1 00:00:00,004 --> 00:00:07,143 [INAUDIBLE] Welcome everybody to the system science seminar series. 2 00:00:07,143 --> 00:00:11,700 And today we're thrilled to have one my colleagues, 3 00:00:11,700 --> 00:00:17,991 one of my colleagues who's gonna talk about urban heat islands. 4 00:00:17,991 --> 00:00:20,255 And at least one of our students has worked a little bit on this as well, so 5 00:00:20,255 --> 00:00:21,870 there's quite a bit of interest in it around here. 6 00:00:22,920 --> 00:00:24,047 >> Cool, well thanks. 7 00:00:24,047 --> 00:00:26,095 >> Take it away. 8 00:00:26,095 --> 00:00:30,549 >> Hi, y'all, this is a interesting opportunity cuz last time I 9 00:00:30,549 --> 00:00:35,169 spoke about water when I was here in the systems science seminar, 10 00:00:35,169 --> 00:00:39,310 so it's good to be back talking about a different topic. 11 00:00:39,310 --> 00:00:43,320 A lot of the work that we're doing really revolves around 12 00:00:43,320 --> 00:00:47,410 using various different spatial analytic platforms to understand 13 00:00:48,940 --> 00:00:53,000 patterns on landscapes, and this is no exception. 14 00:00:53,000 --> 00:00:57,140 In that what we're trying to do is develop, use a set of machine learning 15 00:00:57,140 --> 00:01:02,313 algorithms to help identify specific locations where micro 16 00:01:02,313 --> 00:01:09,350 variations in temperatre can be identified and described throughout an urban area. 17 00:01:10,710 --> 00:01:14,691 Just to give you a little bit of background why urban heat, 18 00:01:14,691 --> 00:01:19,485 it's been mid-20th century, there's been some discussion that 19 00:01:19,485 --> 00:01:23,223 [INAUDIBLE] features in cities, buildings and roads, 20 00:01:23,223 --> 00:01:27,877 actually generate and accumulate a lot of heat throughout the day. 21 00:01:27,877 --> 00:01:31,406 And then throughout the night they don't let go of the heat as would 22 00:01:31,406 --> 00:01:32,736 non-developed areas. 23 00:01:32,736 --> 00:01:37,889 And so in the 80s this guy named was originally 24 00:01:37,889 --> 00:01:43,140 able to identify the phenomenon of urban heat islands. 25 00:01:43,140 --> 00:01:48,150 And that started to get a whole field of urban climate really 26 00:01:48,150 --> 00:01:49,456 going back in the 80s. 27 00:01:49,456 --> 00:01:52,700 And as you might remember, that's when Jim Hanson and others were 28 00:01:52,700 --> 00:01:57,380 talking about a planetary phenomenon of climate destabilization as well. 29 00:01:58,480 --> 00:02:04,283 So this is an attempt to try to say, what happens when we create these roads, 30 00:02:04,283 --> 00:02:08,470 buildings, and other features on a urban landscape? 31 00:02:08,470 --> 00:02:12,930 And how can we go about characterizing their impacts on communities that live in 32 00:02:12,930 --> 00:02:16,495 and around areas that might be disproportionately hot? 33 00:02:16,495 --> 00:02:21,395 And as you might remember, or may remember, in 1995 a big heat wave 34 00:02:21,395 --> 00:02:26,378 came through Chicago and made a big national press around the country and 35 00:02:26,378 --> 00:02:30,476 globe, and it was roughly around 1,000 people died. 36 00:02:30,476 --> 00:02:34,803 And what Chicago, if you look at Chicago's climate action plan, for example, 37 00:02:34,803 --> 00:02:38,431 which many cities are developing now to respond to climate change, 38 00:02:38,431 --> 00:02:42,010 including Portland and the county, Multnomah County. 39 00:02:42,010 --> 00:02:46,780 You'll notice that there's many sections that are emerging in the last five years 40 00:02:46,780 --> 00:02:51,835 around preparedness, and some are daring to call it adaptation because those 41 00:02:51,835 --> 00:02:57,350 words politically are a little bit slippery and tricky and charged. 42 00:02:57,350 --> 00:03:01,306 And so what has created is this preparedness piece, and 43 00:03:01,306 --> 00:03:04,456 that brings directly into this idea of urban heat, 44 00:03:04,456 --> 00:03:09,465 what is the implication of urban heat on the communities in and around Portland. 45 00:03:09,465 --> 00:03:11,630 Phoenix has done a lot of this work. 46 00:03:11,630 --> 00:03:16,240 Chicago, needless to say, because of the impacts of the 95 heat wave, they 47 00:03:16,240 --> 00:03:20,170 were immediate to kind of create a climate action plan that had kind of front and 48 00:03:20,170 --> 00:03:23,620 center how to reduce deaths from heat waves. 49 00:03:23,620 --> 00:03:28,050 In 2003 there were about 40,000 people in France that died through a heat wave. 50 00:03:28,050 --> 00:03:31,970 So it turns out that urban heat is, 51 00:03:33,710 --> 00:03:37,170 unlike hurricanes and tornadoes which get front page news, 52 00:03:37,170 --> 00:03:40,350 urban heat is kind of a silent killer, as it's often called. 53 00:03:40,350 --> 00:03:45,860 It kills more people annually than all other natural disasters combined. 54 00:03:45,860 --> 00:03:49,960 And so it turns out it might actually have a particularly a few applications here in 55 00:03:49,960 --> 00:03:53,750 Portland because only about 33% of households in the region have air 56 00:03:53,750 --> 00:03:58,960 conditioning, which is one immediate way to reduce the impact of urban heat. 57 00:03:59,980 --> 00:04:05,100 So what has occurred is we've taken this work and, 58 00:04:05,100 --> 00:04:07,880 through help from the Institute of Sustainable solutions where I'm serving as 59 00:04:07,880 --> 00:04:13,440 a research director now, we've created a collaboration with the city and 60 00:04:13,440 --> 00:04:15,560 the county in terms of their climate action plan. 61 00:04:15,560 --> 00:04:20,730 And what we've done is start to understand, okay, we have these analytical 62 00:04:20,730 --> 00:04:27,810 approaches and we have a way of gathering temperature data across the city. 63 00:04:27,810 --> 00:04:30,958 It's a relatively easy thing to collect. 64 00:04:30,958 --> 00:04:37,570 It's cheap, you get a bunch of students in cars driving around gathering temperature 65 00:04:37,570 --> 00:04:42,170 relatively easily, and we can get hundreds of thousands of points relatively fast. 66 00:04:42,170 --> 00:04:46,720 We can use some analytics to then start creating surfaces and 67 00:04:46,720 --> 00:04:47,830 trying to look at differences. 68 00:04:47,830 --> 00:04:52,149 So that's really what this presentation is about, little bit of background. 69 00:04:53,150 --> 00:04:54,208 With me so far? 70 00:04:54,208 --> 00:04:56,400 Urban heat, important. 71 00:04:56,400 --> 00:05:02,086 >> Does it ever have a positive side, is there an upside at all to this? 72 00:05:02,086 --> 00:05:03,750 >> Right, that's a really interesting question. 73 00:05:05,020 --> 00:05:08,380 We're trying to look at, most of the work we've done has been in the middle of 74 00:05:08,380 --> 00:05:14,380 the summer during heat waves, and that's when people have died as a result. 75 00:05:14,380 --> 00:05:18,580 So if you flip that over and say does it require, in these heat islands, 76 00:05:18,580 --> 00:05:23,492 could they actually not run their heating as much in the winter time, for example. 77 00:05:23,492 --> 00:05:24,709 We haven't looked at that, though. 78 00:05:24,709 --> 00:05:27,004 It's a really interesting question. 79 00:05:27,004 --> 00:05:30,270 Because there may be reductions in energy costs, for example, 80 00:05:30,270 --> 00:05:34,730 for household if they are in an area that's absorbing a little bit more heat. 81 00:05:34,730 --> 00:05:39,343 There's been no, I mean, Lawrence Berkeley Lab in California has probably done a lot 82 00:05:39,343 --> 00:05:41,886 of the pioneering work since the 80s on this. 83 00:05:41,886 --> 00:05:47,163 And today that we've seen any studies that suggested as much. 84 00:05:47,163 --> 00:05:52,010 So we're in the process now, trying to think about 85 00:05:52,010 --> 00:05:57,100 really the doom and gloom side of things [LAUGH]. 86 00:05:57,100 --> 00:05:59,127 So anyway, I'll jump into this. 87 00:05:59,127 --> 00:06:03,327 Feel free to [INAUDIBLE] I'm going to turn it over to Jackson 88 00:06:03,327 --> 00:06:07,452 who's a geospatial researcher working with us on this work, and 89 00:06:07,452 --> 00:06:10,510 this is also collaboration with a project. 90 00:06:10,510 --> 00:06:16,120 We did this in Portland temperate climate and in incredibly hot, 91 00:06:16,120 --> 00:06:21,540 often upwards to 52 degrees Celsius, in Doha, Qatar, which is in the Middle East. 92 00:06:21,540 --> 00:06:26,088 And we [INAUDIBLE] similar analytical platforms in a place where 93 00:06:26,088 --> 00:06:28,746 there's a lot of data in Portland and 94 00:06:28,746 --> 00:06:34,343 in Qatar where there's very little data we've been able to at least employ. 95 00:06:34,343 --> 00:06:37,795 And so what are the differences in terms of the analytics in these two places? 96 00:06:37,795 --> 00:06:40,498 So that's kind of the background. 97 00:06:40,498 --> 00:06:43,060 Jackson's been doing a lot of the heavy lifting. 98 00:06:43,060 --> 00:06:48,133 So what I'll do is kind of hand it over to him to talk about in a data 99 00:06:48,133 --> 00:06:53,270 poor environment is what we're calling at least the Qatar landscape. 100 00:06:53,270 --> 00:06:56,070 And I can move this forwrad. 101 00:06:56,070 --> 00:07:02,056 >> Okay, I actually was thinking we could skip ahead to. 102 00:07:02,056 --> 00:07:04,453 I focused on Portland. 103 00:07:04,453 --> 00:07:09,350 Another researcher in our lab focused on Qatar. 104 00:07:10,460 --> 00:07:15,395 So I'm going to start with Portland and then we'll follow back up with Doha. 105 00:07:20,989 --> 00:07:26,400 Okay, so introduced you to urban heat islands. 106 00:07:26,400 --> 00:07:30,798 This is a graph, if you Google image search urban heat 107 00:07:30,798 --> 00:07:35,505 islands you'll see 100 different versions of this. 108 00:07:35,505 --> 00:07:38,838 Which it's pretty understood, you go into the city, 109 00:07:38,838 --> 00:07:41,674 it's hotter than the country and the forest. 110 00:07:41,674 --> 00:07:45,177 At least we get to experience it's a lot cooler if you go into Forest Park in 111 00:07:45,177 --> 00:07:47,130 the middle of the summer. 112 00:07:47,130 --> 00:07:51,477 But we wanted to go at a much finer scale. 113 00:07:54,249 --> 00:08:00,800 So the data we used was the minus scale data As LIDAR data. 114 00:08:00,800 --> 00:08:04,755 Has anyone here ever worked with LIDAR data? 115 00:08:04,755 --> 00:08:09,150 The basics of LIDAR, it's laser scanning. 116 00:08:09,150 --> 00:08:09,740 Yeah. >> So 117 00:08:09,740 --> 00:08:13,670 just before we go to that I just want to reiterate the objectives of this 118 00:08:13,670 --> 00:08:16,800 are to create a highly granular description of Earth. 119 00:08:19,140 --> 00:08:23,670 Research questions that aren't on the slides essentially say to what extent 120 00:08:23,670 --> 00:08:28,250 can we describe kind of a some 10 meter 121 00:08:28,250 --> 00:08:33,230 scale variation in temperature across the city and in part because 122 00:08:33,230 --> 00:08:37,480 that can lead to mitigation strategies that the city is trying to pursue. 123 00:08:37,480 --> 00:08:44,740 So just in terms of training LIDAR the objectives, I'm glad you're here. 124 00:08:46,460 --> 00:08:48,920 >> So LIDAR scanning, 125 00:08:48,920 --> 00:08:54,625 it's airplane mounted scanning system, shoots in our case, 126 00:08:54,625 --> 00:09:00,835 I think it shot about 400 pulses per second, laser pulses, rotating mirror. 127 00:09:00,835 --> 00:09:05,015 So it gets a decent swath of the land and 128 00:09:05,015 --> 00:09:10,322 as the plane flies, scans over and we get about 12 points per meter square and 129 00:09:10,322 --> 00:09:14,000 these are the most recent LIDAR collection here in Portland. 130 00:09:14,000 --> 00:09:18,790 They got a region so we're talking over on reservoir over on Horse Road, 131 00:09:18,790 --> 00:09:24,100 Cornelius and from Scott Foose all the way down to. 132 00:09:24,100 --> 00:09:25,430 So we got a huge, 133 00:09:25,430 --> 00:09:30,940 huge area of 63.6 billion points were collected in the data set. 134 00:09:33,330 --> 00:09:35,530 It was quite fun processing all that. 135 00:09:35,530 --> 00:09:40,268 We took over a geography lab that had 30 machines that 136 00:09:40,268 --> 00:09:45,225 all had 8 core processors and we used all of them. 137 00:09:45,225 --> 00:09:52,010 Had them all working over a network and of sharing information with each other. 138 00:09:52,010 --> 00:09:56,377 So we can brute force parallel process all of it over six days. 139 00:09:56,377 --> 00:10:00,542 [CROSSTALK] >> [LAUGH] How much did the data cost 140 00:10:00,542 --> 00:10:01,055 you guys? 141 00:10:01,055 --> 00:10:03,905 >> Close to a million bucks but we didn't pay for a dime. 142 00:10:03,905 --> 00:10:04,693 >> Wow! 143 00:10:04,693 --> 00:10:05,892 >> Yeah. >> I know it's expensive to get 144 00:10:05,892 --> 00:10:06,424 that kind of data. 145 00:10:06,424 --> 00:10:07,346 >> Right. 146 00:10:07,346 --> 00:10:08,537 >> But taxpayer money. 147 00:10:08,537 --> 00:10:10,909 >> Right, thanks to all of you, so. 148 00:10:10,909 --> 00:10:13,960 >> [LAUGH] >> Metro is the largest there's 149 00:10:13,960 --> 00:10:20,430 a consortium of GIS data, people in the region and 150 00:10:20,430 --> 00:10:25,030 they all put it in, but certain amounts to look at to be able to apply this. 151 00:10:25,030 --> 00:10:30,330 And in the contract went to [INAUDIBLE] spatial, which then did 152 00:10:30,330 --> 00:10:34,550 some of the cleaning or directed by him the data and then handed it over to us. 153 00:10:34,550 --> 00:10:38,207 We had to [INAUDIBLE] we have lock it over with hard drive. 154 00:10:38,207 --> 00:10:42,805 Definitely this is very old school in the sense of downloading the data hard drive 155 00:10:42,805 --> 00:10:45,916 walking it over to our lab, putting it on the server and 156 00:10:45,916 --> 00:10:47,690 then quad by quad processing. 157 00:10:49,050 --> 00:10:52,842 >> Five terabytes of data are kind of hard to get across the city quickly, 158 00:10:52,842 --> 00:10:56,976 whether it's electronically or physically carrying a [INAUDIBLE] second. 159 00:10:56,976 --> 00:11:02,684 >> [LAUGH] >> So here is one. 160 00:11:02,684 --> 00:11:04,830 This is a USGS info block. 161 00:11:04,830 --> 00:11:06,905 This whole area has a standard measurement. 162 00:11:06,905 --> 00:11:12,570 The area we studied was, I believe, 18 or 19 of these areas. 163 00:11:12,570 --> 00:11:17,395 You look there's a colored area down here, that's the junction of 217 and. 164 00:11:17,395 --> 00:11:18,799 And so we'll zoom into that. 165 00:11:20,120 --> 00:11:25,200 And what you're looking at here is not an aerial image. 166 00:11:25,200 --> 00:11:29,110 This is LIDAR data that is colored with an aerial image. 167 00:11:29,110 --> 00:11:30,930 We're gonna zoom into the upper right corner. 168 00:11:31,930 --> 00:11:33,050 So you see how it's fuzzy. 169 00:11:33,050 --> 00:11:35,270 That's if we shift it down. 170 00:11:35,270 --> 00:11:37,570 We change our perspective of it. 171 00:11:37,570 --> 00:11:38,870 This is LIDAR data. 172 00:11:38,870 --> 00:11:40,822 This is the. 173 00:11:40,822 --> 00:11:44,106 >> Wow! >> When we wanted to go a very high 174 00:11:44,106 --> 00:11:49,080 resolution, urban heat island map surface. 175 00:11:49,080 --> 00:11:54,690 This of course was the natural choice because we can get 3D, 176 00:11:54,690 --> 00:11:56,920 anything, the possibilities are limitless. 177 00:11:56,920 --> 00:12:04,740 We have every feature and each laser pulse gives us four different returns. 178 00:12:04,740 --> 00:12:06,890 So the footprint of the laser is about this big. 179 00:12:08,270 --> 00:12:12,820 And if part of the laser hits the tree and bounce back and the rest of 180 00:12:12,820 --> 00:12:17,790 the laser keeps going through, we know that it can keep the same laser pulse. 181 00:12:17,790 --> 00:12:21,030 So we can actually see how dense trees are. 182 00:12:21,030 --> 00:12:28,097 We can see through the ground for classification purposes. 183 00:12:28,097 --> 00:12:29,561 >> Was that an isometric view or 184 00:12:29,561 --> 00:12:33,658 was it more like Google satellite where you know it's hitting some angle view. 185 00:12:33,658 --> 00:12:35,760 >> It was an oblique view. 186 00:12:35,760 --> 00:12:39,564 >> [INAUDIBLE] that was a tree, this is blurry because it's actually. 187 00:12:39,564 --> 00:12:40,370 >> Right. 188 00:12:40,370 --> 00:12:44,789 So the next image is looking from about 45 degrees I think, this is correct. 189 00:12:48,206 --> 00:12:49,484 >> Which explains to us. 190 00:12:49,484 --> 00:12:53,283 >> So the LIDAR is very interesting, but 191 00:12:53,283 --> 00:12:58,290 what we create from it is even more interesting. 192 00:12:58,290 --> 00:13:00,500 On the left we have a digital elevation model. 193 00:13:00,500 --> 00:13:03,149 Do you many people on here have experienced GIS? 194 00:13:04,150 --> 00:13:05,402 Some people. 195 00:13:05,402 --> 00:13:08,380 So a DDM is a very common flight art product. 196 00:13:08,380 --> 00:13:10,040 It's just triggered out. 197 00:13:10,040 --> 00:13:13,680 So the lasers penetrate through all the trees, where we find buildings, 198 00:13:13,680 --> 00:13:16,290 we remove those and interpolate the holes. 199 00:13:16,290 --> 00:13:21,270 A digital surface model is very important because it's all the features. 200 00:13:21,270 --> 00:13:27,310 We don't look through the trees, we look at the highest return of any laser pulse, 201 00:13:27,310 --> 00:13:28,850 when we subtract them from each other, 202 00:13:28,850 --> 00:13:32,968 we get a digital tape mark also known as a feature height model. 203 00:13:32,968 --> 00:13:38,480 So when the digital surface model minus the DEM equals how tall 204 00:13:38,480 --> 00:13:42,770 things are on the ground, very important for things like building quality. 205 00:13:44,550 --> 00:13:49,278 It's actually pretty much building volume, tree volume on your data set. 206 00:13:49,278 --> 00:13:53,928 Using that DHM, we looked at the ratio of points that 207 00:13:53,928 --> 00:13:58,684 were tree versus points that were tree on the ground or 208 00:13:58,684 --> 00:14:02,840 anything within a one meter. 209 00:14:02,840 --> 00:14:06,220 And multiply them by each other and come up with a biomass metric so 210 00:14:06,220 --> 00:14:10,650 that we can get about a laser determined 211 00:14:10,650 --> 00:14:15,400 density adjusted volume metric of trees. 212 00:14:15,400 --> 00:14:17,790 So we put biomass in quotes. 213 00:14:17,790 --> 00:14:23,918 I can't directly translate it to kilograms to meter cube. 214 00:14:23,918 --> 00:14:26,390 >> [INAUDIBLE] LIDAR data? 215 00:14:26,390 --> 00:14:28,015 >> Resolution, accuracy. 216 00:14:28,015 --> 00:14:32,317 >> Height? >> Yeah height, I believe it's a plus or 217 00:14:32,317 --> 00:14:34,443 minus few. 218 00:14:34,443 --> 00:14:39,580 In certain instances you'll get a higher one but 219 00:14:39,580 --> 00:14:44,476 I think that was the average, it was very low. 220 00:14:46,262 --> 00:14:47,940 Very, very low error. 221 00:14:49,480 --> 00:14:52,510 Here's a cross section of what LIDAR data looks like. 222 00:14:52,510 --> 00:14:53,840 It's kind of hard to visualize it. 223 00:14:53,840 --> 00:14:58,090 This is if we just drew a line on a LIDAR map and with their profile view. 224 00:14:58,090 --> 00:15:00,035 It's actually the tallest tree in Portland. 225 00:15:00,035 --> 00:15:03,050 >> [LAUGH] >> Wanna ask if anyone [INAUDIBLE]. 226 00:15:03,050 --> 00:15:04,488 >> But I was just gonna ask that. 227 00:15:04,488 --> 00:15:07,070 [LAUGH] >> I'll show you on the map later 228 00:15:07,070 --> 00:15:07,650 where it is. 229 00:15:07,650 --> 00:15:09,760 >> Okay. >> I haven't been out to see it, so. 230 00:15:09,760 --> 00:15:10,525 >> Is it in Forest Park? 231 00:15:10,525 --> 00:15:12,643 >> It's in [INAUDIBLE] >> Yeah, so, 232 00:15:12,643 --> 00:15:16,048 doesn't look tall because it's actually starting from, [INAUDIBLE]. 233 00:15:16,048 --> 00:15:18,160 You can see shorter, trees over here. 234 00:15:18,160 --> 00:15:20,120 >> Yeah. >> But here's the gulch. 235 00:15:20,120 --> 00:15:22,930 So we actually took a cross section of the gulch. 236 00:15:22,930 --> 00:15:26,040 So this tree's actually sprouting up the bottom of the gulch. 237 00:15:26,040 --> 00:15:30,030 So if you were to drive past it and look over at the gulch, you probably wouldn't 238 00:15:30,030 --> 00:15:35,740 even notice them that was bigger than the 25th tallest building in Portland, 239 00:15:35,740 --> 00:15:40,300 which is [INAUDIBLE] towers. 240 00:15:40,300 --> 00:15:42,850 You can see it from that here. 241 00:15:42,850 --> 00:15:46,310 >> Yes, Iron Creek, is it on Iron Creek? 242 00:15:46,310 --> 00:15:49,790 >> No, is actually right by the Audubon. 243 00:15:49,790 --> 00:15:50,790 >> Audubon Society? 244 00:15:50,790 --> 00:15:52,930 >> Audubon Society and Balch Creek. 245 00:15:52,930 --> 00:15:54,880 So right up for now.. 246 00:15:54,880 --> 00:15:56,405 No, it's right within McClay. 247 00:15:56,405 --> 00:16:00,230 [CROSSTALK] It's actually a heritage tree, too, 248 00:16:00,230 --> 00:16:03,883 less than they measured it for a heritage tree. 249 00:16:03,883 --> 00:16:09,130 It was, I believe 10 feet shorter than we measured and that was 96 or 97. 250 00:16:09,130 --> 00:16:10,400 >> Wow. >> So it's still growing. 251 00:16:10,400 --> 00:16:14,610 And this summer we've been promised by a nonprofit group in Portland called Setting 252 00:16:14,610 --> 00:16:15,270 the Giants. 253 00:16:16,600 --> 00:16:18,126 >> So are you gonna climb it? 254 00:16:18,126 --> 00:16:20,700 >> So yeah, we want to validate some of these, so 255 00:16:20,700 --> 00:16:22,950 we essentially take groups up trees. 256 00:16:22,950 --> 00:16:27,860 So we're gonna go hopefully this summer and drop tape measures, 257 00:16:27,860 --> 00:16:32,360 and see if we can actually get it truly validated. 258 00:16:32,360 --> 00:16:35,490 Let's see if it works. 259 00:16:35,490 --> 00:16:39,250 All right, so from those LiDAR products, 260 00:16:39,250 --> 00:16:43,040 this is a big process of building upon data. 261 00:16:43,040 --> 00:16:47,510 A side note here, we do a lot of work with the canopy. 262 00:16:47,510 --> 00:16:51,705 So LiDAR, a lot of these products were already created for 263 00:16:51,705 --> 00:16:58,500 other research purposes and some of these we built specifically for this project. 264 00:16:58,500 --> 00:17:03,420 But a lot of them have many uses, as you probably imagine. 265 00:17:03,420 --> 00:17:06,360 So building data is created from LiDAR. 266 00:17:06,360 --> 00:17:08,670 We actually get that from the city. 267 00:17:08,670 --> 00:17:13,090 They pay a bunch of people to sit in a room and clean up lines and 268 00:17:13,090 --> 00:17:16,950 turn it into a vector polygon, GIS. 269 00:17:16,950 --> 00:17:19,580 But that's created from LiDAR data. 270 00:17:19,580 --> 00:17:23,822 So with that, we get things like building roof height, 271 00:17:23,822 --> 00:17:27,809 we get new volume, [INAUDIBLE], different things. 272 00:17:28,880 --> 00:17:31,160 So we've got that for the entire metro area. 273 00:17:32,210 --> 00:17:37,740 And all of these data sets are gonna be one meter resolution. 274 00:17:37,740 --> 00:17:42,749 We're converting everything to a raster format, so 275 00:17:42,749 --> 00:17:48,328 cell based, needless to say high accuracy, high detail. 276 00:17:48,328 --> 00:17:53,875 The canopy was created, we also obtained three inch, 277 00:17:53,875 --> 00:17:59,301 four channel photography, which is to say, 278 00:17:59,301 --> 00:18:03,282 373 inch resolution photography, 279 00:18:03,282 --> 00:18:08,226 that's with the LiDAR data, is fit adjusted for 280 00:18:08,226 --> 00:18:12,990 the building, so it's super accurate. 281 00:18:12,990 --> 00:18:17,870 And it has red, green, blue and infrared channels in it. 282 00:18:17,870 --> 00:18:22,350 So using a very standard remote sensing index, 283 00:18:22,350 --> 00:18:25,860 the NDVI, we use a slightly adjusted one. 284 00:18:25,860 --> 00:18:29,200 NDVI is channel app. 285 00:18:29,200 --> 00:18:32,210 You use the red or green and infrared channels. 286 00:18:32,210 --> 00:18:34,110 It's a simple equation and 287 00:18:34,110 --> 00:18:38,250 it spits out new raster dataset values of negative 1 to 1. 288 00:18:38,250 --> 00:18:42,630 And usually about 0.2 and above means it's alive at a point. 289 00:18:44,320 --> 00:18:47,030 So how do we find trees? 290 00:18:47,030 --> 00:18:50,780 We know that it's alive and a plant already from the NDVI and 291 00:18:50,780 --> 00:18:54,470 we know how tall everything is from the ground. 292 00:18:54,470 --> 00:18:58,579 So metros definition and therefore the city of Portland's definition, 293 00:18:58,579 --> 00:19:02,156 therefore our definition of a tree is something that is a plant, 294 00:19:02,156 --> 00:19:04,290 that is alive and over 10 feet tall. 295 00:19:06,700 --> 00:19:13,360 So here's the biomass that was actually a direct product of the LIDAR data set. 296 00:19:13,360 --> 00:19:20,036 You can see darker colors represent more, call it HSD biomass. 297 00:19:20,036 --> 00:19:25,681 It's an index [CROSSTALK] 298 00:19:25,681 --> 00:19:31,333 I think that's lumber. 299 00:19:31,333 --> 00:19:34,171 >> [CROSSTALK] That's lumber behind the canopy. 300 00:19:37,161 --> 00:19:42,130 Again, one meter resolution, vegetation is just the inverse of that. 301 00:19:42,130 --> 00:19:44,834 So it's alive, a plant and under 10 feet. 302 00:19:47,833 --> 00:19:49,850 And we made solar radiation. 303 00:19:49,850 --> 00:19:52,480 And I won't talk about that too much because we actually ended up 304 00:19:52,480 --> 00:19:55,565 not using it for a few different reasons. 305 00:19:55,565 --> 00:19:58,660 It was initially tested and we decided not to go with it. 306 00:19:58,660 --> 00:20:03,000 But yeah, one meter resolution of the kilowatt hours per meter squared for 307 00:20:03,000 --> 00:20:06,960 every one meter cell average. 308 00:20:10,560 --> 00:20:14,720 So we've got all of our independent variables. 309 00:20:14,720 --> 00:20:18,070 We've got everything that we want to explore and 310 00:20:18,070 --> 00:20:22,500 test to see if that equals temperature, but we don't have temperature. 311 00:20:22,500 --> 00:20:24,960 We need observational data. 312 00:20:24,960 --> 00:20:28,370 And we obtain that vehicle traverses. 313 00:20:28,370 --> 00:20:30,150 So you can see in blue, 314 00:20:30,150 --> 00:20:35,040 those are routes that we drove over three different time periods throughout the day. 315 00:20:35,040 --> 00:20:38,040 We had six cars, each car, 316 00:20:38,040 --> 00:20:42,276 the engineering folks rigged up some thermal coupling devices, 317 00:20:42,276 --> 00:20:47,210 accurate to about a 10th of a degree centigrade with GPS units on them. 318 00:20:48,540 --> 00:20:53,931 Those were processed into a GIS data set, that was basically 120,000 points. 319 00:20:53,931 --> 00:20:56,940 By the way, they were collecting every one second. 320 00:20:56,940 --> 00:21:01,140 Just to give you a visual, basically it was mounted on the passenger side of a car 321 00:21:01,140 --> 00:21:06,480 and it's a PVC tube that goes up, and it's kind of a hollow shaded PVC tube. 322 00:21:09,400 --> 00:21:13,544 It's a pretty effective device if you're driving at about five miles per hour down 323 00:21:13,544 --> 00:21:17,747 the street, because it gets enough airflow through this little five inch PVC pipe, 324 00:21:17,747 --> 00:21:20,131 where a little thermocoupling unit sticks up. 325 00:21:20,131 --> 00:21:23,552 And that air passing over the thermocoupling unit basically registers 326 00:21:23,552 --> 00:21:24,810 the temperature. 327 00:21:24,810 --> 00:21:28,089 It's been validated, at least for getting local temperatures. 328 00:21:29,530 --> 00:21:34,050 That was the input data set for getting at those readings across the city, 329 00:21:34,050 --> 00:21:36,209 which is what Jackson stands for. 330 00:21:36,209 --> 00:21:41,730 >> And to build onto that, that tube, the aspirator, another 331 00:21:41,730 --> 00:21:45,770 purpose of it was to avoid direct sunlight because we don't care about direct heat. 332 00:21:45,770 --> 00:21:49,270 We want to know that air temperature at human height. 333 00:21:49,270 --> 00:21:51,410 >> Yeah, it's about six feet [INAUDIBLE]. 334 00:21:51,410 --> 00:21:58,130 >> [INAUDIBLE] thing to get samples that are right. 335 00:21:58,130 --> 00:22:00,920 >> Yeah, there's some work to be done to get that right. 336 00:22:00,920 --> 00:22:04,300 >> The clean up is involved in cutting out temperatures above 35 miles an hour, 337 00:22:04,300 --> 00:22:05,220 I believe it was. 338 00:22:05,220 --> 00:22:08,330 Some folks in engineering did it, so it must have been 40. 339 00:22:08,330 --> 00:22:14,041 Stop signs, so they know when someone was stopped based on their GPS. 340 00:22:14,041 --> 00:22:17,867 [CROSSTALK] inaccurate because it's not moving air through the pipe, so 341 00:22:17,867 --> 00:22:20,952 we need it to be aspirated to have accurate measurements. 342 00:22:20,952 --> 00:22:23,320 So those are all cleaned out of it. 343 00:22:23,320 --> 00:22:27,565 And then we're left with about 120,000 temperature observations throughout 344 00:22:27,565 --> 00:22:28,094 the city. 345 00:22:30,482 --> 00:22:33,660 So here's a very GISE thing. 346 00:22:33,660 --> 00:22:39,060 So focal buffers, also known as focal statistics, 347 00:22:39,060 --> 00:22:44,570 we have all these raster data sets of lenses. 348 00:22:44,570 --> 00:22:49,290 If we blur them out, that is to say, change that data set from a binary, 349 00:22:49,290 --> 00:22:52,470 zero, there's no canopy here, one, there is canopy here, 350 00:22:52,470 --> 00:22:55,340 zero, no vegetation, one there is vegetation. 351 00:22:56,340 --> 00:23:01,524 And start looking at within a certain search distance, cell distance, 3 cells, 352 00:23:01,524 --> 00:23:06,650 5 cells, 1000 cells, and start looking at things mean. 353 00:23:06,650 --> 00:23:09,490 So we find out the actual percent for every cell. 354 00:23:09,490 --> 00:23:12,240 There's a new raster data set for each cell. 355 00:23:13,240 --> 00:23:19,940 The number behind it tells us how much of what variable is within what distance. 356 00:23:23,140 --> 00:23:25,583 I have some examples that I can add. 357 00:23:25,583 --> 00:23:28,840 So here's just a really raw, this is three cells. 358 00:23:28,840 --> 00:23:31,890 So this initial one, we can pretend it's canopy. 359 00:23:31,890 --> 00:23:35,490 We're doing a three cell, so this new data set that spits out. 360 00:23:35,490 --> 00:23:41,570 It tells us the new value set, how much canopy is within three cells. 361 00:23:41,570 --> 00:23:43,220 Here's what it looks like with five. 362 00:23:43,220 --> 00:23:45,970 And this is what it looks like with our actual data. 363 00:23:45,970 --> 00:23:48,380 So these are actual data sets that we created. 364 00:23:48,380 --> 00:23:53,153 So this is canopy, and it's sort of looks like you're looking through a camera and 365 00:23:53,153 --> 00:23:59,020 you're unfocusing it So those new data sets aren't 366 00:23:59,020 --> 00:24:03,490 just a blurred version of the original, they actually are really powerful. 367 00:24:03,490 --> 00:24:07,705 Because each cell tells us a lot about it's surrounding area. 368 00:24:07,705 --> 00:24:12,590 A great thing is that all those cells line up. 369 00:24:12,590 --> 00:24:16,510 So we put our points over temperature observations on top of a stack, 370 00:24:16,510 --> 00:24:20,610 all those raster datasets, that ended up being about 100 datasets. 371 00:24:20,610 --> 00:24:26,337 And we can append the value of each search distance for each variable and 372 00:24:26,337 --> 00:24:31,770 append that in a big old table with our temperature observations. 373 00:24:31,770 --> 00:24:35,639 So that's the set up, that's how we got to 374 00:24:35,639 --> 00:24:41,000 the point where we [INAUDIBLE] modeling approaches. 375 00:24:41,000 --> 00:24:45,880 The first modeling approach that we used was multiple linear regression. 376 00:24:45,880 --> 00:24:50,730 That's the standard spatial modeling. 377 00:24:50,730 --> 00:24:53,680 Independent literature uses multiple linear regression. 378 00:24:54,830 --> 00:24:59,812 It's great because you can assign it an exact value, the colocation, 379 00:24:59,812 --> 00:25:04,310 the variable, so you know how much this variable affects this. 380 00:25:05,420 --> 00:25:11,130 But we found with, I'm sorry, 70/30 holdout method is how we validated it. 381 00:25:11,130 --> 00:25:14,581 We took 70% of our data, randomly selected and 382 00:25:14,581 --> 00:25:19,170 built our models off of that 70%, that's how well it worked. 383 00:25:19,170 --> 00:25:24,439 We then took our 30%, set aside, and compared the actual observed 384 00:25:24,439 --> 00:25:30,060 temperatures of that 30% to the model temperature at that point. 385 00:25:30,060 --> 00:25:31,750 >> So you're trying to figure out the temperature? 386 00:25:31,750 --> 00:25:32,250 >> Right. 387 00:25:33,324 --> 00:25:36,891 >> Variables- >> We actually, for 388 00:25:36,891 --> 00:25:41,766 the multiple linear regression, we used SPSS and 389 00:25:41,766 --> 00:25:45,723 used stepwise regression by a good fit. 390 00:25:45,723 --> 00:25:50,882 Or keeping our variance inflation factor [INAUDIBLE] 10, 391 00:25:50,882 --> 00:25:54,184 what ends up being really important, 392 00:25:54,184 --> 00:25:59,230 [INAUDIBLE] ended up being the most important for that. 393 00:25:59,230 --> 00:26:05,112 But it was also r-squared of 0.44 really 394 00:26:05,112 --> 00:26:11,740 less than half of the xs are greater than the y. 395 00:26:11,740 --> 00:26:13,220 We weren't happy with that. 396 00:26:13,220 --> 00:26:16,280 >> Did you put in any interaction effects? 397 00:26:16,280 --> 00:26:19,508 >> We did, and we didn't find any, 398 00:26:19,508 --> 00:26:24,690 it didn't help, not where the variance inflation factor is low enough. 399 00:26:24,690 --> 00:26:30,059 So if we had gone up to a much higher variance inflation factor threshold or 400 00:26:30,059 --> 00:26:33,619 we cut it off, we could have made a better model. 401 00:26:33,619 --> 00:26:36,712 But we wanted to make sure that there was as little interaction 402 00:26:36,712 --> 00:26:40,527 between the variables as possible [INAUDIBLE] multi-linearity purposes. 403 00:26:40,527 --> 00:26:41,027 [CROSSTALK] >> [INAUDIBLE] 404 00:26:41,027 --> 00:26:41,794 >> What's that? 405 00:26:41,794 --> 00:26:42,898 >> [INAUDIBLE] 406 00:26:42,898 --> 00:26:43,829 >> Yes, so 407 00:26:43,829 --> 00:26:49,794 when we built the model with the 70%. There's no close of it because 408 00:26:49,794 --> 00:26:53,527 I'm gonna move on to some other models. This was kind of that initial, 409 00:26:53,527 --> 00:26:57,081 we don't like our results, so let's try something else. 410 00:26:58,590 --> 00:27:01,532 So the next method we used, or 411 00:27:01,532 --> 00:27:07,950 we used a combination of CART trees and linear regression. 412 00:27:07,950 --> 00:27:12,930 So we use classification trees to define areas of the city that were similar, 413 00:27:12,930 --> 00:27:18,180 based on all the land use characteristics within the search distances. 414 00:27:18,180 --> 00:27:23,354 So we ran a huge dataset through our CART library and 415 00:27:24,437 --> 00:27:28,300 and came up with six different nodes. 416 00:27:29,560 --> 00:27:33,440 Those nodes are geographic areas ending. 417 00:27:33,440 --> 00:27:36,216 So here's how we split the nodes. 418 00:27:36,216 --> 00:27:39,142 Sorry, this is. 419 00:27:39,142 --> 00:27:40,396 >> What's ASR? 420 00:27:40,396 --> 00:27:43,350 >> Area solar radiation. 421 00:27:43,350 --> 00:27:48,920 So this is taken from one we built the and without area solar radiation. 422 00:27:50,950 --> 00:27:56,433 So this is from one of the times where we used area solar radiation. 423 00:27:56,433 --> 00:28:00,144 I should say, each model method that we used, 424 00:28:00,144 --> 00:28:05,475 area solar radiation had minimal effect on the overall accuracy, 425 00:28:05,475 --> 00:28:09,948 but also introduced a lot of questions of the validity of 426 00:28:09,948 --> 00:28:13,400 the actual solar radiation model itself. 427 00:28:14,650 --> 00:28:20,530 So we decided to leave it out, because it's really part of the model itself. 428 00:28:22,490 --> 00:28:29,461 Logically, it makes sense but here's what that looks like. 429 00:28:29,461 --> 00:28:35,476 When we look at areas and we meet that classification tree, we're broken 430 00:28:35,476 --> 00:28:39,958 up into many different areas so the same- >> The purple looks 431 00:28:39,958 --> 00:28:43,030 like all the industrial corridors. 432 00:28:43,030 --> 00:28:45,201 And what's the grey look like? 433 00:28:45,201 --> 00:28:47,521 >> That's the tree, the- >> The trees. 434 00:28:47,521 --> 00:28:48,354 >> The trees. 435 00:28:48,354 --> 00:28:50,302 >> How do, [CROSSTALK]. 436 00:28:50,302 --> 00:28:52,389 You've got Tabor. 437 00:28:52,389 --> 00:28:53,031 >> Tabor. 438 00:28:53,031 --> 00:28:58,191 >> [CROSSTALK] and airport, industrial, and Swan Island, [INAUDIBLE] 439 00:28:58,191 --> 00:29:03,903 training compound is actually right here in this very special Marine are. 440 00:29:03,903 --> 00:29:08,474 >> The 205 corridor on the other side and 441 00:29:08,474 --> 00:29:13,050 Rocky Butte is that grey section there. 442 00:29:13,050 --> 00:29:15,778 >> Yes, guess what blue means? 443 00:29:15,778 --> 00:29:18,726 >> [INAUDIBLE] >> Blue is. 444 00:29:18,726 --> 00:29:22,211 Within these areas, so this is just the first step, 445 00:29:22,211 --> 00:29:27,450 we're doing a combination of regression three and linear regression. 446 00:29:27,450 --> 00:29:31,386 So within each of these areas, we then did the last step, 447 00:29:31,386 --> 00:29:34,560 again, the multiple linear regression. 448 00:29:34,560 --> 00:29:40,320 We did it within each one of these defined geographic areas, yeah? 449 00:29:40,320 --> 00:29:43,100 >> What's the temperature difference? 450 00:29:43,100 --> 00:29:47,860 >> The temperature there was about 15 degrees fahrenheit. 451 00:29:47,860 --> 00:29:49,390 So it's very noticeable. 452 00:29:50,770 --> 00:29:56,520 Certain areas get much hotter than the average recorded temperature. 453 00:29:56,520 --> 00:29:58,844 I have some maps at the end that I will show you. 454 00:29:58,844 --> 00:30:02,490 >> [CROSSTALK] we also want, there's a big discussion in the public 455 00:30:02,490 --> 00:30:05,541 health literature around fatalities to urban heat. 456 00:30:05,541 --> 00:30:09,533 And there is a discussion about what constitutes a heat event or 457 00:30:09,533 --> 00:30:11,724 a heat wave as it's often called. 458 00:30:11,724 --> 00:30:16,846 So we looked at this day, August 25, last year, to see where in 459 00:30:16,846 --> 00:30:22,550 the history of temperature that we have for the region, where it falls. 460 00:30:22,550 --> 00:30:26,504 And it's about the 75th percentile, I think, 461 00:30:26,504 --> 00:30:31,776 is what we discovered of the hottest, average day, on that day and 462 00:30:31,776 --> 00:30:37,420 in a much of overall August for that particular day. 463 00:30:37,420 --> 00:30:42,771 So it means, we can quibble about how we define a heat wave, 464 00:30:42,771 --> 00:30:46,303 and so that 15 degrees might be plus or 465 00:30:46,303 --> 00:30:51,040 minus within how hot the hottest period that day is. 466 00:30:51,040 --> 00:30:54,260 >> [COUGH] And this is based on the atmospheric temperature, right? 467 00:30:54,260 --> 00:30:58,380 Not the- >> Not the ground surface temperature, no, 468 00:30:58,380 --> 00:31:00,750 which is what most previous models are built on, 469 00:31:00,750 --> 00:31:05,810 using I forget which landsat, the satellite imagery channel they used. 470 00:31:05,810 --> 00:31:11,040 But there's one that's shortwave, there's one that's visible infrared. 471 00:31:11,040 --> 00:31:14,780 And a lot of people build urban heat models 472 00:31:14,780 --> 00:31:16,860 based off of just the reflection of that. 473 00:31:16,860 --> 00:31:20,764 We were talking about one the other day, about New York City. 474 00:31:20,764 --> 00:31:25,604 And it's very hard to agree with any model they would make based on 475 00:31:25,604 --> 00:31:31,260 satellite imagery, because they're measuring the tops of buildings. 476 00:31:31,260 --> 00:31:37,580 And we wanted to really get as a human, what people are experiencing temperature. 477 00:31:39,110 --> 00:31:44,103 Everything was measured about six feet above the ground in brief periods in 478 00:31:44,103 --> 00:31:47,600 a day, right up to maybe 6 AM to 1 PM, etc, right. 479 00:31:48,620 --> 00:31:50,800 So within each of these defined regions, 480 00:31:50,800 --> 00:31:55,328 we then took the points that fell within there from a traverse of the city. 481 00:31:55,328 --> 00:32:00,883 And we've built more linear regression models within those. 482 00:32:00,883 --> 00:32:04,955 The results were better, 483 00:32:04,955 --> 00:32:09,800 notably better, but still not very high. 484 00:32:10,990 --> 00:32:15,583 We had a pretty good model, 485 00:32:15,583 --> 00:32:20,589 it actually looks like this. 486 00:32:20,589 --> 00:32:24,003 Everyone who looks at it can relate to it. 487 00:32:24,003 --> 00:32:24,669 That's been important. 488 00:32:24,669 --> 00:32:26,900 You see the Brooklyn train yards down here. 489 00:32:26,900 --> 00:32:29,795 You see Foster, you see 82nd. 490 00:32:29,795 --> 00:32:34,020 And then keep in mind there's no road data set in there. 491 00:32:34,020 --> 00:32:38,104 This is building off of lack of canopy building metrics, 492 00:32:38,104 --> 00:32:41,517 vegetation metrics, Northwest industrial. 493 00:32:43,750 --> 00:32:49,816 The airport is [INAUDIBLE] the airport. 494 00:32:49,816 --> 00:32:52,340 >> What a jack of heat there down. 495 00:32:52,340 --> 00:32:57,790 >> It's flat and [INAUDIBLE] very large area. 496 00:32:57,790 --> 00:33:00,900 >> What's on the back side of Mount Scott there? 497 00:33:00,900 --> 00:33:02,340 >> Pall View. 498 00:33:02,340 --> 00:33:03,160 >> That's Pall View? 499 00:33:03,160 --> 00:33:07,180 >> Yeah, so that's the top of this one giant meadow. 500 00:33:07,180 --> 00:33:09,370 Just a big flat barren area. 501 00:33:11,680 --> 00:33:12,997 >> What is it? 502 00:33:12,997 --> 00:33:14,920 Is it meadow or is it oregon? 503 00:33:14,920 --> 00:33:15,716 >> It's meadow. 504 00:33:15,716 --> 00:33:18,024 >> There's a parking lot next to it. 505 00:33:18,024 --> 00:33:19,083 >> I think the parking lot is over here. 506 00:33:19,083 --> 00:33:19,746 >> Okay. 507 00:33:19,746 --> 00:33:24,914 >> You also see a lot of people commented 508 00:33:24,914 --> 00:33:29,948 on Tabor, this big hot spot. 509 00:33:29,948 --> 00:33:34,520 And down here, so the reservoir leading up to 510 00:33:34,520 --> 00:33:38,280 those that if you've ever been to that area where there's no trees and you can 511 00:33:38,280 --> 00:33:43,738 just look straight out into Portland, that's that lines up with this hotspot. 512 00:33:43,738 --> 00:33:48,390 So, there are a lot of things that like the City Portland like this 513 00:33:49,590 --> 00:33:52,510 but we wanted to increase the accuracy. 514 00:33:52,510 --> 00:33:56,460 We didn't settle for something that looked about right, felt about right. 515 00:33:56,460 --> 00:34:05,670 So we wanted to go to all random forest houses. 516 00:34:07,390 --> 00:34:11,565 So with the random forest again, all of this, we use our resolver form. 517 00:34:11,565 --> 00:34:17,080 It's got great GI library so you can actually do some 518 00:34:17,080 --> 00:34:22,570 pretty wonderful geographic spatial analysis with it. 519 00:34:23,840 --> 00:34:27,540 So the service models that I'll show you that we built with the random forest, 520 00:34:29,080 --> 00:34:30,380 not splitting it up and 521 00:34:30,380 --> 00:34:35,145 using something like LFI or CFI, just using one core one process. 522 00:34:35,145 --> 00:34:40,920 It took about 44 hours to apply the model that we created, 523 00:34:40,920 --> 00:34:45,047 to create the model about five minutes with the data set. 524 00:34:45,047 --> 00:34:50,370 The data set was 1.4 billion cells. 525 00:34:50,370 --> 00:34:52,696 >> Now, that of course has some tunable parameters and stuff, 526 00:34:52,696 --> 00:34:56,074 though she had to deal with [CROSSTALK] >> We went through and 527 00:34:56,074 --> 00:34:59,863 we found that the best number to use, 528 00:34:59,863 --> 00:35:05,890 the number of horse views, we want with a thousand trees. 529 00:35:05,890 --> 00:35:10,684 And the M try are the number of variables randomly selected to 530 00:35:10,684 --> 00:35:12,848 build each of those trees. 531 00:35:12,848 --> 00:35:18,786 We tried, we have made iterative script that went through and 532 00:35:18,786 --> 00:35:25,764 find every every employee number we've got, without reasonable set. 533 00:35:25,764 --> 00:35:30,700 But we ended up actually just using the default which is a number of 534 00:35:30,700 --> 00:35:32,690 variables divided by 3. 535 00:35:32,690 --> 00:35:34,520 That worked the best. 536 00:35:34,520 --> 00:35:40,592 That's I guess why is the depot. 537 00:35:40,592 --> 00:35:46,592 [LAUGH] [CROSSTALK] Using when you use or categorizing data, 538 00:35:46,592 --> 00:35:52,224 I know it's recommended to use the number of variables 539 00:35:52,224 --> 00:35:57,131 divided by square root of all those variables. 540 00:35:57,131 --> 00:36:04,438 Sometimes, we get to try that but divided by three, four, pretty well. 541 00:36:04,438 --> 00:36:10,730 So each random tree was built off of 32 or 33 different variables. 542 00:36:12,030 --> 00:36:14,785 Does anyone in here have experience with random forest? 543 00:36:16,100 --> 00:36:16,600 No? 544 00:36:16,600 --> 00:36:21,570 Okay, I really would recommend picking up 545 00:36:21,570 --> 00:36:28,010 a copy of our for free, and do does anyone in here to start? 546 00:36:28,010 --> 00:36:28,970 Okay, good. 547 00:36:28,970 --> 00:36:33,170 >> I think this was brought up as one of the topics for further, 548 00:36:33,170 --> 00:36:36,866 if you're really interested in this kind of stuff in 549 00:36:36,866 --> 00:36:40,906 the environmental data analysis classes on pan offers. 550 00:36:40,906 --> 00:36:45,738 It's mentioned at the end of the class it's one of the cool things you can do 551 00:36:45,738 --> 00:36:48,205 about the stuff as you can teach them. 552 00:36:48,205 --> 00:36:52,568 >> We are was taught r by f. 553 00:36:52,568 --> 00:36:55,180 >> Okay, that's a great class. 554 00:36:55,180 --> 00:36:57,900 >> Yeah, I want to take it. 555 00:36:59,140 --> 00:37:07,530 The library random forest in r is quite simple to use, very tunable. 556 00:37:09,270 --> 00:37:15,670 It's one of the easier machine learning libraries within r. 557 00:37:15,670 --> 00:37:21,360 Things like neural networks and support vector machines, 558 00:37:21,360 --> 00:37:24,960 require a lot more cleansing of the data and 559 00:37:24,960 --> 00:37:29,430 make sure everything's normalized choosing the correct parameters. 560 00:37:29,430 --> 00:37:35,570 Random forest is great because you can put in numbers, letters, symbols, 561 00:37:35,570 --> 00:37:40,820 whatever you want, and it doesn't care if it's qualitative or quantitative data. 562 00:37:41,980 --> 00:37:46,008 And it'll whittle its way down to an answer. 563 00:37:46,008 --> 00:37:50,318 Maybe not the answer you want, but it will do what it's supposed to do. 564 00:37:50,318 --> 00:37:51,295 >> We're waiting to see how it works. 565 00:37:51,295 --> 00:37:51,891 >> Okay. 566 00:37:51,891 --> 00:37:55,870 [LAUGH] I'm morning traverse. 567 00:37:55,870 --> 00:37:59,522 This is the 7030 test over here. 568 00:37:59,522 --> 00:38:05,163 We got an r squared of 0.979 which was quite the increase. 569 00:38:05,163 --> 00:38:09,097 >> [LAUGH] >> I immediately thought that Jack, 570 00:38:09,097 --> 00:38:12,182 that there's something wrong >> [LAUGH] 571 00:38:12,182 --> 00:38:14,794 [CROSSTALK] >> You couldn't find 572 00:38:14,794 --> 00:38:15,950 something wrong with it. 573 00:38:15,950 --> 00:38:18,680 >> Please, please let me know. 574 00:38:18,680 --> 00:38:19,750 >> Yeah. 575 00:38:19,750 --> 00:38:21,680 >> What is it that you're plotting? 576 00:38:21,680 --> 00:38:25,490 >> Right now I'm plotting over here, so on the y yet 577 00:38:25,490 --> 00:38:29,570 is the actual temperature versus the predicted. 578 00:38:29,570 --> 00:38:33,646 So the temperature observed from that 30, 579 00:38:33,646 --> 00:38:38,379 held out against what the 70% model prediction. 580 00:38:38,379 --> 00:38:42,237 >> [INAUDIBLE] >> What models are interpretable in any 581 00:38:42,237 --> 00:38:42,778 sense? 582 00:38:42,778 --> 00:38:45,440 >> It's very hard to explain it to people. 583 00:38:45,440 --> 00:38:48,248 >> [LAUGH] >> It makes sense to me. 584 00:38:48,248 --> 00:38:52,352 [CROSSTALK] >> Dealing with when you're with a bunch 585 00:38:52,352 --> 00:38:56,776 of people like that city quite a bit about it, and most people are used to, okay, 586 00:38:56,776 --> 00:38:59,870 well, how much are trees affecting? 587 00:38:59,870 --> 00:39:03,850 Well, which one of the 1.4 billion cells we want to exam? 588 00:39:03,850 --> 00:39:08,130 It's conditionals but it's a case by case. 589 00:39:09,630 --> 00:39:11,350 Sometimes trees might increase. 590 00:39:11,350 --> 00:39:15,712 I don't know, I've looked at every single cell. 591 00:39:15,712 --> 00:39:17,191 At least you get an average. 592 00:39:17,191 --> 00:39:23,808 [CROSSTALK] Before I came over here, we're averaging out to different geometries, 593 00:39:23,808 --> 00:39:29,790 geographies, to determine which one you want to use for further analysis. 594 00:39:29,790 --> 00:39:34,559 Actually, come up with a number, linear regression is great because I 595 00:39:34,559 --> 00:39:38,055 can say the amount of trees times this coefficient, 596 00:39:38,055 --> 00:39:41,580 there's your production in urban EI temperature. 597 00:39:41,580 --> 00:39:45,510 So for explaining the explanatory power, 598 00:39:45,510 --> 00:39:49,830 explanatory to other humans, linear regression is great, 599 00:39:51,300 --> 00:39:56,450 but for building something that had higher accuracy, but by being He nearly 600 00:39:56,450 --> 00:40:02,640 impossible to verbally explain in less than 40 months. 601 00:40:06,520 --> 00:40:11,149 >> So if it's one of those six zones and 602 00:40:11,149 --> 00:40:15,161 then if it's got buildings and 603 00:40:15,161 --> 00:40:19,680 trees and then- >> This is actually a cool random forest. 604 00:40:19,680 --> 00:40:24,480 We didn't, for this method, did not split it into those nodes. 605 00:40:24,480 --> 00:40:25,640 All of the data at once. 606 00:40:27,320 --> 00:40:28,430 [INAUDIBLE] turn on all of that. 607 00:40:28,430 --> 00:40:35,370 So if there were nodes within there it should figure that out on its own. 608 00:40:35,370 --> 00:40:39,740 So it's the same process that splits it just on a [INAUDIBLE]. 609 00:40:39,740 --> 00:40:46,500 >> So your model is though again, saying if the nearby buildings are blank and 610 00:40:46,500 --> 00:40:52,295 if the nearby trees are blank and if the nearby traffic is blank, 611 00:40:52,295 --> 00:40:56,071 now all of those things like a neural net. 612 00:40:56,071 --> 00:41:03,010 >> The random forest works by the random smaller trees that it built. 613 00:41:03,010 --> 00:41:07,660 It compares all of them and they kind of have a nice democratic vote and 614 00:41:07,660 --> 00:41:13,815 decide which variable is going to be at what point in the classification tree. 615 00:41:15,050 --> 00:41:18,840 Our most important variables were canopy cover, this is for the morning, 616 00:41:18,840 --> 00:41:22,950 canopy cover within 50m, building volume within 900m, 617 00:41:22,950 --> 00:41:28,060 we get some more canopy down at the bottom. 618 00:41:29,850 --> 00:41:35,048 We are happy to see this because we really wanted to know the different scales 619 00:41:35,048 --> 00:41:39,540 at which heat is affected by these landscape variables. 620 00:41:39,540 --> 00:41:41,368 So, some trees right next to you might be okay, but 621 00:41:41,368 --> 00:41:42,991 what about trees within your neighborhood? 622 00:41:42,991 --> 00:41:49,190 What about trees within your census block group different distances? 623 00:41:50,220 --> 00:41:55,860 But afternoon was actually our worst, the R squared was 0.81. 624 00:41:58,770 --> 00:42:03,810 We think that has to do, there's a lot more variation in the afternoon. 625 00:42:03,810 --> 00:42:08,990 We think that it's probably a cause of that direct solar radiation. 626 00:42:08,990 --> 00:42:12,540 But when we put the solar radiation model into this, 627 00:42:12,540 --> 00:42:14,070 we really didn't see that coming back. 628 00:42:14,070 --> 00:42:18,601 Which is one of the reasons we started to question the solar radiation [INAUDIBLE]. 629 00:42:18,601 --> 00:42:23,950 In the afternoon, you find a lot of building metrics. 630 00:42:23,950 --> 00:42:28,433 Now, I'm just showing the top five, but they were all included. 631 00:42:28,433 --> 00:42:32,179 >> See [INAUDIBLE] meters and [INAUDIBLE] meters. 632 00:42:32,179 --> 00:42:37,194 Well directly around you it might hold more heat directly around you, 633 00:42:37,194 --> 00:42:42,636 but then at a larger distance those might contribute to a different amount. 634 00:42:42,636 --> 00:42:47,101 So if you have a skyscraper next to you and nothing else around you, 635 00:42:47,101 --> 00:42:52,445 maybe you'll get very little heat off of that because you're right next to it, 636 00:42:52,445 --> 00:42:56,230 but it's trying to get a feel for everything around it. 637 00:42:57,550 --> 00:42:59,860 With canopy, I was thinking about canopy. 638 00:42:59,860 --> 00:43:04,720 You can have 50 meters around you that's all canopy, but 639 00:43:04,720 --> 00:43:08,839 that's in the middle of the Sahara Desert, it's probably gonna be quite hot. 640 00:43:10,540 --> 00:43:12,530 There's a lot of correlation between those. 641 00:43:12,530 --> 00:43:17,021 But the fact that both are there since you got a little bit more information from 642 00:43:17,021 --> 00:43:19,554 both of them than either one by themselves. 643 00:43:19,554 --> 00:43:22,180 So in [INAUDIBLE] no idea. 644 00:43:22,180 --> 00:43:24,457 Is it just the other things that came out. 645 00:43:24,457 --> 00:43:25,133 >> Right. 646 00:43:25,133 --> 00:43:27,751 >> This is what the alert is making. 647 00:43:27,751 --> 00:43:29,707 Right. We're trying to explain different 648 00:43:29,707 --> 00:43:31,680 variables here in different trees. 649 00:43:31,680 --> 00:43:32,827 >> Right. 650 00:43:32,827 --> 00:43:38,078 The evening reverse we saw very similar to the morning 651 00:43:38,078 --> 00:43:43,678 reverse as far as predictive performing about 97% and 652 00:43:43,678 --> 00:43:50,930 the standard deviation and building height is quite interesting for us. 653 00:43:52,290 --> 00:43:56,230 When we talk to the folks in engineering that were working on this project with us, 654 00:43:56,230 --> 00:43:58,720 they said, yeah, of course. 655 00:43:58,720 --> 00:44:01,250 We know that spec in effective 656 00:44:03,490 --> 00:44:07,930 air mixing from different variations and building heights. 657 00:44:07,930 --> 00:44:12,946 So that was pretty interesting to see because when I first saw that, I was 658 00:44:12,946 --> 00:44:18,390 was just added to the spatial guide and most of the time didn't really know 659 00:44:18,390 --> 00:44:22,530 what to make of it and engineers right away we're just like, yeah of course. 660 00:44:22,530 --> 00:44:26,407 >> So this is the issue of the group and 661 00:44:26,407 --> 00:44:33,493 if any of you are savvy with CF complex fluid dynamics models and 662 00:44:33,493 --> 00:44:37,385 wanna try your CFD skills on this. 663 00:44:37,385 --> 00:44:42,402 I would love to talk to you because we could do something a neighborhood 664 00:44:42,402 --> 00:44:47,334 scale on this as a potential piece, that it's really a emerging area 665 00:44:47,334 --> 00:44:52,540 that has a lot of potential for trying to understand air masses and city. 666 00:44:52,540 --> 00:44:55,181 That's what this standard deviation suggests. 667 00:44:55,181 --> 00:44:57,232 >> And we've got plenty of data. 668 00:44:57,232 --> 00:45:00,180 >> Somebody said [INAUDIBLE] like that in San Francisco. 669 00:45:00,180 --> 00:45:01,630 >> Lots of people have done it, yes. 670 00:45:01,630 --> 00:45:04,330 >> Like with that market. 671 00:45:04,330 --> 00:45:04,830 >> Right. 672 00:45:07,370 --> 00:45:11,846 Follow [INAUDIBLE] we were in [INAUDIBLE] some of this woman did 673 00:45:11,846 --> 00:45:15,338 her PhD dissertation in University of Tokyo, 674 00:45:15,338 --> 00:45:20,366 using a Construct computer to basically run a CMP for all of Tokyo. 675 00:45:20,366 --> 00:45:25,007 Comparing 1888 where they had army massive building heights and 676 00:45:25,007 --> 00:45:28,858 she was able to digitize those good shape 90,000. 677 00:45:28,858 --> 00:45:31,040 >> [LAUGH] >> To 2000. 678 00:45:31,040 --> 00:45:33,328 Well, I think she could. 679 00:45:33,328 --> 00:45:39,094 Change of air masses in Tokyo, which is no small city [INAUDIBLE]. 680 00:45:39,094 --> 00:45:41,194 It was really interesting. 681 00:45:41,194 --> 00:45:45,200 >> I was gonna suggest [INAUDIBLE] questions all along but it might be good 682 00:45:45,200 --> 00:45:49,290 to kind of say is there some wrap up that you definitely want to cover? 683 00:45:49,290 --> 00:45:50,510 Or anything I just- >> Yeah. 684 00:45:51,860 --> 00:45:58,730 Let me let me skip ahead and just show you this in and open it up. 685 00:45:58,730 --> 00:45:59,790 It's kind of hard to see on there. 686 00:45:59,790 --> 00:46:02,810 I can give people a link to this too so they can explore it. 687 00:46:02,810 --> 00:46:05,491 This is something that's publicly put on the website here. 688 00:46:05,491 --> 00:46:06,140 >> Yeah. 689 00:46:06,140 --> 00:46:07,919 >> Share with the people who come to look up. 690 00:46:07,919 --> 00:46:09,460 >> Yeah. >> If there are lakes that you want, 691 00:46:09,460 --> 00:46:13,441 we can add them on to where [CROSSTALK] >> Right, exactly, yeah, so 692 00:46:13,441 --> 00:46:16,500 this is the morning traverse. 693 00:46:16,500 --> 00:46:20,040 It should be noted that temperatures in this 694 00:46:20,040 --> 00:46:24,530 are quite a bit lower than any of the other temperatures I'm about to show you. 695 00:46:24,530 --> 00:46:32,050 The hottest downtown is still cooler than the coldest forests are the acronym. 696 00:46:32,050 --> 00:46:35,800 But what's important to note is its downtown [INAUDIBLE]. 697 00:46:35,800 --> 00:46:40,505 When we go look over in the afternoon we start to see all that 698 00:46:40,505 --> 00:46:42,435 variation popping up. 699 00:46:42,435 --> 00:46:49,189 We start to see the transit corridors coming through. 700 00:46:54,610 --> 00:46:56,290 Mount Taylor. 701 00:46:58,770 --> 00:46:59,840 Now rocky view. 702 00:46:59,840 --> 00:47:05,534 >> Yeah, rocky view >> When we start to look at the evenings, 703 00:47:05,534 --> 00:47:10,712 even more defined areas, we start to see really obviously 704 00:47:10,712 --> 00:47:14,936 transit corridors, downtown though- >> Cools off. 705 00:47:14,936 --> 00:47:21,585 >> Throughout the day, waterfront is very hot, industrial very hot, 706 00:47:21,585 --> 00:47:27,095 but downtown this tends to stay cool throughout the day. 707 00:47:27,095 --> 00:47:32,254 And we're trying to delve into why that is. 708 00:47:32,254 --> 00:47:36,463 We think it's a mix of that mixing [INAUDIBLE] building heights and 709 00:47:36,463 --> 00:47:37,312 also shade. 710 00:47:37,312 --> 00:47:38,384 >> Building shade. 711 00:47:38,384 --> 00:47:39,463 >> Right. 712 00:47:39,463 --> 00:47:44,772 So we think that could cause it to retain more heat, 713 00:47:44,772 --> 00:47:49,598 but it never actually gets that hot downtown. 714 00:47:49,598 --> 00:47:54,359 Relative to the previous literature saying that the more into the city you 715 00:47:54,359 --> 00:47:56,893 get the more downtown more buildings, 716 00:47:56,893 --> 00:48:01,004 the hotter it is As we know, it's not necessarily the case. 717 00:48:01,004 --> 00:48:06,372 >> For the diurnal cycle, did you account for what sunrise or sunset was, 718 00:48:06,372 --> 00:48:11,126 to see even though you weren't tracking installation per se. 719 00:48:11,126 --> 00:48:14,672 Knowing when the sunrises or sets could mitigate, 720 00:48:14,672 --> 00:48:19,706 because the sun is set before 7 PM on the day that you collected the data. 721 00:48:19,706 --> 00:48:24,929 Then it could be cooler downtown because less exposed asphalt, I guess? 722 00:48:24,929 --> 00:48:29,912 >> Of course, the angle, there's a huge variable angle there as well, 723 00:48:29,912 --> 00:48:32,201 is the timing it's coming out. 724 00:48:32,201 --> 00:48:36,098 >> [CROSSTALK] >> This is all summer, August 24, 2014. 725 00:48:36,098 --> 00:48:38,958 That actually is a very good question. 726 00:48:38,958 --> 00:48:43,749 One thing that I've wondered about winter, this is for the winter, to see if we see 727 00:48:43,749 --> 00:48:48,150 the same effects, is that angle of the sun and how that would affect things. 728 00:48:48,150 --> 00:48:52,886 So I think that, just speculating, I think downtown might be cooler when, 729 00:48:52,886 --> 00:48:57,400 yeah, because it's hard to even get down the street on the right because 730 00:48:57,400 --> 00:48:59,782 it's tucked into those hills there. 731 00:48:59,782 --> 00:49:01,998 So for most of the day it will be in the shade. 732 00:49:04,038 --> 00:49:05,696 >> Gil, did you want to say anything? 733 00:49:05,696 --> 00:49:10,692 >> I mean, was it just a piece that was a lot of the existing this isn't in 734 00:49:10,692 --> 00:49:12,426 your area of interest. 735 00:49:12,426 --> 00:49:16,536 A lot of existing research on this is satellite imagery which is usually 736 00:49:16,536 --> 00:49:19,374 one snapshot in a day, it's kind of cross sectional. 737 00:49:19,374 --> 00:49:22,856 And what we're doing with this is trying to get throughout the day, 738 00:49:22,856 --> 00:49:24,210 what is the variability. 739 00:49:24,210 --> 00:49:26,893 Because it turns out when you have an older, 740 00:49:26,893 --> 00:49:31,534 when you have a 75-year old grandmother living in a house by herself, and 741 00:49:31,534 --> 00:49:34,091 there's a heat wave that comes through. 742 00:49:34,091 --> 00:49:37,883 It's actually during the time between about 10 PM and 743 00:49:37,883 --> 00:49:40,877 6 AM when the fatalities actually happen. 744 00:49:40,877 --> 00:49:45,833 So that's, the time difference really turns out to really 745 00:49:45,833 --> 00:49:51,194 matter because of the public health impacts that it has on here. 746 00:49:51,194 --> 00:49:53,295 >> I'm just gonna put a bill here so you can see. 747 00:49:53,295 --> 00:49:54,879 [CROSSTALK] >> And go through this, but 748 00:49:54,879 --> 00:49:55,765 we use the same technique. 749 00:49:55,765 --> 00:49:56,794 >> These are the results. 750 00:49:56,794 --> 00:49:59,913 We went through the same workflow for Doha, but it had different input data 751 00:49:59,913 --> 00:50:02,454 sets, which you can explore on your own when you put the sun. 752 00:50:02,454 --> 00:50:08,667 But it had the same factor countdowns, actually cool. 753 00:50:08,667 --> 00:50:11,156 There's other factors, of course we'll get into that. 754 00:50:11,156 --> 00:50:12,227 >> Is that a waterfront? 755 00:50:12,227 --> 00:50:13,365 >> It's all on a waterfront. 756 00:50:13,365 --> 00:50:17,311 [CROSSTALK] >> That's good to know. 757 00:50:17,311 --> 00:50:21,477 >> You are talking about over the water or something [INAUDIBLE]? 758 00:50:21,477 --> 00:50:29,727 Is that what you're talking about old ladies on the floor. 759 00:50:29,727 --> 00:50:30,820 >> Right. >> [CROSSTALK] 760 00:50:30,820 --> 00:50:33,558 >> So the implication of this work is that 761 00:50:33,558 --> 00:50:39,119 people who have died most frequently through a heat wave usually have certain 762 00:50:39,119 --> 00:50:44,102 pre-existing health condition like respiratory illness likely. 763 00:50:44,102 --> 00:50:47,015 And they tend to be a little bit more vulnerable to heat. 764 00:50:47,015 --> 00:50:48,586 >> [INAUDIBLE] >> Right, so 765 00:50:48,586 --> 00:50:51,071 if they don't have air conditioning in their place. 766 00:50:51,071 --> 00:50:52,143 And in Chicago, 767 00:50:52,143 --> 00:50:57,352 they were finding temperatures of the outside ambient temperature was 110, 768 00:50:57,352 --> 00:51:02,653 and the inside, the house and apartment, was about 140 degrees Fahrenheit. 769 00:51:02,653 --> 00:51:07,212 >> [CROSSTALK] >> These were higher storey apartments. 770 00:51:07,212 --> 00:51:13,786 Nevertheless, that's kind of what was the cause of death in Chicago. 771 00:51:13,786 --> 00:51:18,700 >> I'm just curious if because where we saw the high intensity heat, 772 00:51:18,700 --> 00:51:21,869 were those industrial areas in Portland? 773 00:51:21,869 --> 00:51:28,319 And any of the places where those deaths are in the center or 774 00:51:28,319 --> 00:51:32,807 whatever correlate with the hotspots. 775 00:51:32,807 --> 00:51:39,166 So like cuz, there's not a whole lot of housing around in the. 776 00:51:39,166 --> 00:51:44,718 I mean, this would tell me that you don't wanna live by the airport, 777 00:51:44,718 --> 00:51:51,344 and actually living down here is better than living next to I-205 or I-84. 778 00:51:51,344 --> 00:51:56,776 But is that where- >> This is only one 779 00:51:56,776 --> 00:52:04,262 of the reasons we lost the case. 780 00:52:04,262 --> 00:52:09,573 >> [CROSSTALK] >> Climatecode.org. 781 00:52:09,573 --> 00:52:14,227 >> So right, so what we are doing is trying to get the agencies and 782 00:52:14,227 --> 00:52:16,638 agencies to proactively use this data. 783 00:52:16,638 --> 00:52:20,855 So what we did was put this thing on a site called 784 00:52:20,855 --> 00:52:25,627 Climatecode.org which is [INAUDIBLE] procedural. 785 00:52:25,627 --> 00:52:26,512 >> It should. 786 00:52:26,512 --> 00:52:31,354 The idea is to able to look at those associative demographic factors combined 787 00:52:31,354 --> 00:52:35,821 with the heat, and seeing if there is an overlap of where the most number 788 00:52:35,821 --> 00:52:40,676 of people are, where the most, in this case older adults, younger adults. 789 00:52:40,676 --> 00:52:45,056 Those who have some pre-existing health condition data that we 790 00:52:45,056 --> 00:52:47,499 can get from any health authority. 791 00:52:47,499 --> 00:52:52,123 And seeing if those line up in specific areas so that during a heatwave, 792 00:52:52,123 --> 00:52:55,602 those are the kind of surveillance spots we. 793 00:52:55,602 --> 00:52:56,290 >> Right. 794 00:52:56,290 --> 00:52:59,726 >> Yeah, so yeah, right, it doesn't play out in airports, 795 00:52:59,726 --> 00:53:02,360 it doesn't play out in industries. 796 00:53:02,360 --> 00:53:03,707 >> Great. >> Those are places 797 00:53:03,707 --> 00:53:05,920 you emphasized aviation or. 798 00:53:05,920 --> 00:53:06,842 >> Yeah. 799 00:53:06,842 --> 00:53:07,761 >> That's it. 800 00:53:09,969 --> 00:53:15,106 >> I'II guess, the data in the, it does model for 801 00:53:15,106 --> 00:53:19,356 Portland generalizing well for July? 802 00:53:19,356 --> 00:53:20,936 Is that the same model, 803 00:53:20,936 --> 00:53:24,721 the one that you got the- >> They are, they're different variables 804 00:53:24,721 --> 00:53:28,847 though because we have this light on our data set for Portland which we don't have 805 00:53:28,847 --> 00:53:32,320 for Doha, and so using fundamentally different set of variables. 806 00:53:32,320 --> 00:53:33,816 >> We can't use the same model. 807 00:53:33,816 --> 00:53:36,420 >> We use the same technique but a different, 808 00:53:36,420 --> 00:53:39,397 it turns into a different parameterized model. 809 00:53:39,397 --> 00:53:41,591 >> And we see a lot of same trends. 810 00:53:41,591 --> 00:53:44,363 >> Yeah, I mean can also last pair. 811 00:53:44,363 --> 00:53:49,752 >> Before and it's pretty awesome to do this, but like, people die in a heat wave, 812 00:53:49,752 --> 00:53:55,480 do you really think if the temperature was a couple degrees lower they wouldn't die? 813 00:53:55,480 --> 00:53:56,976 >> That doesn't seem to be the problem, 814 00:53:56,976 --> 00:53:58,951 the problem is they don't have anyone to talk to. 815 00:53:58,951 --> 00:54:00,245 >> Right, right, right. 816 00:54:00,245 --> 00:54:02,900 So that's exactly the kind of direction where we're going with this, 817 00:54:02,900 --> 00:54:05,748 there are some mitigation measures you can do in the physical environment. 818 00:54:05,748 --> 00:54:09,304 Then it turns out that social programs are checking 819 00:54:09,304 --> 00:54:13,032 in on vulnerable populations in your neighborhood. 820 00:54:13,032 --> 00:54:17,357 That's the direction the county is considering in terms of the heatwave. 821 00:54:17,357 --> 00:54:22,340 So we're trying to point out areas where it's higher likelihood that 822 00:54:22,340 --> 00:54:25,559 it might happen, at least physically. 823 00:54:25,559 --> 00:54:27,990 But yeah, social programs are a big part of it. 824 00:54:27,990 --> 00:54:34,553 >> The temperature inside your house is not gonna be directly correlated. 825 00:54:34,553 --> 00:54:36,625 >> That's something we're looking at now. 826 00:54:36,625 --> 00:54:42,537 >> Cuz we do have data of AC that's not window mounted. 827 00:54:42,537 --> 00:54:45,323 [CROSSTALK] Right? 828 00:54:45,323 --> 00:54:45,969 >> Right. 829 00:54:45,969 --> 00:54:48,545 Maybe that's what this gets more at, the resiliency- 830 00:54:48,545 --> 00:54:49,874 >> We should thank these guys for 831 00:54:49,874 --> 00:54:51,742 this really interesting presentation. 832 00:54:51,742 --> 00:54:53,976 >> Thank you. >> [APPLAUSE] 833 00:54:53,976 --> 00:54:56,391 >> Give you a chance to get out of 834 00:54:56,391 --> 00:55:02,646 the chairs and take off, go to the meeting, whatever you have to do, so. 835 00:55:02,646 --> 00:55:08,051 >> There's a lot of privacy aspects, that you might be able to ask because 836 00:55:08,051 --> 00:55:13,144 of very obvious ways heat waves power bill graph indicating an AC. 837 00:55:13,144 --> 00:55:16,089 Records are, we've been talking with energy transfer from Oregon about some 838 00:55:16,089 --> 00:55:17,258 data, so that's [INAUDIBLE]. 839 00:55:17,258 --> 00:55:22,279 [CROSSTALK] >> [INAUDIBLE] this key 840 00:55:22,279 --> 00:55:27,160 problem in urban areas in terms of 841 00:55:27,160 --> 00:55:31,119 making policy [CROSSTALK] >> Yeah, 842 00:55:31,119 --> 00:55:32,581 maybe there needs to be more trees. 843 00:55:32,581 --> 00:55:37,747 >> I guess something 844 00:55:37,747 --> 00:55:45,328 has to be done [INAUDIBLE]. 845 00:55:55,075 --> 00:56:05,075 [CROSSTALK] 846 00:58:03,425 --> 00:58:13,425 [INAUDIBLE]