1 00:00:15,396 --> 00:00:15,826 Speaker 1: Pushkin. 2 00:00:31,606 --> 00:00:33,946 Speaker 2: I'm Jacob Goldstein, and this is What's Your Problem? My 3 00:00:33,986 --> 00:00:38,346 Speaker 2: guest today is Catherine Nakalembe. Catherine is an assistant professor 4 00:00:38,426 --> 00:00:41,396 Speaker 2: at the University of Maryland. And she runs the Africa 5 00:00:41,446 --> 00:00:45,006 Speaker 2: program at NASA Harvest, which is a project that uses 6 00:00:45,106 --> 00:00:49,766 Speaker 2: satellite images to track agriculture around the world. And it 7 00:00:49,786 --> 00:00:53,076 Speaker 2: wouldn't be wrong to say that Catherine's problem is using 8 00:00:53,196 --> 00:00:57,716 Speaker 2: satellite imagery and GoPro cameras and AI to predict seasonal 9 00:00:57,756 --> 00:01:03,046 Speaker 2: food shortages in sub-Saharan Africa. But in fact, her work, 10 00:01:03,346 --> 00:01:06,786 Speaker 2: the problem she's trying to solve, is problematic. ultimately more 11 00:01:06,886 --> 00:01:10,646 Speaker 2: human than that, ultimately more interesting, I would say, because 12 00:01:10,706 --> 00:01:13,446 Speaker 2: the problem she's really trying to solve is, how do 13 00:01:13,466 --> 00:01:17,516 Speaker 2: you use evidence to convince leaders to act to prevent 14 00:01:17,596 --> 00:01:21,996 Speaker 2: people from starving? Catherine grew up in Uganda. She went 15 00:01:22,016 --> 00:01:24,976 Speaker 2: to the University of Maryland to start a PhD, and 16 00:01:25,016 --> 00:01:27,936 Speaker 2: then she went back to Uganda to do her graduate research. 17 00:01:28,276 --> 00:01:32,546 Speaker 2: And the project was combining satellite data and on-the-ground research 18 00:01:32,566 --> 00:01:35,116 Speaker 2: to to predict when the crops were likely to fail, 19 00:01:35,456 --> 00:01:38,796 Speaker 2: leading to widespread hunger. And she was working on this 20 00:01:38,816 --> 00:01:41,356 Speaker 2: for a few years. And then in the summer of 2015, 21 00:01:41,356 --> 00:01:45,056 Speaker 2: when she was still a graduate student, there was a 22 00:01:45,116 --> 00:01:48,716 Speaker 2: particularly bad drought in the region she studied. And she knew, 23 00:01:48,916 --> 00:01:51,196 Speaker 2: she knew from her work that the people there would 24 00:01:51,256 --> 00:01:53,286 Speaker 2: not have enough to eat in a few months. And 25 00:01:53,316 --> 00:01:56,066 Speaker 2: she decided she had to get that information to someone 26 00:01:56,086 --> 00:01:57,756 Speaker 2: with the power to do something about it. 27 00:01:58,226 --> 00:02:03,306 Speaker 1: I realized really quickly that I knew more about the severity, 28 00:02:03,786 --> 00:02:07,846 Speaker 1: basically the distribution of drought across this region more than 29 00:02:07,926 --> 00:02:10,306 Speaker 1: anybody because I was able to look at it from space. 30 00:02:11,006 --> 00:02:13,326 Speaker 1: And one year I finished my work and I was like, 31 00:02:13,366 --> 00:02:15,546 Speaker 1: I can't just go back with this information because it 32 00:02:15,566 --> 00:02:17,246 Speaker 1: had been an extreme drought year. I went to the 33 00:02:17,286 --> 00:02:20,146 Speaker 1: office of the prime minister to share my My data 34 00:02:20,186 --> 00:02:23,046 Speaker 1: and evidence in that year had taken lots of photographs 35 00:02:23,326 --> 00:02:25,926 Speaker 1: and videos, et cetera. So I brought that. And then 36 00:02:25,946 --> 00:02:28,546 Speaker 1: they had a meeting with the prime minister that led to— Wait. 37 00:02:28,586 --> 00:02:32,106 Speaker 2: Can I just pause? So you're, what, a grad student 38 00:02:32,186 --> 00:02:33,606 Speaker 2: or a young professor. 39 00:02:33,146 --> 00:02:34,266 Speaker 1: At this point? No, I'm a grad student. 40 00:02:35,056 --> 00:02:38,416 Speaker 2: And so it's an unusual thing for a grad student 41 00:02:38,456 --> 00:02:40,536 Speaker 2: to be like, oh, my God, I have to tell 42 00:02:40,596 --> 00:02:44,896 Speaker 2: the prime minister about this. What happened? How did that happen? 43 00:02:45,416 --> 00:02:49,126 Speaker 1: So I've been coming for the last three years. People 44 00:02:49,146 --> 00:02:51,206 Speaker 1: are trying to do the same things and crops fail. 45 00:02:52,186 --> 00:02:54,826 Speaker 1: And I'd already started to kind of figure out how to, 46 00:02:55,106 --> 00:02:57,226 Speaker 1: you know, predict what's going to happen, you know, early 47 00:02:57,266 --> 00:03:00,176 Speaker 1: in the season. And there's always news around this issue, 48 00:03:00,316 --> 00:03:03,456 Speaker 1: you know, Karamoja drought again, etc. 49 00:03:03,676 --> 00:03:07,516 Speaker 2: So Karamoja is the region in, I guess, northern Uganda 50 00:03:07,576 --> 00:03:08,896 Speaker 2: where you were studying. 51 00:03:09,176 --> 00:03:13,496 Speaker 1: Yeah. And so it's always contentious. Is it happening? Are they, 52 00:03:13,556 --> 00:03:16,996 Speaker 1: you know, just pretending? And this year, I had so 53 00:03:17,026 --> 00:03:20,756 Speaker 1: much evidence, you know. First, because I'd taken so many photographs, 54 00:03:20,866 --> 00:03:24,386 Speaker 1: I'd recorded video. And I wanted this to be sort 55 00:03:24,406 --> 00:03:27,726 Speaker 1: of complementary to my PhD research. So I went and 56 00:03:27,746 --> 00:03:31,586 Speaker 1: I shared it to the commissioner. And then he asked me, 57 00:03:32,326 --> 00:03:35,566 Speaker 1: can you stay an extra few days? Because we have 58 00:03:35,606 --> 00:03:39,116 Speaker 1: this food security meeting on Thursday with the prime minister 59 00:03:39,276 --> 00:03:43,006 Speaker 1: and his steering committee, which are the other ministers. And 60 00:03:43,066 --> 00:03:45,226 Speaker 1: I was like, of course. So I show up and 61 00:03:45,306 --> 00:03:47,846 Speaker 1: I share the evidence and they asked me to say 62 00:03:47,906 --> 00:03:50,036 Speaker 1: a few words. And I was like, you see this photograph? 63 00:03:50,756 --> 00:03:52,796 Speaker 1: You see all of this thing that's dead here? This 64 00:03:52,876 --> 00:03:57,716 Speaker 1: map basically represents the same thing everywhere. And it's eminent 65 00:03:57,916 --> 00:04:00,716 Speaker 1: and people will be hungry pretty soon. And three days 66 00:04:00,756 --> 00:04:03,966 Speaker 1: later or two days later, they sent food trucks. I 67 00:04:03,986 --> 00:04:06,446 Speaker 1: have this photo that was sent to me from the 68 00:04:06,486 --> 00:04:10,306 Speaker 1: guy who was the secretary to the prime minister or 69 00:04:10,386 --> 00:04:15,706 Speaker 1: personal assistant. He was like, the trucks left this morning. 70 00:04:15,726 --> 00:04:19,496 Speaker 1: I have this email, and I was amazed that I 71 00:04:20,116 --> 00:04:23,356 Speaker 1: did that, and it went that far, basically. 72 00:04:23,496 --> 00:04:26,216 Speaker 2: And that fast. So when you say people will be 73 00:04:26,256 --> 00:04:30,796 Speaker 2: hungry pretty soon, that is an understated way, it seems, 74 00:04:30,876 --> 00:04:33,336 Speaker 2: of saying what's going on. What was going on? 75 00:04:34,056 --> 00:04:38,156 Speaker 1: You have subsistence farming where you produce what you eat. 76 00:04:38,216 --> 00:04:41,836 Speaker 1: Very little is left over. And at the end of 77 00:04:41,896 --> 00:04:44,836 Speaker 1: the season, you're supposed to sort of start picking up 78 00:04:45,196 --> 00:04:47,996 Speaker 1: or eating what will be available to you through to 79 00:04:48,056 --> 00:04:50,956 Speaker 1: the end of the next season. So this is typically 80 00:04:50,976 --> 00:04:56,916 Speaker 1: around September. And so by August, everything had failed. And 81 00:04:56,976 --> 00:05:01,146 Speaker 1: so there wasn't going to be anything harvested. But if 82 00:05:01,246 --> 00:05:05,856 Speaker 1: you've kind of been out in the field a collecting data, 83 00:05:06,016 --> 00:05:09,356 Speaker 1: and you run into kids eating sorghum shoots, for example, 84 00:05:09,396 --> 00:05:10,186 Speaker 1: you know, the middle part of a. 85 00:05:10,196 --> 00:05:14,556 Speaker 2: It's kind of like eating grass. 86 00:05:15,106 --> 00:05:17,586 Speaker 1: Exactly. So like the middle part of a corn stem, 87 00:05:18,386 --> 00:05:22,426 Speaker 1: it's sugary. And so, you know, you find kids eating 88 00:05:22,466 --> 00:05:26,526 Speaker 1: that and they don't have anything else. They don't have livestock. 89 00:05:26,586 --> 00:05:28,726 Speaker 1: Their family does not have any other source of income. 90 00:05:29,696 --> 00:05:32,546 Speaker 1: It's pretty dire. So I think the other thing that 91 00:05:32,586 --> 00:05:35,286 Speaker 1: might have made a huge difference in this particular instance 92 00:05:35,306 --> 00:05:39,686 Speaker 1: is the photo evidence that I had. I take thousands 93 00:05:39,726 --> 00:05:41,666 Speaker 1: of photos. I think I have over 300,000 photos. 94 00:05:41,866 --> 00:05:47,576 Speaker 2: That's such a low-tech thing. Do you know what I mean? 95 00:05:47,616 --> 00:05:52,436 Speaker 2: It's not like you were doing some wild satellite AI something. 96 00:05:52,476 --> 00:05:54,396 Speaker 2: You were just walking around and taking pictures. 97 00:05:54,416 --> 00:05:58,846 Speaker 1: Mm-hmm. And I had my map, my drought analysis map, 98 00:05:59,046 --> 00:06:02,346 Speaker 1: that was very simple. And then we called it NDVI, 99 00:06:02,446 --> 00:06:05,476 Speaker 1: Normalized Difference Vegetation Index, the anomaly of it. 100 00:06:06,296 --> 00:06:09,016 Speaker 2: Basically, how much drier than usual. Exactly. 101 00:06:09,276 --> 00:06:09,656 Speaker 1: Yeah. 102 00:06:10,836 --> 00:06:13,276 Speaker 2: So, okay, so you do this, and the result in 103 00:06:13,316 --> 00:06:15,636 Speaker 2: that first year is that they get food aid to 104 00:06:15,676 --> 00:06:16,976 Speaker 2: people sooner. 105 00:06:16,996 --> 00:06:23,116 Speaker 1: Right. Almost immediately. I have this text message from the commissioner. 106 00:06:23,156 --> 00:06:26,336 Speaker 1: He said, this is the first time that nobody argued 107 00:06:26,376 --> 00:06:27,156 Speaker 1: with the evidence. 108 00:06:28,346 --> 00:06:32,746 Speaker 2: Did they then integrate your work in an ongoing way? 109 00:06:32,766 --> 00:06:35,686 Speaker 2: How did that play out in the years after that? 110 00:06:36,166 --> 00:06:40,586 Speaker 1: So the project that's called Disaster Risk Financing was basically 111 00:06:40,626 --> 00:06:43,566 Speaker 1: being designed the following week. So I stayed that extra 112 00:06:43,606 --> 00:06:46,836 Speaker 1: week and was like, I can not only develop the methods, 113 00:06:46,876 --> 00:06:49,476 Speaker 1: but I can train the office how to utilize the 114 00:06:49,496 --> 00:06:51,536 Speaker 1: data that's available in order for them to do that 115 00:06:51,616 --> 00:06:55,336 Speaker 1: early warning, as well as write summaries, reports, et cetera. 116 00:06:56,576 --> 00:06:59,816 Speaker 2: Take pictures. It sounds like one big lesson is take pictures. 117 00:06:59,956 --> 00:07:03,676 Speaker 1: It's a validation for sure, but they'd already understood. So 118 00:07:03,756 --> 00:07:06,756 Speaker 1: I kind of created this method to understand, compared to 119 00:07:06,816 --> 00:07:10,116 Speaker 1: the history, if we hit this threshold, things are going 120 00:07:10,136 --> 00:07:13,506 Speaker 1: to fail. Considering what has happened in the last 30 121 00:07:13,506 --> 00:07:16,676 Speaker 1: years or last 20 years, If we hit this threshold, 122 00:07:16,756 --> 00:07:19,096 Speaker 1: vegetation won't recover and we need to start planning. 123 00:07:19,256 --> 00:07:22,966 Speaker 2: Basically a threshold of drought. If rain is below some level. 124 00:07:23,326 --> 00:07:27,286 Speaker 1: So initially it was like if vegetation conditions drop below 125 00:07:27,806 --> 00:07:32,386 Speaker 1: this level within this period, a few months, months within 126 00:07:32,426 --> 00:07:34,526 Speaker 1: the growing season, then we need to do something. 127 00:07:35,106 --> 00:07:38,946 Speaker 2: So it's interesting. I mean, I might naively have thought 128 00:07:38,996 --> 00:07:43,316 Speaker 2: that the constraint, the fundamental constraint was just money. 129 00:07:43,726 --> 00:07:43,886 Speaker 1: Right. 130 00:07:43,906 --> 00:07:45,626 Speaker 2: That there's just not the money to get the food 131 00:07:45,646 --> 00:07:49,496 Speaker 2: to the people or infrastructure or something. But it's information 132 00:07:49,656 --> 00:07:51,716 Speaker 2: at some level, like reliable evidence. 133 00:07:52,676 --> 00:07:58,136 Speaker 1: Reliable evidence that is accessible. But I think money and 134 00:07:58,176 --> 00:08:03,646 Speaker 1: the willingness to use that money is also really important. 135 00:08:03,746 --> 00:08:07,406 Speaker 1: So that region where I was studying, even though it 136 00:08:07,466 --> 00:08:11,606 Speaker 1: had the lowest economic indicators, you know, household income was 137 00:08:11,646 --> 00:08:16,786 Speaker 1: very low. Infrastructure is very poor. The cost of food 138 00:08:16,846 --> 00:08:20,326 Speaker 1: was so much higher than it is in the capital. Yes. 139 00:08:20,746 --> 00:08:23,486 Speaker 2: Well, that goes with poor infrastructure, right? Exactly. When it 140 00:08:23,566 --> 00:08:27,596 Speaker 2: costs a lot to move food, the food costs a lot. 141 00:08:27,916 --> 00:08:30,996 Speaker 1: Exactly. Now, add the fact that there is a drought 142 00:08:31,236 --> 00:08:34,176 Speaker 1: and traders can do whatever they want. And so it 143 00:08:34,216 --> 00:08:37,196 Speaker 1: becomes much more expensive. And when the government is trying 144 00:08:37,256 --> 00:08:40,626 Speaker 1: to acquire this water and food, et cetera, they end 145 00:08:40,666 --> 00:08:43,966 Speaker 1: up paying a lot more. Uh-huh. And so if you're 146 00:08:44,006 --> 00:08:47,146 Speaker 1: doing it in anticipation, it means that then you can 147 00:08:47,206 --> 00:08:49,366 Speaker 1: acquire things at a much lower cost. 148 00:08:49,466 --> 00:08:51,946 Speaker 2: Oh, so if they know in August that there is 149 00:08:51,986 --> 00:08:54,976 Speaker 2: going to be great need in September, they can buy 150 00:08:55,166 --> 00:08:56,196 Speaker 2: before there's a run. 151 00:08:56,436 --> 00:09:00,976 Speaker 1: Exactly. And the other thing was, because it was in anticipation, 152 00:09:00,996 --> 00:09:06,176 Speaker 1: what ended up happening is they would have projects that 153 00:09:06,236 --> 00:09:10,036 Speaker 1: could hire people to work on that were community development projects. 154 00:09:10,076 --> 00:09:13,496 Speaker 1: It could be tree planting. It could be road construction, 155 00:09:13,516 --> 00:09:16,796 Speaker 1: it could be improving wells, and then they will be 156 00:09:16,856 --> 00:09:19,816 Speaker 1: paid for those jobs. Yeah, that then they could use 157 00:09:19,836 --> 00:09:21,856 Speaker 1: that money to buy food and other things. 158 00:09:22,336 --> 00:09:24,256 Speaker 2: So it's interesting. I thought we were going to be 159 00:09:24,326 --> 00:09:27,526 Speaker 2: talking about satellites and AI, and I guess we will be. 160 00:09:28,006 --> 00:09:31,006 Speaker 2: But it's interesting that this conversation starts in such a 161 00:09:32,086 --> 00:09:35,026 Speaker 2: low-tech way. I mean, obviously, digital cameras were not quite 162 00:09:35,066 --> 00:09:37,756 Speaker 2: as ubiquitous as they are today. But this is 2016. 163 00:09:37,756 --> 00:09:40,916 Speaker 2: It's not 2000 or something. Lots of people already had 164 00:09:40,936 --> 00:09:41,816 Speaker 2: digital cameras. 165 00:09:44,336 --> 00:09:50,256 Speaker 1: Why? Yeah, why? I'd say people believe what they see. 166 00:09:50,566 --> 00:09:57,706 Speaker 1: And the NASA administrator in 2019, Jim Bernstein, he used 167 00:09:57,746 --> 00:10:02,406 Speaker 1: to share my photos in his presentations. He'd talk about 168 00:10:02,866 --> 00:10:06,246 Speaker 1: from NASA to the moon and beyond. But he would 169 00:10:06,286 --> 00:10:09,086 Speaker 1: still use the photos and not the maps or the images. 170 00:10:09,196 --> 00:10:11,996 Speaker 1: the numbers because it tells it you know we can 171 00:10:12,116 --> 00:10:14,516 Speaker 1: see it from above and this is what it means 172 00:10:14,676 --> 00:10:17,836 Speaker 1: on the ground this is the reality of what it 173 00:10:17,936 --> 00:10:19,876 Speaker 1: is on the ground and that's i think much more 174 00:10:19,896 --> 00:10:22,536 Speaker 1: accessible and understandable. 175 00:10:21,716 --> 00:10:25,586 Speaker 2: Um so I feel a little weird now after you're 176 00:10:25,646 --> 00:10:28,486 Speaker 2: talking compellingly about how important photos are. But I do 177 00:10:28,546 --> 00:10:31,146 Speaker 2: want to talk to you about satellite imagery and AI. 178 00:10:31,166 --> 00:10:32,306 Speaker 1: But they're photos. 179 00:10:32,606 --> 00:10:36,016 Speaker 2: They're photos. Very good. They're photos. I find them moving 180 00:10:36,056 --> 00:10:40,096 Speaker 2: in a different kind of way. I mean, you know, 181 00:10:40,196 --> 00:10:43,306 Speaker 2: I've read this sort of big hand-wavy thing, which is, 182 00:10:44,256 --> 00:10:46,276 Speaker 2: You know, there's lots of satellite imagery of the Earth 183 00:10:46,336 --> 00:10:49,226 Speaker 2: now and lots of AI models designed to analyze crops, 184 00:10:49,326 --> 00:10:54,826 Speaker 2: but those are largely designed for kind of big monocultural agribusiness, 185 00:10:54,906 --> 00:10:57,366 Speaker 2: and they don't apply so well, certainly, to the kind 186 00:10:57,386 --> 00:11:01,546 Speaker 2: of smallholder farms in the places you have worked. Like, 187 00:11:01,586 --> 00:11:05,056 Speaker 2: I get that that's kind of the big picture. But 188 00:11:05,116 --> 00:11:08,216 Speaker 2: more narrowly, like, tell me about your work to deal 189 00:11:08,236 --> 00:11:09,736 Speaker 2: with that. Like, first of all, what is it that 190 00:11:09,776 --> 00:11:13,116 Speaker 2: you want to know from satellite imagery? 191 00:11:13,766 --> 00:11:18,016 Speaker 1: We want to understand what's growing where and how it's doing. 192 00:11:19,796 --> 00:11:23,176 Speaker 1: The other is the complexity of the agriculture system. So 193 00:11:23,196 --> 00:11:28,206 Speaker 1: when we're looking for crop fields, we're looking for shapes 194 00:11:28,926 --> 00:11:31,886 Speaker 1: and then we're looking for some sort of seasonal pattern. 195 00:11:32,586 --> 00:11:36,066 Speaker 1: So if you had 10 images over a growing season, 196 00:11:36,766 --> 00:11:41,596 Speaker 1: you should see something like it's cleared now. it browns, 197 00:11:41,676 --> 00:11:45,256 Speaker 1: it's like brown and wet, and then something green pops up, 198 00:11:45,856 --> 00:11:48,536 Speaker 1: it matures, and then it's harvested. So it sort of 199 00:11:48,576 --> 00:11:51,476 Speaker 1: like becomes, you know, dry. And so we should be 200 00:11:51,516 --> 00:11:56,736 Speaker 1: able to see that signal. However, how low the brownness 201 00:11:56,896 --> 00:12:01,066 Speaker 1: might be, like how dry something might be, is contextual. 202 00:12:02,006 --> 00:12:07,166 Speaker 1: And while a reduction might seem like minor in some places, 203 00:12:07,246 --> 00:12:11,656 Speaker 1: it's actually a very serious problem. And so the first 204 00:12:11,676 --> 00:12:15,756 Speaker 1: thing was create a cropland map, basically exclude everything that's 205 00:12:15,796 --> 00:12:21,496 Speaker 1: not cropland, estimate, understand vegetation conditions historically, like what happens 206 00:12:21,636 --> 00:12:25,216 Speaker 1: every season, what months matter the most in terms of vegetation, 207 00:12:25,576 --> 00:12:28,756 Speaker 1: in terms of crop production, and then try to understand 208 00:12:28,876 --> 00:12:31,996 Speaker 1: at what point recovery doesn't happen. So to be able 209 00:12:32,016 --> 00:12:35,276 Speaker 1: to do that at a really large scale, satellites are 210 00:12:35,296 --> 00:12:39,036 Speaker 1: a phenomenal tool for this. At the time, even today, 211 00:12:39,536 --> 00:12:41,746 Speaker 1: the only satellite that allows us to be able to 212 00:12:41,826 --> 00:12:45,146 Speaker 1: do this at a much higher frequency is MODIS. And 213 00:12:45,206 --> 00:12:49,726 Speaker 1: MODIS is a 250-meter resolution. That could be multiple crop 214 00:12:49,746 --> 00:12:51,726 Speaker 1: fields in a single pixel. 215 00:12:52,386 --> 00:12:56,136 Speaker 2: So the farms are too small or too varied. 216 00:12:56,436 --> 00:12:57,396 Speaker 1: Too varied, yep. 217 00:12:58,206 --> 00:13:01,856 Speaker 2: Tell me, what is a typical farm in this region like? 218 00:13:02,096 --> 00:13:07,216 Speaker 1: So in Karamoja, people grow sorghum, millet, a little bit 219 00:13:07,256 --> 00:13:09,316 Speaker 1: of maize. There's a lot of sunflower, too. 220 00:13:10,236 --> 00:13:13,346 Speaker 2: And will they grow all that in one place, more 221 00:13:13,366 --> 00:13:13,526 Speaker 2: or less? 222 00:13:13,546 --> 00:13:17,566 Speaker 1: So you will find sunflower mixed in with sorghum, for example. 223 00:13:17,966 --> 00:13:19,886 Speaker 2: And how big is a typical farm? 224 00:13:20,146 --> 00:13:26,476 Speaker 1: So farm plot... On average, less than 2.5 hectares. So 225 00:13:26,756 --> 00:13:29,726 Speaker 1: a football field, a soccer field is about an acre. 226 00:13:31,546 --> 00:13:37,076 Speaker 1: That's the average over most of sub-Saharan Africa, but they're 227 00:13:37,116 --> 00:13:39,636 Speaker 1: much smaller. So like in a single soccer field, you 228 00:13:39,656 --> 00:13:43,076 Speaker 1: might find maybe four plots as an example. 229 00:13:43,756 --> 00:13:45,316 Speaker 2: Like four different people's little farms? 230 00:13:45,336 --> 00:13:49,346 Speaker 1: Four different people's plots with hedges, and they're not like 231 00:13:49,406 --> 00:13:51,386 Speaker 1: distinct perfect shapes like you would see. 232 00:13:51,606 --> 00:13:56,326 Speaker 2: Quite small. Hard to see from a satellite, presumably, especially 233 00:13:56,366 --> 00:13:58,106 Speaker 2: if there's a bunch of different crops there. 234 00:13:58,466 --> 00:14:01,736 Speaker 1: So hard to see their boundaries. So a single modus 235 00:14:01,796 --> 00:14:04,196 Speaker 1: pixel might include four of those fields that might have 236 00:14:04,256 --> 00:14:05,736 Speaker 1: different things, for example. 237 00:14:06,076 --> 00:14:12,076 Speaker 2: Yeah. Let's talk about the GoPros. So fun. Tell me 238 00:14:12,116 --> 00:14:13,206 Speaker 2: about your work with GoPros. 239 00:14:13,906 --> 00:14:17,456 Speaker 1: So one of the issues is in order to get 240 00:14:17,596 --> 00:14:20,556 Speaker 1: from cropland, where all the crops are growing, to go 241 00:14:20,596 --> 00:14:23,916 Speaker 1: to crop type. So the USDA, for example, produces a 242 00:14:23,996 --> 00:14:27,576 Speaker 1: cropland data layer, which if you take a look at it, 243 00:14:27,656 --> 00:14:30,206 Speaker 1: you can see where corn, wheat, soy is growing. And 244 00:14:30,236 --> 00:14:32,426 Speaker 1: now that's been changing over the last couple of years. 245 00:14:32,506 --> 00:14:35,686 Speaker 1: In the US. In the US, yes. This also exists 246 00:14:35,806 --> 00:14:40,166 Speaker 1: for Europe. They have something similar. There's, of course, some simplicity. 247 00:14:40,246 --> 00:14:44,576 Speaker 1: There's a single crop. The fields are homogeneous. They're kind 248 00:14:44,596 --> 00:14:47,706 Speaker 1: of easy to map. In East Africa, in addition to 249 00:14:47,746 --> 00:14:51,646 Speaker 1: fields being small, I already mentioned the intercropping, mixed cropping, 250 00:14:51,866 --> 00:14:55,336 Speaker 1: and all the heterogeneity, how people plant at different days. 251 00:14:56,176 --> 00:14:58,036 Speaker 1: To be able to get a crop type map, you 252 00:14:58,096 --> 00:15:00,996 Speaker 1: need a lot of training data. So, you know, one 253 00:15:01,016 --> 00:15:04,076 Speaker 1: of the biggest advances, I think, you know, the reason 254 00:15:04,096 --> 00:15:07,076 Speaker 1: why we have huge leaps in terms of AI is 255 00:15:07,096 --> 00:15:10,516 Speaker 1: the availability of training data. You know, same as face detection, 256 00:15:10,676 --> 00:15:13,856 Speaker 1: all people's faces on Facebook, et cetera, makes that technology 257 00:15:13,876 --> 00:15:16,856 Speaker 1: a lot easier to develop. And so this is true 258 00:15:16,956 --> 00:15:19,616 Speaker 1: also for crop type mapping. So it's like face detection, 259 00:15:19,676 --> 00:15:20,316 Speaker 1: but for crops. 260 00:15:20,616 --> 00:15:22,376 Speaker 2: Why do you want a crop type map? 261 00:15:22,776 --> 00:15:26,466 Speaker 1: There's so many benefits to knowing exactly what crop is 262 00:15:26,506 --> 00:15:29,706 Speaker 1: growing where. You want this map in order to understand, 263 00:15:29,786 --> 00:15:32,966 Speaker 1: for example, what kind of fertilizer will be relevant in 264 00:15:33,006 --> 00:15:37,586 Speaker 1: a particular region. So if you know there's places dominated 265 00:15:37,636 --> 00:15:40,886 Speaker 1: by rice, just like in Ghana, a particular place in Ghana, 266 00:15:41,366 --> 00:15:45,266 Speaker 1: where could irrigation infrastructure make a difference? The other and 267 00:15:45,366 --> 00:15:48,506 Speaker 1: most important, most basic thing is, if you want to 268 00:15:48,666 --> 00:15:51,236 Speaker 1: understand how much will be produced at the end of 269 00:15:51,296 --> 00:15:54,016 Speaker 1: the season, you need to know the total area by 270 00:15:54,076 --> 00:15:57,216 Speaker 1: specific crop. So if we have 25,000 hectares of corn planted... 271 00:16:00,286 --> 00:16:03,906 Speaker 1: And the average condition or the average yield is 2.5 272 00:16:03,906 --> 00:16:07,146 Speaker 1: tons per hectare. You multiply that and you have your production, 273 00:16:07,236 --> 00:16:10,496 Speaker 1: which will allow you to understand how much corn is available. 274 00:16:10,616 --> 00:16:13,276 Speaker 1: If it's a critical crop in a particular area, you 275 00:16:13,316 --> 00:16:16,856 Speaker 1: would know that, you know, food security would be okay. 276 00:16:17,296 --> 00:16:21,476 Speaker 1: So we had a big issue in Illinois with corn, 277 00:16:21,796 --> 00:16:25,256 Speaker 1: et cetera. If Germany had a big issue with wheat production, 278 00:16:25,376 --> 00:16:26,796 Speaker 1: bread being so important. 279 00:16:27,176 --> 00:16:27,376 Speaker 2: Yeah. 280 00:16:27,716 --> 00:16:30,336 Speaker 1: They would have to plan accordingly. So you need to 281 00:16:30,376 --> 00:16:32,896 Speaker 1: figure out, what do I need to import to have 282 00:16:32,946 --> 00:16:34,036 Speaker 1: some sort of stability? 283 00:16:34,276 --> 00:16:38,326 Speaker 2: And so for Sub-Saharan Africa in general, or the regions 284 00:16:38,366 --> 00:16:41,426 Speaker 2: you study, are there crop-type maps now? 285 00:16:42,086 --> 00:16:47,306 Speaker 1: There are some crop-type maps, yes. Some are done on 286 00:16:47,366 --> 00:16:50,736 Speaker 1: a very small scale, regional scale. But what this project 287 00:16:50,776 --> 00:16:53,866 Speaker 1: was trying to do was try to, one, just figure 288 00:16:54,066 --> 00:16:57,006 Speaker 1: out the fastest possible way we can collect data over 289 00:16:57,126 --> 00:17:00,316 Speaker 1: really large areas. One, because in some countries there are 290 00:17:00,336 --> 00:17:03,816 Speaker 1: three seasons. So Rwanda has three growing seasons. So you're 291 00:17:03,836 --> 00:17:07,356 Speaker 1: looking at three crops. So to cover the whole country 292 00:17:07,496 --> 00:17:13,646 Speaker 1: and get an understanding of did they grow corn more generally, 293 00:17:13,786 --> 00:17:16,886 Speaker 1: bananas more generally? No. You need to be able to 294 00:17:16,926 --> 00:17:18,226 Speaker 1: do the survey continuously. 295 00:17:18,806 --> 00:17:22,016 Speaker 2: And the satellites don't have the resolution, essentially, to do it. 296 00:17:22,016 --> 00:17:22,956 Speaker 1: To be able to do it, yeah. 297 00:17:23,176 --> 00:17:24,516 Speaker 2: You can't tell from satellite data. 298 00:17:24,556 --> 00:17:27,376 Speaker 1: So we need to train a model to tell the model, 299 00:17:27,456 --> 00:17:31,776 Speaker 1: this pixel, this particular spot is banana. Find me banana 300 00:17:31,876 --> 00:17:33,536 Speaker 1: everywhere else in the image, yeah. 301 00:17:34,296 --> 00:17:39,016 Speaker 2: So you have this idea of like, okay, we need 302 00:17:39,076 --> 00:17:42,436 Speaker 2: more on-the-ground data. So what do you do? What is 303 00:17:42,496 --> 00:17:43,416 Speaker 2: the GoPro project? 304 00:17:44,796 --> 00:17:49,606 Speaker 1: GoPros are perfect because they have a GPS. So when 305 00:17:49,626 --> 00:17:52,146 Speaker 1: we take a photo, the photo has a location. So 306 00:17:52,466 --> 00:17:56,096 Speaker 1: typically when I'm collecting field data, I need a photo, evidence, 307 00:17:56,576 --> 00:17:59,476 Speaker 1: but also I need to be clear when somebody labels 308 00:17:59,736 --> 00:18:03,586 Speaker 1: in the form that maize is in this growing stage 309 00:18:03,866 --> 00:18:05,486 Speaker 1: and you have to stand in the middle of the field. 310 00:18:06,006 --> 00:18:08,126 Speaker 1: Because I want to use that location to train my 311 00:18:08,166 --> 00:18:10,176 Speaker 1: model to predict where else maze might be. 312 00:18:10,326 --> 00:18:13,216 Speaker 2: This is the old school, slow, never going to scale 313 00:18:14,316 --> 00:18:17,956 Speaker 2: technique that you're trying to kind of quasi-automate here. 314 00:18:17,976 --> 00:18:21,816 Speaker 1: Yes. At this point, even if all I had were 315 00:18:21,976 --> 00:18:27,166 Speaker 1: images that had a longitude and latitude, and I know 316 00:18:27,186 --> 00:18:30,346 Speaker 1: where the camera is facing, I could go through all 317 00:18:30,386 --> 00:18:34,146 Speaker 1: these individual photos in Google Earth Pro and add the 318 00:18:34,186 --> 00:18:34,986 Speaker 1: label myself. 319 00:18:36,216 --> 00:18:38,276 Speaker 2: So at least you would have a bunch of images 320 00:18:38,876 --> 00:18:43,656 Speaker 2: with a very specific geolocation on latitude and longitude. 321 00:18:43,876 --> 00:18:46,176 Speaker 1: And I know what the crops look like. I know 322 00:18:46,216 --> 00:18:47,056 Speaker 1: what the crops are. 323 00:18:47,536 --> 00:18:50,196 Speaker 2: So this is phase one. But you don't want to 324 00:18:50,216 --> 00:18:54,236 Speaker 2: have to label a thousand things corn, whatever. 325 00:18:54,256 --> 00:18:58,866 Speaker 1: Yeah, I would have been okay with that too, though. Considering, 326 00:18:58,946 --> 00:19:01,506 Speaker 1: so I did this in Mali. I did it in Senegal. 327 00:19:02,336 --> 00:19:02,976 Speaker 1: In Tanzania. 328 00:19:02,996 --> 00:19:07,376 Speaker 2: Wait, this meaning driving around with the GoPro? No, the phone. Oh, 329 00:19:07,396 --> 00:19:08,136 Speaker 2: the walking around. 330 00:19:08,196 --> 00:19:09,236 Speaker 1: Walking around, yeah. 331 00:19:09,336 --> 00:19:11,996 Speaker 2: Yeah, okay. I mean, it sounds kind of charming in 332 00:19:12,036 --> 00:19:12,516 Speaker 2: a way. 333 00:19:12,536 --> 00:19:16,556 Speaker 1: I learn a lot from doing field work. You learn 334 00:19:16,856 --> 00:19:19,116 Speaker 1: what it means for someone's field to be washed out 335 00:19:19,136 --> 00:19:22,276 Speaker 1: when you go. But if the idea is to try 336 00:19:22,336 --> 00:19:25,226 Speaker 1: and create a map that will be useful for this task, 337 00:19:26,606 --> 00:19:29,386 Speaker 1: there's only so much you can do when you cover 338 00:19:29,406 --> 00:19:32,886 Speaker 1: a very small area. Yeah. So 2020, I call this 339 00:19:32,926 --> 00:19:36,236 Speaker 1: project a COVID-safe project. 2020 was like, oh, well, now 340 00:19:36,246 --> 00:19:38,616 Speaker 1: we have this funding. We have to figure this out. 341 00:19:38,696 --> 00:19:41,596 Speaker 1: I go buy a GoPro at a store, a fashion 342 00:19:43,656 --> 00:19:47,056 Speaker 1: sort of amount with my dad because I couldn't order 343 00:19:47,096 --> 00:19:51,206 Speaker 1: it from Amazon. And then we make this magnetic thing 344 00:19:51,356 --> 00:19:53,846 Speaker 1: on the car and just drive. I go to, there's 345 00:19:53,866 --> 00:19:57,546 Speaker 1: an agricultural research station. And the director was like, I 346 00:19:57,566 --> 00:19:59,106 Speaker 1: described to him what I'm trying to do. He was like, oh, 347 00:19:59,166 --> 00:20:02,086 Speaker 1: course you can drive around here, whatever. So we try 348 00:20:02,106 --> 00:20:03,996 Speaker 1: it out. So the first thing was, if we get 349 00:20:04,036 --> 00:20:07,256 Speaker 1: the images, are they useful? Is the GPS good enough? 350 00:20:07,836 --> 00:20:08,396 Speaker 2: And it was. 351 00:20:09,116 --> 00:20:12,436 Speaker 1: And so then we were like, oh, this could actually work. 352 00:20:17,256 --> 00:20:29,206 Speaker 2: We'll be back in just a minute. That's the end 353 00:20:29,226 --> 00:20:31,046 Speaker 2: of the ads. We're going back to the show. And 354 00:20:31,266 --> 00:20:33,786 Speaker 2: as you probably recall, when we left off before the break, 355 00:20:34,186 --> 00:20:37,546 Speaker 2: Catherine had just tested out a GoPro herself and decided 356 00:20:37,586 --> 00:20:40,066 Speaker 2: that GoPros could be a good way to gather crop 357 00:20:40,086 --> 00:20:41,506 Speaker 2: data from small farms. 358 00:20:42,106 --> 00:20:46,146 Speaker 1: So I have a friend in Kenya. I asked him 359 00:20:46,266 --> 00:20:48,456 Speaker 1: if he can buy, if he can order things on 360 00:20:48,496 --> 00:20:53,336 Speaker 1: Amazon and to work with a team there to see 361 00:20:53,356 --> 00:20:56,896 Speaker 1: if they can collect data in Kenya. So we get 362 00:20:56,936 --> 00:21:00,016 Speaker 1: a team of two. So basically four. So it's a 363 00:21:00,036 --> 00:21:03,706 Speaker 1: team of two people. One team drives south, the other 364 00:21:03,726 --> 00:21:08,006 Speaker 1: drives north. And we cover all of Western Kenya in 365 00:21:08,046 --> 00:21:08,406 Speaker 1: a week. 366 00:21:09,186 --> 00:21:12,206 Speaker 2: Wow. And they're just driving everywhere they can drive, in 367 00:21:12,226 --> 00:21:13,486 Speaker 2: a car, on a motorcycle? 368 00:21:13,846 --> 00:21:15,846 Speaker 1: So you can see from some of the original data 369 00:21:15,886 --> 00:21:18,326 Speaker 1: that we have from Kenya that we're kind of covering 370 00:21:18,346 --> 00:21:22,326 Speaker 1: these big... There are lots of fields, but they're proximal 371 00:21:22,376 --> 00:21:27,716 Speaker 1: to these big roads, which... It's great, but not good enough. 372 00:21:28,436 --> 00:21:31,596 Speaker 1: And so it became pretty quickly obvious that a motorcycle 373 00:21:31,616 --> 00:21:33,836 Speaker 1: might be better because a motorcycle can go to a 374 00:21:33,896 --> 00:21:37,556 Speaker 1: level two, level three, you know, smaller paths and go through. 375 00:21:37,596 --> 00:21:39,446 Speaker 1: And in Tanzania, they went a little bit crazy because 376 00:21:39,526 --> 00:21:42,206 Speaker 1: I feel like they're driving into rice paddies and going 377 00:21:42,246 --> 00:21:42,626 Speaker 1: like this. 378 00:21:42,666 --> 00:21:45,366 Speaker 2: Well, people live far off of roads that you could 379 00:21:45,386 --> 00:21:48,006 Speaker 2: drive a car on. Exactly. The kinds of farms you're 380 00:21:48,046 --> 00:21:50,826 Speaker 2: talking about exist far from drivable roads. 381 00:21:50,926 --> 00:21:55,626 Speaker 1: Exactly. So... But just from the Kenya team, we had over, 382 00:21:55,646 --> 00:22:01,166 Speaker 1: I think the first iteration might have been maybe 300,000 383 00:22:01,166 --> 00:22:06,086 Speaker 1: images or more. Then the team of two in Uganda, 384 00:22:06,106 --> 00:22:11,266 Speaker 1: I kind of equipped them. And during COVID, there's nothing 385 00:22:11,286 --> 00:22:12,866 Speaker 1: else you could do and asked them to kind of 386 00:22:12,906 --> 00:22:16,266 Speaker 1: just go drive around. And they went crazy because then 387 00:22:16,306 --> 00:22:19,116 Speaker 1: we had over 2 million images from them. Okay. And 388 00:22:19,176 --> 00:22:22,066 Speaker 1: so then it's just too many images. Even I got 389 00:22:22,186 --> 00:22:24,706 Speaker 1: tired of just like trying to figure it out. But 390 00:22:24,726 --> 00:22:27,466 Speaker 1: I had an undergraduate computer science student who worked over 391 00:22:27,506 --> 00:22:31,526 Speaker 1: this and developed Street to Sat, the first iteration of it. 392 00:22:32,676 --> 00:22:38,136 Speaker 1: And remove clouds, adjust images automatically so that they're kind 393 00:22:38,176 --> 00:22:44,776 Speaker 1: of like straight. And then remove faces and ensure that 394 00:22:44,836 --> 00:22:47,816 Speaker 1: they're actually crops in the image. 395 00:22:48,376 --> 00:22:50,896 Speaker 2: Crop and not crop, you called it in the paper. 396 00:22:50,916 --> 00:22:52,806 Speaker 2: It reminded me, I don't know if you watched the 397 00:22:52,846 --> 00:22:56,866 Speaker 2: show Silicon Valley, but there's a guy in that show 398 00:22:57,006 --> 00:22:57,866 Speaker 2: who invents AI. 399 00:22:59,596 --> 00:23:02,536 Speaker 2: And they show it a hot dog and it says 400 00:23:02,596 --> 00:23:04,556 Speaker 2: hot dog. And everybody's like, wow, it can detect food. 401 00:23:04,576 --> 00:23:05,816 Speaker 2: And then they show it a hamburger and it just 402 00:23:05,836 --> 00:23:07,156 Speaker 2: says not hot dog. 403 00:23:07,236 --> 00:23:07,416 Speaker 1: Yeah. 404 00:23:07,536 --> 00:23:09,236 Speaker 2: And all it does is say hot dog, not a 405 00:23:09,256 --> 00:23:10,616 Speaker 2: hot dog. This is where you are now. 406 00:23:10,656 --> 00:23:12,196 Speaker 1: This is where we are. It's like, is there a 407 00:23:12,216 --> 00:23:15,916 Speaker 1: crop in this field? Yes. If there is, then that's phenomenal. 408 00:23:15,996 --> 00:23:19,156 Speaker 1: And then we kind of did a couple of iterations 409 00:23:19,576 --> 00:23:25,016 Speaker 1: of what we call labeling. Yeah. Basically, go through a 410 00:23:25,176 --> 00:23:28,816 Speaker 1: couple of images to create a label dataset, which is, 411 00:23:29,016 --> 00:23:32,056 Speaker 1: this is, we draw bounding boxes, maize, maize, this is 412 00:23:32,096 --> 00:23:34,576 Speaker 1: what maize looks like, this is what banana looks like. 413 00:23:34,816 --> 00:23:37,596 Speaker 2: So now you're saying you're, this is training, you're creating training. 414 00:23:37,616 --> 00:23:38,676 Speaker 1: We're creating a training dataset, yeah. 415 00:23:38,696 --> 00:23:39,596 Speaker 2: Training data, yeah. 416 00:23:39,876 --> 00:23:42,896 Speaker 1: And then with the first iteration, then we run it 417 00:23:42,936 --> 00:23:48,126 Speaker 1: through the initial data from Kenya to do sugarcane maize. Okay. Okay. 418 00:23:48,446 --> 00:23:51,706 Speaker 1: And it was pretty good. It was pretty good. It 419 00:23:51,746 --> 00:23:54,866 Speaker 1: was like, you know, when it's actually working and you 420 00:23:54,906 --> 00:23:57,726 Speaker 1: can't believe that it's working, but it is working and 421 00:23:57,786 --> 00:24:00,386 Speaker 1: it's unbelievable. So you're like, you're in this phase where 422 00:24:00,406 --> 00:24:03,666 Speaker 1: you're like, oh my God, is this actually happening? 423 00:24:05,006 --> 00:24:09,466 Speaker 2: What are people doing now with this basic technique of, 424 00:24:09,506 --> 00:24:11,846 Speaker 2: you know, driving around with GoPro cameras where there are 425 00:24:11,886 --> 00:24:15,436 Speaker 2: these small farms? Like what is one specific thing someone 426 00:24:15,476 --> 00:24:18,656 Speaker 2: is doing that is helping them take action in the 427 00:24:18,696 --> 00:24:20,236 Speaker 2: world in a way they would not have been able 428 00:24:20,276 --> 00:24:20,536 Speaker 2: to do? 429 00:24:21,136 --> 00:24:25,986 Speaker 1: So I give one example. So in Kenya, when we 430 00:24:25,996 --> 00:24:29,926 Speaker 1: did the first iteration of it as a test pilot, 431 00:24:30,566 --> 00:24:32,626 Speaker 1: and then were able to actually work with the Ministry 432 00:24:32,666 --> 00:24:34,786 Speaker 1: of Agriculture for them to go out and collect the 433 00:24:34,806 --> 00:24:37,186 Speaker 1: data themselves, they were like, do we get to keep 434 00:24:37,226 --> 00:24:42,106 Speaker 1: the cameras? I was like, yeah. And I asked, you know, 435 00:24:42,246 --> 00:24:44,466 Speaker 1: what sorts of things are you thinking about? I was like, well, 436 00:24:44,606 --> 00:24:48,156 Speaker 1: you know, that would be kind of just Phenomenal photographic 437 00:24:48,256 --> 00:24:53,616 Speaker 1: evidence because a person can go back whenever and send 438 00:24:53,656 --> 00:24:57,256 Speaker 1: me those photos while they're out there in the middle 439 00:24:57,276 --> 00:25:00,486 Speaker 1: of nowhere. So remember going back to the low-tech technology 440 00:25:00,686 --> 00:25:03,766 Speaker 1: is that when the ministry reports, so we have these 441 00:25:03,806 --> 00:25:08,306 Speaker 1: WhatsApp groups where extension agents share constantly what's happening. If 442 00:25:08,346 --> 00:25:10,826 Speaker 1: they share that there was a flood and the flood 443 00:25:10,866 --> 00:25:13,566 Speaker 1: was serious, everybody knows a flood is serious, but how 444 00:25:13,686 --> 00:25:17,286 Speaker 1: serious is a different story. when it comes to things 445 00:25:17,306 --> 00:25:20,146 Speaker 1: like beans, water staying on the surface for a long 446 00:25:20,206 --> 00:25:23,456 Speaker 1: time and the beans turn yellow is because all the 447 00:25:23,496 --> 00:25:27,016 Speaker 1: nutrients have leached. And so for the extension agents to 448 00:25:27,056 --> 00:25:29,896 Speaker 1: be able to go through and collect these images without 449 00:25:29,936 --> 00:25:32,296 Speaker 1: needing to get off, you know, take a photo, et cetera, 450 00:25:32,336 --> 00:25:35,556 Speaker 1: et cetera, is something that is used as kind of 451 00:25:35,666 --> 00:25:38,566 Speaker 1: evidence and as part of their reporting in the ministry. 452 00:25:38,786 --> 00:25:40,566 Speaker 2: So it gives you a baseline. 453 00:25:40,586 --> 00:25:44,046 Speaker 1: It's like a baseline, but also people believe what they see. 454 00:25:44,606 --> 00:25:47,616 Speaker 2: It goes back to you walking around taking pictures, but 455 00:25:47,636 --> 00:25:48,936 Speaker 2: just a lot more pictures. 456 00:25:48,996 --> 00:25:49,236 Speaker 1: Yeah. 457 00:25:50,236 --> 00:25:53,956 Speaker 2: So if you sort of distill it down, what is 458 00:25:53,976 --> 00:25:56,236 Speaker 2: your big project? What are you trying to do in 459 00:25:56,276 --> 00:25:57,336 Speaker 2: the world in your work? 460 00:25:58,176 --> 00:26:04,456 Speaker 1: So data is an essential, right? Evidence is essential. But 461 00:26:04,556 --> 00:26:07,996 Speaker 1: try to use it to improve, you could say, the 462 00:26:08,036 --> 00:26:12,616 Speaker 1: human condition through better decision making. So If I know 463 00:26:12,656 --> 00:26:16,356 Speaker 1: for sure that something is coming and it will be 464 00:26:16,406 --> 00:26:20,206 Speaker 1: this bad, it means that the persons or people responsible 465 00:26:20,246 --> 00:26:24,626 Speaker 1: for decision-making, they have no excuse to not do the 466 00:26:24,666 --> 00:26:28,386 Speaker 1: right thing. I think that's kind of what kind of 467 00:26:28,406 --> 00:26:30,326 Speaker 1: drives me. And I was going to talk about the 468 00:26:30,326 --> 00:26:33,676 Speaker 1: 2.0 of the disaster risk financing program is because we've 469 00:26:33,716 --> 00:26:37,836 Speaker 1: gone so far now since then that what I've built 470 00:26:37,976 --> 00:26:43,406 Speaker 1: out now gives a three-month forecast of what things will 471 00:26:43,426 --> 00:26:47,146 Speaker 1: be three months from now based on a model that 472 00:26:47,186 --> 00:26:51,656 Speaker 1: has learned. So because I can process more data, I 473 00:26:51,716 --> 00:26:56,836 Speaker 1: can combine multiple complex data sets than before, thanks to compute, 474 00:26:58,676 --> 00:27:02,636 Speaker 1: and increase access to data. But I understand the problem 475 00:27:02,656 --> 00:27:06,736 Speaker 1: so deeply that that threshold now is specific to the 476 00:27:06,856 --> 00:27:10,556 Speaker 1: specific research It's no longer one threshold for all of them. 477 00:27:11,216 --> 00:27:14,816 Speaker 1: It is a district, like a county-specific threshold that could 478 00:27:14,856 --> 00:27:19,126 Speaker 1: even be lowered to be sub-county that could be zip 479 00:27:19,166 --> 00:27:25,506 Speaker 1: code-specific threshold that in this zip code, this threshold is it. 480 00:27:26,786 --> 00:27:28,206 Speaker 1: It doesn't apply to everywhere else. 481 00:27:28,466 --> 00:27:32,126 Speaker 2: It's sort of zip code by zip code in Uganda. 482 00:27:32,186 --> 00:27:34,766 Speaker 2: You can tell how much drought is going to be 483 00:27:34,826 --> 00:27:37,136 Speaker 2: really bad. Three months in advance. 484 00:27:37,296 --> 00:27:39,726 Speaker 1: Exactly. And it's not only in Uganda. The method is 485 00:27:40,016 --> 00:27:43,746 Speaker 1: largely scalable. I could apply the same methodology to the U.S. 486 00:27:43,806 --> 00:27:47,166 Speaker 2: Do you have the data? Yeah. Can you do it anywhere, 487 00:27:47,326 --> 00:27:49,746 Speaker 2: or do you need some amount of data collection in 488 00:27:49,786 --> 00:27:50,326 Speaker 2: some places? 489 00:27:50,886 --> 00:27:56,046 Speaker 1: So for the forecasting component, all I need is satellite rainfall, 490 00:27:56,166 --> 00:28:02,466 Speaker 1: satellite temperature, satellite vegetation conditions. And then the validation component 491 00:28:02,706 --> 00:28:05,386 Speaker 1: is where I would need people to actually go and confirm, 492 00:28:05,566 --> 00:28:09,146 Speaker 1: you know, am I actually predicting the right thing? As 493 00:28:09,186 --> 00:28:13,106 Speaker 1: an example, you could use auxiliary things, you know, you 494 00:28:13,136 --> 00:28:16,056 Speaker 1: could use images people are posting on social media about 495 00:28:16,096 --> 00:28:19,156 Speaker 1: events to sort of as an in-between. But in order 496 00:28:19,296 --> 00:28:23,516 Speaker 1: for the evidence, you know, to be strengthened, then you 497 00:28:23,556 --> 00:28:26,646 Speaker 1: want to go, you know, take those photos. And then 498 00:28:26,666 --> 00:28:28,826 Speaker 1: the other thing is, you know, have a routine component 499 00:28:28,866 --> 00:28:32,326 Speaker 1: around collecting and gathering that evidence so that it's not 500 00:28:32,676 --> 00:28:35,096 Speaker 1: whenever you feel like or whenever you have money, but 501 00:28:35,136 --> 00:28:37,456 Speaker 1: it's just that at this point, we need to check 502 00:28:37,616 --> 00:28:41,216 Speaker 1: in and see how things are going. 503 00:28:41,656 --> 00:28:45,856 Speaker 2: Like, if I were going to say, Catherine's problem is this, like, 504 00:28:45,916 --> 00:28:46,976 Speaker 2: what would the sentence be? 505 00:28:48,336 --> 00:28:53,836 Speaker 1: There are lots of quite simple things that we could 506 00:28:53,906 --> 00:28:57,686 Speaker 1: do already, but we just don't do them because they 507 00:28:57,706 --> 00:29:02,626 Speaker 1: seem so Maybe we overcomplicate them. I always think about these, 508 00:29:02,786 --> 00:29:06,306 Speaker 1: a minister will not, you know, read your F1 score table, 509 00:29:07,006 --> 00:29:10,506 Speaker 1: you know, the confidence of your model, et cetera, but 510 00:29:10,526 --> 00:29:13,886 Speaker 1: the minister will read a memo that's usually like three paragraphs. 511 00:29:14,646 --> 00:29:17,766 Speaker 1: What I learned in the risk financing work is that 512 00:29:18,086 --> 00:29:21,616 Speaker 1: it has to be distilled into a half-page memo. The 513 00:29:21,676 --> 00:29:29,686 Speaker 1: memo is evidence points that we've hit the thresholds, about 514 00:29:29,686 --> 00:29:34,266 Speaker 1: 70,000 people will need emergency aid, and we're going to 515 00:29:34,306 --> 00:29:38,866 Speaker 1: release funding on this day. That's it. And then the 516 00:29:38,906 --> 00:29:44,306 Speaker 1: minister says, I agree. Right, so whether or not it 517 00:29:44,346 --> 00:29:53,696 Speaker 1: took me 100 months, $ 5 billion... 25 analysts to create 518 00:29:53,756 --> 00:29:58,376 Speaker 1: the evidence or whatever it is, usually it's something that 519 00:29:59,276 --> 00:30:05,516 Speaker 1: is accessible, concise, that helps that decision making. And accessibility 520 00:30:05,576 --> 00:30:08,846 Speaker 1: to that information is so critical in this case. So 521 00:30:08,886 --> 00:30:10,866 Speaker 1: I'd spend a lot of time, you know, training and 522 00:30:11,226 --> 00:30:13,566 Speaker 1: building capacity so people can reproduce things. 523 00:30:13,466 --> 00:30:14,026 Speaker 2: On their own. 524 00:30:15,666 --> 00:30:18,136 Speaker 1: Like you can run the helmet's work all on your own. 525 00:30:18,156 --> 00:30:21,676 Speaker 1: All the methods, everything you need in the kit. And 526 00:30:21,696 --> 00:30:23,276 Speaker 1: a lot of people have done this as a team 527 00:30:23,316 --> 00:30:25,116 Speaker 1: that's done it in Nigeria. People have done it in 528 00:30:25,176 --> 00:30:26,036 Speaker 1: India on their own. 529 00:30:26,776 --> 00:30:29,956 Speaker 2: Basically building crop-type maps using GoPro cameras? 530 00:30:30,136 --> 00:30:36,016 Speaker 1: Creating crop-type labels, yeah, using GoPros. But also the workflow itself. 531 00:30:37,076 --> 00:30:39,996 Speaker 1: We have a notebook available that anybody can run with 532 00:30:40,016 --> 00:30:45,116 Speaker 1: a Jupyter notebook without anything else. So the problem is 533 00:30:47,336 --> 00:30:49,576 Speaker 1: some things are a lot simpler than we make them 534 00:30:49,656 --> 00:30:54,456 Speaker 1: out to be. And if the thing is to solve hunger... 535 00:30:54,636 --> 00:30:58,836 Speaker 1: I think those solutions for solving hunger already exist. And 536 00:30:58,856 --> 00:31:01,356 Speaker 1: they have nothing to do with my model or my technology. 537 00:31:02,076 --> 00:31:05,776 Speaker 1: It has to be moving food where it is produced 538 00:31:05,836 --> 00:31:09,626 Speaker 1: to where it's needed, making it available when it's needed 539 00:31:09,686 --> 00:31:13,906 Speaker 1: the most, for example. And I can provide the evidence 540 00:31:13,946 --> 00:31:16,326 Speaker 1: to be like, okay, it's absolutely needed here. So I 541 00:31:18,586 --> 00:31:19,626 Speaker 1: think that's my problem. 542 00:31:24,476 --> 00:31:26,296 Speaker 2: We'll be back in a minute with The Lightning Round. 543 00:31:37,796 --> 00:31:41,136 Speaker 2: We're going to finish with The Lightning Round, which is 544 00:31:41,176 --> 00:31:43,436 Speaker 2: just a little bit more playful than the rest of 545 00:31:43,456 --> 00:31:47,836 Speaker 2: the show. What's one thing your dad taught you about 546 00:31:47,906 --> 00:31:48,736 Speaker 2: fixing cars? 547 00:31:49,386 --> 00:31:50,726 Speaker 1: Oh, my God, that guy. 548 00:31:52,586 --> 00:31:54,686 Speaker 2: My dad my dad told me um. 549 00:31:56,346 --> 00:31:58,946 Speaker 1: We had my car battery in 2020 of it was 550 00:31:59,006 --> 00:32:03,516 Speaker 1: overheating whatever and he was like how do you drive 551 00:32:03,576 --> 00:32:05,796 Speaker 1: a car without opening the hood and i was like 552 00:32:07,396 --> 00:32:10,456 Speaker 1: like seriously and then i said this to him uh 553 00:32:10,956 --> 00:32:14,006 Speaker 1: because his phone is runs out of memory and i 554 00:32:14,026 --> 00:32:16,296 Speaker 1: was like How do you work with a phone that 555 00:32:16,336 --> 00:32:18,436 Speaker 1: has no memory? So, you know, I was trying to 556 00:32:18,456 --> 00:32:20,396 Speaker 1: kind of like paint this problem. My dad, what I've 557 00:32:20,416 --> 00:32:23,226 Speaker 1: learned from him is you can figure it out. And 558 00:32:23,276 --> 00:32:28,306 Speaker 1: he has this terminology. It's, I think, a Luganda thing. 559 00:32:28,406 --> 00:32:34,206 Speaker 1: It says chito chige, which is kill it to learn it. So, 560 00:32:34,446 --> 00:32:36,786 Speaker 1: you know, in terms of like fixing a radio, you 561 00:32:36,806 --> 00:32:38,786 Speaker 1: take it apart and then you have to put it 562 00:32:38,826 --> 00:32:39,476 Speaker 1: back together. 563 00:32:39,496 --> 00:32:41,836 Speaker 2: It's like break it, break it and fix it and 564 00:32:41,876 --> 00:32:42,636 Speaker 2: you'll understand it. 565 00:32:42,936 --> 00:32:45,906 Speaker 1: Break it to fix it. basically, because then you will 566 00:32:45,946 --> 00:32:48,996 Speaker 1: understand it. And so he always took apart, you know, 567 00:32:49,016 --> 00:32:51,696 Speaker 1: car engines, and then he will put them all back together. 568 00:32:51,796 --> 00:32:54,316 Speaker 1: So I think what I learned is that you can learn, 569 00:32:54,376 --> 00:32:56,916 Speaker 1: you can figure it out, whatever it is. 570 00:32:58,296 --> 00:33:00,866 Speaker 2: What's one thing you learned from your mom about running 571 00:33:00,906 --> 00:33:01,466 Speaker 2: a restaurant? 572 00:33:02,886 --> 00:33:08,246 Speaker 1: My mom is a workaholic. She's told me recently that 573 00:33:09,706 --> 00:33:12,426 Speaker 1: She works, she likes, she keeps working because she likes 574 00:33:12,466 --> 00:33:14,726 Speaker 1: to work. She's no longer, it's no longer profitable, but 575 00:33:14,786 --> 00:33:16,876 Speaker 1: she likes to work and she believes she stops working. 576 00:33:17,416 --> 00:33:20,576 Speaker 1: You know, she'll just like fall over and die. It's 577 00:33:20,616 --> 00:33:22,996 Speaker 1: just an extreme, it's an extreme statement. But what she 578 00:33:23,036 --> 00:33:27,356 Speaker 1: means is that whatever the work is, it's important and 579 00:33:27,396 --> 00:33:29,936 Speaker 1: do it to the best of your ability. And so 580 00:33:30,116 --> 00:33:32,856 Speaker 1: she does, you know, has done so many different things 581 00:33:32,896 --> 00:33:36,996 Speaker 1: to kind of sustain it and keep it running. And so... 582 00:33:37,806 --> 00:33:41,646 Speaker 1: Keep on moving, keep on working, and you can figure 583 00:33:41,686 --> 00:33:43,126 Speaker 1: it out, I guess, also in this case. 584 00:33:43,926 --> 00:33:46,586 Speaker 2: Who's a better badminton player, you or your sister? 585 00:33:46,626 --> 00:33:47,926 Speaker 1: My sister, she's left-handed. 586 00:33:48,366 --> 00:33:49,506 Speaker 2: Oh, is that a big advantage? 587 00:33:49,566 --> 00:33:51,716 Speaker 1: Of course, you have no idea where her left from 588 00:33:51,756 --> 00:33:54,656 Speaker 1: her right is. Her backhand is your, what you think 589 00:33:54,716 --> 00:33:59,096 Speaker 1: is her backhand is her forehand. And then she worked 590 00:33:59,176 --> 00:34:02,696 Speaker 1: so hard to improve her backhand that her backhand feels 591 00:34:02,716 --> 00:34:05,156 Speaker 1: like a forehand, right? It's like the one when you 592 00:34:05,196 --> 00:34:09,236 Speaker 1: hit it and it comes back. You're like, oh my God. 593 00:34:09,566 --> 00:34:13,116 Speaker 2: Yeah. That's the sound it makes too, right? In badminton. 594 00:34:13,936 --> 00:34:16,016 Speaker 2: What's one thing I should do if I visit Uganda? 595 00:34:21,296 --> 00:34:21,876 Speaker 1: Eat a mango. 596 00:34:22,936 --> 00:34:24,396 Speaker 2: Okay. Are they better there? 597 00:34:24,556 --> 00:34:26,116 Speaker 1: Absolutely. And an avocado. 598 00:34:26,816 --> 00:34:28,096 Speaker 2: And an avocado. Also better? 599 00:34:28,256 --> 00:34:33,726 Speaker 1: Yeah. Never refrigerated. Fresh. Maybe harvested yesterday. 600 00:34:34,986 --> 00:34:40,006 Speaker 2: Um... What about GoPros? Are you dreaming of other uses 601 00:34:40,066 --> 00:34:42,686 Speaker 2: of GoPros for work or not for work or things 602 00:34:42,786 --> 00:34:44,386 Speaker 2: other people could be doing with GoPros? 603 00:34:44,946 --> 00:34:49,066 Speaker 1: Yeah. It's funny. I always thought about them. I used 604 00:34:49,106 --> 00:34:52,226 Speaker 1: to wear one. The first, first iteration of GoPros, when 605 00:34:52,266 --> 00:34:54,566 Speaker 1: I went to field work, I wanted to record everything. 606 00:34:55,166 --> 00:34:59,666 Speaker 1: But about GoPros, if I was to design one, I'd 607 00:34:59,726 --> 00:35:02,886 Speaker 1: include one that has a distance measurement system. 608 00:35:02,946 --> 00:35:03,926 Speaker 2: I would do that. 609 00:35:05,866 --> 00:35:08,136 Speaker 1: So I can know how far my maze is from 610 00:35:08,756 --> 00:35:09,696 Speaker 1: where I'm standing. 611 00:35:11,336 --> 00:35:13,356 Speaker 2: You're trying to do that with software, essentially. 612 00:35:13,396 --> 00:35:15,976 Speaker 1: Yes. But I'd like for it to give me an 613 00:35:16,076 --> 00:35:21,586 Speaker 1: estimate of how far something is. It's kind of easy 614 00:35:21,686 --> 00:35:24,426 Speaker 1: to derive that if you're in a city because you 615 00:35:24,446 --> 00:35:28,366 Speaker 1: have corners and things, very strict structures. But you can't 616 00:35:28,406 --> 00:35:30,666 Speaker 1: do that when you're doing it for crops. I did 617 00:35:30,756 --> 00:35:33,536 Speaker 1: win a GoPro for good, but they sent me a GoPro. 618 00:35:34,896 --> 00:35:37,316 Speaker 1: which I already have a lot of GoPros. 619 00:35:37,356 --> 00:35:39,656 Speaker 2: You won an award, and the award was another GoPro? 620 00:35:39,676 --> 00:35:39,916 Speaker 1: Yes. 621 00:35:40,456 --> 00:35:41,696 Speaker 2: What do you wish they'd sent you? 622 00:35:41,996 --> 00:35:46,996 Speaker 1: I wish they had a conversation and I'd suggest what 623 00:35:47,016 --> 00:35:47,876 Speaker 1: I want in the GoPro. 624 00:35:48,436 --> 00:35:52,186 Speaker 2: Yeah, right. A GoPro that could measure distance. 625 00:35:52,606 --> 00:35:53,726 Speaker 1: Yeah. 626 00:35:53,946 --> 00:35:56,626 Speaker 2: I appreciate your time. It was lovely to talk with you. 627 00:35:56,666 --> 00:35:57,386 Speaker 2: Thank you so much. 628 00:35:57,566 --> 00:35:59,986 Speaker 1: Lovely talking with you, too. Thank you for having me. 629 00:36:07,236 --> 00:36:10,296 Speaker 2: Catherine Nakalembe is an assistant professor at the University of 630 00:36:10,336 --> 00:36:15,076 Speaker 2: Maryland and Africa program director at NASA Harvest. Please email 631 00:36:15,136 --> 00:36:18,696 Speaker 2: us at problem at pushkin.fm. Tell us what kinds of 632 00:36:18,716 --> 00:36:20,356 Speaker 2: shows we should do more of or less of or 633 00:36:20,416 --> 00:36:22,316 Speaker 2: particular people you think would be good on the show. 634 00:36:23,196 --> 00:36:25,476 Speaker 2: You can also find me on X and on LinkedIn. 635 00:36:26,336 --> 00:36:29,696 Speaker 2: Our show is produced by Gabriel Hunter Chang and Trina Menino. 636 00:36:30,056 --> 00:36:32,756 Speaker 2: Our editor is Lydia Jean Cott and our engineer is 637 00:36:32,816 --> 00:36:35,836 Speaker 2: Sarah Bruguere. I'm Jacob Goldstein, and we'll be back next 638 00:36:35,876 --> 00:36:37,546 Speaker 2: week with another episode of What's Your Problem?