1
00:00:02,268 --> 00:00:03,770
In 1972,

2
00:00:03,770 --> 00:00:06,573
the first Landsat satellite
was launched into orbit,

3
00:00:07,107 --> 00:00:10,510
ushering in a revolutionary
new era of earth observation.

4
00:00:11,211 --> 00:00:13,780
More than 50 years
and eight satellites later,

5
00:00:14,080 --> 00:00:15,815
the Landsat program
has collected

6
00:00:15,815 --> 00:00:17,417
an immense amount of data

7
00:00:17,417 --> 00:00:19,252
that's proven
an invaluable resource

8
00:00:19,252 --> 00:00:20,086
to scientists

9
00:00:20,086 --> 00:00:22,689
studying the complexities
of our planet's surface.

10
00:00:23,223 --> 00:00:24,290
The abundance of data

11
00:00:24,290 --> 00:00:26,493
provides insights,
but it can also pose

12
00:00:26,493 --> 00:00:27,594
a daunting challenge

13
00:00:27,594 --> 00:00:30,363
for researchers to extract
and analyze information.

14
00:00:30,697 --> 00:00:33,166
Landsat 8 and 9 alone each gather

15
00:00:33,166 --> 00:00:35,502
close to a terabyte of data per day.

16
00:00:36,236 --> 00:00:38,471
Enter Artificial Intelligence.

17
00:00:38,471 --> 00:00:41,341
AI's popularity has taken off
in recent years

18
00:00:41,641 --> 00:00:42,475
with new tools

19
00:00:42,475 --> 00:00:45,845
that allow users to generate
imagery, transcribe audio,

20
00:00:46,046 --> 00:00:48,715
and even compose music
at the click of a button.

21
00:00:49,549 --> 00:00:51,818
This is nothing new
for the scientific community,

22
00:00:51,818 --> 00:00:53,486
however, which have been using

23
00:00:53,486 --> 00:00:55,889
artificial intelligence
methods for decades.

24
00:00:56,356 --> 00:00:56,856
When it comes

25
00:00:56,856 --> 00:00:59,926
to working with Landsat data,
one of the most popular A.I.

26
00:00:59,926 --> 00:01:01,861
tools is machine learning.

27
00:01:01,861 --> 00:01:03,930
Machine learning is a subset of AI

28
00:01:04,130 --> 00:01:04,764
that can train

29
00:01:04,764 --> 00:01:05,832
computer programs

30
00:01:05,832 --> 00:01:08,935
to recognize patterns
and analyze imagery, skills

31
00:01:08,935 --> 00:01:10,503
that prove
exceptionally useful

32
00:01:10,503 --> 00:01:12,806
in the application of Landsat data.

33
00:01:12,806 --> 00:01:14,841
In fact, when combined with Landsat

34
00:01:14,841 --> 00:01:16,609
machine learning models have led

35
00:01:16,609 --> 00:01:18,545
to a number of advances across a variety

36
00:01:18,545 --> 00:01:20,013
of scientific fields,

37
00:01:20,013 --> 00:01:21,347
granting further insight

38
00:01:21,347 --> 00:01:24,050
into our planet's
past, present and future.

39
00:01:24,851 --> 00:01:26,653
One of the major challenges
of working

40
00:01:26,653 --> 00:01:27,654
with satellite imagery

41
00:01:27,654 --> 00:01:30,924
like Landsat can actually be
found up in the sky,

42
00:01:31,524 --> 00:01:32,559
clouds

43
00:01:32,559 --> 00:01:34,394
obscuring Earth's surface
and casting

44
00:01:34,394 --> 00:01:36,396
shadows
that reduce visibility.

45
00:01:36,396 --> 00:01:38,431
A cloudy day
can be a downright nuisance

46
00:01:38,431 --> 00:01:41,000
when it comes to analyzing
certain satellite imagery.

47
00:01:41,434 --> 00:01:44,170
Pinpointing these clouds
helps to improve data quality

48
00:01:44,170 --> 00:01:46,106
by removing
noise and artifacts,

49
00:01:46,106 --> 00:01:48,775
making it easier
to detect changes over time.

50
00:01:49,309 --> 00:01:50,910
But while
accurate cloud detection

51
00:01:50,910 --> 00:01:53,246
across a massive dataset
such as Landsat

52
00:01:53,313 --> 00:01:54,948
would be a tall
task for any one

53
00:01:54,948 --> 00:01:57,417
human, it's a piece of cake
for a computer.

54
00:01:57,951 --> 00:02:01,287
In 2019, researchers
from Oregon State University

55
00:02:01,387 --> 00:02:03,256
constructed
a deep convolutional

56
00:02:03,256 --> 00:02:04,958
neural network model

57
00:02:04,958 --> 00:02:06,759
a machine learning tool
that excels

58
00:02:06,759 --> 00:02:08,928
at recognizing patterns
in imagery.

59
00:02:08,928 --> 00:02:11,231
With the help of existing
Landsat 8 data,

60
00:02:11,464 --> 00:02:12,732
they taught their neural network

61
00:02:12,732 --> 00:02:15,335
to automatically detect clouds
in satellite imagery

62
00:02:15,401 --> 00:02:18,872
with an amazing 97.1%
accuracy rate.

63
00:02:19,672 --> 00:02:21,508
The researchers
believe in the future

64
00:02:21,508 --> 00:02:22,909
cloud detection algorithms

65
00:02:22,909 --> 00:02:24,777
like this one
could even be harnessed

66
00:02:24,777 --> 00:02:28,348
to identify clouds across the
entire Landsat 8 archive.

67
00:02:30,683 --> 00:02:32,485
Machine learning's
benefits don't just end

68
00:02:32,485 --> 00:02:34,687
when the clouds
clear - down on the ground

69
00:02:34,721 --> 00:02:36,656
there's plenty
to keep an eye on.

70
00:02:36,656 --> 00:02:38,791
Our planet's one
constant is change.

71
00:02:39,192 --> 00:02:39,893
Earth's surface

72
00:02:39,893 --> 00:02:43,329
is perpetually evolving due
to human and natural forces.

73
00:02:43,997 --> 00:02:46,366
Landsat its ability
to track these changes over

74
00:02:46,366 --> 00:02:47,333
time has proven

75
00:02:47,333 --> 00:02:49,869
to be an incredible asset
to the scientific community,

76
00:02:50,470 --> 00:02:53,206
especially when used in
concert with machine learning.

77
00:02:53,840 --> 00:02:56,643
For example, researchers
from the University of Texas

78
00:02:56,643 --> 00:02:59,646
at Austin used Landsat data
with a random force

79
00:02:59,646 --> 00:03:01,981
classifier, yet
another type of machine

80
00:03:01,981 --> 00:03:03,249
learning tool that combines

81
00:03:03,249 --> 00:03:05,552
multiple decision trees
to make predictions.

82
00:03:06,252 --> 00:03:08,521
Using data from Landsats 4
through 8,

83
00:03:08,821 --> 00:03:11,524
They used the classifier
to map changes in land use

84
00:03:11,524 --> 00:03:14,661
in northwestern Belize
from the 1980s to the present.

85
00:03:16,029 --> 00:03:18,464
The results showed that
tropical forests and wetlands

86
00:03:18,464 --> 00:03:20,767
that don't have
a designated protection status

87
00:03:21,000 --> 00:03:23,836
are increasingly vulnerable
to deforestation due

88
00:03:23,836 --> 00:03:26,406
to Belize's expanding
industrial agriculture.

89
00:03:27,040 --> 00:03:28,942
By combining
these new advances in machine

90
00:03:28,942 --> 00:03:29,943
learning with Landsat

91
00:03:29,943 --> 00:03:31,911
to capacity
for looking back in time,

92
00:03:31,911 --> 00:03:33,680
researchers
believe in the future,

93
00:03:33,680 --> 00:03:35,248
this approach would make it
possible

94
00:03:35,248 --> 00:03:38,251
to provide robust estimates
of deforestation in Belize.

95
00:03:41,454 --> 00:03:44,190
As climate change drives our
planet's temperatures higher,

96
00:03:44,457 --> 00:03:44,991
so does the

97
00:03:44,991 --> 00:03:48,428
prevalence of extreme events
that put ecosystems at risk.

98
00:03:48,995 --> 00:03:49,996
Wildfires across

99
00:03:49,996 --> 00:03:52,865
the globe have increased
in frequency and intensity.

100
00:03:53,199 --> 00:03:56,002
Australia is no stranger
to these types of fires.

101
00:03:56,202 --> 00:03:58,771
The 2019-2020 bushfire season

102
00:03:58,771 --> 00:04:00,840
was one of the most
destructive on record.

103
00:04:01,341 --> 00:04:03,142
Using satellite imagery
to pinpoint

104
00:04:03,142 --> 00:04:05,044
when and where
wildfires are burning

105
00:04:05,044 --> 00:04:06,145
could be an important tool

106
00:04:06,145 --> 00:04:08,281
for assessing damage
in future fires.

107
00:04:09,349 --> 00:04:11,217
Researchers
at the University of Western

108
00:04:11,217 --> 00:04:13,019
Australia
have developed a new machine

109
00:04:13,019 --> 00:04:15,521
learning approach with
the help of Landsat 8 data

110
00:04:16,022 --> 00:04:19,158
to generate a 16-year
history of wildfire severity

111
00:04:19,158 --> 00:04:21,761
in the eucalyptus forests
of the continent's southwest.

112
00:04:22,762 --> 00:04:26,132
They fed data gathered by
satellites, including Landsat 8

113
00:04:26,132 --> 00:04:29,802
from 2005 to 2020
into a supervised classifier

114
00:04:29,869 --> 00:04:31,571
a type of machine
learning algorithm

115
00:04:31,571 --> 00:04:32,772
that learns to classify

116
00:04:32,772 --> 00:04:35,808
data based on labeled examples
provided during training.

117
00:04:36,776 --> 00:04:39,245
By teaching the algorithm
with examples from the past.

118
00:04:39,279 --> 00:04:41,481
This method of machine
learning could be used

119
00:04:41,481 --> 00:04:43,683
to predict the severity
of future wildfires.

120
00:04:43,950 --> 00:04:44,884
Critical data

121
00:04:44,884 --> 00:04:47,053
that could aid in the
management and conservation

122
00:04:47,053 --> 00:04:49,722
of Australia's
extensive eucalyptus forests.

123
00:04:50,690 --> 00:04:52,992
These are just a few examples
of the remarkable ways

124
00:04:52,992 --> 00:04:55,728
Landsat data and machine
learning tools are unlocking

125
00:04:55,728 --> 00:04:57,964
new possibilities
for understanding our planet.

126
00:04:58,731 --> 00:05:00,233
This combination
has already led

127
00:05:00,233 --> 00:05:02,235
to significant advances
in agriculture,

128
00:05:02,268 --> 00:05:04,570
forestry,
urban planning, climate change

129
00:05:04,570 --> 00:05:06,372
research and more

130
00:05:06,372 --> 00:05:08,908
as more satellite data
becomes available and machine

131
00:05:08,908 --> 00:05:11,110
learning techniques
continue to improve.

132
00:05:11,110 --> 00:05:12,445
So too will the potential

133
00:05:12,445 --> 00:05:14,580
for applications
in additional fields

134
00:05:14,580 --> 00:05:17,050
critical to the health
of our planet's ecosystems.

135
00:05:17,517 --> 00:05:18,084
In tackling

136
00:05:18,084 --> 00:05:20,386
the complex challenges
of today and tomorrow,

137
00:05:20,553 --> 00:05:22,622
the blending of Landsat data
and machine

138
00:05:22,622 --> 00:05:24,357
learning will be vital to help
people

139
00:05:24,357 --> 00:05:26,626
make better decisions
to protect our planet.

140
00:08:30,243 --> 00:08:31,878
In 1972, the

141
00:08:31,878 --> 00:08:34,780
Landsat mission launched its
first satellite into orbit,

142
00:08:35,147 --> 00:08:37,950
ushering in a revolutionary
new era of earth

143
00:08:37,950 --> 00:08:39,619
observation.

144
00:08:42,121 --> 00:08:43,623
The Landsat mission
has collected

145
00:08:43,623 --> 00:08:45,091
an immense amount of data

146
00:08:45,091 --> 00:08:45,625
that's proven

147
00:08:45,625 --> 00:08:47,627
an invaluable resource
to scientists

148
00:08:47,627 --> 00:08:50,096
studying the complexities
of our planet's surface.

149
00:08:50,563 --> 00:08:53,266
The incredible abundance
of data provides insights,

150
00:08:53,466 --> 00:08:55,401
but it can also pose
a daunting challenge

151
00:08:55,401 --> 00:08:58,170
for researchers to extract
and analyze information.

152
00:08:58,738 --> 00:09:01,307
Landsat eight and nine alone,
each gather

153
00:09:01,307 --> 00:09:03,543
close to a terabyte of data
per day.

154
00:09:04,310 --> 00:09:08,214
Enter artificial intelligence

155
00:09:09,649 --> 00:09:10,516
with new tools

156
00:09:10,516 --> 00:09:13,653
that allow users to generate
imagery, transcribe audio,

157
00:09:13,786 --> 00:09:16,656
and even compose music
at the click of a button.

158
00:09:17,523 --> 00:09:19,892
This is nothing new
for the scientific community,

159
00:09:19,892 --> 00:09:21,327
however, which have been using

160
00:09:21,327 --> 00:09:23,863
artificial intelligence
methods for decades.

161
00:09:24,497 --> 00:09:26,399
When it comes to working
with Landsat data.

162
00:09:26,399 --> 00:09:27,800
One of the most popular A.I.

163
00:09:27,800 --> 00:09:29,802
tools is machine learning.

164
00:09:29,802 --> 00:09:32,038
Machine
learning is a subset of A.I.

165
00:09:32,071 --> 00:09:33,573
that can train
computer programs

166
00:09:33,573 --> 00:09:36,642
to recognize patterns
and analyze imagery skills

167
00:09:36,642 --> 00:09:38,210
that prove exceptional useful

168
00:09:38,210 --> 00:09:40,513
in the application of Landsat
data.

169
00:09:40,513 --> 00:09:42,548
In fact,
when combined with Landsat

170
00:09:42,848 --> 00:09:44,984
machine learning models
have led to a number

171
00:09:44,984 --> 00:09:47,820
of advances across a variety
of scientific fields,

172
00:09:48,120 --> 00:09:50,656
granting further insight
into our planet's past,

173
00:09:50,756 --> 00:09:52,858
present and future.

174
00:09:52,858 --> 00:09:54,627
One of the major challenges
of working

175
00:09:54,627 --> 00:09:56,529
with satellite imagery
like Landsat

176
00:09:56,529 --> 00:10:00,032
can actually be
found up in the sky, clouds

177
00:10:00,466 --> 00:10:02,401
obscuring the Earth's surface
and casting

178
00:10:02,401 --> 00:10:04,103
shadows
that reduce visibility.

179
00:10:04,103 --> 00:10:06,305
A cloudy day
can be a downright nuisance

180
00:10:06,305 --> 00:10:08,874
when it comes to analyzing
certain satellite imagery.

181
00:10:10,042 --> 00:10:11,310
Accurately detecting these

182
00:10:11,310 --> 00:10:14,347
pesky clouds would be a tall
task for any one human,

183
00:10:14,580 --> 00:10:18,150
but a piece of cake
for a computer.

184
00:10:25,992 --> 00:10:29,161
In 2019, researchers
from Oregon State University

185
00:10:29,161 --> 00:10:31,263
constructed
a deep convolutional neural

186
00:10:31,263 --> 00:10:32,565
network model

187
00:10:32,565 --> 00:10:33,699
a machine learning tool

188
00:10:33,699 --> 00:10:36,268
that excels at recognizing
patterns in imagery.

189
00:10:36,836 --> 00:10:39,171
With the help of existing
Landsat eight data,

190
00:10:39,305 --> 00:10:40,539
they taught
their neural network

191
00:10:40,539 --> 00:10:43,309
to automatically detect clouds
and satellite imagery

192
00:10:43,309 --> 00:10:46,979
with an amazing 97.1%
accuracy rate.

193
00:10:47,713 --> 00:10:49,682
The researchers
believe in the future,

194
00:10:49,682 --> 00:10:51,150
this machine learning
technique

195
00:10:51,150 --> 00:10:53,352
could even be harnessed
to identify clouds

196
00:10:53,352 --> 00:10:55,955
across the entire Landsat
eight archive.

197
00:10:57,456 --> 00:11:00,026
Machine learning benefits
don't just end when the clouds

198
00:11:00,026 --> 00:11:01,594
clear down on the ground.

199
00:11:01,594 --> 00:11:03,529
There's plenty
to keep an eye on.

200
00:11:03,529 --> 00:11:05,731
Our planet's one
constant is change.

201
00:11:06,098 --> 00:11:06,799
Earth's surface

202
00:11:06,799 --> 00:11:10,302
is perpetually evolving due
to human and natural forces.

203
00:11:10,836 --> 00:11:13,706
Landsat its ability to track
these changes over time

204
00:11:13,939 --> 00:11:15,641
has proven to be
an incredible asset

205
00:11:15,641 --> 00:11:17,309
to the scientific community,

206
00:11:17,309 --> 00:11:20,046
especially when used in
concert with machine learning.

207
00:11:20,680 --> 00:11:23,416
For example, researchers
from the University of Texas

208
00:11:23,416 --> 00:11:25,317
at Austin
used the Landsat data

209
00:11:25,317 --> 00:11:27,520
with a random forest
classifier,

210
00:11:27,520 --> 00:11:30,289
yet another type of machine
learning tool that combines

211
00:11:30,289 --> 00:11:32,558
multiple decision trees
to make predictions.

212
00:11:33,092 --> 00:11:35,461
Using data from Landsat four
through eight.

213
00:11:35,695 --> 00:11:38,664
They used the classifier
to map the changes in land use

214
00:11:38,664 --> 00:11:41,934
in northwestern Belize
from the 1980s to the present.

215
00:11:42,902 --> 00:11:45,337
The results showed that
tropical forests and wetlands

216
00:11:45,571 --> 00:11:47,640
don't have
a designated protection status

217
00:11:47,807 --> 00:11:50,409
or are increasingly vulnerable
to deforestation

218
00:11:50,643 --> 00:11:53,212
due to Belize's expanding
industrial agriculture.

219
00:11:53,846 --> 00:11:55,748
By combining
these new advances in machine

220
00:11:55,748 --> 00:11:57,283
learning with Landsat
to capacity

221
00:11:57,283 --> 00:11:59,485
for looking back in time,
researchers

222
00:11:59,485 --> 00:12:01,387
believe this approach
makes it possible

223
00:12:01,387 --> 00:12:04,623
to provide robust estimates
of deforestation in Belize.

224
00:12:04,623 --> 00:12:06,192
Going forward,

225
00:12:08,294 --> 00:12:10,563
as climate change drives
our planet's temperatures

226
00:12:10,563 --> 00:12:11,197
higher,

227
00:12:11,197 --> 00:12:12,231
so does the prevalence

228
00:12:12,231 --> 00:12:14,767
of extreme events
that put ecosystems at risk.

229
00:12:15,234 --> 00:12:17,870
Wildfires across
the globe has seen an increase

230
00:12:17,870 --> 00:12:20,039
in frequency and intensity.

231
00:12:20,039 --> 00:12:22,508
Australia is no stranger
to these types of fires.

232
00:12:22,842 --> 00:12:23,843
The 2019

233
00:12:23,843 --> 00:12:25,411
2020 bushfire season

234
00:12:25,411 --> 00:12:27,513
was one of the most
destructive on record.

235
00:12:27,847 --> 00:12:29,749
Using satellite imagery
to pinpoint

236
00:12:29,749 --> 00:12:31,851
when and where
wildfires are burning

237
00:12:31,851 --> 00:12:32,985
can be an important tool

238
00:12:32,985 --> 00:12:35,688
for assessing damage
and preventing future fires.

239
00:12:36,288 --> 00:12:38,891
Researchers at the University
of Western Australia

240
00:12:43,095 --> 00:12:46,398
to generate a 16 year
history of wildfire severity

241
00:12:46,398 --> 00:12:49,235
in the eucalyptus forests
of the continent's southwest.

242
00:12:49,935 --> 00:12:51,203
They said data gathered by

243
00:12:51,203 --> 00:12:54,974
satellites, including Landsat
eight from 2005 to 2020

244
00:12:55,107 --> 00:12:57,810
into a supervised classifier,
a type of machine

245
00:12:57,810 --> 00:12:59,912
learning algorithm
that learns to classify

246
00:12:59,912 --> 00:13:02,882
data based on labeled examples
provided during training

247
00:13:03,849 --> 00:13:06,252
by teaching the algorithm
with examples from the past.

248
00:13:06,285 --> 00:13:08,020
This method of machine
learning could be used

249
00:13:08,020 --> 00:13:10,189
to predict the severity
of future wildfires.

250
00:13:10,389 --> 00:13:11,857
Critical data
that could aid in

251
00:13:11,857 --> 00:13:13,425
the management
and conservation

252
00:13:13,425 --> 00:13:15,861
of Australia's
extensive eucalyptus forests.

253
00:13:17,463 --> 00:13:19,965
These were just a few examples
of the remarkable ways

254
00:13:20,199 --> 00:13:23,102
the fusion of Landsat data
with machine learning tools

255
00:13:23,269 --> 00:13:24,937
are unlocking
new possibilities

256
00:13:24,937 --> 00:13:26,872
for understanding our planet.

257
00:13:26,872 --> 00:13:28,474
This collaboration
has already led

258
00:13:28,474 --> 00:13:31,343
to significant advances
in agriculture, forestry,

259
00:13:31,377 --> 00:13:34,079
urban planning, climate change
research and more

260
00:13:34,680 --> 00:13:36,982
as satellite data
such as Landsat becomes

261
00:13:36,982 --> 00:13:38,951
increasingly available
and machine

262
00:13:38,951 --> 00:13:40,986
learning techniques
continue to improve.

263
00:13:40,986 --> 00:13:43,389
So too will
the potential for applications

264
00:13:43,389 --> 00:13:44,423
in additional fields

265
00:13:44,423 --> 00:13:46,959
critical to the health
of our planet's ecosystems.

266
00:13:47,526 --> 00:13:48,093
In tackling

267
00:13:48,093 --> 00:13:50,563
the complex challenges
of today and tomorrow,

268
00:13:50,763 --> 00:13:52,598
the partnership
between Landsat data

269
00:13:52,598 --> 00:13:54,667
and machine
learning will be vital to help

270
00:13:54,667 --> 00:13:57,269
people make better decisions
to protect our planet.