Exploring In-Situ Data: Definitions, Sources, and Training Applications
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 |
| |
| Title of Series | ||
| Number of Parts | 44 | |
| 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 | 10.5446/66860 (DOI) | |
| Publisher | ||
| Release Date | ||
| Language | ||
| Producer | ||
| Production Year | 2023 | |
| Production Place | Wageningen |
Content Metadata
| Subject Area | ||
| Genre | ||
| Abstract |
| |
| Keywords |
8
9
10
11
15
17
19
22
23
25
27
29
32
36
37
41
00:00
Meeting/Interview
00:11
WikiSource codeGeometryMathematical analysisSystem programmingVertex (graph theory)Computer programEmailTranslation (relic)Term (mathematics)Latin squareSatelliteInternetworkingComputing platformSurfaceMeasurementIntegrated development environmentWorkstation <Musikinstrument>RadarComputer networkEndliche ModelltheorieDigitizingMeasurementArithmetic meanSlide ruleTerm (mathematics)Field (computer science)Multiplication signType theoryLatin squarePosition operatorReading (process)Latent heatState observerDigital photographyMathematicsSatelliteComputer animation
03:29
EmailService (economics)Open setArchaeological field surveyCovering spaceAreaFluxNetwork topologyData storage deviceForestDatabasePlot (narrative)Computer networkFile formatVertex (graph theory)GradientInformationGeometryAddress spaceUniform resource locatorPrice indexData managementInformation securityMathematicsWeb pageWebsiteState of matterHessian matrixPresentation of a groupBasis <Mathematik>Digital signalDigitizingModel theoryComputing platformMobile WebMobile appWikiIntegrated development environmentInterface (computing)Communications protocolMultiplication signDatabaseForestService (economics)Reference dataFluxDeclarative programmingSingle-precision floating-point formatStatisticsSatelliteSampling (statistics)Form (programming)Product (business)Process (computing)Latent heatInformation securityVirtual machineWave packetBitInformation systemsCovering spaceSoftwareEndliche ModelltheorieField (computer science)State observerValidity (statistics)Uniform resource locatorSet (mathematics)Direction (geometry)Computer animation
08:43
Point (geometry)Mobile appPoint (geometry)Decision theoryNetwork topologyField (computer science)
09:09
Level (video gaming)Decision theoryLevel (video gaming)Type theoryArchaeological field surveyCovering spaceRootSoftware testingComputer animation
09:41
Outline of industrial organizationProcess (computing)Covering spaceLevel (video gaming)Mechanism designComputer animationDiagram
10:18
Point (geometry)Alpha (investment)InformationLevel (video gaming)Data managementMappingState observerLevel (video gaming)Wave packetWebsiteInformationDirection (geometry)FreewareUniform resource locatorType theoryComputer animation
11:23
Computer-generated imageryBuildingMedical imagingSoftware testingMessage passingGoogle Street ViewType theoryImage resolutionSatelliteState observerBuildingCategory of beingDirection (geometry)Level (video gaming)Game controllerCASE <Informatik>Computer animation
Transcript: English(auto-generated)
00:11
I'm planning to just give you a brief introduction and then my colleague Linda C will deep dive into the GEOviki as such. Just to remind us what actually in-situ data is
00:26
and where it comes from. I think we're all struggling with the term itself. We have to comply to in-situ data specifications and so forth, but it's coming originally from the Latin and it means in its original place or position and it has been a lot
00:45
used also in archaeology where this is really critical when you have to determine at what time and age something was found like an old vase was found somewhere and it's really critical to know was this vase moved to another place or was it at its original place and this is actually where
01:05
the term comes from and how the term was used quite a bit. But this is now also used a lot in the field of earth observation and in the field of earth observation in-situ means data collected
01:23
adjacent to the measuring instrument like temperature readings and so forth. However, we also need to know that this term has been widened a lot now with respect to what EEA, the European Environmental Agency understands under in-situ data and the new definition by EEA for in-situ
01:48
data means all data not collected by satellites. So any other data which is not collected by satellites is referred to as in-situ data. I think that is a really a big change and widening of
02:04
the actual term. So the European Environmental Agency uses two types of in-situ data. So on the one, the more traditional one, environmental measurements from measuring stations, weather balloons,
02:23
sensors abroad, airplay aboard airplanes, ships, floats, moorings, radars, so ground-based radars, river ganches, air quality centers and so forth. But also any other data as we know now such as
02:42
topographic maps, land surface, man-made features, hydrology, transport networks, land cover, digital elevation models, even aerial imagery, aerial photographs are referred to as in-situ data. And again, the scope is even widening more because of new sensors continuously becoming
03:07
available, new sensor technologies where we start to be able to measure everything with low-cost sensors. But also crowdsourcing is taking off and for that we also have prepared
03:22
some more slides to give you some insights on what data can be collected via citizen science and crowdsourcing. There is no doubt that the holy grail as my friend and colleague from the GRC Rafael Dantrimo refers to it as holy grail. I refer to it as new goat for remote sensing.
03:45
In-situ data is absolutely critical, especially in deep learning, in machine learning models where they apply now deep learning. It needs a lot of training data and initially this training data cannot be automatically generated. It needs some human-based labels and they come many times
04:05
they are in-situ data and they are absolutely relevant for training these models. But also you can use in-situ data if you have a robust sample over a country you can produce statistics for a specific country such as traditional forest inventories. They are samples
04:26
which produce statistical relevant forest data for a specific country. Examples of existing in-situ data, those are mostly authoritative here, weather data, you know, weather
04:41
services, then the land, the lucas data produced by Eurostat, land parcel information systems data. These are farmer declarations where countries are responsible and then they make it openly available or not. This depends a lot on the country. That has been nicely summarized in a so-called
05:03
Eurocrops database where, I think it was, Nünschen University has done this job compiling this LPI as in a nice standardized form. Then we have the national forest inventories. Also we are running at the other kind of in-situ network for biomass which is called GeoTrees. Then, as you know, the
05:34
there's flux data, for example, managed by several institutions. So you can question is this now
05:40
authoritative or not authoritative. That can be debated but so forth. These are just some examples of what in-situ data is currently out there. This is just an example of this GeoTrees database which you can use. This is globally available to get biomass data and
06:03
this data set was set up also to train the future missions of ESA, the biomass missions and other missions that there is some reference data to calibrate and to validate satellite data. There is also a little bit hidden for the general public and for many institutions the
06:25
Corda database that is only accessible to Copernicus service providers, mainly companies in Europe working on remote sensing data. It's quite an interesting database. It has a lot of data on land monitoring, marine monitoring, atmosphere monitoring, emergency management, security
06:46
and climate change, a lot of the Copernicus services. If you deep dive into it so we got access for some time but we don't have access anymore but you can find many different also
07:00
topographic data sets, national land cover maps, a single layer of specific products such as forests and so forth. Now coming to the more non-authoritative data, citizens data. As you might know, NASA has been involved in this field for quite some years and we've developed a GeoViki
07:26
tool which Linda will particularly talk about but also many other tools around GeoViki. Also some rapid assessment tool called PicturePile and also quite a lot of tools where you can
07:41
collect data on the ground such as for example PhotoQuest. So I just want to pick out two which is PhotoQuest Go and PicturePile to give you a bit of a flavor on what we have been doing and what you can do with these tools. Coming to PhotoQuest Go, we've
08:01
also tried to understand what is actually the quality of citizen science and crowdsourcing. This is a repeatedly asked question and people, you know, they are very skeptical. Some people say, oh your quality might not be good enough. So we thought let's just do an experiment and direct people. So this is what we call directed data collection on the ground versus
08:26
opportunistic. We direct people to certain locations. These are Lucas locations which are stable. Some of them are stable over time. We know where they are. We know when they are measured in which year and we can check how good is the quality of the citizen observation.
08:43
We designed for that an app which directs you to those specific points. We're interested, it tells people if a point has been visited or not. We try different incentives, prices, but also payments. One euro if it's close to your accessibility to a road or field
09:06
street and three euros when you are further away from it. We designed the decision trees of the basic land cover types level one going into level three which is very detailed and goes down to crop types actually in the end. It goes then to root crop cereals but it goes even
09:25
down at level three to the specific crop type. So we mimic that kind of Lucas survey and we thought let's test how good are citizens. We just throw it out, we run a campaign, we go through the media, advertise it and see what we get and what we can test.
09:44
We run two different campaigns. In 2018 we actually improved quite a lot the methodology. We also more widely engaged the public through this payment mechanism versus a price mechanism and price incentive and we got up to roughly 90 percent accuracy on the broader
10:06
land cover land use level which was actually pretty good news. So this shows that citizens are able with quite basic land cover to do a reasonably good job. Other apps, this is an
10:21
example of an opportunistic data collection, is to observe crops. You can go to any place, you can use this app, it's free and freely available to you. You can then also download the data, it goes straight to the website and you can collect crop type information geo-referenced at the location and we also, and this is I think different to many other apps, we provide you
10:46
with a background map where you can exactly pinpoint where your observation is rather than normally it just takes observations along the roads. When you're standing along the road then it will not tell you where, in which direction you're taking for example a picture
11:04
and what specific crop you are observing. We record information on crop type, phenological stage, if the crop is damaged or not. So this information is also very useful for global crop monitoring and for providing training data for global and European crop type mapping.
11:25
Another example is this picture pile, this very quick a rapid assessment tool. You get the picture, you are asked do you see for example cropland in this red box, we can use very high resolution satellite images, yes, no or maybe and we can rapidly collect lots of classification very very
11:44
quickly. We had one example of more than three thousand players where we collected more than four million observations so this was really interesting that this was actually possible to collect that many observations. You can also apply this for deforestation monitoring when you
12:00
have very high resolution available or for disaster assessment if buildings have been damaged after an earthquake or after an hurricane Matthew in this case where we did some tests. You can also apply it for more categorical classifications what type of crop do you see
12:25
you and this is much faster it's not anymore asked do you see maize yes or no but you can actually slide then this into the specific category of what type of crop you see and you can use for this street view imagery, mapillary information, other street level photography,
12:44
your own photography you take on the phone and you can more use this tool to rapidly classify and label collect labels for crop type classification geolocated. We did a test on this and we found relatively high accuracy when we compared the land use sorry for the LPIS data
13:07
we compared with that and we we found that this data is actually very reliable when you use street view images and compared with the official LPIS data so the accuracy was
13:21
94 percent though this was because we probably had too many vineyards in there so people were very good in recognizing vineyards but in general the take-home message is that the accuracy is really accurate when you use this street view images where you also know and can have control over the direction the image has been taken yeah that's it from my side