Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
PolicyViz Podcast Episode #7: Rob Simmon
140 Plays10 years ago
In this week’s episode, I speak with color-guru Rob Simmon from Planet Labs. If you don’t know Rob, you should absolutely read is famous 6-part series on color hosted at the Visual.ly blog. Nowadays, Rob is busy launching satellites and...
The post PolicyViz Podcast Episode #7: Rob Simmon appeared first on PolicyViz.
Recommended
Transcript
Introduction to Rob Simmons and Planet Labs
00:00:11
Speaker
Welcome to the Policy Vis podcast. I'm your host, John Schwabisch. I'm here with my good friend, ex-DC Patriot, Rob Simmons from Planet Labs. Rob, welcome to the show.
00:00:21
Speaker
Hi, John. Thanks. Thanks for coming on. Great to see you again, man. It's been what you've been in in California Planet Labs now for two years. A year? No, not even a year. We moved out in September. Nice. And you're at NASA before that. And now doing some fun stuff at Planet Labs.
Planet Labs' Mission and Technology
00:00:41
Speaker
You want to maybe just explain for listeners what Planet Labs does and what your role is there?
00:00:48
Speaker
Sure. So Planet Labs is in some ways a classic Silicon Valley startup, but more focused on hardware and space than the traditional startup internet app type of thing. So it was founded by three NASA scientists who were working at Ames. And they were interested in using satellites and taking pictures of the earth to help people do things like manage natural disasters,
00:01:16
Speaker
or improve agriculture in poor areas of the world, and basically find ways to use data to help people who didn't have the resources to use the traditional data that comes from either the government, like NASA, USGS, or from private industries like Digital Globe. So they wanted to find something that was in between those two spaces and make something more usable. And they basically figured out that to do that, you actually had to
00:01:44
Speaker
find a way to make money on the data as well as use it to help people. So they started developing a fleet of satellites based on the CubeSat format, which is a 10 centimeter by 10 centimeter satellite. But you can actually stack several of those together. So ours are 10 centimeters by 10 centimeters by 30
Daily Imaging and Discoveries
00:02:05
Speaker
centimeters. So they're literally about the size of a bread box. They're pretty much all telescope with a camera on one end.
00:02:14
Speaker
And they take imagery that's three to five meters per pixel. So you can just barely resolve cars, but you can see individual houses. And so what you can do with these small low-cast satellites is you don't necessarily get super high-quality data, but you can get the entire Earth every day by launching between 100 and 200.
00:02:40
Speaker
And so you can see things that the traditional model, which relies on specifically tasking a satellite, so taking a picture of a specific place at a specific time, if you're not doing that, you're just getting the whole world, you can see things that you miss otherwise.
Algorithm Development and Innovations
00:02:55
Speaker
And you can see things that you didn't know you wanted to be looking for. So in effect, you're actually building a time machine where you can go back and look at, you know, before a disaster, you can say, hey, was there a landslide that blocked part of a river upstream of where this flood happened or something like that? And you also end up with data that can be sold less expensively per unit area. So it's basically,
00:03:23
Speaker
you know, helps fulfill that obligation or that desire to build a system and acquire data that actually helps people in addition to making money. So it's this very ambitious project.
00:03:36
Speaker
We've gone through several cycles of satellites. I think we're currently launching what's considered to be build 12 and designing build 13. And so this was over a period of about three and a half years since the first one was designed and built. And we've been launching for about two years at this point. So we're still in more of a research and development phase than an operational phase and have a few dozen satellites operating in orbit right now.
Potential of Continuous Imagery
00:04:06
Speaker
But by this time next year should be at or close to the goal of getting the whole world every day.
00:04:15
Speaker
So before we talk about your specific role, I wonder, are there folks there, or maybe you're doing this, that are sort of looking for the outlier sort of thing? Because it was interesting how you said that we're taking pictures of everything. So we sort of have, we can look at everything. So that means there may be things about the world in remote places that we haven't seen before. Are there folks there sort of looking for those
00:04:44
Speaker
outlier type formations or agriculture that, you know, you know, that maybe are not sort of shown up in sort of traditional satellite images. Sure. So right now it's a little informal. Like literally we have, you know, hip chat, which is an instant messaging client. And there's a channel that people are just like, Hey, I found this really cool thing. And, you know, it's like, Oh, are we gonna publish on the gallery? Are we gonna
00:05:11
Speaker
you know, look at it a little bit more in depth and we'll have to like evaluating the quality of the satellites and the imagery through here. But at the same time, we're developing some algorithms to basically do automated change detection. So do you see a new set of fields being developed in the Saudi Arabian desert or do you see, you know, the latest image we put up is logging in the Pacific Northwest. Right. And so you can do automated change detection on that.
00:05:39
Speaker
as well as building an API so that you can do automated querying of the database so that other people can develop techniques to look at our dataset and
00:05:53
Speaker
basically discover things that we haven't thought of yet.
Mapping and Data Visualization
00:05:55
Speaker
And so are you guys, are you really sort of visualizing those data as you sort of build these algorithms and go through, are there data visualization techniques you guys are using to sort of visualize the actual data aside from like the image itself? Does that make sense? Oh yeah, so right now we're doing it pretty simply because it's all for internal consumption.
00:06:20
Speaker
Yeah, the eventual goal is to make very polished, very well designed maps that are information rather than data, if that makes sense. So basically, you can see the change, you can see what type of change it is without having to pour through the data on your own or even like, you know, compare two images side by side. Just you look at one map that says, okay, we've got, you know, the health of these crops has improved 50% over last year or
00:06:48
Speaker
you know, that this glacier has moved or, you know, whatever, whatever we're looking at. But you definitely want to do things where, you know, like any type of data visualization, you're focusing on the information that you want to transmit. Right. And like distill it down into the important bits instead of throwing everything out there. So that that's not what I'm working on now. It's something I hope to be working on.
00:07:11
Speaker
when I get the basic imaging part nailed down. Right.
Color Use in Visualization
00:07:16
Speaker
So for listeners who have heard of you, they probably know you best from your, I will call it a legendary series on color that you wrote for visually. Thank you. And I suspect you're doing a lot of work on color at Planet Labs. Do you want to talk a little bit about what that role entails? Sure. So right now, actually, what I'm mostly focused on is
00:07:41
Speaker
working on the color of the satellite imagery. So when a satellite takes an image of the surface of the Earth, it's above the atmosphere. And so you get all of the effects of the light coming from the sun, traveling through the atmosphere, hitting the surface, and then traveling back through the atmosphere to the sensor. So the image that an astronaut or a satellite sees is not quite the same thing as what we see from ground level.
00:08:07
Speaker
So what I'm trying to do, starting on an individual basis where you look at a single scene and then try to match what we would expect it to see, how we expect things to look, and then sort of grow that out so we can do it on a systematic level and do it globally every day. Yeah. And piece all those images together so that they sort of look
00:08:31
Speaker
Right. So yeah, it's challenging. It's work that I had done previously with colleagues, including Rado Stokely at NASA with the Blue Marble, where you basically take an algorithmic approach. And by having access to a lot of data, you can do things to homogenize it and throw out bad data.
00:08:57
Speaker
And there we were doing it at 500 meters per pixel. Now we're going to be doing it at 5 meters per pixel. So it's this huge increase in data volume and also, in some ways, variability because you're looking at much smaller scale. So it could be considered a much bigger challenge. But we can use a lot of the techniques that we used earlier to try to apply that to the data set. Right. Interesting. Interesting. Well, that sounds great.
00:09:27
Speaker
I want to switch gears a little bit and talk about color more generally for folks who are out there creating visualizations or creating products.
Common Visualization Mistakes
00:09:36
Speaker
And to ask you, based on the work that you've done on color, what do you view as the biggest mistakes people make when it comes to using color in their visualization work?
00:09:51
Speaker
I think it's still the rainbow palette. If I had money, I would have put down. Rob was going to say the rainbow palette. Because it's the obvious answer. So Trenish and Rugowitz published a paper, I think even in the mid-90s, about what scientists and engineers need to understand about color, where very commonly the rainbow palette is used. And that's basically just a straight
00:10:20
Speaker
red, green, and blue palette that's basically presenting colors in a way that computers represent colors. So just, you know, red is 255, 0, 0, green is 0, 255, 0, and blue is 0, 0, 255. And just basically ramping between those to go from red, orange, yellow, green, blue. And you get sort of this very striking, very bright, very rich and saturated color palette that
00:10:49
Speaker
when you're using it to represent data becomes misleading because it's not showing colors in the way that our eyes and brains interpret color.
00:10:59
Speaker
So you end up with bands in the color bar, so you see areas where there's a lot, it seems like there's a lot of contrast and a lot of change, and that's just due to the palette. And there's other areas where all those variations might be smoothed out and obscured. Again, not because there's anything, any lack of detail in the data you're trying to show, but merely because in, say, the green region, you get this huge band of sort of uniform color.
00:11:27
Speaker
both because it's a default in many types of software, and because people are so used to seeing data presented this way, there's a lot of resistance to changing. I mean, it takes a little bit more work, although I was very pleased to see MATLAB actually developed a perceptual color palette and switched so that new versions of MATLAB use, I think it's called Perula, and published a very cool white paper about why they made the change and perceptual issues.
00:11:56
Speaker
which a lot of people hadn't been giving enough consideration to earlier. So I would say that it's still a problem. But it's one that there's slow improvement. And I think there seems to be a growing design consciousness. And so a lot of younger scientists are definitely taking this type of thing more seriously.
Choosing the Right Color Palettes
00:12:22
Speaker
Yeah, I mean, I think it's
00:12:25
Speaker
partly you have this for researchers and academics, you know, who are using SPSS or MATLAB or Stater or Excel, they sort of just, they open the tool and whatever the color palettes are there, they're just going to use those defaults, trusting that the software has done the right job. And I, in my experience, a lot of people feel
00:12:48
Speaker
not comfortable going out and picking a color palette or creating another color palette. Not because of the technical challenges of putting it into the computer, but how do I know that the color palette I've chosen is a good palette? Do you have thoughts on what researchers or scholars or people who are just not familiar enough with how to pick a good color palette, what approach they should take to that?
00:13:16
Speaker
Yeah, I think the most straightforward way is to just go to Cynthia Brewer's site called Color Brewer. And she basically built a tool that will, it shows a little demo map. It has a bunch of pre-selected palettes. And you can just go click on the palette. It'll give you the RGB values for the color ramp. She talks about the theory of why she's doing it, the different types of imagery
00:13:42
Speaker
or not the different types of imagery, the different types of data and how you use different palettes that are optimized to those different types of data sets and just walks you through it. You could also go to the subtleties of color series of blog posts. One of them is a list of tools where I explain what each one's good for, a little bit about how to use them. There's another one that's come up since I wrote that called HCL Wizard, which is
00:14:11
Speaker
a more refined version of some of those earlier tools and I've only used it a few times but it may be the best of what's out there for doing a custom palette. So you can pick your start color, you can pick your end color and have a little bit of control over how the gradation occurs between them and is fairly powerful and also will output the
00:14:35
Speaker
both for like R and Python, but also, you know, just in ASCII, you know, the red, green, and blue values, or even in Hacks. And allows you to set the number of steps very flexibly, so up to like 40 or something. So you can get finer gradations than you can with Color Brewer.
00:14:54
Speaker
So, I really like that one. Yeah, I mean, I recommend lots of other, you know, there's lots of tools out there obviously to help people pick, you know, their own color palettes. I mean, one thing I tell people is if you like the advertisement in your magazine or on that website, you know, just use that color palette and you can, you know, there are lots of tools you can use to actually go in and take, you like that combination of blue and
00:15:17
Speaker
gold and black or whatever it is, just go in and take it and use some of these tools out there to do the checks to make sure that they're consistent and they have the colorblind consistency and all those other sorts of things. But I feel like people, especially folks who are not confident in their design skills, they get a little uncomfortable going in and picking a color palette because they're not designers. Yeah, I can definitely sense that.
Concluding Thoughts on Visualization and Future Developments
00:15:46
Speaker
Yeah, so another problem with the rainbow palette, which I really didn't talk about much, is that because it uses so many colors and they're so saturated, it sort of like uses up the whole spectrum. It literally uses up the whole spectrum. And that doesn't leave any room for either overlaying a different data set or highlighting different parts of the data or anything like that. So if you do something like you suggested, where you take pieces from an existing palette or
00:16:11
Speaker
the hues that are in the other elements on a page or a website. And then use some of these tools to just ramp in between them so you can make sure that basically the trick for a good palette is that every step is equivalent. So if you're going from one to two or 100 to 101, that change in value is still the same. So you're not exaggerating any steps, you're not minimizing any steps.
00:16:38
Speaker
And so as long as you do that, there's actually a fair amount of flexibility in what the start and end colors are. So if you match or otherwise make a coherent color scheme, you end up with something that is both perceptually accurate and pleasing on a larger sense. And so that's a really good way to look at things is just, OK, what are you already working with or what do you like already? And then apply that.
00:17:04
Speaker
Very good. Well, I'm always hopeful people will stay away from the rainbow color palette and be careful of their reds and greens for the folks with color vision deficiencies, but we'll see. It's an evolution. Well, thanks so much for coming on the show. I appreciate it. It sounds like you guys are doing some exciting work there at Planet Labs, and I look forward to seeing what you guys come up with over the next year or so. All right. You're welcome, John. All right. This has been the Policy Viz Podcast. Thank you for listening.