1
00:00:02,460 --> 00:00:03,633
Thanks for the intro.

2
00:00:05,220 --> 00:00:07,140
We'll jump right into it here.

3
00:00:07,140 --> 00:00:10,590
So, a little background for this work,

4
00:00:10,590 --> 00:00:13,260
and really why would we need

5
00:00:13,260 --> 00:00:17,100
fine scale carbon accounting at all.

6
00:00:17,100 --> 00:00:21,810
In New York State, a big
legislation was passed,

7
00:00:21,810 --> 00:00:25,202
the Climate Leadership and
Community Protection Act of 2019,

8
00:00:25,202 --> 00:00:27,480
which is a bundle of
energy, sustainability,

9
00:00:27,480 --> 00:00:28,743
climate legislation.

10
00:00:29,850 --> 00:00:31,080
And in that bundle,

11
00:00:31,080 --> 00:00:33,912
they set really, really
aggressive emissions targets,

12
00:00:33,912 --> 00:00:36,510
the state's planning
to be net zero by 2050,

13
00:00:37,470 --> 00:00:41,460
and they're looking to offset
15% of the statewide emissions

14
00:00:41,460 --> 00:00:42,333
in some way.

15
00:00:43,272 --> 00:00:46,200
With 60% of the state being forested,

16
00:00:46,200 --> 00:00:49,110
policy makers are really gonna be leaning

17
00:00:49,110 --> 00:00:51,090
on the forest carbon sink,

18
00:00:51,090 --> 00:00:53,610
they're expecting it, or depending
on it to more than double

19
00:00:53,610 --> 00:00:55,020
in the next 30 years,

20
00:00:55,020 --> 00:00:58,680
which we know to be a tall, tall order,

21
00:00:58,680 --> 00:01:01,290
very likely impossible.

22
00:01:01,290 --> 00:01:03,090
But to this end,

23
00:01:03,090 --> 00:01:06,390
we're trying to help the
state understand the where,

24
00:01:06,390 --> 00:01:08,550
the how, and to what extent forests

25
00:01:08,550 --> 00:01:10,680
are gonna be able to contribute here.

26
00:01:10,680 --> 00:01:14,970
And so, you know, like Sarah
and the panel talked about,

27
00:01:14,970 --> 00:01:19,290
like we are right at the
peak of this hype cycle,

28
00:01:19,290 --> 00:01:24,290
and hopefully this talk is gonna be about

29
00:01:24,630 --> 00:01:27,270
maybe how we can jump over
that gap a little bit,

30
00:01:27,270 --> 00:01:30,240
but really being humble
about what we're able to do

31
00:01:30,240 --> 00:01:32,313
in this context at a state scale.

32
00:01:34,710 --> 00:01:36,360
So, stock changes,

33
00:01:36,360 --> 00:01:40,623
you know, what is this accounting
method in general, right?

34
00:01:41,820 --> 00:01:42,653
All it really is

35
00:01:42,653 --> 00:01:44,490
is about changes between
two points in time.

36
00:01:44,490 --> 00:01:46,590
So as trees grow,

37
00:01:46,590 --> 00:01:48,270
they're pulling carbon
outta the atmosphere,

38
00:01:48,270 --> 00:01:51,630
that's assumed to be a
removal from the atmosphere.

39
00:01:51,630 --> 00:01:54,750
And if you have a disturbance or a loss

40
00:01:54,750 --> 00:01:57,270
in above ground biomass, that's
assumed to be an emission.

41
00:01:57,270 --> 00:02:01,113
And so we're operating at
a super coarse level here.

42
00:02:02,940 --> 00:02:04,590
Really, the fate of that woody material

43
00:02:04,590 --> 00:02:05,610
is gonna have a big impact

44
00:02:05,610 --> 00:02:08,250
on the immediacy of those emissions.

45
00:02:08,250 --> 00:02:12,450
Right? Is the wood being
transferred to long-lived products,

46
00:02:12,450 --> 00:02:14,910
or is it being burned for fuel right away?

47
00:02:14,910 --> 00:02:16,563
We're above that, right?

48
00:02:17,460 --> 00:02:19,440
For sake of simplicity, and in scale,

49
00:02:19,440 --> 00:02:22,073
we're obscuring those details

50
00:02:22,073 --> 00:02:24,720
and just assuming any losses
in aboveground biomass

51
00:02:24,720 --> 00:02:26,400
are gonna be emissions.

52
00:02:26,400 --> 00:02:29,310
So this kind of information,
stock change information,

53
00:02:29,310 --> 00:02:33,000
is already implemented at
state and national scale

54
00:02:33,000 --> 00:02:34,620
by the United States Forest Service,

55
00:02:34,620 --> 00:02:37,203
with their Forest Inventory
and Analysis program.

56
00:02:39,030 --> 00:02:41,963
Folks go out and measure
trees on the ground,

57
00:02:41,963 --> 00:02:43,800
and they repeat those measurements

58
00:02:43,800 --> 00:02:45,303
every five to seven years.

59
00:02:46,890 --> 00:02:48,960
That information is absolutely invaluable,

60
00:02:48,960 --> 00:02:51,120
we couldn't do this kind
of thing without it,

61
00:02:51,120 --> 00:02:53,730
it's one of the best programs in the world

62
00:02:53,730 --> 00:02:55,380
for that kind of thing,

63
00:02:55,380 --> 00:02:56,820
but it's limited by the density

64
00:02:56,820 --> 00:02:59,190
of that sample spatially, right?

65
00:02:59,190 --> 00:03:00,270
There's only one plot

66
00:03:00,270 --> 00:03:02,970
for every 6,000 acres in New York,

67
00:03:02,970 --> 00:03:05,610
and you're limited temporally
by that remeasurement cycle,

68
00:03:05,610 --> 00:03:07,050
five to seven years.

69
00:03:07,050 --> 00:03:09,861
So we're trying to
translate these measurements

70
00:03:09,861 --> 00:03:13,110
into a map space, into wall to wall maps,

71
00:03:13,110 --> 00:03:17,010
and we're trying to do it
at an annual time-step,

72
00:03:17,010 --> 00:03:19,560
going back in time to 1990,
which is our baseline.

73
00:03:19,560 --> 00:03:22,350
The idea being that you could
take any parcel of land,

74
00:03:22,350 --> 00:03:23,760
any unit of land,

75
00:03:23,760 --> 00:03:25,920
and get this stock change information

76
00:03:25,920 --> 00:03:28,530
for any slice of time
in this 30 year period

77
00:03:28,530 --> 00:03:30,033
from 1990 up until now.

78
00:03:30,870 --> 00:03:33,330
That's the big goal here.

79
00:03:33,330 --> 00:03:35,373
So what are we doing specifically?

80
00:03:36,510 --> 00:03:40,710
We are leveraging the above
ground biomass measurements

81
00:03:40,710 --> 00:03:42,720
at FIA plots across New York State,

82
00:03:42,720 --> 00:03:46,350
and we're combining that with
Landsat satellite imagery,

83
00:03:46,350 --> 00:03:49,500
which offers that historical
perspective, importantly,

84
00:03:49,500 --> 00:03:52,950
and offers that regular, in
our case, annual, set of data.

85
00:03:52,950 --> 00:03:55,290
So both of these things,

86
00:03:55,290 --> 00:03:59,070
federally funded, really well
supported, widely adopted,

87
00:03:59,070 --> 00:04:02,250
and there's a whole host
of open source tooling

88
00:04:02,250 --> 00:04:06,480
and existing applications
that have made this possible.

89
00:04:06,480 --> 00:04:09,120
And so we relate these things spatially,

90
00:04:09,120 --> 00:04:11,580
and we train models based
on these relationships

91
00:04:11,580 --> 00:04:13,770
to make predictions statewide

92
00:04:13,770 --> 00:04:15,150
where we have Landsat information

93
00:04:15,150 --> 00:04:16,700
but we don't have measurements.

94
00:04:17,640 --> 00:04:20,850
So some nuts and bolts here, quickly.

95
00:04:20,850 --> 00:04:24,903
We used 2000 FIA plots,
inventoried since 2002,

96
00:04:26,190 --> 00:04:29,280
that gets at Charlotte's
point about, right,

97
00:04:29,280 --> 00:04:32,310
we kind of have this missing
piece of historical data

98
00:04:32,310 --> 00:04:33,723
before 2000.

99
00:04:35,310 --> 00:04:36,150
Plots were inventoried,

100
00:04:36,150 --> 00:04:38,580
but the methods are not the same,

101
00:04:38,580 --> 00:04:40,500
and so they don't really match up,

102
00:04:40,500 --> 00:04:43,263
so we're stuck using post-2000 data.

103
00:04:44,580 --> 00:04:46,020
Landsat spectral indices,

104
00:04:46,020 --> 00:04:49,080
which are various combinations
of red, green, blue,

105
00:04:49,080 --> 00:04:51,960
and near infrared imagery,

106
00:04:51,960 --> 00:04:53,793
use topographic and climatic data.

107
00:04:55,110 --> 00:04:57,450
Right, this is pretty coarse information,

108
00:04:57,450 --> 00:05:01,530
especially compared to what the
last talk was talking about,

109
00:05:01,530 --> 00:05:04,743
right, we don't have any
diameter diversity information.

110
00:05:05,700 --> 00:05:07,320
This is what we have wall to wall,

111
00:05:07,320 --> 00:05:09,360
and so that's what we have to use.

112
00:05:09,360 --> 00:05:12,540
We feed all this information
into a set of models

113
00:05:12,540 --> 00:05:13,950
that we ensemble together.

114
00:05:13,950 --> 00:05:17,460
The idea being, right, we're
taking three individual models

115
00:05:17,460 --> 00:05:20,190
that are all making their own predictions,

116
00:05:20,190 --> 00:05:22,320
and we average them together in a sense,

117
00:05:22,320 --> 00:05:24,120
hopefully getting us

118
00:05:24,120 --> 00:05:25,500
something that's closer to the true value

119
00:05:25,500 --> 00:05:28,623
than any of these individual
models produce on their own.

120
00:05:29,820 --> 00:05:33,060
And so, right, we got our
models, we can make maps,

121
00:05:33,060 --> 00:05:36,450
but then we compare them
back to a set of FIA plots

122
00:05:36,450 --> 00:05:38,910
that were not part of
this modeling framework,

123
00:05:38,910 --> 00:05:40,620
that are independent in that regard,

124
00:05:40,620 --> 00:05:43,803
to get a sense of how
accurate our predictions are.

125
00:05:44,776 --> 00:05:48,060
So if we do that, we get
this kind of scatter plot.

126
00:05:48,060 --> 00:05:51,630
So comparing our predictions at FIA plots

127
00:05:51,630 --> 00:05:53,160
to the measurements at FIA plots.

128
00:05:53,160 --> 00:05:55,830
And, right, the biggest
thing that jumps out

129
00:05:55,830 --> 00:05:58,410
is this wall, this like, saturation point,

130
00:05:58,410 --> 00:06:02,340
that 200 metric tons per
hectare of above ground biomass,

131
00:06:02,340 --> 00:06:04,440
beyond which our model
can't make predictions.

132
00:06:04,440 --> 00:06:07,240
That's that saturation problem
that Collin talked about,

133
00:06:08,610 --> 00:06:10,593
and it's problematic, right?

134
00:06:12,150 --> 00:06:15,090
Imagine a forest is continuing to grow,

135
00:06:15,090 --> 00:06:19,320
but our models can't see
that because of this problem.

136
00:06:19,320 --> 00:06:22,740
And really it's about
the spectral signature,

137
00:06:22,740 --> 00:06:26,130
the color of a forest, is
saturating or maxing out

138
00:06:26,130 --> 00:06:29,490
well before the size of that
forest is gonna max out.

139
00:06:29,490 --> 00:06:31,410
So it's sort of an inherent problem

140
00:06:31,410 --> 00:06:35,790
to modeling forest structure
with spectral information.

141
00:06:35,790 --> 00:06:39,180
And then generally looking
at some of our error metrics,

142
00:06:39,180 --> 00:06:40,890
it's quite a bit of error.

143
00:06:40,890 --> 00:06:43,890
40% relative root mean squared error,

144
00:06:43,890 --> 00:06:46,230
that's relative to the mean FIA value.

145
00:06:46,230 --> 00:06:48,393
30% relative mean absolute error.

146
00:06:50,910 --> 00:06:52,108
That is what it is,

147
00:06:52,108 --> 00:06:54,300
that's what we're dealing with here.

148
00:06:54,300 --> 00:06:57,240
The mean error, which is
just average plus or minus,

149
00:06:57,240 --> 00:06:58,863
is actually pretty good.

150
00:06:59,850 --> 00:07:01,530
The idea being that if you're pooling

151
00:07:01,530 --> 00:07:02,910
a bunch of these predictions up, right,

152
00:07:02,910 --> 00:07:05,070
you have misses below and misses above,

153
00:07:05,070 --> 00:07:06,750
they're gonna average out

154
00:07:06,750 --> 00:07:09,930
and get something pretty close
to the true value, hopefully.

155
00:07:09,930 --> 00:07:13,230
So we can push maps statewide,

156
00:07:13,230 --> 00:07:14,910
like the one I'm showing here,

157
00:07:14,910 --> 00:07:16,860
it's just a single year in time,

158
00:07:16,860 --> 00:07:19,920
and it's pretty hard to
gather any meaningful insights

159
00:07:19,920 --> 00:07:21,423
spatially at this scale,

160
00:07:22,320 --> 00:07:23,729
but I just wanted to show that.

161
00:07:23,729 --> 00:07:26,370
You know, we can do this
statewide, before I zoom down

162
00:07:26,370 --> 00:07:30,270
way into Huntington Wildlife
Forest in Newcomb, New York,

163
00:07:30,270 --> 00:07:31,500
which is in Adirondacks there,

164
00:07:31,500 --> 00:07:33,453
situated within that black box.

165
00:07:34,320 --> 00:07:36,570
So, zoomed in, again,

166
00:07:36,570 --> 00:07:40,350
we have a year's worth of
aboveground biomass predictions

167
00:07:40,350 --> 00:07:41,944
for 2015,

168
00:07:41,944 --> 00:07:44,220
we have a LIDAR derived
canopy height model there

169
00:07:45,360 --> 00:07:47,610
as just added reference information

170
00:07:47,610 --> 00:07:50,070
to try to get a sense for
how well our predictions

171
00:07:50,070 --> 00:07:54,450
are characterizing the structure,

172
00:07:54,450 --> 00:07:56,730
for structure spatially
across this landscape.

173
00:07:56,730 --> 00:07:58,500
So, some features to point out

174
00:07:58,500 --> 00:08:00,453
that are that are perhaps interesting.

175
00:08:01,320 --> 00:08:05,820
You can see that pocket
of really tall vegetation

176
00:08:05,820 --> 00:08:06,653
in the canopy height model,

177
00:08:06,653 --> 00:08:10,380
in yellow there in the
northeast of the map,

178
00:08:10,380 --> 00:08:12,540
sort of missing from from that set

179
00:08:12,540 --> 00:08:16,230
of above ground biomass
predictions there on the left,

180
00:08:16,230 --> 00:08:19,110
right, and that gets at
that saturation issue.

181
00:08:19,110 --> 00:08:21,480
And we have this snippet
of the high peaks,

182
00:08:21,480 --> 00:08:24,720
with high elevation,
low stature vegetation,

183
00:08:24,720 --> 00:08:26,730
that is pretty well
captured by our model there,

184
00:08:26,730 --> 00:08:29,490
as well as some of these
cadastral boundaries

185
00:08:29,490 --> 00:08:30,870
in the southwest there,

186
00:08:30,870 --> 00:08:33,873
especially that boundary running
diagonally along the lake.

187
00:08:35,640 --> 00:08:37,710
To the east of that boundary
is is a natural area

188
00:08:37,710 --> 00:08:38,970
in ESS property,

189
00:08:38,970 --> 00:08:41,250
it's primary forest that's
never been harvested.

190
00:08:41,250 --> 00:08:44,970
And to the west is a more
heavily managed working forest

191
00:08:44,970 --> 00:08:46,827
that we know have been
harvested in the past 20 years.

192
00:08:46,827 --> 00:08:49,680
And you can see some of those
effects of management there,

193
00:08:49,680 --> 00:08:50,850
as well as those property boundaries

194
00:08:50,850 --> 00:08:53,010
in our prediction service.

195
00:08:53,010 --> 00:08:55,980
So I'm gonna zoom into this working forest

196
00:08:55,980 --> 00:08:57,450
and show you a little bit

197
00:08:57,450 --> 00:08:59,343
of our our time series capabilities,

198
00:09:00,660 --> 00:09:02,400
as opposed to just the single snapshot.

199
00:09:02,400 --> 00:09:05,760
And, of course, Collin scooped
this for his presentation,

200
00:09:05,760 --> 00:09:08,010
so it's a bit stale at this point,

201
00:09:08,010 --> 00:09:11,100
but, right, we're mapping year over year,

202
00:09:11,100 --> 00:09:13,203
and then we're totalling up the biomass

203
00:09:13,203 --> 00:09:15,540
in that slice of forest

204
00:09:15,540 --> 00:09:17,820
and showing that as a function of time.

205
00:09:17,820 --> 00:09:21,210
So as that red dot moves,
the map's gonna change.

206
00:09:21,210 --> 00:09:22,560
And really what this is getting at is,

207
00:09:22,560 --> 00:09:24,990
this slice of forest is accumulating

208
00:09:24,990 --> 00:09:27,690
above ground biomass
in the first 20 years,

209
00:09:27,690 --> 00:09:29,280
followed by a large harvest event,

210
00:09:29,280 --> 00:09:30,390
which you can see pretty well,

211
00:09:30,390 --> 00:09:31,560
and then that period of recovery,

212
00:09:31,560 --> 00:09:35,553
which might be too fast to be realistic,

213
00:09:36,960 --> 00:09:38,400
given what we know.

214
00:09:38,400 --> 00:09:43,400
So this is really nice,
this year over year mapping,

215
00:09:43,980 --> 00:09:48,840
but it's actually just the fuel
for the stock change, right?

216
00:09:48,840 --> 00:09:50,940
Now that we have every year mapped,

217
00:09:50,940 --> 00:09:53,640
we can take any slice in this time window

218
00:09:53,640 --> 00:09:55,530
and do some differencing.

219
00:09:55,530 --> 00:09:58,830
So, back out to Huntington
Wildlife Forest,

220
00:09:58,830 --> 00:10:00,750
I've got some decadal differences here.

221
00:10:00,750 --> 00:10:02,943
So on the far left is that first decade,

222
00:10:03,780 --> 00:10:06,480
2000 aboveground biomass minus 1990,

223
00:10:06,480 --> 00:10:10,380
and second decade in the
final 2019 aboveground biomass

224
00:10:10,380 --> 00:10:12,780
minus 2010 aboveground biomass.

225
00:10:12,780 --> 00:10:14,310
Each decade, interestingly,

226
00:10:14,310 --> 00:10:15,600
shows a bit of a different pattern.

227
00:10:15,600 --> 00:10:19,470
So the first, the area is
characterized by stability

228
00:10:19,470 --> 00:10:21,360
and biomass accumulation.

229
00:10:21,360 --> 00:10:22,650
Then you see some of those harvests

230
00:10:22,650 --> 00:10:24,550
starting to occur in the southern area

231
00:10:25,470 --> 00:10:27,870
that we saw in that time series movie.

232
00:10:27,870 --> 00:10:30,030
And then finally, in the third decade,

233
00:10:30,030 --> 00:10:32,940
some of those harvests that
you saw in the second decade

234
00:10:32,940 --> 00:10:37,200
are starting to recover and
accumulate biomass again.

235
00:10:37,200 --> 00:10:39,870
And then just a note, right,

236
00:10:39,870 --> 00:10:43,740
some of this pixel level
speckling that you're seeing here,

237
00:10:43,740 --> 00:10:47,610
those could be real
small scale disturbances

238
00:10:47,610 --> 00:10:49,200
and changes in the forest,

239
00:10:49,200 --> 00:10:51,510
or it could be noise in
our model predictions

240
00:10:51,510 --> 00:10:54,090
based on noise in the satellite data,

241
00:10:54,090 --> 00:10:56,610
so it's important to remember.

242
00:10:56,610 --> 00:11:00,120
So again, back to this
working forest slice here

243
00:11:00,120 --> 00:11:01,500
that I showed you,

244
00:11:01,500 --> 00:11:03,210
instead of scale differences,

245
00:11:03,210 --> 00:11:05,370
we're gonna see some annual differences.

246
00:11:05,370 --> 00:11:09,900
So, same map, and now this
time series plot is showing you

247
00:11:09,900 --> 00:11:11,010
year over year changes.

248
00:11:11,010 --> 00:11:13,650
So current years above ground biomass

249
00:11:13,650 --> 00:11:16,380
minus the previous years
above ground biomass,

250
00:11:16,380 --> 00:11:17,700
anything above that line

251
00:11:17,700 --> 00:11:20,970
is gonna be a net accumulation of biomass,

252
00:11:20,970 --> 00:11:24,780
and anything below that dotted
line is gonna be an emission.

253
00:11:24,780 --> 00:11:26,520
So the pattern's the same,

254
00:11:26,520 --> 00:11:27,960
but what this tells us

255
00:11:27,960 --> 00:11:30,570
is that the rate of biomass accumulation

256
00:11:30,570 --> 00:11:35,190
is slowing in the first 20
years as we approach 2010.

257
00:11:35,190 --> 00:11:37,530
You have this large harvest event,

258
00:11:37,530 --> 00:11:38,760
right, with that emission,

259
00:11:38,760 --> 00:11:39,930
and then you have that period

260
00:11:39,930 --> 00:11:43,230
of really rapid recovery
and biomass accumulation

261
00:11:43,230 --> 00:11:45,153
in the first five years post harvest,

262
00:11:46,680 --> 00:11:48,930
but that rate of recovery slows down

263
00:11:48,930 --> 00:11:51,080
in the second five
years after the harvest.

264
00:11:52,590 --> 00:11:55,980
So I showed off quite a
bit of what we can do here,

265
00:11:55,980 --> 00:11:58,443
but I'm trying to be humble, right?

266
00:11:58,443 --> 00:12:00,300
That this is far from perfect,

267
00:12:00,300 --> 00:12:02,070
there's a lot that's missing from this,

268
00:12:02,070 --> 00:12:04,350
and I wanna highlight
some of those things.

269
00:12:04,350 --> 00:12:06,900
First, that saturation
effect that we talked about,

270
00:12:07,980 --> 00:12:10,583
you kind of can't miss it
when you're doing this thing.

271
00:12:11,580 --> 00:12:13,384
Right, a forest could be,

272
00:12:13,384 --> 00:12:16,620
there's a question in the last
talk that was asked, right,

273
00:12:16,620 --> 00:12:19,350
are the old growth forests
still sequestering carbon,

274
00:12:19,350 --> 00:12:20,280
or are they stagnating?

275
00:12:20,280 --> 00:12:22,950
Like we can't really answer that,

276
00:12:22,950 --> 00:12:26,100
right, once you get up to a
high enough point of biomass,

277
00:12:26,100 --> 00:12:29,013
our model is sort of imposing
an artificial ceiling,

278
00:12:30,030 --> 00:12:33,723
and so that limits what kind
of insights we can gather.

279
00:12:35,070 --> 00:12:37,860
Then this problem of subtle changes,

280
00:12:37,860 --> 00:12:39,273
steady growth and decline.

281
00:12:40,140 --> 00:12:41,550
We think our models are pretty good

282
00:12:41,550 --> 00:12:43,560
at predicting big changes,

283
00:12:43,560 --> 00:12:45,420
like those harvests that we showed,

284
00:12:45,420 --> 00:12:48,300
but when it comes to smaller changes,

285
00:12:48,300 --> 00:12:49,440
were more or less unproven,

286
00:12:49,440 --> 00:12:52,470
especially given that
the size of our errors

287
00:12:52,470 --> 00:12:56,220
are going to outpace the
actual annual changes

288
00:12:56,220 --> 00:12:58,720
that we would expect to
occur in terms of biomass.

289
00:13:00,090 --> 00:13:03,390
Right? That's problematic
given that the rates,

290
00:13:03,390 --> 00:13:04,950
the rates are really what matter,

291
00:13:04,950 --> 00:13:06,120
maybe there's some wiggle room there

292
00:13:06,120 --> 00:13:08,010
in terms of what we can do.

293
00:13:08,010 --> 00:13:08,880
And lastly,

294
00:13:08,880 --> 00:13:10,680
there's some pretty strong relationships

295
00:13:10,680 --> 00:13:13,050
that we can derive from
above ground biomass

296
00:13:13,050 --> 00:13:14,490
to above ground carbon

297
00:13:14,490 --> 00:13:16,410
and below ground live carbon,

298
00:13:16,410 --> 00:13:19,920
but we're really quite
limited when it comes to soil

299
00:13:19,920 --> 00:13:22,020
and litter carbon pools.

300
00:13:22,020 --> 00:13:24,750
Information we're using,
the spectral information,

301
00:13:24,750 --> 00:13:26,760
doesn't really get at that,

302
00:13:26,760 --> 00:13:31,533
and there's just different
drivers behind those two pools.

303
00:13:32,460 --> 00:13:35,250
And then, you know, some
of the upshots here,

304
00:13:35,250 --> 00:13:36,510
some of the upsides of why we think

305
00:13:36,510 --> 00:13:39,963
this is a pretty good first
approach for this kind of thing.

306
00:13:41,460 --> 00:13:42,930
It's really cheap and efficient, right?

307
00:13:42,930 --> 00:13:46,983
We're using federally funded
data that's free to us,

308
00:13:48,120 --> 00:13:51,630
open source tooling, and we're
able to turn around some,

309
00:13:51,630 --> 00:13:54,270
we think are pretty nice
products in a few short years,

310
00:13:54,270 --> 00:13:55,653
with a pretty small team.

311
00:13:57,270 --> 00:13:59,280
We're pretty confident that
our maps and our models

312
00:13:59,280 --> 00:14:02,820
are representing landscape
patterns and processes

313
00:14:02,820 --> 00:14:04,110
at a large scale,

314
00:14:04,110 --> 00:14:05,193
that's really nice.

315
00:14:06,060 --> 00:14:08,940
And then we have this
flexible capacity here,

316
00:14:08,940 --> 00:14:13,940
we can zoom way down to a
landowner or parcel level,

317
00:14:14,340 --> 00:14:18,990
and we can also zoom out to
county scale, township scale,

318
00:14:18,990 --> 00:14:21,690
all using the same set of underlying data,

319
00:14:21,690 --> 00:14:23,400
which is powerful.

320
00:14:23,400 --> 00:14:26,070
And then I've showed you
the retrospective capacity,

321
00:14:26,070 --> 00:14:28,200
looking back towards 1990,

322
00:14:28,200 --> 00:14:30,990
but this same framework
could be turned around

323
00:14:30,990 --> 00:14:35,130
and used in a monitoring sense
as we get more satellite data

324
00:14:35,130 --> 00:14:37,113
and take new FIA emissions.

325
00:14:38,550 --> 00:14:39,720
So then just a couple of thoughts

326
00:14:39,720 --> 00:14:41,913
based on the way the
panel discussion went.

327
00:14:43,140 --> 00:14:45,120
Right, like, this is far from perfect,

328
00:14:45,120 --> 00:14:47,970
I hope I've sort of described that,

329
00:14:47,970 --> 00:14:50,943
but the state is asking
for this, they want this,

330
00:14:51,870 --> 00:14:54,630
and we're doing our best
to give them what they need

331
00:14:54,630 --> 00:14:56,970
while also sticking true to our guns

332
00:14:56,970 --> 00:14:59,700
in terms of the rigor
and the accuracy here.

333
00:14:59,700 --> 00:15:02,793
So we're sort of in
this push-pull position.

334
00:15:04,290 --> 00:15:07,950
Yeah, thank you all for
coming out and listening.

335
00:15:07,950 --> 00:15:08,910
It was fun.

336
00:15:08,910 --> 00:15:12,990
Thanks to the support I've gotten from ESF

337
00:15:12,990 --> 00:15:15,570
the DC FIA program,

338
00:15:15,570 --> 00:15:18,907
and yeah, it's a great
opportunity, thank you.

339
00:15:18,907 --> 00:15:22,740
(audience members applauding)

340
00:15:26,550 --> 00:15:27,383
Yeah?

341
00:15:27,383 --> 00:15:28,216
[Attendant] Thanks, Lucas,

342
00:15:28,216 --> 00:15:29,820
that was a really great talk.

343
00:15:29,820 --> 00:15:32,490
I'm wondering if you
could expand a little bit?

344
00:15:32,490 --> 00:15:33,450
I think what the state

345
00:15:33,450 --> 00:15:35,160
is also probably really interested in

346
00:15:35,160 --> 00:15:38,340
is what should they do with
their forest moving forward?

347
00:15:38,340 --> 00:15:39,173
Yes.

348
00:15:39,173 --> 00:15:40,350
[Attendant] And maybe if
they are looking for them

349
00:15:40,350 --> 00:15:42,660
for offsets, for it in setting,

350
00:15:42,660 --> 00:15:44,250
it's what can we do with our forests

351
00:15:44,250 --> 00:15:45,760
to make them suck up more carbon?

352
00:15:45,760 --> 00:15:46,593
Yep.

353
00:15:46,593 --> 00:15:48,930
[Attendant] How do you
see the work you're doing

354
00:15:48,930 --> 00:15:51,330
in like, hind-casting, translating,

355
00:15:51,330 --> 00:15:55,749
to confident forecasting or future-casting

356
00:15:55,749 --> 00:15:57,180
of different management actions?

357
00:15:57,180 --> 00:16:01,800
So like, you're layering on
now multiple uncertainties.

358
00:16:01,800 --> 00:16:02,790
That's a great question,

359
00:16:02,790 --> 00:16:05,160
and I'm gonna do a bit of hand waving,

360
00:16:05,160 --> 00:16:08,010
as I buried my head in the weeds here

361
00:16:08,010 --> 00:16:09,183
on this side of things.

362
00:16:10,650 --> 00:16:12,990
I think that the golden ticket is

363
00:16:12,990 --> 00:16:15,663
you can look at our data set and say,

364
00:16:16,537 --> 00:16:19,140
"Right, this plot of land
was owned by these people,

365
00:16:19,140 --> 00:16:20,220
and this is what they did."

366
00:16:20,220 --> 00:16:21,660
They know what they did, you know,

367
00:16:21,660 --> 00:16:25,080
and here is the above ground
biomass and carbon implications

368
00:16:25,080 --> 00:16:25,980
of what they did.

369
00:16:25,980 --> 00:16:29,370
So this data alone isn't
gonna answer that question,

370
00:16:29,370 --> 00:16:33,810
we need folks to tell us
what they did on their land

371
00:16:33,810 --> 00:16:36,720
to match that up with these
outcomes before we can say,

372
00:16:36,720 --> 00:16:38,820
you know, "that was good," "that was bad."

373
00:16:43,560 --> 00:16:45,006
[Attendant 2] Just kind of curious,

374
00:16:45,006 --> 00:16:47,640
the tool that might exist

375
00:16:47,640 --> 00:16:50,820
to kind of see through
that saturation point,

376
00:16:50,820 --> 00:16:52,560
you kind of indicated was LIDAR?

377
00:16:52,560 --> 00:16:53,393
Yeah.

378
00:16:53,393 --> 00:16:54,330
[Attendant 2] Is that what you were using

379
00:16:54,330 --> 00:16:56,453
to kind of show the areas activity?

380
00:16:57,752 --> 00:16:59,850
That comparison.

381
00:16:59,850 --> 00:17:00,990
[Attendant 2] And that the issue

382
00:17:00,990 --> 00:17:03,150
with kind of using that as tools,

383
00:17:03,150 --> 00:17:04,860
we don't have full data
sets for the state,

384
00:17:04,860 --> 00:17:07,200
we only have maybe one
or two snapshots in time.

385
00:17:07,200 --> 00:17:08,500
Yep.

386
00:17:08,500 --> 00:17:09,540
[Attendant 2] What's the potential

387
00:17:09,540 --> 00:17:10,890
for like photo AR using,

388
00:17:10,890 --> 00:17:12,630
like aerial imagery data sets

389
00:17:12,630 --> 00:17:14,790
going back decades over time

390
00:17:14,790 --> 00:17:17,253
to show vegetation heights
and correlate that?

391
00:17:18,723 --> 00:17:20,510
So, I think you're talking about like,

392
00:17:20,510 --> 00:17:22,290
is the layering of the imagery, right,

393
00:17:22,290 --> 00:17:23,370
you can get a point cloud?

394
00:17:23,370 --> 00:17:24,203
[Attendant 2] Yeah.

395
00:17:24,203 --> 00:17:26,640
The way I've heard it, as
digital aerial photogrammetry,

396
00:17:26,640 --> 00:17:29,370
DAP point clouds, I think
it's the same thing.

397
00:17:29,370 --> 00:17:33,663
[Attendant 2] Yeah, just using
kind of the stereo imagery

398
00:17:35,730 --> 00:17:37,740
to derive height, vegetation height,

399
00:17:37,740 --> 00:17:40,980
you have a good model of ground elevation,

400
00:17:40,980 --> 00:17:44,160
so now you can just
kind of go back in time

401
00:17:44,160 --> 00:17:45,330
looking at vegetation height.

402
00:17:45,330 --> 00:17:50,330
So I've seen that kind of of
effort done really recently.

403
00:17:52,320 --> 00:17:53,852
You can get at that,

404
00:17:53,852 --> 00:17:54,900
I'm not sure if you're familiar with NIP,

405
00:17:54,900 --> 00:17:56,880
it's the National Imagery Program,

406
00:17:56,880 --> 00:18:00,663
if you can get the data that's
behind that, you can do that.

407
00:18:01,860 --> 00:18:03,690
I don't know how far back.

408
00:18:03,690 --> 00:18:06,810
Right, they're changing,
there's different resolutions.

409
00:18:06,810 --> 00:18:08,790
Not every year, but they upgrade,

410
00:18:08,790 --> 00:18:13,290
I really don't know about
the hind-casting capabilities

411
00:18:13,290 --> 00:18:14,700
and matching those things up.

412
00:18:14,700 --> 00:18:17,523
But you can get a point
cloud from aerial imagery,

413
00:18:18,870 --> 00:18:22,620
not as dense a point cloud
from like a true LIDAR sensor,

414
00:18:22,620 --> 00:18:26,160
but you can get a vegetation
height pretty accurately

415
00:18:26,160 --> 00:18:28,953
from that kind of sensor.

416
00:18:30,150 --> 00:18:33,060
So, I don't know that I
fully answered your question.

417
00:18:33,060 --> 00:18:34,170
[Attendant 2] Yeah, I was
just kind of wondering

418
00:18:34,170 --> 00:18:37,350
if it could be a good way to
see through that saturation

419
00:18:37,350 --> 00:18:38,970
at older time sets.

420
00:18:38,970 --> 00:18:39,803
Yeah.

421
00:18:39,803 --> 00:18:40,980
I mean that kind of information

422
00:18:40,980 --> 00:18:42,900
is the best for this kind of thing,

423
00:18:42,900 --> 00:18:43,733
we just don't,

424
00:18:43,733 --> 00:18:46,590
like, that was the first
thing we did, was use LIDAR,

425
00:18:46,590 --> 00:18:48,540
and we covered 60% of the state.

426
00:18:48,540 --> 00:18:51,810
So, it's a patchwork, right,

427
00:18:51,810 --> 00:18:53,280
this was acquired at this time,

428
00:18:53,280 --> 00:18:55,380
and this was acquired at another time, so.

429
00:18:57,665 --> 00:18:58,530
[Attendant 3] Yes, just a question about,

430
00:18:58,530 --> 00:19:01,860
you did a great job kind of
describing the limitations

431
00:19:01,860 --> 00:19:03,840
and general error,

432
00:19:03,840 --> 00:19:05,670
was wondering what your thoughts are

433
00:19:05,670 --> 00:19:08,760
with the, you know,
applicability of the data,

434
00:19:08,760 --> 00:19:10,170
at what spatial scale.

435
00:19:10,170 --> 00:19:12,793
So as you know, in New York state,

436
00:19:12,793 --> 00:19:14,730
about 50% of the forest

437
00:19:14,730 --> 00:19:16,950
is owned by family forest landowners.

438
00:19:16,950 --> 00:19:19,560
The ownerships are around
the average ownership size,

439
00:19:19,560 --> 00:19:21,180
it's quite small.

440
00:19:21,180 --> 00:19:24,570
You know, do you think that
some of these technologies

441
00:19:24,570 --> 00:19:25,770
are gonna be appropriate

442
00:19:25,770 --> 00:19:28,350
at some of those kind
of stand based scales

443
00:19:28,350 --> 00:19:29,520
or smaller ownerships?

444
00:19:29,520 --> 00:19:31,377
Yeah, that's a great question,

445
00:19:31,377 --> 00:19:33,113
we've been thinking a lot about that.

446
00:19:35,400 --> 00:19:37,980
Right, our error is all at FIA plots,

447
00:19:37,980 --> 00:19:42,450
and that scale is gonna
be the hardest comparison

448
00:19:42,450 --> 00:19:43,283
you can make.

449
00:19:43,283 --> 00:19:45,250
(audience members applauding)

450
00:19:45,250 --> 00:19:46,320
The idea is that as you aggregate up,

451
00:19:46,320 --> 00:19:47,640
right, like I said,

452
00:19:47,640 --> 00:19:50,643
that mean error starts
to get closer to zero.

453
00:19:51,480 --> 00:19:55,140
We're really trying to work through

454
00:19:55,140 --> 00:19:56,490
aggregating uncertainties

455
00:19:56,490 --> 00:19:58,890
up from these individual
pixel level predictions,

456
00:19:58,890 --> 00:20:02,433
and that's an uncracked nut for us.

457
00:20:03,930 --> 00:20:06,130
But in terms, like, that's
why we're making this,

458
00:20:06,130 --> 00:20:08,370
is so that, at a landowner level,

459
00:20:08,370 --> 00:20:10,293
you can get this kind of information.

460
00:20:13,350 --> 00:20:14,183
Yeah.

461
00:20:15,322 --> 00:20:17,370
I don't think I nailed it,

462
00:20:17,370 --> 00:20:22,233
but do you have any
follow ups based on that?

463
00:20:23,310 --> 00:20:24,390
[Attendant 3] Yeah, I guess,

464
00:20:24,390 --> 00:20:25,650
no, not necessarily,

465
00:20:25,650 --> 00:20:27,300
but I guess that's just
the error is coming,

466
00:20:27,300 --> 00:20:28,980
so I'm just trying to think of like,

467
00:20:28,980 --> 00:20:31,980
how the error is generated, or
how you're thinking about it,

468
00:20:31,980 --> 00:20:35,163
whether those individual
FIA plots represent,

469
00:20:36,720 --> 00:20:39,120
probably like an acre or a point space,

470
00:20:39,120 --> 00:20:41,760
but then how you're
relating that to your model,

471
00:20:41,760 --> 00:20:43,275
which is a 30 by 30 meter.

472
00:20:43,275 --> 00:20:44,356
30 meter, yeah, yeah, yeah.

473
00:20:44,356 --> 00:20:45,189
[Attendant 3] Right.

474
00:20:45,189 --> 00:20:48,030
So like, that's what I'm trying to kind of

475
00:20:48,030 --> 00:20:49,770
wrap my head around, yeah,

476
00:20:49,770 --> 00:20:53,072
as a non-like spatial analyst.

477
00:20:53,072 --> 00:20:54,499
Yeah, yeah, yeah.

478
00:20:54,499 --> 00:20:55,500
[Attendant 3] A support practitioner.

479
00:20:55,500 --> 00:20:56,333
No. Yeah.

480
00:20:56,333 --> 00:20:58,860
I think, you know, the more you zoom out,

481
00:20:58,860 --> 00:20:59,730
and the more you aggregate,

482
00:20:59,730 --> 00:21:03,240
the more you're averaging up
those misses above and below,

483
00:21:03,240 --> 00:21:04,073
and that's a good thing.

484
00:21:04,073 --> 00:21:06,660
But we haven't quite cracked

485
00:21:06,660 --> 00:21:09,930
the aggregating uncertainties
yet to those scales.

486
00:21:09,930 --> 00:21:10,943
[Attendant 3] Thanks.

487
00:21:11,790 --> 00:21:13,500
[Attendant 4] I think
that's all the time we have,

488
00:21:13,500 --> 00:21:14,333
at lunch now,

489
00:21:14,333 --> 00:21:16,846
but want catch up with them after?

490
00:21:16,846 --> 00:21:20,679
(audience members applauding)