Deforestation Monitoring using Change Detection
This is a modal window.
The media could not be loaded, either because the server or network failed or because the format is not supported.
Formal Metadata
Title |
| |
Subtitle |
| |
Title of Series | ||
Number of Parts | 295 | |
Author | ||
Contributors | ||
License | CC Attribution 3.0 Germany: 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 | 10.5446/43363 (DOI) | |
Publisher | ||
Release Date | ||
Language |
Content Metadata
Subject Area | ||
Genre | ||
Abstract |
| |
Keywords |
FOSS4G Bucharest 2019220 / 295
15
20
28
32
37
38
39
40
41
42
43
44
46
48
52
54
57
69
72
75
83
85
87
88
101
103
105
106
108
111
114
119
122
123
126
129
130
131
132
137
139
140
141
142
143
144
147
148
149
155
157
159
163
166
170
171
179
189
191
192
193
194
195
196
197
202
207
212
213
214
215
216
231
235
251
252
263
287
00:00
Mathematical analysisProcess (computing)SatelliteMathematicsMathematical analysisProcess (computing)Level (video gaming)ImplementationMathematicsArithmetic meanAddress spaceMeeting/InterviewComputer animation
00:56
Traffic reportingComputing platformClient (computing)InformationData typeData managementAddress spaceInheritance (object-oriented programming)EmailService (economics)SoftwareProduct (business)Data management1 (number)Meeting/Interview
01:20
Common Language InfrastructureConvex hullInterior (topology)ForestArc (geometry)Programmer (hardware)InformationSatelliteRange (statistics)AreaProcess (computing)Point cloudHost Identity ProtocolPointer (computer programming)Open sourceOpen setType theoryScale (map)Mathematical analysisMenu (computing)Service (economics)Forest10 (number)Insertion lossObject (grammar)Projective planeAlgorithmSatelliteOpen setInterface (computing)Product (business)MultiplicationClient (computing)MathematicsMappingObservational studyComputing platformData managementBitLibrary (computing)MereologyRange (statistics)Computer programmingProcess (computing)Function (mathematics)InformationAreaOpen sourceMotion captureLevel (video gaming)Physical systemCASE <Informatik>Universe (mathematics)CuboidMusical ensembleMultiplication signForcing (mathematics)Near-ringSpacetimeReal-time operating systemState observerMathematical analysisDemo (music)Point cloudKeyboard shortcutType theoryComputer animationLecture/Conference
04:03
Medical imagingAreaMultiplication signPoint (geometry)SatelliteFrequencyPosition operator
04:33
Computer-generated imageryRun-time systemOpen setWindows RegistryGEDCOMFluxEstimationHydraulic jumpMenu (computing)Graphical user interfaceRange (statistics)FreewareBitMedical imagingSolid geometryStructural loadLevel (video gaming)Different (Kate Ryan album)Physical systemConcurrency (computer science)Internet service providerTime seriesMathematical analysisMathematicsLoginProcess (computing)ResultantGraph (mathematics)Visualization (computer graphics)BefehlsprozessorError messageValidity (statistics)Standard deviationMappingComputing platformSoftware developerPixelInternet forumNumerical analysisWindows RegistryThresholding (image processing)Java appletLibrary (computing)Function (mathematics)FrequencyRule of inferenceAreaPoint (geometry)Query languageRaster graphicsDoubling the cubeDebuggerOpen setForestSet theoryMultiplication signDecision theoryStability theoryCentralizer and normalizer1 (number)MultiplicationWrapper (data mining)PreprocessorSubsetUniform resource locatorCubeMetric systemLine (geometry)Figurate numberState of matterLogic gatePresentation of a groupAmsterdam Ordnance DatumQuantum stateDirected graphShape (magazine)Group actionView (database)Address spaceDiscrete element methodWebsiteForcing (mathematics)Computer animation
09:56
Execution unitGraphical user interfacePoint cloudInstallable File SystemGamma functionSineEstimationTexture mappingComputing platformVolumeInformationInterface (computing)Data storage deviceComputer fileTask (computing)Multiplication signLimit (category theory)PlanningSet theoryMappingMobile appMaxima and minimaComputing platformUniform resource locatorLogic gateNeuroinformatikRange (statistics)Software developerBuildingDemo (music)Default (computer science)Error messageMedical imagingGoodness of fitInstance (computer science)AdditionType theoryService (economics)Computer configuration1 (number)Row (database)MultiplicationSoftware testingStapeldateiAreaEntire functionPhysical systemResultantGame theoryIntegrated development environmentData managementPoint cloudSingle-precision floating-point formatVirtual machineDressing (medical)Configuration spaceWindows RegistryDifferent (Kate Ryan album)Open setSemiconductor memoryCentralizer and normalizerInterface (computing)Heat transferVolume (thermodynamics)InformationMereologyBitWechselseitige InformationQueue (abstract data type)FamilyWeb browserBefehlsprozessorElasticity (physics)State observerLambda calculusOpen sourceLocal ringExpert systemVariety (linguistics)MathematicsLevel (video gaming)Query languageLecture/Conference
15:18
Presentation of a groupInformationResultantCuboidCellular automatonBitWindowForestDemo (music)AreaMobile appMultiplication signMathematicsComputer animationLecture/Conference
16:15
Latent heatMathematicsInterface (computing)AreaRing (mathematics)MappingBitFunction (mathematics)Context awarenessFront and back endsInsertion lossForest
16:52
ForestRight angleTrailMathematicsVideo gameAreaMenu (computing)Open sourceDemo (music)
17:37
ForestFrequencyBitLevel (video gaming)Correspondence (mathematics)Computing platformMappingAreaResultantElectronic visual displayFunction (mathematics)PolygonMobile appType theoryQuery languagePopulation densityInsertion lossFront and back endsMultiplication signSheaf (mathematics)2 (number)Different (Kate Ryan album)Context awarenessWeb pageMathematicsCartesian coordinate systemCalculationSet theoryWebsitePerfect groupMenu (computing)Square numberMixed realityDirection (geometry)Dependent and independent variablesComputer animation
19:25
Interface (computing)Texture mappingStability theoryAreaMathematicsMenu (computing)Cartesian coordinate systemContext awarenessState of matterBitPolygonResultantQuery languageUniform resource locatorFunction (mathematics)Mobile appForestFunktionalanalysisMappingComputing platformField (computer science)LageparameterSpacetimeFrequencyExploratory data analysisInformationDecision theoryChainIntegrated development environmentClosed setMultiplication signInsertion lossLevel (video gaming)InjektivitätDigital photographyStatisticsLecture/ConferenceComputer animation
22:20
Process (computing)Latent heatForestEmailNetwork topologyIntegrated development environmentProduct (business)Coefficient of determinationPhysical lawFrequencySoftware developerVariety (linguistics)Library (computing)Different (Kate Ryan album)Multiplication signResultantBitNeuroinformatikAuditory maskingInsertion lossFormal verificationMetric systemAnalytic setRange (statistics)Type theoryAreaProjective planeMathematicsOnline helpInformation securityDirection (geometry)QuicksortLie groupForcing (mathematics)WordTowerTask (computing)MassMeeting/Interview
26:49
Traffic reportingComputing platformClient (computing)InformationData typeData managementTexture mappingDialectForestData managementMereologyTraffic reportingInstance (computer science)Content (media)Figurate numberProjective planeWordMultiplication signVariety (linguistics)Control flowDifferent (Kate Ryan album)Type theoryTelecommunicationEstimatorCalculationField (computer science)Meeting/Interview
29:18
Raster graphicsVisualization (computer graphics)Computing platformFunction (mathematics)PixelThresholding (image processing)MaizeMathematicsForestRule of inferenceDecision theoryMetric systemJava appletBefehlsprozessorMixed realityScalabilityComputer-generated imageryInternet forumGraph (mathematics)Texture mappingForestControl flowComputer animationLecture/Conference
Transcript: English(auto-generated)
00:07
Thanks everybody. Thanks for coming. My name is Damon Rand, and today we're going to be presenting a talk on something we've been working on the past few years, which is
00:21
developing a processing and analysis pipeline for monitoring deep fire station via satellite. With me is Poruk Harley and Jill Bornasal, who will also be sharing the stage with me today. Oh, absent is, unfortunately, is Ryan Elfman, who was originally scheduled to be one of the
00:45
speakers today. He is in Kenya, so couldn't join us, but he did a lot of the technical implementation of the change detection itself. So if you do have any questions, we've set up an email address that you can email us, and if you have super technical questions,
01:04
we'll try as best as we can to answer them. But if not, we have an email address for you, and you can send them there. So we are Ecometrica. We're a software as a service company founded in climate science, and we're passionate about clear, correct, meaningful sustainability
01:21
management tools to help businesses meet their sustainability challenges. To that end, we have two main products. One is a sustainability reporting platform that is basically a greenhouse gas accounting system, and we also have a mapping platform, which we're going to be talking about more today. And Jill will be giving a bit of a demo on for combining
01:43
client spatial information with Earth observation data. So yeah, if you do have any questions for us, you can email us at phosphorgy.ecometrica.com, and we'll get you an answer to your question as soon as we can. So the project that this all came about out of was called FOREST 2020, and it's a major investment by the UK Space Agency as part of an international
02:04
partnership program to help protect and restore up to 300 million hectares of tropical forests by improving forest monitoring in six partner countries using advanced uses of satellite data. Those are the partner countries there, Mexico, Ghana, a few others, Brazil.
02:21
And these are the partners, various companies involved, various academic institutions as well. And the challenge for us was to develop an end-to-end pipeline for processing radar imagery to detect forest loss. Essentially, given an area and a range of time, we want
02:40
to automate the rest of the process. So we want a cloud-based service, we need to be able to fetch terabytes of data, we want to pre-process and clean the data, run the change detection algorithm on it, and diff the output against the forest baseline map. In this case, we used Hansen. Some of the open source tools and open data we're using, Ubuntu, Python,
03:05
Docker, the ESA Snap tool, GDAL, Araster.io, Fiona, ND, which is the change detection library by Johannes Hansen at the University of Edinburgh. X-Ray, we use Mapbox GLJS and React on our front-end interface so you can visualize it, and we use Sentinel
03:26
and the Hansen dataset. So the objective was to scale the service out to be able to handle tens of thousands of areas, generalize across multiple forest types ideally, and provide historical and near real-time analysis and be an end-to-end service. Not all of these
03:45
objectives have been met yet, we're still working on this for this project as well as other clients as well. So where we're getting our data from is Sentinel. This is the Sentinel satellite and we're working with the C-band synthetic aperture radar
04:02
data. And you can see Sentinel works basically like this, there's two satellites that go around the Earth and create these swaths. And so imagine if you're looking at areas within these swaths at certain points in time, you know, the satellites move and
04:21
so depending on when you want, when you're taking your time, the satellites are going to be in different positions. So you have to basically calculate all the different images you're going to need from the satellite. So if looking at my colleague Porek's homeland here in Ireland, if you wanted to get all the images for a one-year period from August
04:44
to 2018 to 2019, you're looking at 4.5 gigs per zipped image and it's going to take 2065 images as an example. So the change detection itself, how do we do it? So we get the
05:01
clean SAR VRTs, one for each date in the time series and then we process those using the ND library. Basically what this does is it outputs a frequency of change per pixel. Then we clip that against a forest, non-forest map baseline to measure the change in forest
05:24
and we extract the pixels that have a change of a threshold of greater than two. This was basically, that number was a result of a lot of trial and error and analysis after the fact and the output is a raster which is then vectorized to a shapefile and prepped
05:46
for visualization in our mapping platform. And now I'm going to turn it over to Porek who's going to talk to you about the AWS pipeline that we built to go along with this. Thank you. Hello everybody. So, can everybody hear me okay? Is it good? Yeah.
06:06
So yeah, here's the schematic for how we actually get all of the setup going. You first get your area of interest. We serialize it to just a standard text wicket from our front end and we send it back to the change detection system from there. So you give
06:22
it this particular date range that you're actually interested in and then we do the search and filter against Copernicus or one of the DS providers to actually pull down those particular images that are required for that. We upload those zip files that we do get up onto S3 and then we do, we use snap to then do the pre-processing for
06:42
the subsets of the areas from those images that we actually care about. So once we have all of that and we mosaic it into actual useful images for us, we then use our data cube with x-ray which you've been hearing a lot about I think during the conference and we use the forest baseline then to actually figure out what the metrics are and get the decision
07:02
rules and at that point we also do some manual validation if the thresholds are kind of in the middle area we'll double check those ones if we're happy with the from the automated validation then we're okay with it and then we send the results back to the website. So first thing is the actual Sentinel APIs. So I know there's a lot of ESA and Copernicus crew here and some people from I think
07:21
are in the lobby as well so Copernicus itself is fairly slow mainly because it's handling it for everybody for free so they actually have to keep it at a fairly solid level and make sure they're not giving everybody too much. There's limited to concurrent downloads per person so we don't want to throttle it from that set of things as well but it's free it's stable and
07:43
the API is really really good and there's a lot of Python APIs and stuff or wrappers for it as well which really helps. Dias like creatives they're here they're fast they have multiple locations and providers so it means you're not always knocking out Copernicus if you're actually trying to make these queries. The pricing's mixed some things are still available for free
08:02
others are they have their own platforms as well which costs money but you can you can see which providers are actually giving what you need for it. They're in development there's a lot of APIs they're not all the same as Copernicus I think don't know if every one of them supports the open search API yet but they also have a lot of stuff for their
08:21
actual individual logins and everything as well which makes it a bit difficult when you have to manage a whole load of different ones so if anybody else is working on a library to access all of them at the same time maybe talk to us afterwards because we're also doing that so we can pull resources. Also the other one is the AWS open data registry that's on EU central
08:43
one request or pay so don't request more than you need because it'll cost you more money. It'll only have central one's GRDs and SL1Cs and L2As for the central two and it's the S3 API that you're going to have to use again there for that. The ESO snap tool for myself it was pretty difficult figuring out Java because I've always
09:03
done a lot more Python work so when something went wrong it took a while to figure out what was going on. The documentation for the same reason is mixed because it's a lot of Java API documentation though I know they got a lot of tutorials online as well but normally they're not in the areas that we've had curiosities in. Make sure that you have at least
09:20
eight gigs of ram for whatever processing you're needing because the Java JVM needs that much to just get going with the amount of data that we're pushing through it but the support forum is good and the devs are active and I think Monday or Wednesday they had the talk about the snap 7 which just came out in July as well and the GUI is really good for generating
09:41
those process graphs for terrain correction and subsetting and everything and it's a really good tool to make sure that you know what you're getting first. Docker allows you to actually do that scoping of CPU and ram so you know you need at least eight gigs of ram or anything like that you can provision it there you can make sure it's not taking over the rest of your system if you're testing locally. It also means that when you're running in the cloud you can
10:03
actually set very fixed amounts of ram and you can run it's not going to eat up the entire thing so that's good. It's scalable, the build is fixed and it's reproducible which means that we all the different devs and the analysts can all check it out and we can scale it out quite horizontally on the cloud which has been very useful. Dependencies can cause the images to be
10:21
really really big that's one thing we're hoping to fix this year and you might need more cleanup steps and it's another tool for people to learn at the same time. So we use elastic container service on AWS for managing this. There's two main options you have with that, there's Fargate and there's EC2 for self-managing it. So Fargate you choose how much CPU you want, you choose how
10:41
much ram you want, you choose where in the world you want to do it and it'll provision it themselves away from the rest of your things and they just manage it for you. The problem you have there is that the container size is maxed out and fixed at the 10 gig limit which is a Docker default limit as well. You can try and expand it with EBS and things but configuring that with
11:03
Fargate has been an issue so we gave up on that part of it but we still use it for actually doing the download management from Copernicus and Sentinel. And it's also a bit more expensive because they're managing more so they charge you more. The EC2 part, at self-manage you have to choose basically everything, the EC2 instance types, you have to handle your own errors,
11:24
you're using AWS's base AMIs for it or either the bun-to-buns or their own ones, so they handle the actual putting the Docker containers running on every single instance for you so that's great and then you choose what kind of storage you want actually on the machine as well. You can and you should use spot fleet whenever you're doing this, bid low and try and
11:45
get as many as you can depending on what resources you need. There is also things like AWS batch and there's more Lambda things coming but batch we couldn't debug easily, we needed to be able to be on the system to actually double check that everything was working as we expected and sometimes you need to clean out some old dead containers before the auto one normally would.
12:04
And Lambda at the time didn't have enough of a ability for us on that. So lessons learned on this, probably one that you already know, use S3 a lot, use it heavily if you're on AWS, try and avoid using EBS and snapshots for storing all of your data, make sure that
12:22
S3 is your main storage location, you do a computation you get your cached results from that, that either goes into ElastiCache or you're dumping the file computed back into S3 as soon as possible. S3 to AWS as long as you're in the same region, the costs are basically nothing for the actual transfers so make sure you're doing wherever your files are and you're computing
12:43
in the same region to avoid that cost, especially if you're using the open data registry stuff, that's EU central so make sure you're also doing your compute in EU central to try and avoid some extra costs there. EBS is your on instance storage or the actual instance store itself, use that during the runs but make sure you're clearing everything
13:01
out as well to avoid a lot of extra crap just lying around afterwards and temp files and you don't want to be paying more for it than you have to. EC2 spot fleet, use it, avoid on demand for this because unless you're I need this task right now and I don't want to fiddle around with thinking about it. Even reserved instancing, if you buy a reserved instancing you're paying
13:24
for the entire time you've reserved even if you're not using it, most of the time you're probably going to be using kind of big spikes of queries so spot fleet is really really good for that so plan your budget set a max price that's still fairly sane. EC2 fleet is a new system for choosing multiple different instance types as well so you can choose many different families from EC2
13:44
at the same time to choose your ram and memory across what's available and configure your queues so that each instance should be downloading one image and checking all the areas for that one image rather than downloading the same image to many many containers because that will just make all of your processing a lot slower and that's me and Jill. Thanks okay so cool pipeline but
14:10
now we want to see a map so what we're doing is that we display those change detection results on our mapping platform so before I do a quick demo of the mapping platform I just wanted to give
14:22
you a really brief background on it so it's intended as a platform for a wide range of environmental social economic data and it aims to simplify the extraction of information of increasing volumes of data that come from earth observation or other sources and it provides a
14:42
simple interface for non-user experts to get answers to variety of questions and basically it makes special data accessible to the public to stakeholders and to non-GIS specialists and the data can be stored and backed up locally but the platform runs on the cloud so that we can
15:01
ensure it's always performing and it works with browser updates and you don't need a local IT person to look after it and we use agile development so it's frequently updated with new features this said we're just going to move on to a little demo of the mapping platform I've
15:21
recorded it just to avoid live demo issues but it's probably going to be epic anyway so bear with me and all right so when you just first log in onto the app that's what you see all the apps are designed around a specific theme to answer a specific question and this one was designed to present the results of the cell change detection that Porec and
15:44
Damon just presented to you it features results that were produced in window areas for three of our partner countries from forest 2020 which were in Brazil Mexico and Kenya and today I'll only have time to present Mexico and Brazil results so yeah first thing you see is this
16:03
introduction box so that's fully customizable by the user and it just gives a little bit of information on what this app is designed for and some kind of narrative info and then we move on to this the interface menus first one is the layers so layers are just visual maps of the data that
16:25
was uploaded in the back end and that just gives a bit of context and this first layer we're looking at here is the change detection output and for an area called Brandonia in Brazil and so in orange here what you see is forest loss that was detected by a change detection
16:44
between February and September 2018 in this specific monitoring area no I knew that was gonna happen gosh right okay so we're back on track kind of and so
17:10
yeah the change detection and then we're just switching another layer which is the forest baseline which we extract from hands-on forest data but it could come from you know other sources
17:25
and that just shows the forest that was available that was not available but present in your studio area before you start monitoring for forest change and then the other kind of menu is the areas of interest so for this demo we're just focusing on this subsection
17:43
of Randonia which is where we run the change detection so each area on the application has a results section and basically when you click results you just get a summary of what can the platform tell you about this queried polygon and so the results page display results that
18:07
were extracted by the mapping platform from data layers uploaded in the backends and basically the platform the mapping platform runs query calculations on data sets on this polygon that's queried here and in this app we've set up two queries the first one returns the amount of
18:27
forest loss detected by your change detection so in this subsection we found out that 555 hectares of forest were lost in this short amount of time for the monitoring period so that corresponds
18:41
to this orange bits we see on the map and that corresponds also to 1.5 percent of this subsection that was deforested and then the second query is just the results for the forest baseline so you can see how much forest was in this area at the beginning so it was about
19:01
34 percent of this polygon that was forest at the beginning of 2018 so that's kind of a first output for this type of like dense forest in Brazil and and then we're just gonna look at a second type of forest a different context in Mexico so just moving a bit so we're just
19:27
gonna look at the change detection in the context of avocado farming so we can just select our Mexico areas here in the application menu and just yeah using the filter area stable we
19:41
just select the avocado area of interest um so that's uh where some of the avocado farms are located in the state of Jalisco in Mexico so the first layer we're gonna look at is the forest baseline just to get a bit of context of what's happening in this area and most of the
20:02
forest is located on the west on the west of the area of interest and the rest of it like kind of close from the forest is mostly avocado fields and then we're looking at the change detection output so similarly to what we just saw it's in orange means there was forest loss between February and September 2018 there was some loss in the north but the majority of
20:26
the loss was kind of detected in this area here in the southwest and we can use the mapping platform to kind of explore the area a bit more so when you look at the aerial photography in the base map you see that the forest in this area is pretty fragmented which is usually means like
20:45
high risk of deforestation and also all around it is avocado fields so there's quite a high assumption here that the forest is slowly being cut off to make more space for um new avocado fields and just one more tool I wanted to show you on the platform today which is quite useful
21:03
in this kind of exploratory work is that users can hand draw their own polygon of interest so let's say you're a producer in this area and you just want to look at the state of the forest near your fields you just hand draw this polygon you can just give it a name save it and then similarly to application polygons you get a results button and when you click on it
21:26
the platform runs the queries on the fly and you just get a summary statistics of how much forest was lost as per change detection in your area of interest and how much forest was also present at the beginning of the monitoring period right so and that's the url of the app we just
21:46
presented to you so feel free to just go and explore it a bit more by yourself you can send us any questions if you if you have any and yeah that was just a short overview of what the mapping platform can do there's many more functionalities in this tool but I hope that
22:01
gave a better idea of what the change detection results look like and how they can be used to make informed decisions on supply chain impacts on the environment thank you very much and yeah we have time for some questions I think so I'm very interesting I'm very interested I have
22:22
many questions but I'll leave it to the to the to the floor yes please speak up so people can you hear it okay sorry sorry so the forest mask we use in this project is from hands-on
23:01
data so that's already available you know for us data that's free to download and it's a global forest data so we just download the data and just process it to have a forest baseline that's up to 2018 or whenever it's the beginning of our monitoring period so we don't do any of this processing ourselves we just use freely available data for the forest mask but you could use any
23:23
mask you have if you have your own forest mask that you know is more accurate than the hands-on data then you could just feed it into the the change detection to answer to answer your other question the for the change detection itself for the actual metrics that's mostly been Ryan
23:40
who did the actual analytics back and forth and testing out a whole different variety of them but I know he used a lot of there was random forest being used and a lot in the ND library that Johannes Hansen different Hansen made so he's doing that in Edinburgh and he's collaborating back and forth with Ryan on on that so that's that's an active development so that tool allows you to actually choose from a wide range of things you can test on yeah if you have a
24:06
very specific technical question please email it to us and we'll get it to Ryan and he can answer it directly for you sorry deforestation in brazil is obviously a hot topic these days a little work in brazil and the thing that surprised me the most is how bad the
24:21
disasters are there because they're self-reported and without topology and without verification and so forth but they're the primary regulatory instrument that says has somebody violated the law by deforestation deforesting more than apex percentage or specific areas have you guys given thought as to what tools might be made available to kind of help on that
24:48
aspect we have done some stuff as a pastor well I guess what we're doing here could maybe be seen
25:01
as a different type of accuracy because we're taking into account different type of forest and terrain and things like that and we spend quite a lot of time validating the results but if you have the like you know the national kind of monitoring agency that is producing results and those results are like used in policy then you can fully overcome that so I think it depends
25:24
yeah exactly yeah I mean it's like you can you can just yeah yeah yeah how do you fit in into so the work we do with forest 2020 actually we work with inpe which is the agency that produces products which is the national data in used in brazil and so yeah
25:41
we're trying to just collaborate a bit more and see if like we can help with improving the methodologies to detect forest change so yeah that's that's all we do and hopefully that's going to be enough in the end but we'll see so my question is when you do these computations
26:14
run them in the cloud and spawn all these processes how much forest loss did you occur
26:21
through your computations I mean you know equivalent is that and you know it's kind of advocate question this has factored into like tenders these days like when when somebody's tendering you allow maximally to emit you know these we as a company we do our own environmental greenhouse gas assessment using our own product we we dog food our our products
26:45
ourselves so we do a full greenhouse gas accounting assessment of our own company every year and we do we do offset all our emissions including our travel to this conference
27:29
we we do as part of the sustainability we actually because we were we report and we calculate on our estimations on how much actual computationally cost-wise we are as part of our
27:41
reports if we are being asked for that we can actually provide it to any but any any content it's not okay I get your question it's I don't think it is I can double check with our project manager who's looking at forest 2020 and she's the one who
28:00
worked on the tender and the proposal I haven't heard of anything like that so I don't think it's in the tender follow-up it's related which is just like since you've done the calculations like how much is how much is field work flying in brazil versus no just the bc2 instances you know how much is like where your overall compute versus overall travel I would imagine the travel is
28:24
the travel is typically our biggest our biggest we will use it wherever people want us to use it basically yes we're not we're not geographically
28:45
bound we're using it for a variety of companies right now as well in different regions in gana and argentina and as well yeah we do a lot we do a lot on tropical forest right now we don't have much many projects in europe but this is like a universal methodology so
29:05
you could apply it to any type of forest one last question quick and I think we'll have plenty of time in the coffee break because there's a lot of questions here also because
29:21
we had a wind break we had damage from forest as well and maybe nice to check this thank you everybody