speaker 1: what would you spend it on? And satellites is not

704
00:34:41,520 --> 00:34:42,480
Speaker 1: their top priority.

705
00:34:42,840 --> 00:34:45,640
Speaker 2: Huh okay, all right, so maybe twenty is the right number.

706
00:34:47,280 --> 00:34:49,319
Speaker 1: So you have all these different kinds of data. You

707
00:34:49,360 --> 00:34:52,320
Speaker 1: have automatic weather stations, you have radar, you have buois,

708
00:34:52,400 --> 00:34:54,920
Speaker 1: you have ship bucket data, you have weather, balloons, aircraft,

709
00:34:54,960 --> 00:34:57,920
Speaker 1: you have satellites. What you need for your model are

710
00:34:57,960 --> 00:35:00,160
Speaker 1: the initial conditions. What you need for your model as

711
00:35:00,160 --> 00:35:03,000
Speaker 1: a set of what is the temperature and the pressure

712
00:35:03,040 --> 00:35:06,160
Speaker 1: and the humidity everywhere on the planet right now, so

713
00:35:06,200 --> 00:35:08,120
Speaker 1: that I can run it and predict it in the future.

714
00:35:08,800 --> 00:35:11,319
Speaker 1: And there's not a trivial step from like here I

715
00:35:11,360 --> 00:35:14,160
Speaker 1: have all this data to what are the initial conditions?

716
00:35:14,360 --> 00:35:17,600
Speaker 1: Because the data can disagree, right you have multiple measurements,

717
00:35:17,640 --> 00:35:21,040
Speaker 1: sometimes nearby, using different kinds of sensors. How do you

718
00:35:21,080 --> 00:35:23,880
Speaker 1: incorporate that, How do you clean this data, how do

719
00:35:23,920 --> 00:35:26,080
Speaker 1: you decide what to use? How do you merge all

720
00:35:26,120 --> 00:35:29,560
Speaker 1: of this into your best prediction? And so there's a

721
00:35:29,600 --> 00:35:32,160
Speaker 1: lot of work in this area. It's called data assimilation

722
00:35:32,760 --> 00:35:35,880
Speaker 1: of running sort of mini models fluid dynamics to do

723
00:35:36,000 --> 00:35:39,840
Speaker 1: like physics informed interpolations between the places where you don't

724
00:35:39,840 --> 00:35:43,600
Speaker 1: have measurements, and to factor in the various uncertainties from

725
00:35:43,640 --> 00:35:47,160
Speaker 1: the various different kinds of measurements. So sometimes you like

726
00:35:47,280 --> 00:35:49,840
Speaker 1: back up the model a little bit and feed in

727
00:35:49,920 --> 00:35:52,520
Speaker 1: some data and then use it to predict the current

728
00:35:52,520 --> 00:35:55,759
Speaker 1: initial conditions before you go to your full model, and

729
00:35:55,840 --> 00:35:58,320
Speaker 1: then you do what we talked about earlier, which is ensembling.

730
00:35:58,360 --> 00:36:01,000
Speaker 1: You say, well, here's my best guess for the weather,

731
00:36:01,080 --> 00:36:03,759
Speaker 1: like right now, before we even run the model. But

732
00:36:03,800 --> 00:36:05,839
Speaker 1: I'm going to make like one hundred versions of it,

733
00:36:05,920 --> 00:36:07,960
Speaker 1: and each one I'm going to tweak my assumptions a

734
00:36:08,040 --> 00:36:10,640
Speaker 1: little bit. So I get an envelope where I hope

735
00:36:10,719 --> 00:36:13,520
Speaker 1: reality somehow is described by one of these models, or

736
00:36:13,560 --> 00:36:15,840
Speaker 1: is near one of these models, or the spread in

737
00:36:15,880 --> 00:36:19,000
Speaker 1: these models describes my uncertainty in the state of the

738
00:36:19,040 --> 00:36:21,600
Speaker 1: initial conditions. We haven't even done any predictions yet. This

739
00:36:21,680 --> 00:36:23,960
Speaker 1: is just like measuring what's happening.

740
00:36:23,600 --> 00:36:27,080
Speaker 2: Now right well, and what's so stressful for me to

741
00:36:27,120 --> 00:36:30,160
Speaker 2: thinking about this is like your data are coming in constantly,

742
00:36:30,239 --> 00:36:31,880
Speaker 2: and so it's not like you do this once and

743
00:36:31,920 --> 00:36:33,879
Speaker 2: then you're like, okay, good, now we will project. It's

744
00:36:33,880 --> 00:36:36,359
Speaker 2: like every second more data are coming in. So this

745
00:36:36,400 --> 00:36:39,000
Speaker 2: has to be like a constant process that's happening over

746
00:36:39,040 --> 00:36:41,920
Speaker 2: and over again. And integrating the information into bigger models

747
00:36:41,920 --> 00:36:44,440
Speaker 2: in exactly, amazing, exactly.

748
00:36:44,920 --> 00:36:46,480
Speaker 1: And yeah, and we haven't even talked about how the

749
00:36:46,520 --> 00:36:47,080
Speaker 1: models work.

750
00:36:47,200 --> 00:36:49,240
Speaker 2: And so let's take a break and when we get back,

751
00:36:49,520 --> 00:37:10,719
Speaker 2: we'll talk about how those models work. All right, So

752
00:37:10,800 --> 00:37:13,640
Speaker 2: now we have all of this data and you've got

753
00:37:13,680 --> 00:37:16,480
Speaker 2: it into an ensemble and you sort of maybe know

754
00:37:16,560 --> 00:37:20,359
Speaker 2: what's happening right now plus some uncertainty. How do you

755
00:37:20,400 --> 00:37:21,960
Speaker 2: now predict what's going to happen next?

756
00:37:22,360 --> 00:37:24,719
Speaker 1: Yeah, so simple. You just break out your pencil and

757
00:37:24,760 --> 00:37:27,000
Speaker 1: paper and you do a bunch of strength theory calculations

758
00:37:27,040 --> 00:37:29,319
Speaker 1: and that's it. Right, it's just like physics. It into

759
00:37:29,320 --> 00:37:29,840
Speaker 1: the future.

760
00:37:30,320 --> 00:37:32,160
Speaker 2: Finally, strength theory is useful.

761
00:37:33,360 --> 00:37:37,160
Speaker 1: Yeah, unfortunately not, as we said earlier, like you can't

762
00:37:37,160 --> 00:37:40,399
Speaker 1: describe everything. You have to make assumptions about what you're

763
00:37:40,400 --> 00:37:43,239
Speaker 1: going to calculate and what you're going to simplify. Otherwise

764
00:37:43,280 --> 00:37:45,399
Speaker 1: you're never going to be able to make a prediction, right,

765
00:37:46,080 --> 00:37:47,520
Speaker 1: or your predictions are going to be done in a

766
00:37:47,560 --> 00:37:50,279
Speaker 1: thousand years for the weather that's happening in an hour,

767
00:37:50,320 --> 00:37:52,800
Speaker 1: and that's not useful. And so it's always a question

768
00:37:52,840 --> 00:37:56,479
Speaker 1: of how to judiciously make those assumptions. So the current

769
00:37:56,520 --> 00:37:58,759
Speaker 1: state of the art for weather modeling has basically two

770
00:37:58,840 --> 00:38:03,879
Speaker 1: big pieces. One is directly model the atmosphere itself as

771
00:38:03,920 --> 00:38:06,440
Speaker 1: if it's a big fluid. So you use like navea

772
00:38:06,480 --> 00:38:09,200
Speaker 1: Stokes equations and think about how it flows and how

773
00:38:09,200 --> 00:38:13,480
Speaker 1: temperature moves through it. That's the dynamical core of the model.

774
00:38:13,760 --> 00:38:16,960
Speaker 1: But there's a bunch of stuff that influences the atmosphere

775
00:38:17,120 --> 00:38:21,200
Speaker 1: that you don't explicitly include in the model. The clouds,

776
00:38:21,239 --> 00:38:25,319
Speaker 1: the convection, the ocean, the radiation, the surface temperature, all

777
00:38:25,360 --> 00:38:28,479
Speaker 1: this kind of stuff. Your model doesn't explicitly include that stuff.

778
00:38:28,520 --> 00:38:30,680
Speaker 1: We don't have like a complete model of the ocean

779
00:38:31,120 --> 00:38:34,560
Speaker 1: or the clouds, etc. And so we have like various

780
00:38:34,560 --> 00:38:38,560
Speaker 1: inputs to this core piece that they call parameterizations that

781
00:38:38,680 --> 00:38:42,240
Speaker 1: like capture the big picture effects of all these pieces

782
00:38:42,239 --> 00:38:45,560
Speaker 1: that are not included directly in our model but are

783
00:38:45,640 --> 00:38:46,760
Speaker 1: influencing us.

784
00:38:47,320 --> 00:38:50,520
Speaker 2: So I feel like this is a question where you

785
00:38:50,520 --> 00:38:52,480
Speaker 2: think to yourself, am I about to ask a really

786
00:38:52,520 --> 00:38:54,560
Speaker 2: stupid question? But I'm going to move forward because that's

787
00:38:54,560 --> 00:38:58,880
Speaker 2: my job in this podcast. The atmosphere is not a

788
00:38:58,920 --> 00:39:04,840
Speaker 2: fluid though, right, So like guy, am, so, what why

789
00:39:04,920 --> 00:39:07,120
Speaker 2: are we doing? Are we modeling it as a fluid

790
00:39:07,200 --> 00:39:10,040
Speaker 2: because we just can't model it as something else because

791
00:39:10,040 --> 00:39:15,600
Speaker 2: it's too complicated and fluids are a simplification or Daniel,

792
00:39:15,640 --> 00:39:17,480
Speaker 2: I don't think the atmosphere is a fluid.

793
00:39:18,320 --> 00:39:20,040
Speaker 1: Well, it depends on what you mean by a fluid,

794
00:39:20,600 --> 00:39:22,760
Speaker 1: and you know, when it comes to like how things

795
00:39:22,800 --> 00:39:26,600
Speaker 1: flow and pressure, et cetera, the fluid dynamic equations do

796
00:39:26,719 --> 00:39:31,600
Speaker 1: describe the atmosphere. And so you know, fluid doesn't mean liquid, right,

797
00:39:31,719 --> 00:39:35,000
Speaker 1: Fluid is about how things flow and move. Right. So,

798
00:39:35,080 --> 00:39:38,360
Speaker 1: for example, like the mantle of the Earth is a fluid.

799
00:39:38,760 --> 00:39:42,560
Speaker 1: It flows. It's not a liquid, right, It's this weird

800
00:39:43,000 --> 00:39:45,920
Speaker 1: solidy kind of state and it moves, but it also flows,

801
00:39:45,960 --> 00:39:48,280
Speaker 1: and so you can describe it and it has convection.

802
00:39:48,440 --> 00:39:51,759
Speaker 1: You can describe it with fluid equations. And so the

803
00:39:51,840 --> 00:39:55,440
Speaker 1: Navi or Stokes equations are these famous equations that describe

804
00:39:55,440 --> 00:39:58,760
Speaker 1: fluid dynamics and they're pretty good at modeling the atmosphere.

805
00:39:58,760 --> 00:40:01,280
Speaker 1: They're not perfect, right, They're not perfect, but they're pretty

806
00:40:01,280 --> 00:40:03,239
Speaker 1: good at it. So, yeah, I think fluid is not

807
00:40:03,320 --> 00:40:05,160
Speaker 1: a liquid. It's just things that flow.

808
00:40:05,400 --> 00:40:08,799
Speaker 2: Okay, in my head, fluid is synonymous with liquid. And

809
00:40:08,840 --> 00:40:11,480
Speaker 2: so I have learned something today that will probably help

810
00:40:11,520 --> 00:40:13,319
Speaker 2: me not look silly in the future. That's great.

811
00:40:13,360 --> 00:40:15,840
Speaker 1: No, it was a great question. And so this is

812
00:40:15,880 --> 00:40:18,560
Speaker 1: the big picture. You have the dynamical core, and then

813
00:40:18,600 --> 00:40:21,080
Speaker 1: you have these parameterizations and we'll dig into that and

814
00:40:21,120 --> 00:40:23,200
Speaker 1: we'll describe sort of the US approach to it. But

815
00:40:23,239 --> 00:40:27,080
Speaker 1: there are three sort of major weather communities. There's the US,

816
00:40:27,120 --> 00:40:31,880
Speaker 1: the UK, and the Japanese, and they have slightly different approaches,

817
00:40:32,239 --> 00:40:35,280
Speaker 1: which is good because you know, different predictions can crosscheck

818
00:40:35,320 --> 00:40:38,200
Speaker 1: each other. But then some people think it's bad because hey,

819
00:40:38,239 --> 00:40:40,120
Speaker 1: let's pool all of our resources and make one big

820
00:40:40,160 --> 00:40:42,400
Speaker 1: global model, and that's awesome, but then you only have

821
00:40:42,440 --> 00:40:44,480
Speaker 1: the one and you're not sure. Maybe it's all wrong.

822
00:40:45,000 --> 00:40:46,799
Speaker 1: There's a lot of debate about, you know, how to

823
00:40:46,840 --> 00:40:49,520
Speaker 1: deal with global questions and global resources.

824
00:40:49,560 --> 00:40:50,440
Speaker 2: But who's the best.

825
00:40:53,040 --> 00:40:55,120
Speaker 1: Oh, I'll give you a ranking at the end. Okay,

826
00:40:55,160 --> 00:40:58,440
Speaker 1: all right, So the dynamical core, Right, think of the atmosphere.

827
00:40:58,600 --> 00:41:01,480
Speaker 1: We're going to treat the atmosphere basically like a spherical cow. Right,

828
00:41:01,680 --> 00:41:05,319
Speaker 1: It is a spherical fluid, right, the atmosphere. Yeah, it's

829
00:41:05,400 --> 00:41:09,680
Speaker 1: a thin shell around the Earth, and you know the

830
00:41:09,680 --> 00:41:13,440
Speaker 1: temperature and the pressure, and then you can describe how

831
00:41:13,480 --> 00:41:15,600
Speaker 1: it's going to flow, how the temperature and pressure are

832
00:41:15,600 --> 00:41:19,600
Speaker 1: going to change using the Navier Stokes equations. So Navia

833
00:41:19,640 --> 00:41:23,840
Speaker 1: Stokes is a set of really gnarly equations. They're nonlinear

834
00:41:24,160 --> 00:41:28,560
Speaker 1: partial differential equations. A differential equation is one where like

835
00:41:29,080 --> 00:41:33,040
Speaker 1: the value depends on how quickly it's changing. For example,

836
00:41:33,080 --> 00:41:36,680
Speaker 1: like antecology, you have differential equations that describe like predator

837
00:41:36,680 --> 00:41:39,680
Speaker 1: and prey. Right, these two things are coupled, and so

838
00:41:39,719 --> 00:41:43,160
Speaker 1: these are nonlinear partial differential equations, which means like that

839
00:41:43,280 --> 00:41:46,360
Speaker 1: depends on things squared or cubed. All that to say,

840
00:41:46,480 --> 00:41:49,799
Speaker 1: they're very, very difficult to solve. In fact, differential equations

841
00:41:49,840 --> 00:41:51,520
Speaker 1: in general are hard to solve. If you've taken a

842
00:41:51,560 --> 00:41:55,000
Speaker 1: differential equations class, it's basically like differential equations are not

843
00:41:55,080 --> 00:41:58,840
Speaker 1: solvable except for these four examples that we have answers

844
00:41:58,880 --> 00:42:01,160
Speaker 1: to and we know how to solve them, and so

845
00:42:01,200 --> 00:42:03,319
Speaker 1: you just got to memorize those. It's a little bit

846
00:42:03,360 --> 00:42:03,960
Speaker 1: like chemistry.

847
00:42:04,000 --> 00:42:05,000
Speaker 2: I gotta say, oh no.

848
00:42:05,080 --> 00:42:09,080
Speaker 1: It's mostly unsolved, right. And the Navias Stokes equations we've

849
00:42:09,080 --> 00:42:11,160
Speaker 1: known about them for like two hundred years that were

850
00:42:11,200 --> 00:42:14,239
Speaker 1: initially developed to try to answer these questions about like

851
00:42:14,280 --> 00:42:17,280
Speaker 1: how do things flow and how does momentum and mass

852
00:42:17,280 --> 00:42:21,640
Speaker 1: flow through pipes, et cetera. Essentially, people took Newton's second

853
00:42:21,719 --> 00:42:25,279
Speaker 1: law ethicals MA and applied it to fluids and then

854
00:42:25,400 --> 00:42:29,160
Speaker 1: added terms for like stress and pressure and viscosity. And

855
00:42:29,280 --> 00:42:31,319
Speaker 1: it's like a real triumph that we can describe this

856
00:42:31,400 --> 00:42:35,120
Speaker 1: at all. But calculationally it's a real bear. You can't

857
00:42:35,160 --> 00:42:38,400
Speaker 1: like sit down and derive a solution and say, here's

858
00:42:38,440 --> 00:42:40,640
Speaker 1: my pressure and temperature. Let me crunch it through the

859
00:42:40,719 --> 00:42:43,560
Speaker 1: Naviastokes equation. It's going to give me a formula. It's

860
00:42:43,640 --> 00:42:47,120
Speaker 1: all numerical approximations, which means it takes a lot of

861
00:42:47,120 --> 00:42:50,160
Speaker 1: computing to go from now to one second from now

862
00:42:50,280 --> 00:42:54,440
Speaker 1: or two seconds from now, and that computing means approximating things.

863
00:42:54,480 --> 00:42:59,080
Speaker 1: You're like doing numerical derivatives instead of exact analytical derivatives.

864
00:42:59,320 --> 00:43:02,200
Speaker 2: Okay, so some of that got pretty complicated, But what

865
00:43:02,320 --> 00:43:03,719
Speaker 2: I guess what I want to know is when this

866
00:43:03,760 --> 00:43:06,120
Speaker 2: is all done, I feel like, if we are trying

867
00:43:06,160 --> 00:43:08,600
Speaker 2: to model fluids, does this just tell us that, like,

868
00:43:09,040 --> 00:43:12,080
Speaker 2: the wind is now over here going this fast but

869
00:43:12,120 --> 00:43:14,440
Speaker 2: before it was over there? And how many are we

870
00:43:14,480 --> 00:43:16,319
Speaker 2: going to get to like how you get from that

871
00:43:16,440 --> 00:43:18,800
Speaker 2: to like and it's raining, Because that seems like a

872
00:43:18,840 --> 00:43:22,120
Speaker 2: different problem sort of than how fluid is moving around.

873
00:43:22,560 --> 00:43:24,000
Speaker 1: Yeah, so there's a couple of things to know there.

874
00:43:24,000 --> 00:43:26,120
Speaker 1: You're exactly right. It takes the current conditions and tries

875
00:43:26,120 --> 00:43:29,080
Speaker 1: to predict the future conditions. And those conditions are pressure

876
00:43:29,160 --> 00:43:33,520
Speaker 1: and temperature, wind speed, humidity, right, these kinds of things.

877
00:43:33,920 --> 00:43:37,680
Speaker 1: But because we're solving this numerically, we can't solve it everywhere.

878
00:43:38,239 --> 00:43:40,799
Speaker 1: If you have a formula and analytics description you can

879
00:43:40,800 --> 00:43:43,600
Speaker 1: write down, for like, where is my ball as I've

880
00:43:43,640 --> 00:43:45,120
Speaker 1: thrown it? I can write that down on a piece

881
00:43:45,120 --> 00:43:46,680
Speaker 1: of paper a formula. I can tell you where the

882
00:43:46,680 --> 00:43:49,080
Speaker 1: ball is at any point in time. You ask me

883
00:43:49,200 --> 00:43:52,080
Speaker 1: any point literally any value of T, I could plug

884
00:43:52,120 --> 00:43:54,759
Speaker 1: it into my formula give you an answer. But if

885
00:43:54,760 --> 00:43:57,440
Speaker 1: I don't have a formula, that's called an analytics description.

886
00:43:58,000 --> 00:44:00,160
Speaker 1: If all I have is a numerical estimate, then I've

887
00:44:00,200 --> 00:44:02,440
Speaker 1: made a grid. I've said I'm going to sample it

888
00:44:02,680 --> 00:44:04,839
Speaker 1: at time one time, two time, three time four, I'm

889
00:44:04,840 --> 00:44:06,960
Speaker 1: going to make an estimate of those times, and it

890
00:44:07,040 --> 00:44:10,000
Speaker 1: don't have an answer everywhere. And that's the situation we

891
00:44:10,040 --> 00:44:11,840
Speaker 1: have with weathers that they put a grid on the

892
00:44:11,840 --> 00:44:16,040
Speaker 1: planet and they estimate what's going to be the weather, temperature,

893
00:44:16,120 --> 00:44:19,799
Speaker 1: et cetera, in a grid of points, not everywhere over

894
00:44:19,840 --> 00:44:22,120
Speaker 1: the planet. And you might think, oh, I bet that

895
00:44:22,160 --> 00:44:24,680
Speaker 1: grid's pretty small, right, maybe they measure down to the

896
00:44:24,719 --> 00:44:28,120
Speaker 1: centimeter or something. No, the grid sizes are like ten

897
00:44:28,239 --> 00:44:29,520
Speaker 1: kilometer cubes.

898
00:44:29,800 --> 00:44:31,720
Speaker 2: What, Yes, that's too big.

899
00:44:31,800 --> 00:44:35,640
Speaker 1: It's too big, right. They are averaging the temperature and

900
00:44:35,640 --> 00:44:39,920
Speaker 1: the humidity over cubes of atmosphere ten kilometers on a side,

901
00:44:40,200 --> 00:44:42,920
Speaker 1: it's crazy and I think that's way too big. On

902
00:44:42,960 --> 00:44:45,279
Speaker 1: the other hand, that's still a lot of cubes, right,

903
00:44:45,400 --> 00:44:48,400
Speaker 1: Like the atmosphere is a lot of ten kilometer sized cubes.

904
00:44:48,920 --> 00:44:52,840
Speaker 1: And then the time steps are tens of minutes, right,

905
00:44:52,920 --> 00:44:55,319
Speaker 1: And this is awesome that we can even do this.

906
00:44:55,560 --> 00:44:58,640
Speaker 1: It requires massive supercomputers. We'll talk about it in a minute.

907
00:44:59,000 --> 00:45:01,080
Speaker 1: But the problem is that it ignores a lot of

908
00:45:01,080 --> 00:45:04,399
Speaker 1: little details like how big is a cloud? Usually they're

909
00:45:04,480 --> 00:45:08,160
Speaker 1: like a kilometer or less. And so you're missing out

910
00:45:08,200 --> 00:45:11,040
Speaker 1: on a lot of stuff by making your grid. Anything

911
00:45:11,040 --> 00:45:15,319
Speaker 1: that happens that's subgrid. That's crucial and important, but it's

912
00:45:15,320 --> 00:45:17,359
Speaker 1: small than the size of your grid is not being

913
00:45:17,400 --> 00:45:20,600
Speaker 1: described by your model. But your question was like, is

914
00:45:20,640 --> 00:45:23,160
Speaker 1: this directly outputting? Like, hey, it's going to rain on

915
00:45:23,239 --> 00:45:26,719
Speaker 1: Kelly's picnic. In a sense, yes, the direct outputs are

916
00:45:26,719 --> 00:45:30,320
Speaker 1: things like temperature, pressure, humidity, and those are enough to

917
00:45:30,360 --> 00:45:32,799
Speaker 1: tell you like, okay, it's going to rain because the

918
00:45:32,880 --> 00:45:35,680
Speaker 1: pressure and humidity are above some threshold or whatever. So

919
00:45:35,719 --> 00:45:39,239
Speaker 1: it's not directly outputting like three centimeters of snow. There's

920
00:45:39,239 --> 00:45:41,400
Speaker 1: another step you have to take after that, but it

921
00:45:41,440 --> 00:45:43,480
Speaker 1: feeds into that. So those are the inputs you need

922
00:45:43,520 --> 00:45:45,560
Speaker 1: to the next step, which says how much snow is

923
00:45:45,600 --> 00:45:46,040
Speaker 1: going to fall?

924
00:45:46,239 --> 00:45:47,360
Speaker 2: Okay, So let me see if I can do a

925
00:45:47,360 --> 00:45:50,560
Speaker 2: super simplified version of this. You get all of the

926
00:45:50,640 --> 00:45:53,160
Speaker 2: data that you have about a square in the grid,

927
00:45:53,840 --> 00:45:55,759
Speaker 2: and you do the best job you can to sort

928
00:45:55,760 --> 00:45:58,200
Speaker 2: of summarize it and ensemble it, and then you put

929
00:45:58,239 --> 00:46:02,520
Speaker 2: it in the model the runs through the equations. Than

930
00:46:02,600 --> 00:46:07,640
Speaker 2: does the information from the surrounding grids feed into your

931
00:46:07,680 --> 00:46:09,680
Speaker 2: grid as well, because you would, okay, because you would

932
00:46:09,680 --> 00:46:13,319
Speaker 2: expect there to be similarities between closely related squares in

933
00:46:13,320 --> 00:46:13,720
Speaker 2: the grid.

934
00:46:13,920 --> 00:46:15,640
Speaker 1: Yeah, you can't solve one grid at a time. You

935
00:46:15,640 --> 00:46:18,359
Speaker 1: have to solve all the grids. To grids touch each

936
00:46:18,360 --> 00:46:21,560
Speaker 1: other and influence each other and wind flows right right,

937
00:46:21,600 --> 00:46:25,319
Speaker 1: And that's why Siberia affects Manhattan over time because you've

938
00:46:25,360 --> 00:46:28,200
Speaker 1: propagated these things from grid cell to grid cell.

939
00:46:28,239 --> 00:46:32,840
Speaker 2: Absolutely, So does Siberia have bigger grid cells or just

940
00:46:32,960 --> 00:46:35,480
Speaker 2: the same number of small grid cells each with poorer

941
00:46:35,560 --> 00:46:36,279
Speaker 2: data in them.

942
00:46:36,560 --> 00:46:40,120
Speaker 1: Yeah, great questions. So some of these models are adaptive, right,

943
00:46:40,160 --> 00:46:42,880
Speaker 1: they have bigger grid cells where we have more uncertainty,

944
00:46:42,920 --> 00:46:46,200
Speaker 1: and smaller we have more data. The most precise ones

945
00:46:46,280 --> 00:46:49,680
Speaker 1: are the UK supercomputers. They go down to two kilometers

946
00:46:50,080 --> 00:46:52,840
Speaker 1: in some cases. Some of them are like fixed grids,

947
00:46:52,840 --> 00:46:55,400
Speaker 1: and some of them are adaptive exactly. It depends a

948
00:46:55,440 --> 00:46:58,000
Speaker 1: little bit on the model. But you know, there's lots

949
00:46:58,040 --> 00:47:00,680
Speaker 1: of details that are not described here, and these are

950
00:47:00,680 --> 00:47:04,840
Speaker 1: called the parameterizations, like especially subgrid stuff and exchanges with

951
00:47:04,920 --> 00:47:07,399
Speaker 1: other parts of the system. They're not just the fluid.

952
00:47:07,600 --> 00:47:10,720
Speaker 1: And one important thing are the clouds. Like you cannot

953
00:47:10,760 --> 00:47:13,799
Speaker 1: model every individual cloud because clouds are smaller than your

954
00:47:13,800 --> 00:47:15,759
Speaker 1: grid size. We do not have the compute to do that.

955
00:47:15,800 --> 00:47:19,080
Speaker 1: People have tried, and you can like do dedicated runs

956
00:47:19,120 --> 00:47:22,120
Speaker 1: on subsets to try to resolve clouds, but then you

957
00:47:22,160 --> 00:47:24,480
Speaker 1: don't have enough computing to do like ensembles. So you

958
00:47:24,520 --> 00:47:27,240
Speaker 1: can be like one prediction you're like, well, here's a prediction,

959
00:47:27,360 --> 00:47:29,279
Speaker 1: but I don't know what the uncertainties are on it

960
00:47:29,320 --> 00:47:32,160
Speaker 1: at all. And so instead what you tend to do

961
00:47:32,560 --> 00:47:36,840
Speaker 1: is parameterize the bulk outcomes, you know, the vapor, the clouds,

962
00:47:36,840 --> 00:47:39,839
Speaker 1: the liquid, the ice, the rain, the snow, etc. The condensation,

963
00:47:40,520 --> 00:47:42,760
Speaker 1: all this kind of stuff. You try to like grab

964
00:47:42,880 --> 00:47:46,040
Speaker 1: all that average over what's happening in that grid cell

965
00:47:46,040 --> 00:47:49,000
Speaker 1: and use that to inform your naveor Stokes equation. So

966
00:47:49,360 --> 00:47:52,680
Speaker 1: things you're not explicitly modeling, you're sort of like averaging

967
00:47:52,760 --> 00:47:55,719
Speaker 1: over You're losing all the details and saying like, well,

968
00:47:55,760 --> 00:47:57,640
Speaker 1: on average, this is going to be the effect of

969
00:47:57,719 --> 00:47:59,279
Speaker 1: clouds on my grid cell.

970
00:48:00,000 --> 00:48:01,960
Speaker 2: Do you think that as Oh, well, I was going

971
00:48:02,040 --> 00:48:04,000
Speaker 2: to say, do you think that as we continue to

972
00:48:04,040 --> 00:48:06,840
Speaker 2: have more and more computing power and more and more supercomputers,

973
00:48:06,840 --> 00:48:09,359
Speaker 2: at some point we'll be doing better here? But we

974
00:48:09,360 --> 00:48:13,160
Speaker 2: were just talking about how More's law. We've maybe hit

975
00:48:13,239 --> 00:48:16,120
Speaker 2: the end of that. So are we like, is this

976
00:48:16,160 --> 00:48:17,600
Speaker 2: as good as wherever we're going to get at it?

977
00:48:17,640 --> 00:48:19,719
Speaker 2: This is probably an end of the podcast question, but

978
00:48:19,760 --> 00:48:20,880
Speaker 2: I'm thinking it right now.

979
00:48:22,239 --> 00:48:24,920
Speaker 1: No, I think that there's lots of possibilities for making

980
00:48:24,960 --> 00:48:28,239
Speaker 1: this faster and more efficient, and not just wait till

981
00:48:28,280 --> 00:48:31,200
Speaker 1: computers get faster. Okay, there's definitely clever ideas and we'll

982
00:48:31,239 --> 00:48:33,840
Speaker 1: get there. Yeah, But there are lots of parts of

983
00:48:33,880 --> 00:48:36,600
Speaker 1: the weather that are not directly described in the dynamical core,

984
00:48:36,719 --> 00:48:39,480
Speaker 1: and not just the clouds, but also things like convection,

985
00:48:39,680 --> 00:48:44,359
Speaker 1: like vertical transport of heat. You know, especially there is

986
00:48:44,480 --> 00:48:49,160
Speaker 1: complex boundary mixing near the surface, like the lowest kilometer

987
00:48:49,239 --> 00:48:51,439
Speaker 1: or so of the atmosphere, where you have like heat

988
00:48:51,480 --> 00:48:55,279
Speaker 1: from the surface and turbulent momentum exchanges as wind is

989
00:48:55,320 --> 00:48:58,480
Speaker 1: like hitting mountains and stuff. These things. You can't model

990
00:48:58,520 --> 00:49:01,880
Speaker 1: all of those details, and so you have like parameterization

991
00:49:02,080 --> 00:49:05,640
Speaker 1: schemes that model the turbulence and the boundary level mixings.

992
00:49:06,160 --> 00:49:10,120
Speaker 1: There's radiation from the surface also, right that changes from

993
00:49:10,160 --> 00:49:14,239
Speaker 1: day to night. You have models of vegetation and snow

994
00:49:15,080 --> 00:49:17,920
Speaker 1: how those things couple. But then the biggest one is

995
00:49:17,960 --> 00:49:21,280
Speaker 1: the ocean. Right, Like we would love for our models

996
00:49:21,320 --> 00:49:25,640
Speaker 1: to include also a Navio Stoke simulation of the whole ocean, right,

997
00:49:25,920 --> 00:49:27,560
Speaker 1: might as well do that because the ocean it plays

998
00:49:27,600 --> 00:49:29,640
Speaker 1: a big role. But we don't have the compute for

999
00:49:29,680 --> 00:49:32,480
Speaker 1: that at all, So we just like use a slab

1000
00:49:32,520 --> 00:49:35,240
Speaker 1: ocean model. We just say, let's just assume the ocean

1001
00:49:35,320 --> 00:49:38,239
Speaker 1: is like simple, and we have a certain temperature, and

1002
00:49:38,280 --> 00:49:42,560
Speaker 1: we assume like how the energy transfers from the boundaries,

1003
00:49:43,000 --> 00:49:46,480
Speaker 1: and it's really quite simplified. But that's we're just limited, right.

1004
00:49:46,520 --> 00:49:48,440
Speaker 1: We don't have great data in the ocean, and we

1005
00:49:48,480 --> 00:49:50,759
Speaker 1: don't have the compute to also model the ocean as

1006
00:49:50,800 --> 00:49:54,360
Speaker 1: well as the atmosphere. So places where we don't do

1007
00:49:54,400 --> 00:49:57,239
Speaker 1: our best approximation, which is like Navia Stokes equations of

1008
00:49:57,239 --> 00:50:00,959
Speaker 1: the atmosphere, we have simplified versions, which are called parmetization. Says,

1009
00:50:01,000 --> 00:50:03,759
Speaker 1: feed in to the core. But in the end you

1010
00:50:03,800 --> 00:50:06,120
Speaker 1: got to take it to the computers. And this is

1011
00:50:06,160 --> 00:50:09,799
Speaker 1: why we have like massive supercomputers to make weather predictions.

1012
00:50:10,120 --> 00:50:13,040
Speaker 1: So Noah in the US has a couple of really

1013
00:50:13,080 --> 00:50:16,279
Speaker 1: big facilities. They're called Dogwood and Cactus. One of them

1014
00:50:16,320 --> 00:50:16,600
Speaker 1: is in.

1015
00:50:16,600 --> 00:50:18,560
Speaker 2: Virginia, You're welcome, everyone.

1016
00:50:18,280 --> 00:50:21,720
Speaker 1: And one of them is in Arizona, and they're huge,

1017
00:50:21,880 --> 00:50:26,640
Speaker 1: amazing computers. Those two have like twelve point one petaflops.

1018
00:50:26,960 --> 00:50:27,839
Speaker 2: You made that word up.

1019
00:50:28,160 --> 00:50:31,759
Speaker 1: It sounds like a made up word. Peta means quadrillion

1020
00:50:31,880 --> 00:50:35,759
Speaker 1: and flops are floating point operations. So you know it

1021
00:50:35,800 --> 00:50:39,440
Speaker 1: takes the computer time to add like three point nine

1022
00:50:39,480 --> 00:50:42,799
Speaker 1: to one to fourteen point four two and floating point

1023
00:50:42,880 --> 00:50:45,360
Speaker 1: numbers those numbers with a dot in them, right, not

1024
00:50:45,520 --> 00:50:50,279
Speaker 1: integers are more computationally expensive to add or subtract. And

1025
00:50:50,320 --> 00:50:52,120
Speaker 1: that's what most of these models do. They're like, add

1026
00:50:52,120 --> 00:50:54,480
Speaker 1: this number, multiply by this, and so this is like

1027
00:50:54,600 --> 00:50:57,319
Speaker 1: the way you measure the speed of a computer. And

1028
00:50:57,400 --> 00:51:01,280
Speaker 1: so these computers can each do twelve point one quadrillion

1029
00:51:01,560 --> 00:51:03,880
Speaker 1: floating point operations per second.

1030
00:51:04,080 --> 00:51:04,360
Speaker 2: Wow.

1031
00:51:04,440 --> 00:51:06,839
Speaker 1: Right, imagine the guys back in the nineteen twenties, they're

1032
00:51:06,880 --> 00:51:09,640
Speaker 1: like adding two numbers. It probably takes them a minute, right,

1033
00:51:09,719 --> 00:51:13,040
Speaker 1: or they're super good, takes them twenty seconds. The computer

1034
00:51:13,160 --> 00:51:18,480
Speaker 1: does twelve quadrillion a piece per second. Right. So together

1035
00:51:18,760 --> 00:51:22,160
Speaker 1: with all of their computers, Noah has about fifty petaflops

1036
00:51:22,520 --> 00:51:25,000
Speaker 1: and that's what it uses to run its model. And

1037
00:51:25,040 --> 00:51:27,719
Speaker 1: so that's the state of the art. In the United States.

1038
00:51:28,360 --> 00:51:32,080
Speaker 1: The Europeans have a couple of computers. Is one really

1039
00:51:32,080 --> 00:51:35,720
Speaker 1: big one in Bologna called Bull Sequanya and has about

1040
00:51:35,719 --> 00:51:40,279
Speaker 1: thirty petaflops. But the biggest, most powerful weather computer in

1041
00:51:40,320 --> 00:51:43,120
Speaker 1: the world is in the UK. It's at the Met

1042
00:51:43,160 --> 00:51:47,120
Speaker 1: Office and it's built by Microsoft and has sixty petaflops.

1043
00:51:47,239 --> 00:51:49,520
Speaker 1: And this is why the UK has some of the

1044
00:51:49,520 --> 00:51:52,320
Speaker 1: best weather prediction in the world because they have the

1045
00:51:52,360 --> 00:51:55,759
Speaker 1: biggest computer. They beat us. Yeah exactly, they just spent

1046
00:51:55,840 --> 00:51:58,359
Speaker 1: more money. They bought more computer. This is literally like

1047
00:51:58,640 --> 00:51:59,800
Speaker 1: money equals computing.

1048
00:52:00,719 --> 00:52:02,000
Speaker 2: Tea drinking bastards.

1049
00:52:02,400 --> 00:52:06,399
Speaker 1: Good day. Yeah, well, you know, they got tricky weather

1050
00:52:06,480 --> 00:52:09,400
Speaker 1: over there, and so they need it. It's an island

1051
00:52:09,400 --> 00:52:13,319
Speaker 1: after yet guys and the Japanese. The Japanese have a

1052
00:52:13,360 --> 00:52:16,439
Speaker 1: big investment in weather prediction computers. Also, this one called

1053
00:52:16,480 --> 00:52:19,960
Speaker 1: Prime HBC. It has thirty one petaflops. So these are

1054
00:52:19,960 --> 00:52:23,359
Speaker 1: really powerful devices and they run these huge models. And

1055
00:52:23,480 --> 00:52:25,799
Speaker 1: you know, think about what the model does. It predicts

1056
00:52:26,080 --> 00:52:30,080
Speaker 1: the state of the atmosphere on these pretty chunky grids.

1057
00:52:30,120 --> 00:52:32,319
Speaker 1: But it's still it's a huge amount of data, like

1058
00:52:32,440 --> 00:52:35,880
Speaker 1: every few minutes, every ten kilometers. My friend Jane was

1059
00:52:35,880 --> 00:52:38,880
Speaker 1: telling me that sometimes the data is so big that

1060
00:52:39,000 --> 00:52:41,280
Speaker 1: you just throw it away. You run it, you get

1061
00:52:41,280 --> 00:52:43,640
Speaker 1: like a summary number, but you can't keep all of

1062
00:52:43,640 --> 00:52:45,560
Speaker 1: the data because it would just like fill up all

1063
00:52:45,600 --> 00:52:48,400
Speaker 1: of the hard drives. Everywhere. And this is familiar for

1064
00:52:48,440 --> 00:52:50,399
Speaker 1: me because like at the LHC, we also we run

1065
00:52:50,440 --> 00:52:54,040
Speaker 1: these experiments every twenty five danoseconds. We throw away most

1066
00:52:54,040 --> 00:52:56,479
Speaker 1: of the data from that because it would just fill

1067
00:52:56,560 --> 00:52:59,400
Speaker 1: up all of our storage. And they're in a similar situation.

1068
00:52:59,440 --> 00:53:01,480
Speaker 1: They produce more data than they can store.

1069
00:53:01,960 --> 00:53:07,279
Speaker 2: So are these facilities where the Navier Stokes equations are

1070
00:53:07,320 --> 00:53:11,200
Speaker 2: being run or are these facilities where you have the

1071
00:53:11,239 --> 00:53:14,319
Speaker 2: output from each grid and now you are translating that

1072
00:53:14,400 --> 00:53:16,440
Speaker 2: into information about where the rain is falling?

1073
00:53:17,200 --> 00:53:21,640
Speaker 1: Both? Yeah, okay, So these programs do the data similation,

1074
00:53:22,000 --> 00:53:24,360
Speaker 1: come up with the current initial conditions, and then also

1075
00:53:24,640 --> 00:53:28,000
Speaker 1: run the model forward to make those predictions, and from

1076
00:53:28,040 --> 00:53:32,560
Speaker 1: that glean things like weather details, snowfall, et cetera. And

1077
00:53:32,600 --> 00:53:34,439
Speaker 1: so what you're getting on your phone, what you're hearing

1078
00:53:34,440 --> 00:53:37,400
Speaker 1: on TV is not just like what Jane, your local forecaster,

1079
00:53:37,560 --> 00:53:42,480
Speaker 1: came up with. She's relying heavily on these central predictions

1080
00:53:42,800 --> 00:53:47,160
Speaker 1: from major resources. Right, So, for example, if worldwide governments

1081
00:53:47,200 --> 00:53:50,160
Speaker 1: decide we don't need these computers anymore, we don't need

1082
00:53:50,200 --> 00:53:52,760
Speaker 1: these satellites, it's not like you could be like, that's cool,

1083
00:53:53,000 --> 00:53:55,560
Speaker 1: I got my local weather forecaster I don't need you.

1084
00:53:55,560 --> 00:53:58,719
Speaker 1: No your local weather forecaster is getting that information from

1085
00:53:58,800 --> 00:54:01,600
Speaker 1: these big models that are being run by the government.

1086
00:54:02,080 --> 00:54:06,359
Speaker 2: Oh wow, yeah, and did Noah get cuts recently? I'm

1087
00:54:06,360 --> 00:54:07,160
Speaker 2: going to bet they did.

1088
00:54:07,400 --> 00:54:09,120
Speaker 1: There were some talk about cuts. I don't know how

1089
00:54:09,160 --> 00:54:11,440
Speaker 1: much of that is going through. It's all kind of scary.

1090
00:54:12,160 --> 00:54:12,840
Speaker 1: It's hard to know.

1091
00:54:13,200 --> 00:54:15,839
Speaker 2: Yeah, okay, all right, we won't get into that. Moving on.

1092
00:54:16,800 --> 00:54:21,160
Speaker 1: But amazingly, currently we can pretty accurately predict the weather

1093
00:54:21,320 --> 00:54:24,040
Speaker 1: five to six days in the future, you know, and

1094
00:54:24,080 --> 00:54:26,680
Speaker 1: you mostly remember when the weather prediction is wrong. You

1095
00:54:26,719 --> 00:54:29,600
Speaker 1: mostly don't realize that most of the time it's right. Yeah,

1096
00:54:29,640 --> 00:54:30,840
Speaker 1: you know, it tells you it's going to rain, it

1097
00:54:30,840 --> 00:54:32,840
Speaker 1: tells you it's going to be study. It's mostly correct.

1098
00:54:32,920 --> 00:54:36,399
Speaker 1: It's amazing, but you know, there's still challenges. Things are

1099
00:54:36,440 --> 00:54:40,400
Speaker 1: not perfect. One of the biggest challenge is just incomplete information.

1100
00:54:41,000 --> 00:54:43,719
Speaker 1: You know, we don't have sensors in enough places, and

1101
00:54:43,760 --> 00:54:48,240
Speaker 1: we don't have enough sensors, and sometimes this data availability changes,

1102
00:54:48,280 --> 00:54:50,560
Speaker 1: you know, things go offline or come online. Now your

1103
00:54:50,560 --> 00:54:53,200
Speaker 1: model has to compensate for that. I don't have the data.

1104
00:54:53,320 --> 00:54:55,480
Speaker 1: Do I assume it's similar to the past. Do I

1105
00:54:55,520 --> 00:54:58,200
Speaker 1: try to ignore that kind of data. It's not easy

1106
00:54:58,280 --> 00:55:01,520
Speaker 1: to be running a model if the puts are constantly changing.

1107
00:55:01,760 --> 00:55:03,279
Speaker 2: The bucket had a hole in it, So now you

1108
00:55:03,320 --> 00:55:06,960
Speaker 2: don't have good bucket data exactly exactly.

1109
00:55:07,080 --> 00:55:08,600
Speaker 1: They got a new kind of bucket. You don't know

1110
00:55:08,680 --> 00:55:11,279
Speaker 1: how to calibrate it. I spoke to John Martin, he's

1111
00:55:11,280 --> 00:55:13,960
Speaker 1: a professor of meteorology, and he said that this might

1112
00:55:14,000 --> 00:55:16,759
Speaker 1: be the biggest challenge is how to combine the data

1113
00:55:16,880 --> 00:55:20,760
Speaker 1: to make a high quality initial state. That's one challenge.

1114
00:55:20,800 --> 00:55:24,760
Speaker 1: The other are these subgrid parameterizations. Can we develop better

1115
00:55:24,840 --> 00:55:28,040
Speaker 1: models for turbulent flow at the boundaries or for latent

1116
00:55:28,120 --> 00:55:32,120
Speaker 1: he's released back into the environment. And another limiting factor

1117
00:55:32,200 --> 00:55:35,520
Speaker 1: is just the computing cost. More computes, more GPUs from

1118
00:55:35,600 --> 00:55:40,040
Speaker 1: Nvidia means smaller grids, which means the effect of these approximations,

1119
00:55:40,040 --> 00:55:45,279
Speaker 1: these parameterizations is less. Another continuing challenge are rare and

1120
00:55:45,400 --> 00:55:49,920
Speaker 1: extreme events, like we're pretty good at predicting the bigger picture,

1121
00:55:50,120 --> 00:55:51,600
Speaker 1: like is it going to be sunny here, is it

1122
00:55:51,600 --> 00:55:55,560
Speaker 1: going to be rainy here? But like small, rare extreme events,

1123
00:55:55,560 --> 00:55:58,719
Speaker 1: like there's a tornado right here, that's more challenging because

1124
00:55:58,719 --> 00:56:01,600
Speaker 1: they depends in detail well on things that happen within

1125
00:56:01,680 --> 00:56:05,160
Speaker 1: the grid that we're averaging over. And so there's a

1126
00:56:05,200 --> 00:56:07,879
Speaker 1: lot of work being done right now. One thing we're

1127
00:56:07,880 --> 00:56:10,080
Speaker 1: hoping to do is like, let's reduce the grid size,

1128
00:56:10,160 --> 00:56:14,560
Speaker 1: get more computing, more accurate. Right. But another really promising

1129
00:56:14,640 --> 00:56:18,759
Speaker 1: error of research is using machine learning. Oh, there's this

1130
00:56:18,960 --> 00:56:21,879
Speaker 1: movement in many fields of science to use machine learning

1131
00:56:21,880 --> 00:56:25,239
Speaker 1: to make predictions by essentially skipping the physics. Like, the

1132
00:56:25,280 --> 00:56:27,520
Speaker 1: physics is hard, it takes a lot of time to

1133
00:56:27,560 --> 00:56:30,839
Speaker 1: push the initial conditions through these equations. In the end

1134
00:56:31,120 --> 00:56:34,640
Speaker 1: you have an input and an output. And the idea is, well,

1135
00:56:34,680 --> 00:56:38,320
Speaker 1: can we train machine learning, not a chatbot, not LMS,

1136
00:56:39,000 --> 00:56:43,160
Speaker 1: it's AI, but it's not LMS to map the initial

1137
00:56:43,160 --> 00:56:46,200
Speaker 1: conditions to the output because in the end it's just

1138
00:56:46,239 --> 00:56:49,200
Speaker 1: a mapping and one could learn it. And so we

1139
00:56:49,280 --> 00:56:52,480
Speaker 1: have these machine learning models that are simple functions that

1140
00:56:52,560 --> 00:56:55,040
Speaker 1: take the input and give you the output, and they

1141
00:56:55,040 --> 00:56:57,360
Speaker 1: don't have the physics encoded in them, but they learn

1142
00:56:57,480 --> 00:57:00,319
Speaker 1: from the simulations, they learn the patterns, they learn what

1143
00:57:00,400 --> 00:57:03,440
Speaker 1: the rules are implicitly, and so you don't have to

1144
00:57:03,480 --> 00:57:07,160
Speaker 1: go through all the detailed calculations. So this can dramatically

1145
00:57:07,200 --> 00:57:09,840
Speaker 1: speed up your predictions. We use these the large handroom

1146
00:57:09,840 --> 00:57:12,160
Speaker 1: collider all the time so that we don't have to,

1147
00:57:12,360 --> 00:57:15,279
Speaker 1: for example, model every single particle that might hit the

1148
00:57:15,320 --> 00:57:18,200
Speaker 1: detector and create another particle and another particle. We can

1149
00:57:18,320 --> 00:57:21,360
Speaker 1: learn to predict the final thing we're interested in and

1150
00:57:21,400 --> 00:57:24,240
Speaker 1: to sort of leapfrog over all the tiny details.

1151
00:57:24,600 --> 00:57:27,400
Speaker 2: And his machine learning being used right now for weather predictions,

1152
00:57:27,480 --> 00:57:29,160
Speaker 2: or they're just starting to work on how you would

1153
00:57:29,160 --> 00:57:29,360
Speaker 2: do that.

1154
00:57:29,920 --> 00:57:32,240
Speaker 1: They're using that now. There's sort of experimental. But there's

1155
00:57:32,240 --> 00:57:34,360
Speaker 1: a guy here at you see Irvine, Mike Pritchard, who

1156
00:57:34,400 --> 00:57:36,680
Speaker 1: is an expert in this kind of stuff, and it's

1157
00:57:36,800 --> 00:57:40,560
Speaker 1: very powerful, absolutely cool. Yeah. So I asked John Martin,

1158
00:57:40,600 --> 00:57:43,480
Speaker 1: if I give you a billion dollars to improve weather predictions,

1159
00:57:43,560 --> 00:57:45,800
Speaker 1: what would you do, And he said he would spend

1160
00:57:45,800 --> 00:57:48,800
Speaker 1: a billion dollars on ocean probes, like he wanted a

1161
00:57:48,840 --> 00:57:52,120
Speaker 1: more substantial understanding of how water is circulating in the

1162
00:57:52,160 --> 00:57:55,400
Speaker 1: ocean and temperature in the ocean and how that's all working.

1163
00:57:55,400 --> 00:57:58,160
Speaker 1: Because his suspicion was like, we're right next to this

1164
00:57:58,240 --> 00:58:00,800
Speaker 1: other big fluid that's affecting our temperaatereure, and we don't

1165
00:58:00,800 --> 00:58:03,000
Speaker 1: have much enough data about it. If we just knew

1166
00:58:03,320 --> 00:58:06,520
Speaker 1: more about the ocean, and this just highlights like how

1167
00:58:06,680 --> 00:58:09,400
Speaker 1: little information we have. It's not just a question of

1168
00:58:09,440 --> 00:58:11,840
Speaker 1: like puzzling out the rules of the universe, but just

1169
00:58:11,880 --> 00:58:15,640
Speaker 1: like knowing what's happening. If we had more data everywhere

1170
00:58:15,960 --> 00:58:19,720
Speaker 1: about temperature, pressure, about cosmic rays, we would just learn

1171
00:58:19,800 --> 00:58:22,760
Speaker 1: so much about the universe. And we have so few

1172
00:58:22,800 --> 00:58:25,480
Speaker 1: ways to probe. But we're really just like taking the

1173
00:58:25,560 --> 00:58:29,720
Speaker 1: tiniest teaspoon out of this massive river of data and

1174
00:58:29,760 --> 00:58:32,240
Speaker 1: trying to use that to understand the whole river. It's crazy.

1175
00:58:32,480 --> 00:58:34,360
Speaker 2: How good do you think weather prediction would have to

1176
00:58:34,360 --> 00:58:36,880
Speaker 2: be before people stopped complaining about weather prediction?

1177
00:58:37,800 --> 00:58:40,440
Speaker 1: I asked John that question, and his prediction was, quote,

1178
00:58:40,560 --> 00:58:44,040
Speaker 1: the complaining will never stop amazing. I think that, you know,

1179
00:58:44,080 --> 00:58:46,960
Speaker 1: weather prediction has improved a lot over the last few decades.

1180
00:58:47,720 --> 00:58:49,760
Speaker 1: It used to be you couldn't get any reliable prediction

1181
00:58:49,840 --> 00:58:52,200
Speaker 1: more than a day in advance. Now five six days,

1182
00:58:52,240 --> 00:58:55,600
Speaker 1: it's pretty reliable. But people expect that and they get

1183
00:58:55,680 --> 00:58:57,680
Speaker 1: used to it, and they're like, what, you didn't predict

1184
00:58:57,680 --> 00:59:00,360
Speaker 1: the weather or my ski trip in two weeks? I'm you,

1185
00:59:01,000 --> 00:59:03,240
Speaker 1: And so yeah, the complaining will never stop because we

1186
00:59:03,280 --> 00:59:07,080
Speaker 1: always just get used to the level of technological prowess

1187
00:59:07,720 --> 00:59:10,920
Speaker 1: that we've had, and so people want more because it's

1188
00:59:10,960 --> 00:59:13,680
Speaker 1: so important and it's a hard problem. There's so much

1189
00:59:13,720 --> 00:59:17,240
Speaker 1: physics here, there's instrumental science, there's so many different kinds

1190
00:59:17,280 --> 00:59:19,720
Speaker 1: of science at interface with each other. It's really an

1191
00:59:19,720 --> 00:59:22,160
Speaker 1: exciting field. And let me throw a special thanks to

1192
00:59:22,240 --> 00:59:24,920
Speaker 1: Professor Jane Baldwin here you see I who told me

1193
00:59:24,960 --> 00:59:27,520
Speaker 1: a lot about weather predictions, and Professor John Martin at

1194
00:59:27,520 --> 00:59:30,240
Speaker 1: Wisconsin who answered a lot of naive questions of mine.

1195
00:59:30,280 --> 00:59:31,000
Speaker 1: Thanks to both of you.

1196
00:59:31,200 --> 00:59:33,880
Speaker 2: Thank you community. All right, see you all next time.

1197
00:59:33,920 --> 00:59:35,440
Speaker 2: I hope the weather is nice where you are.

1198
00:59:35,920 --> 00:59:46,400
Speaker 5: It always will be nice where I am.

1199
00:59:46,520 --> 00:59:50,360
Speaker 2: Daniel and Kelly's Extraordinary Universe is produced by iHeartRadio. We

1200
00:59:50,400 --> 00:59:51,800
Speaker 2: would love to hear from you.

1201
00:59:51,920 --> 00:59:54,840
Speaker 1: We really would. We want to know what questions you

1202
00:59:55,080 --> 00:59:57,680
Speaker 1: have about this Extraordinary Universe.

1203
00:59:57,800 --> 01:00:00,760
Speaker 2: We want to know your thoughts on recent shows, suggestions

1204
01:00:00,760 --> 01:00:03,760
Speaker 2: for future shows. If you contact us, we will get

1205
01:00:03,800 --> 01:00:04,200
Speaker 2: back to you.

1206
01:00:04,480 --> 01:00:08,000
Speaker 1: We really mean it. We answer every message. Email us

1207
01:00:08,040 --> 01:00:11,240
Speaker 1: at questions at Danielandkelly.

1208
01:00:10,280 --> 01:00:12,400
Speaker 2: Dot org, or you can find us on social media.

1209
01:00:12,480 --> 01:00:16,280
Speaker 2: We have accounts on x Instagram, Blue Sky, and on

1210
01:00:16,360 --> 01:00:18,320
Speaker 2: all of those platforms. You can find us at D

1211
01:00:18,760 --> 01:00:20,280
Speaker 2: and K Universe.

1212
01:00:20,480 --> 01:00:22,040
Speaker 1: Don't be shy, write to us,