 Hello everybody. Thank you so much for having me. It's been a great pleasure to be here. Absolutely great pleasure. I need to see if I can get my eyes wonderful. I can get my cursor over here if I want to. This is really sophisticated. So thank you so much again. So my topic is our planet scene from space and I am actually a PhD candidate at the University of Alaska Fairbanks, and I moved to Fairbanks in 2011. But in a previous life, I worked for a software company with marketing software, and then something attracted me to Alaska or someone attracted me to Alaska and I stuck there. Really people either stick there or don't and I really fell in love with the North. There's so much going on there and people don't know about it. I monitor wildfires with satellite imagery. And there are a few people who give me money so that I can do that. So I should give a shout out to NASA and several organizations. There was not very much seen from space in the in the title slide earlier. So I put in a space photo. We see a lot of space, pictures from space. This is from the International Space Station. It shows the Aurora Borealis, the green northern light that we had a lot of in where I live in the winter. And the premise of my talk is a really simple question. The question is there is so much web mapping there. There is no one here in the audience who will bog at me saying projection or something like that, even if you work in a different field. But hardly anybody uses raster imagery, satellite imagery, even though there's a lot of it there. We send students all the time out to get like GIS vector data from, for example, a local government and then put it on a map as an exercise. This is very common to do, but we very rarely give them some satellite imagery and tell them put it on a map and do something with it. For those who are unfamiliar, the differences are quite quite stark. So for vector data, you have files that could be proprietary like the SV shape file or GeoJSON, which is open, or it could be a simple flat CSV file. It's basically a database table where each record has a geometry associated with it. Could be a point, a polygon, a line, and some variations on it. First, a raster image, however, is just an image, like it comes out of your digital camera. So it is an array where each cell has a value. And I've been wondering why is everybody going for vector and no one works, not no one, but very few work with raster. I've been wondering is there something, has it something to do with vector data being sort of an abstraction and then you do something concrete, you make a concrete representation of it and for raster imagery, it's kind of the other way around, right? You already start with a representation and then you analyze it and extract something to omit. I'm not sure. Maybe this is just a speculation. This is, however, a data visualization conference, not an image analysis conference. You see it's kind of one in the one direction, the other in the other direction, but maybe who knows. But both are just data. We can deal with data, right? I mean, if you're a developer or can program, I'm self-taught. You know, no one gave me a stamp. I can deal with data. You know, I can deal with data. Vector tables, there is geolocation data associated with it and metadata. Metadata is extremely important because it tells you what's in your data. Raster data is 2D arrays stacked. Sometimes you get really 5D things that have time and height and all sorts of dimensions associated with it with geolocation data and metadata. So we can deal with this, but should we? Question is, should we care that there are so, like, oodles of free satellite data available? Well, obviously I wouldn't be here if I didn't think we should care, but I think we should kind of really give it some thought. And, for example, here, let me just start with something. You will recognize this as a satellite image of some sort. I actually downloaded it from scratch. I got it from the archive and made this with it just to show you. And what is this? Oh, let's just see if that actually works from this speaker view. So you can see it's something, huh? Yes, it does. It does not work. So I have to change this over. I'm sorry. I apologize for this. All right, do it like this. Ha! This will work. If I find this, there's a marker here and it's fairing. This is where I live. So I can zoom into this. If I find it, I can zoom into it like this. Right. So I can zoom into it. I can go around in it. I can see there is a river. There is something weird going on over here and this is that there was a fire. This image was from 2013 July and we had a huge fire here. I don't live in Fairbanks. I live here. So a 30 miles outside Fairbanks. We had to evacuate. So I might actually want to know these things and know what is out there at imagery that tells us about them. If I go for the south, something interesting happens, I will see glaciers. There are big glaciers here. You see that these glaciers, once they went up to all the way here, now they're here. And I can do even more. I can go over here and I can switch this imagery over and suddenly this is a bit dark. We saw this earlier. I'm sorry. But this glacier actually, this is an image from 1987. I also got this one. Put it all together and the glacier is a little bit further ahead in 1987 than it is now. These glaciers retreat really fast. So this sort of thing is not unfamiliar to people who live where I live and we want to work with this. And you might want to too. You know, you, where's my mouse? You might have a mine outside your town that might have a problem somewhere. For example, this mine in 2014 is in Canada. On the left, it was July 29th and there is a tailing pond. You see it there in the south. And the right August 5th, the tailing pond is gone because it drained into this system here through a little creek that then became a huge mud bath. So you might actually want to have images of this and follow through what's going on in your environment. And some people have now started to do this professionally. One of the motivations I had for giving a talk about is what seeing this in journalistic investigation. It's a piece by ProPublica that was widely published over a year ago about the disappearing coastline of Louisiana losing ground. And I heard an interview on NPR with one of the developers who works with this team and he basically said he had never learned how to deal with satellite data. They went and got this Landsat data and made this visualization. They're wonderful. Google it, find it. And that tells me that we, as a science educator and communicator, of course we use science imagery, but not just us. There might be citizen scientists around. There might be people who want to monitor their urban or rural environment, their town, their city, their state, industrial development, agricultural development. Does my city get more parks? Will they disappear? What's actually going on? Is it growing? Is it shrinking? And of course some of us help planning for disasters. I happen to work with the fire service and they also need to understand this better. So this brings me to my third question. What data and tool kits do we actually have to do this? And if you think of this image I had from the Louisiana investigation in the bottom right corner, there was this little thing there, this little map telling you where New Orleans is. And it says source NASA USGS Landsat. So that tells you where you could go. But we should know a little bit more about satellite imagery. What satellites are we talking about? We're talking about something called polar orbiting satellites. There are others. We heard earlier from someone who was tracking satellite overpasses and these were communication satellites. They have very different orbits. You also have certainly heard of geostationary satellites. Geostationary satellites just look at the same part of the earth all the time and they are very far away from the earth because they orbit with the earth. They are about 10% all the way to the moon, which is really far. Whereas the polar orbiting satellites are just about 800, 700 kilometers above us. The International Space Station is even lower. So they actually get a pretty good image of the earth. And as you can see, well, this one here has orbited itself out. Let's start it again. Oh, well, I don't know where the orbit went. Anyhow, as you can see still here is that it will acquire the image in a band that goes around and at the poles, those bands overlap more. Which is why at the poles you put down or close to the poles in the high latitudes, you put down receiving stations that get the data by radio link. At my institutions in Alaska in Fairbanks, we actually have two that are operated by the university. One belongs to NASA or I mean we operated for NASA. And the other one just uses their software, which actually is also open source. If you have a thing like that, you can also download it. You don't need any permission for it. It can be a smaller one. But yes, institutions can actually do that. You know, and some do it commercially. We also have downloading stations by NOAA. That's the National Atmospheric and Oceanographic Organization that does weather service. So they have satellites and run it there. And then we also sometimes see a dish like this in the woods. And if we dig a little deeper, it turns out it's a private company that runs satellites. So as I said, the easiest satellite data to use by far actually is Landsat. The nice thing is it's all free. Spatial resolution is 30 meters. Like each pixel is pretty much a square 30 by 30 meters, which is good enough for many things. Good enough for what I said earlier, the images I showed were Landsat. Not good enough for monitoring your house. But anything a little bit bigger, yeah. It has a data archive. It goes back to the 70s, which is great. And the data is published in a format called the GeoTIFF, which is basically a TIFF image with geo information. Sometimes there is a bit more to it, but we'll come to that. If you want satellite data, you need to work with data portals. And this is, I think, one of the biggest obstacles because data portals are written by scientists, for scientists, and that's not always a good thing. But the people who run these data portals have become a lot more aware of questions of usability. There are now organizations who talk about data usability, discoverability, metadata standardization, and things like that. I think this is a great development. Some of the easier-to-use portals are actually not hard to use. This one has a lot of public data, not just from Landsat but from other satellites, including also non-satellite data like aerial imagery, historical aerial imagery from coastal surveys that they did in the 50s, a lot of really interesting stuff. And the use is really get an account, simple registration. You don't have to belong to any particular, you just have to have an email address. And then you do a selection by location, you have a relatively easy-to-use, well, that's a Google map, clearly, right? And then, well, just leave open everything you don't understand, and this one actually still works. Others don't, how to say. But this one is really good. And when you go there, you can make your selection and hear the images for some reason they have disappeared. But when you want to download it, you get some offers. And you see there are something called Landsat look images and they are about a few megabytes. So if you just want to illustrate something, they're good. If you want to do geospatial mapping, you need something slightly better. Well, the Landsat look with a geographic reference comes with a little file that is called a world file, which is also already better than nothing. Many applications do work with that. But if you want the full thing, you get the level one product. And I recommend do it. For the older data, the older Landsat had a lower pixel depth, so it's about 150 megabytes for one scene. The newest one's about 900 megabytes, which unzips to 2 gigabytes. So we'll go back here. So you have to be able to do with this amount of data, but it's still quite manageable. Alternatives, there is a similar satellite. I will go deeper about what they are from the European Space Agency, Sentinel missions. And the Europeans, until recently, had, like the Japanese and others, have more of the philosophy that, yes, the data should be available, but for registered users in a well-recognized institution. But I think there was really a good influence of the United States, who has always this idea that public produced data should be public. And I think this is one of the really good principles here. They have really, that has made some headway. And ISA data, Sentinel data, is now also available, which is great. It's in a format called JPEG 2000, which is a bit like a geotip, except that it's based on a JPEG. And I have less experience with it, but it's kind of, it works workable. And then if you want to map the whole world, if you wanted to climate mapping, you need something like modus of years, just search for them. But you may have to deal with slightly more arduous data formats. But the tools we have do work with them. So I'm really happy to say that if you try to sign up for a Sentinel account, it says Sentinel data access is free and open to all, which I think is really wonderful. It used to be not like this. And Sentinel data is going to be 10, they just started with the newest one. They're still going to be 10 to 20 meters resolution, which is extremely nice. There are a few more, one called Spot is a French one. These are not free, but for example MyState, the state of Alaska, has bought this data that they have for MyState. So that might be the case for your state or for your country too. So check, just check and look. A good source is also NASA worldview, where you can, well, it's a very rich interface where you can overlay all sorts of layers and download directly from there. But as you can see, the download names are really complicated and the format is called HDF. So there's something learning to be done here. So what is a satellite image? That image is actually, as I said, not very different from any kind of RGB image that you have, except that it may have a lot of channels. The current land set has 11 channels. The older one had seven. Modus has 36. There are even some satellites that have satellite sensors that have 120 channels. You do spectroscopy with those, but you can use some if you want there too. Some have just four. What you can then do is you select some and make a false color RGB image from them. And depending on what you select, you will get different results. So here's what you get from land set. Current scene, you just get it as a zip file, unzip it, and you give tiff files. Wonderful. Band 1, band 10, band 11, well, okay. Of course, it sorts it in this order. And down to band 9. And then there is a quality band. You can forget about that. And the MTL file has the metadata. A library I will talk about in a moment will actually help you read that metadata, but it's a completely human readable file. It's not even XML. It's really quite nice. Well, it's not nice to parse, but it's okay to read. So then, depending on which of those channels you combine here, I've got a fire I worked on. I worked on wildfires. This was last year. We had a lot of them. We're going to be having a lot of them this year. This was in a town called Willow, which is there under the smoke. There is a beautiful river to the left. This is Sydney River. You can see this is a color combination that is most close to our eyes. So this is a natural color image. The river is a glacial river. It has a lot of silt, which is why it looks so gray. And you can't see the fire because there's smoke there. There is a road, which we don't have many in Alaska. But if you take the bands that are in the near and shortwave infrared, which is, if you know micrometers, it's about one to two micrometers, and combine those with visible bands. You don't know, just think near infrared is good for looking through smoke. You get actually an image of where the fire front is at that moment, and most of the smoke has disappeared because it has a capacity of looking through the smoke. This landsat also has a thermal band, which works at even a higher wavelength, which is real thermal radiation emitted by the fire. And you can plot this too. In this case, I just open it in a free image processing software and added a color ramp. We could do it with a command line too. That's what I did here. So, the question is now, what can we do with this? How do we get that far anyway? And what is our infrastructure here? So, I'm a Python person. You have similar things in R, and many of those underlying libraries, like Proge4 and Geos and GDAL, they are at C libraries or C++ libraries that have bindings in R. They have bindings in Python, they have bindings in other languages. NumPy is, of course, Python specific. You have also special libraries there to either plot or to read these IGDF files if you need them. You don't need those for landsat. And PyGARs is my library. I will tweet a link to my... It makes just the opening of the files quite a bit easier. But there are others. Sean Gillies is someone who has contributed enormously in making this easier with his tools Shapely, Piona and either Rasterio or Rasterio. I never know. And these are built on the underlying libraries. They need them installed, but you don't have to deal with the pretty awkward interface and API, have a much nicer API. So, to come together, in order to get just these images I just showed you, what you need is these GDAL, it's a big library, it comes with command line utilities. And these command line utilities allow you to just take in those band files I showed you earlier, combine them into RGBs, cut them to the size you want. You will need the project approach for library to figure out where your corner should be because you may know the latitude and longitude, but landsat does not come in a latitude longitude. It comes in a projection. And if you have several landsat scenes and you want to kind of be in between and want all of them together, you can use GDAL merge.py to very easily mosaic stuff together and then combine all these. And frankly, shell scripting is your best guess here. I do think it's a good skill, you don't need much more than just kind of associating commands, but I have a few tools, just a basic knowledge of shell scripting is a good idea here. Now you have these things, what do you want to do with them? You've downloaded your landsat data and now have RGB composites in all sorts of combinations that you want. You can actually kind of do all of them. Sometimes you see something really interesting by using a combination that no one told you to do. There is a lot of pretty artistic stuff that comes out of this. Vegetation has different color from sand. Forest has different colors from pasture or grass. You will see a lot of things there. But now we want to put stuff towards an audience. We want to make it usable. I want to make something with it. I'm following kind of two approaches here. One is more like you want to put stuff on the web. Well, you have HTML, you have CSS, you have more than HTML5 and all these things that you probably know a lot better than I do. I know you do. But what you can do is, for example, I have pictures of my file that I had earlier. Now I have a before, a during and an after picture. I did what I did before here. I took this. This is the during picture and I have a before and an after. And I combined them using nothing more fancy than... Well, I transformed them into PNGs and then I combined them into a sprite using image magic. And then with HTML5, I can do a keyframe animation just to show the before, during and after, just to make change visible. So change detection is a really great thing. And one way of... It's very simply doing this is this. There are more complex ones. Obviously, you have a very large number of libraries that use before, after sliders. And here I do have... Ah, I can get it. Try to get it. So this was a situation where we had a flood. The Yukon River each year freezes, of course. It's Alaska. It's cold. And then we have a moment in time that's called ice out or snow melt. And the ice out can lead to horrible ice jams, which then can lead to horrible flooding. And last two years ago was that. The town of Galina, which you see in the middle, shown here in 2012 in a completely normal situation in May 2012, was victim of a flood because of an ice jam. So I can slide over and can make visible how much flooding there was a year later. It was 2013 then. Yes, 2013 happened. So as you can see, this... Here's a little scale here. I did not process this. These were processed by the Earth Observatory. But this is... What is it, 20 km? So this is 10 km. This is like six miles of just water. It's a small town, but it was pretty much wiped out. People had to fly out. And one thing you don't see here is roads. There is no roads going there. So you travel on the river or you travel by train here. So this is an example of what you want to use. Or, of course, the example I had earlier. I need to get back. My leaflet map, as you see, this is nothing more than a regular map made with leaflet, the leaflet library. I could use open layers just as well. So I took my things here, and Gidal has even a little helper that will make tiles. It's one command line, one line, Gdel to tiles. I get tiles and I just have to figure out how to get them into leaflet. And then I can use all the tools I have from leaflet, including where is my thing here. The roads visible. So as you can see, you can just make a normal web map like this. But the before and the after slide is I had earlier. So this is one thing I would like to put towards you to think doing with satellite maps. But not everybody wants to do all that much. Well, this is the recapitulation. So satellite mapping with HTML5, CSS and JavaScript, you can use Gdel to tiles to make tiles. You can use leaflet or open layer, or whatever else. There are others, of course. You can use image magic if you just want to manipulate the image. And you can have your favorite libraries, favorite plugins to make sliders. You can use canvases, whatever. You just have the images there. And as long as you have preserved the projection and the geo information before you ended up going towards PNG images, you have perfectly aligned images. And very well aligned images, really. And even over, the processing of these agencies is so good that even if you compare some things from the 1970s to now, they overlap. You saw it here. 1987 and 2015, 2013, 25 years in between, and I had a near-perfect overlap. But this was really easy and not quite gratifying. So, things that I, however, do more often is to analyze the data a little bit more, to get them inside some software, inside a script, manipulate them, and then use libraries in the script to plot them on maps, or wherever I want them. And maybe to animate them, make movies out of them. So, the toolkit for this one, I call it toolkit 2B, is a little bit less webby. And you can do this in Python, you can do it in R or Matlab or whatever you like. You use NumPy, because then you have Rusters. You use Matplip for plotting, Basemaps, again, GDAL. And you can use FFMPag to show you, to finish this up, an example about what I did, or some examples of what I do, and one of the examples has a little bit of code. So, an example of what I do without code right now is something called the burn index, or actually the differential normalized burn index. But it means, basically, you take a before image and an after image and you look what certain band combinations are typical for freshly burned soil. So, you need maybe an intern who knows a little bit about that, but it's really, really not much. We teach that in introductory remote sensing. The burn index, before and after, if I compare it, in Brown for the fire I had earlier, the most intensely, most severely burned places. Then I also have low resolution data that I personally transformed into a GIS polygon dataset, but you could get this polygon dataset from someone else, or a point dataset that tells you fire detections, where was fire detected by satellites that fly over more often, but with a lower resolution. So, I can overlay those with Python and plot them, and then the more red things are more detections, and then I can compare, does it agree with where there's more burning, and I can take it from here. This is just one example. It might be something entirely different, maybe you have demographic data that you want to overlay over a map and see, for example, how does a demographic make up of my city coincide with green spaces. You detect the green spaces from a Landsat image, and you have demographic data that is in polygon form, for example. And now for my other, my more worked example, there is something I really have a very fondness for. It's the National Snow and Ice Data Center. They are doing, I think, the best job that I have ever seen for distributing data. They have good documentation that is actually readable, that tells you in words that you can understand what does a pixel mean and so on and so forth. So, they deal with many datasets. One is the one that I have the page here, sea ice concentration from, well, from some satellite or other, which happens to be someone that looks at microwave radiation coming up from the Earth. And microwave radiation shows you a lot about cold things, because microwave radiation is actually relatively cold compared to light or the near infrared. So, we have ice in the North and it's shrinking. So, they have the best datasets about sea ice concentration around the North Pole. And if I take one of these images I don't have this. So, the concentration goes up to 100%. I don't have anything right at the pole, which is typical for a polar orbiting satellite because it actually orbits slightly next to the pole, always around the pole. And we had sea ice all the way halfway around Greenland, along the northern Canadian coast, along the Bering Strait, down to the Aleutians and Alaska. This is some data that I forget. So, if I want to, however, want to combine a lot of a long time series of sea ice concentration to a visualization, what do I need to do? So, I need to open this thing in Python. I need to import NumPy so that I can manipulate rasters. I take it from my library of PyGars just to show how easy it is. You can do the same with GDAL or RasterIO. It takes maybe a few more lines. I just have to say Raster.geotiff of my filename and I have a data property that has the data in it. And in this case, reading the documentation, I know that I have to divide it by 2.5 because 250 data value is 100%. I have to normalize it to 2%. And then I can do a little bit about where it's my... I have to say, where should my center be? Where would I actually... I don't want to see the whole North Pole. I have a particular area that I'm interested in. In this case, it's north of the Aleutian Islands. I will show you that. So, I give it latitude and longitude of that. I get the coordinate transformation, which is a function that I can use to transform coordinates from latitude and longitude in the coordinate system of my data. And I can do a few tests about which IJ coordinate of my Raster corresponds to which XY coordinate of my projected data. And then there we go. I can use the base map toolkit from Matplotlib, get a figure. You may not be familiar with this syntax, but you've probably seen similar things, right? Even if you never use this one, you define a map which has some variables in there. You give it a latitude and longitude of your center. And then you say, okay, there's a scatterplot in there. We'll see what it's for. Give me coastlines. Give me rivers. And then you put it all together. You get a lot of data for each scene. And then I use the FFM pack to make a movie out of it. And I get this. So here is the coast of Alaska, the Aleutians. I always happen to be interested in this yellow and the red spot, which are the Pribilove Islands. Someone I know was doing research on foxes that are there, and he wanted to know how much sea ice was there during these years. So I made a video for him. See, the sea ice doesn't really usually reach those Pribilove Islands, George and St. Paul, but you can see how it develops, how it retreats, etc. So this was not very much lines of code to get to a really nice thing. And with this observation I'm drawing to the end of my talk. I hope I've been giving you something to think about and some inspiration maybe to take the problems that you have and that people you work with have and think about not just is there a government agency or a business or a citizen signs outfit or a sensor out there that would give me some point data but maybe is there a satellite that is up there in the air in the space looking down on us that would give me some images that I can use to do something with it and then combine it and mash it up. I think that term has become a little bit out of fashion, but it just came up to me. It's kind of a mash-up of things that seem to be people specialized in one or the other, right? So I thank you very much for your attention I believe we have some time for questions, right? Any questions? Well the US has the tendency to open data eventually so a lot of interesting data that used to be closed, the data that is closed from the US is all, typically that's what's classified and data does get declassified. So for the stuff that is classified currently no one actually knows what it is and people who work with it. So for the US I can't really answer. Canada has a lot of radar data and they monetize it so the radar sat data is most that would be really nice, a lot of that would be free because we don't actually have good radar satellites. I did not talk about radar at all it's harder to deal with because you need a little bit of knowledge beforehand to interpret a radar data set. And the Europeans are on the way I think there are a lot of sensors that would be really nice to have integrated data sets of all the sensors that are available and for example I say boulders and veers. Europe has similar satellites but this data is only available if you're a member of a European research organization or I think that is kind of the condition it's not very obvious but most people aren't, right? So there's no short answer for you but I would really like to see the Canadian for the radar data and the Japanese and Europeans for the data that is comparable mostly to the MODIS data because it would be nice to have a longer time series there and go back further and fill the gaps start from behind and then here the dot density do you have this in your roster? Actually they do this quite a bit by treatment and my resolution as grid cells it's sort of pretty straight I would just go ahead and play with these scatter capabilities of the libraries in Python it's I think not that hard you might have to calculate the densities yourself, yeah by hand. But people have sometimes similar problems when they have LiDAR data because LiDAR gives you point clouds, it basically shoots down a very thin beam and reflects it so you have to basically have very thick scatter plots that you need to deal with and think about that. Well I working on one actually I'm working on one but there is a lot out there actually I know some that are more exciting than what I do that the applications go deep into for example coastal monitoring monitoring of the density of algae there because you can see it in the ocean color it's a whole field in itself so do you have experience with ocean color? I don't but the more bands you have the better because then you have a whole spectrum and it gets extremely rich and you can use you can throw so I could learn at it which I'm doing for kind of not exactly this application yes so it's a very very lively field and I think which will grow yeah I fight against you know not against the user interface at all but institutions tend to have site licenses for something that someone was convinced they should need and ArcGIS is of course a big you know big gorilla in that field. I do use ArcGIS I use grass I actually started out doing more with it I use it specifically I think it's most but by far the most useful thing it is to define polygons of areas of interest so you can prioritize something by hand so that you can then easily export your result and then use it further because you can just you have it there and you can click and it tells you exactly where you are and it will assemble those points into whatever kind of shape you want points or line or polygon and then you can use it further however you get very quickly to the point where you would like to process something and I tend to even if these all these tools are scriptable I tend to kind of think that having a command line script tends to be faster and more efficient in the end so I do use them but I use them for these specific things and I admit I use ArcGIS if I need a publication for the final map with like a nicely formatted legend and a beautiful little arrow at exactly the location I need but that's like the last, last, last step awesome thank you so much thank you