We're sorry but this page doesn't work properly without JavaScript enabled. Please enable it to continue.
Feedback

"Earth in Colour" with EarthDaily Analytics

00:00

Formal Metadata

Title
"Earth in Colour" with EarthDaily Analytics
Title of Series
Number of Parts
351
Author
License
CC Attribution 3.0 Unported:
You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor.
Identifiers
Publisher
Release Date
Language
Production Year2022

Content Metadata

Subject Area
Genre
Abstract
EarthDaily Analytics is building a powerful new constellation that will collect scientific-grade, 5 meter resolution imagery of the planet in a unique combination of 22 spectral bands using 3 different camera types, covering a broad spectral range from visible to thermal wavelengths. The mission will be launched in 2023 and will allow us to see the Earth’s global land mass each day in a wholly new way with more spectral bands, higher revisit, and at a higher resolution than ever before. It will allow us to monitor, detect changes, alert, and predict what is happening anywhere on the planet to help with some of world’s most pressing challenges in agriculture, Environmental, Social and Governance (ESG), and disaster prevention and recovery. This mission has been made possible by a near-perfect convergence of three major technology breakthroughs in the last 10 years: 1) lower cost satellite launch and manufacturing, 2) advancements in computer vision and machine learning to support automation of petabyte scale processing, and 3) cloud compute power and storage necessary to drive the processing and calibration of trillions of pixels each day. Together these three emerging technologies are key to driving next generation geospatial insights, but to bring them together requires a software solution capable of handing the complexity of raw satellite with automation driven by machine learning, and cloud-based Big Geo Data pipelines for cost-effective scale and latency. At EarthDaily Analytics, our software solution has been made possible by leveraging many open source software packages to form the backbone for our satellite processing, calibration and quality services called the EarthPipeline. Together with open source packages and custom machine learning and computer vision approaches, we are working on delivering true scientific satellite image products that can be applied directly to algorithms without the need for very costly (and dreaded) end user data normalization and correction procedures. This talk will focus on how EarthDaily Analytics uses open source packages and machine learning to create normalized scientific quality data, and will also provide some example applications of how the data can be used.
Keywords
Open sourcePredictabilityMeeting/Interview
Open sourceSoftwareSpacetimeService (economics)SatelliteProcess (computing)Machine learningIntegrated development environmentScale (map)PredictionInformation securityWater vaporMathematical optimizationForestElectronic program guideDependent and independent variablesNatural numberInsertion lossMathematicsEnterprise architecturePhysical systemSoftwarePresentation of a groupTime seriesGradientAnalytic setForestBuildingSign (mathematics)Generating set of a groupOpen sourceSatelliteOffice suiteProcess (computing)Multiplication signComputer animation
OrbitMassMusical ensembleSpectrum (functional analysis)SatelliteMusical ensemblePoint cloudSynchronizationVideo gameSpectrum (functional analysis)MetreMessage passingGraph coloringView (database)Operator (mathematics)Musical ensembleMedical imagingEndliche ModelltheorieSpecial unitary groupAreaChainInternet forumLink (knot theory)Maxima and minimaNatural numberRange (statistics)OrbitCloud computingComputerComputer animation
ConsistencyData streamSatelliteSpectrum (functional analysis)Physical systemInheritance (object-oriented programming)Category of beingSoftwarePoint cloudVirtual machineOpen sourceMeasurementUniform resource locatorPixelSatelliteEntire functionSoftwareVideo gameDiagramMedical imagingRadiometryAutomationProduct (business)Nichtlineares GleichungssystemMultiplication signOpen sourceTerm (mathematics)MassVolume (thermodynamics)Loop (music)OrbitGraph coloringRight angleComputer animation
Analytic setSpacetimeComputer hardwareMedical imagingData structureMereologySoftwareComputer animation
Virtual machineSocial classSoftwareScale (map)GradientSimilarity (geometry)Degree (graph theory)SatellitePolygonOpen sourceExploratory data analysisCloud computingService (economics)Data analysisSinguläres IntegralImage processingElement (mathematics)Stack (abstract data type)VideoconferencingInterpolationOrbitRaster graphicsSimulationNP-hardCorrelation and dependenceDigital image correlationProcess modelingGeometryCross-correlationComputer-generated imageryCore dumpOpen setRadiometryPixelSpacetimeRaw image formatGradientAutomationAuditory maskingMetreAnalytic setOpen sourceSocial classMereologyDifferent (Kate Ryan album)Type theoryCausalityPolygonPerspective (visual)Software industryOpen setCore dumpSpline (mathematics)OrbitReal numberPixelPropagatorBuildingDampingComputer fileVideoconferencingVector spaceComputer simulationLibrary (computing)CodeSimulationComputer programSatelliteModal logicSoftware developerRadiometryProjective planeElectronic mailing listPoint (geometry)ChainSoftwareCloud computingEquivalence relationScaling (geometry)Internet service providerPoint cloudPublic domainProcess (computing)Level (video gaming)PlotterFrame problemDistortion (mathematics)Matching (graph theory)Shape (magazine)Traffic reportingMedical imagingUniform resource locatorExecution unitMusical ensembleSigma-algebraDegree (graph theory)Patch (Unix)Computer animation
Artistic renderingMusical ensembleMathematicsVideoconferencingStack (abstract data type)Open sourceRaster graphicsComputer-generated imageryContent (media)LaptopOpen sourceBitInformationSoftwareMultiplication signNatural numberVolumenvisualisierungVideoconferencingContent (media)Computer animation
Open sourceBuildingSoftwareSatelliteTask (computing)Open setStandard deviationProcess modelingSimulationContent (media)OnlinecommunityMultiplicationMereologyOpen sourceStandard deviationPlanningOpen setLatent heatSoftwareExtension (kinesiology)WärmestrahlungOperator (mathematics)Computer animation
Computer animation
Transcript: English(auto-generated)
That's great. Hi, everyone. Thanks for attending this talk. It was a great talk by Dr. Nicholas Wachowski. A lot of the concepts he presented with prediction and big data are going to apply directly to what we're doing here at Earth Daily Analytics as well. So what I'm going to do is talk about a big mission we have up and coming. And it's all really been enabled by open source
geospatial data. So I'm going to tie in some of the lessons we learned and how we do it at Earth Daily. So I'll quickly go over what our mission is at Earth Daily Analytics and what we're trying to achieve. And then I'll show you how open source was a key enabler for all the software technologies that we're building in-house.
So I'm Chris Rampersad, VP of Engineering. For those of you who don't know us, we are a company called Earth Daily Analytics. You may know us from our previous incarnation as EarthCast. But we are a company that specializes in software and analytics. We deliver satellite processing solutions
for our own satellites but also for other satellites. And we generate agriculture analytics out of our office in Toulouse, France. And I'm from Vancouver, BC, where we do most of the satellite processing. So we have a belief at Earth Daily that if you have a system that could scientifically view
the world each day, it could give you a system that could deliver change anywhere on the planet in an enterprise-grade way. So this means reliable alerts for a change anywhere on land. And if you could do those changes and do those alerts, getting back to Nicholas's presentation, you can do that reliably.
You can go one step further and then start predicting what could happen next. So once you have that time series, you could imagine if you're looking at a forest and a road starts getting built in a place that it shouldn't be, that could be a telltale sign that illegal deforestation might be happening. Likewise, we're looking to monitor methane throughout the whole planet. So if a large methane plume starts appearing,
that will be a sign that there's something that should be addressed, and then you can send in the correct people to address that problem. So our solution is a constellation of 10 satellites. It's equispaced at 10.30 a.m. It's a sun synchronous orbit,
and these are moderate-sized satellites. They're about 200 kilograms. They're roughly a meter on edge, so I like to think of them as a large floating oven. They also have a 10-year design life, and so what that means is we've designed the propellant, the electronics, how we downlink all to be compatible with a long-living satellite.
So hopefully it'll last a little longer than 10 years, maybe 15, but at a minimum, 10. And this constellation, we like to compare it to David Attenborough's Life in Color. Just like David Attenborough goes out into nature, takes pictures of animals and uses different frequencies like ultraviolet, we're doing the same thing of the Earth with our satellites.
We're using a broad range of spectrums, 22 super spectral bands to be exact. We have 11 bands in the near-visible spectrum. We've aligned those to be very well aligned with central to even the spectral band passes are nearly identical. We also have shortwave infrared.
This is mainly for doing things like atmospheric correction, cloud detection, as well as methane detection. And then we also have thermal bands. This is really useful for wildfire monitoring as well as evapotranspiration modeling. And just to give you a quick view of how this looks, this is our satellite constellation. What you'll immediately notice
is that it's a very simple concept of operations. We just turn the camera on whenever you're imaging over land. We also acquire some maritime areas to see boats for supply chain and also potentially illegal activities. I should mention, I forgot to mention, we're launching in late 2023.
So when you look at the technologies to make this possible, it really comes down to four key capabilities. On the right, we have daily global coverage. It's about, I should say, of land. It's about 150 million square kilometers. We also have an orbit that is controlled accurately for 10 years. So we're always imaging
with an equator crossing time of 1030 a.m. And that's throughout the entire 10-year design life. Those two capabilities are fundamentally driven by the satellite itself. It's not really a software problem. But when you look at the diagram, on the left side of that diagram, those two bubbles, those relate to software and also our open source technology stack.
So the first is high image quality. How do you achieve accurate radiometry, which is a measure of the color or the scientific signals on the ground, and also location, the geolocation, the accuracy of every pixel in terms of latitude and longitude? Those both are fundamentally driven by how you process the data.
And then the low latency. So that is delivering pixels fast. This is a massive amount of data. It's about 20 terabytes of downlink data per day. When you decompress that and process it, you get to about 100 terabytes of data. So it's an amazing volume. And we need to process that fast for our customers. And that includes automation all the way from the tasking
down to the delivery of products. So there can't be humans in the loop. So shown here are our suppliers. This is our space segment. We have Airbus building our structure for the satellites. ABB is our camera supplier. They're building the scientific instruments
to image the Earth. Loft-Orbels integrating all the different various parts. The one thing you won't see up here is the Earth Daily Analytics logo. We're not a hardware company. We don't build satellites. What we do is we focus on the software. So for us, just like there's a big challenge on building satellites, there's a really heavy challenge
on the ground side as well. For ESA, they invest heavily into building a ground sigma technology, and that's what gives them their scientific grade quality. For us, we also have this challenge of delivering true scientific quality. So we're trying to match Sentinel-2 and Landsat, and we actually have a few requirements that are a little more tough. In agriculture, you need really accurate cloud masks
for doing agriculture analytics, so we gotta take that even further. We also have this enormous scale and latency. So again, that drives us to automation. Every part has to be fully automated so that we can handle that enormous scale and do it in a cloud-native way. And then these are smallsats. I mentioned they're about one meter on edge.
They're not the size of a billion dollar Sentinel-class mission. So they're smaller. That means you have lower cost star trackers, GPS, all the componentry is not quite as accurate. And so again, we have to do more processing on the ground, more calibration of the sensors to achieve that same scientific quality. And NASA and ESA have budgets upwards
of 100 million to build their ground segments. A new space company can't afford that. So we have to take a different approach. And I'm gonna walk you through the process of how we got here to be able to do this mission to deliver this type of data. So I'm gonna go back to the beginning, and I'm talking way back to the beginning.
When I was a child, and I think most of you probably have stories like this, my mother told me that she used to walk six miles in the snow in minus 30 degree weather. And this is Celsius. I'm from Winnipeg, Manitoba. And she was just a child, eight years old. This is basically three miles there and back. And as a kid, I couldn't even fathom what that meant. Like that's, as a child going out in the cold
for more than 100 meters, it would seem like a distance, unsurmountable. So I couldn't put this into perspective. I couldn't even fathom how this was done. But back in the older days, there was things you had to do to get to school. Full disclosure, that's not my mom walking. That is a picture of Winnipeg. The snow is real.
But that's not that old. So myself, I realized I actually have a story just like this that I can't fathom how it used to work. But when I first started my career in the space industry about 20 years ago, all the software we created was all from scratch. This is everything from the packages that handle file names,
your vector manipulation libraries. I worked on polygon intersection code in C++. And this is extremely difficult stuff. And full disclosure, that's not me 20 years ago. It's more like 15 years ago, but the poor programming posture is real. So today, no space company, no software company,
geospatial or otherwise, would consider starting from literally scratch without being enabled by the open source community. And here's a small smattering of some of the packages we use from, and I just got a few names from the hard developers. But I apologize if I missed a few geospatial projects that we're using, but there's a whole array.
And the usual suspects are there. We used QGIS, GDAL, Resterio. We have Matplotlib, PyTorch, and a bunch of others. Our complete stack is driven by open source. And using this, when we started, when I started back at Earthcast, which is now Earth daily analytics, nine years ago,
we started using open source from day one. We knew it would be impossible for us to build this automated pipeline with scientific quality using just in-house brain knowledge. Like the geometric patch alone is a huge amount of work, and now there's things like Shapely, which can replace that. And so we've developed this tool that allows us to process satellite imagery,
recalibrate it, QA it, all automatically leveraging open source software. And now we can handle this enormous scale of 20 terabytes of processed, of raw data per day. So I'm going to tie in now how we use open source for our Earth pipeline. So this is just a high level view,
and this isn't a complete list of the open source packages. But the point I want to make here is every single step along the way in our processing chain, it leverages open source and geospatial open source packages all the way from the downlink bytes that you get from the satellite. They're typically compressed with JPEG 2000. We use open JPEG.
When we do the orbit propagation, we can use SGP4. Every step of the way, we are leveraging open source, and this is basically like having thousands of developers helping us build our pipeline, and this is how we can do something like ESA. ESA has an incredible ground segment, and for us to be able to build something equivalent,
it's really leveraging the community. And the other lesson here is this works really well for niche activities. There's not that many satellite data providers out there, so it's relatively a niche domain to do raw, downlink satellite processing, and open source fits really well into that mold.
So I'm going to walk through a few of the things that we do. Again, every step of the way, we use open source, so this is just a few of the examples. When you have a satellite in orbit, one of the first things you have to do is simulate your data, and there's different ways you can do this. There's commercial packages that allow you to simulate your data, but we leverage open source, first and foremost, and so we emulate what EarthDaily will do.
We use GDAL, Rasterio, Matplotlib to render the video frames, and you can kind of see one of our simulations. What we do is we simulate the raw data and feed it to our processor, and that's how we validate that. We know how to handle the data, and when we simulate it, it's not just a core simulation. It's simulating all the weebles
and wobbles of the satellite, the optical distortion, how the star trackers aren't perfectly aligned. All those things get simulated, and then we use that within our pipeline. Another place where we use it, which is really key to us achieving high scientific quality, is in correlating satellite imagery, and so typically what new space companies do,
if you don't have really expensive star trackers, you take your image and correlate it to, say, a Sentinel-2 or a WorldView image because those satellites are beautiful. They're exquisite without any specialized ground control point marking, but often lining up your image to another image can be really difficult. Your eyes can do a fairly good job,
but once you get into the modalities like near-infrared, they don't match well to red, so we developed in-house some capabilities to automatically correlate these bands especially for satellite data, and so we call it MLCore. It basically handles radiometric agnostic features and looks at more of the geometric shapes.
We also use geospatial open source for doing automatic geometric and radiometric QA reports. Geometric QA is simply validating the location of the pixels that you can get sub-pixel accuracy, and then radiometric QA is validating the subtle scientific quantitative units
and making sure we're accurate, and we often use Sentinel-2 for validating our satellites. Again, open source stack, we use Matplotlib, a lot of SciPy, GDAL, Rasterio, and some OpenCV, and this one was a bit of a surprise.
This came up recently. We were trying to create some marketing information to support a marketing team, and they were having a hard time handling the large nature of geospatial data, and so they were trying to render 4K videos, and so it ended up the engineering team started creating some of the marketing content, so we used QGIS, GDAL, Jupyter Notebooks.
We actually have a Jupyter Notebook pipeline specifically for our content creation now, and you can see just some little clip. It's a visualizing imagery. This is all done with FFmpeg and some other tools, but going forward, when we release our TikTok videos, you'll know that everything behind the hood is open source software helping us,
and lastly, I just want to close with some final thoughts of our experience with using open source geospatial software. We've noticed that the community evolves very rapidly, and this is a great thing, but what it also means as a user of open source, you want to make sure you're constantly
updating all the packages, because as you bring in things, if you have multiple people working or multiple teams, could easily go stale, so it's good to both contribute if you can, but also, if you're not contributing, at the very least, always be taking the latest, because things are moving at such a rapid pace. I think I mentioned this before, but niche activities,
open source software works extremely well. Some of the content creation was a real surprise that we had to use open source to make it happen, but again, using geospatial data and marketing don't always go hand in hand. The other part is, open source is very vital for operational commercial companies' activities,
and it's really a win-win by using those, using open source as a commercial company and contributing back, so our plan in the future is to contribute to open source. We've done it very lightly in the past, but we want it to be more active in the stack community for our satellites. They have probably some specificities around thermal
and some of our sensor technology, which would fit well into some new stack extensions. And lastly, open standards and open software are not only changing how we develop, but it's really changing how customers work with the data. There's an interesting story that we just encountered. As we're selling our Earth Daily data, the last six customers asked for one thing,
all in common, and this is quite rare, but it was unanimous among the last six customers. They all were asking for stack specification, every single one of them, and so we were planning to use stack, but we were just so thrilled at how successful the open source community was of showing a great standard and getting adoption, not just from developers, but among the end user community.
And that summarizes my talk. I'd be happy to take any questions.