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

Exploratory study of urban resilience in the region of Stuttgart based on OpenStreetMap and literature resilience indicators

00:00

Formal Metadata

Title
Exploratory study of urban resilience in the region of Stuttgart based on OpenStreetMap and literature resilience indicators
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
Publisher
Release Date
Language

Content Metadata

Subject Area
Genre
Abstract
Training spatio-temporal OSM-indicators based on the resilience core from Cutter (2016) and exploring the implications for urban planning in the light of revealed thematic tags in the region of Stuttgart. “Nobody on this planet is going to be untouched by the impacts of climate change” Rayendra Pachauri (2014) The overarching nature of building resilience across disciplines and its inherent positive mutual understanding due to the association with the immune system, also amongst the non-scientific community, makes it an attractive and increasing popular concept which everybody seems able to grasp its necessity. Hence there is an exponential increase, even limited down to the key words “urban resilience”, in scientific literature over the last decade. Moreover the concept is also taken up by the New Urban Agenda – Habitat III, the SDG goals and also the IPCC. Hand in hand with this development the definitions and operationalizations are innumerable and starting to lay a smoke screen above it. Conjoined, there is a clear lack of validation of resilience measures, including spatio-temporal aspects but also of the single component of it (Bakkensen 2017). Moreover, traditional data sources like census or governmental data miss out on certain important facets making empirical validation impossible and lack the spatio-temporal resolution necessary to cover the characteristics of resilience (Burton 2014). Hence, this experimental study explores and develops new spatial indicators through machine learning methods derived from OpenStreetMap data to replicate conventional core indicators. In order to cover all spatial attributes indicators for points, lines and areas will be deduced and separately as well as in a combined analysis investigated by means of supervised and unsupervised algorithms. The outcome is expected to uncover hidden spatial relations and patterns of urban resilience. Moreover, Burton (2014) stresses the need for new data sources to better understand the multifaceted phenomena of urban resilience. Therefore this study is contributing in developing robust and reliable socio-economic indicators contributing to this challenge to clear up the smoke.
Keywords
129
131
137
139
Thumbnail
28:17
Observational studyPressure volume diagramPlanningThomas KuhnComputer fontGrass (card game)Physical systemPerspective (visual)Variable (mathematics)System programmingVariety (linguistics)Moving averageView (database)Level (video gaming)Control flowUniverse (mathematics)Open setSoftware developerTerm (mathematics)State of matterBitState observerMeasurementMathematicsSpeciesPerspective (visual)Context awarenessObject (grammar)Focus (optics)Stress (mechanics)Price indexObservational studyPhysical systemDivisorComputer fontContent (media)Contrast (vision)Greatest elementDialectField (computer science)Theory of relativityAreaHypermediaWordGroup actionMultiplication signLeakDivision (mathematics)Database normalizationResultantComputer animation
Finite element methodChannel capacityHazard (2005 film)Gamma functionIntrusion detection systemField (computer science)Hazard (2005 film)Content (media)Electronic program guideChannel capacityState of matterComputer animation
Continuous functionPhysical systemEstimationPressure volume diagramLevel (video gaming)Archaeological field surveyFile formatSet (mathematics)FrequencyTemporal logicFeasibility studyWitt algebraStatisticsSoftware testingSystem identificationCharacteristic polynomialSource codePrice indexComputer programmingDenial-of-service attackPrice indexOrder (biology)Source codePhysical systemFile formatImage resolutionLimit (category theory)Slide ruleAuthorizationDifferent (Kate Ryan album)Machine learningState of matterOpen setAngular resolutionMeasurementDatabaseTerm (mathematics)Level (video gaming)BitTemporal logicComputer fileStandard deviationFeasibility studyMereologyDimensional analysisBasis <Mathematik>Computer animation
Price indexSummierbarkeitAreaDatabaseMoore's lawDatabaseCentralizer and normalizerMereologyMappingTerm (mathematics)Channel capacityElectronic data processingMilitary base
AreaSummierbarkeitBuildingStatisticsFluid staticsDatabaseBoundary value problemSet (mathematics)Table (information)Term (mathematics)PolygonAreaGeometryLine (geometry)StatisticsAuthorizationCountingInformationLengthDatabasePoint (geometry)Moore's lawKey (cryptography)MathematicsComputer animationProgram flowchart
AlgorithmSimilarity (geometry)Virtual machineMathematical analysisLinear regressionLeast squaresLinear mapForestPressure volume diagramComputer-generated imagerySource codeHausdorff dimensionHuman migrationData structureVariable (mathematics)Dependent and independent variablesFunction (mathematics)Price indexHuman migrationRight angleSource codeWave packetStatisticsDimensional analysisSelectivity (electronic)Chemical equationBasis <Mathematik>Pairwise comparisonSet (mathematics)Data structureOrder (biology)Computer animation
Pressure volume diagramHausdorff dimensionHuman migrationDependent and independent variablesVariable (mathematics)Data structureSource codeGrass (card game)CircleDifferent (Kate Ryan album)Dimensional analysisArithmetic meanTable (information)ResultantArrow of timeNumberSquare numberBuildingOpen setVapor barrierLevel (video gaming)Key (cryptography)Price indexEndliche ModelltheorieDeutscher FilmpreisObservational studyTerm (mathematics)Contrast (vision)Product (business)PredictabilityError messageRandomizationBitFunction (mathematics)Forest
AreaWorkstation <Musikinstrument>Vapor barrierSurfaceBuildingOrder (biology)Connected spaceMapping
Random numberForestStatisticsHuman migrationMassIntrusion detection systemData structureData structureResultantPrice indexEndliche ModelltheorieHuman migrationSlide ruleDifferent (Kate Ryan album)Attribute grammarFunction (mathematics)Chemical equationSimilarity (geometry)BitSocial classRandomizationForestPredictabilityMereologyStatisticsGoodness of fitProduct (business)Lie groupMultiplication signAreaRule of inference
Grass (card game)Point (geometry)Pressure volume diagramLine (geometry)StatisticsCharacteristic polynomialDecision theoryService (economics)Complex (psychology)Integrated development environmentLevel (video gaming)Link (knot theory)Goodness of fitNeuroinformatikOpen setPower (physics)ResultantCombinational logicOrder (biology)QuicksortClosed setPredictabilityPresentation of a groupDatabaseConnected spaceCASE <Informatik>Attribute grammarFreewareEndliche ModelltheoriePhysicalismMoore's law
MappingLevel (video gaming)QuicksortFunction (mathematics)TwitterReflection (mathematics)Point (geometry)SphereResultantData structureClosed setBlack boxStatisticsPhysicalismHuman migrationAreaFocus (optics)Execution unitPrice indexDivisorField (computer science)Lecture/Conference
Data structureLemma (mathematics)Maxima and minimaInclusion mapGrass (card game)Linear partial informationBitElectronic visual displayProof theoryParameter (computer programming)Multitier architectureChemical equationPrice indexHypothesisEndliche ModelltheorieHuman migrationContrast (vision)Point (geometry)Different (Kate Ryan album)Perspective (visual)Complete metric spaceFunction (mathematics)PredictabilityDistribution (mathematics)Lie groupOpen setMultiplication signConnected spaceObservational studyAreaLecture/Conference
Endliche ModelltheorieOrder (biology)State of matterTrailPrice indexDifferent (Kate Ryan album)Process (computing)Term (mathematics)Software testingSoftwareControl flowVector potentialSensitivity analysisCartesian coordinate systemPredictabilityValidity (statistics)Level (video gaming)Open sourceGeometryPresentation of a groupImage resolutionoutputData structurePersonal identification numberStorage area networkOpen setSource code40 (number)Product (business)Similarity (geometry)Forcing (mathematics)Lecture/Conference
Transcript: English(auto-generated)
Okay, we can have the last talk for this session and it's my pleasure to introduce Holger from the University of Stuttgart. He will give us the last talk of this session on OpenStreetMap. Again, you have 20 minutes, 5 minutes for questions and answers and then we will move to the coffee break.
Okay, thank you Marco. Let me introduce myself first. My name is Holger Sauter. I'm as Daniel at the Institute of Spatial and Regional Planning at the University of Stuttgart. We're colleagues. We worked both on this topic and as you can see at the title of my talk now, it's related.
We wanted to discover some hidden factors of OpenStreetMap, some hidden indicators. My approach was more, or the approach of this study was more from the bottom up to look exactly at region, at regional level
and the content of OpenStreetMap in the region of Stuttgart in contrast to the international perspective that Daniel just previously presented. So let me start with some, yeah, with a basic outline of the talk first.
I want to give you a brief background on resilience, on urban resilience because I'm not sure if at this conference with this technical focus or some technical stress on the methodology if you're aware about the concept.
So a few definitions at the beginning, then I come to the research objectives and will then introduce the methods we applied and come up with some results and conclusions. There will be a bit of redundancy, of course, because the topic is very similar to the one Daniel just presented.
But on the other hand, I'm happy I don't have to explain everything in depth because it was already explained. Okay, as I warned you, some definitions. What is resilience? Maybe you know Holling introduced this term, this concept in 1973 by observation of ecosystems,
ecological systems with populations of various species and he defined it as a measure of persistence of systems and their ability to absorb change and disturbance and still maintain the same relationship between population or state variables.
So he introduced the system perspective on terms like robustness, flexibility of system, how much can a system cope before it collapses or breaks down. This system perspective went into various research disciplines. It's a kind of buzzword nowadays in many research arenas you find it.
And it also entered many political agendas. One of it, the most important one each of you is aware of is the Sustainable Development Goals of the United Nations
where you find at least in these five goals specifically the term resilience in the goal. And we at the institute, at the University of Stuttgart work basically in the field of goal 13, climate action, which states that resilience has to be strengthened
and the adaptive capacity to climate related hazards and natural disasters has to be increased. This general goal specifically guides some goals of goal 11 or some
main contents of goal 11 to create resilient cities, sustainable cities and communities. So measuring resilience is very important in urban planning and spatial planning in this regard.
So if we go a bit more specifically to urban resilience, you see here our definition I like very well because the first term is very important to have a measurable ability of any urban system. And the rest is a bit like the definition of hauling. What means measurable? It means we need indicators.
We need something to measure on the urban level or the regional level to define the state of resilience to, as Daniel explained before as well, to determine how successful measures are, programs, initiatives or even single infrastructures that are built in order to increase resilience.
For example, what is the impact of this dam, of this flood prevention infrastructure to the overall resilience of the region? So we need these indicators and we have them already. This is not a very new topic to have these indicators for resilience.
So many regions or countries already have accepted and tested resilience indicator sets available which are in use.
But most of them have the shared problem. You need a lot of data because they consist of many, usually many indicators, a few dozens. Each of them belongs to other authorities or comes from other data sources which means different file formats, different spatial resolution, different temporal resolution.
And so it's always a big deal to standardize, to bring it together and compute it accordingly to an index. So we have a limited feasibility and standardization which led us to this idea how
can we use a standardized global database which OSM is to derive these indicators for resilience. So we did basically the same, almost the same as we did for
the World Risk Index with four recognized resilience indicators that we derived from literature. I will introduce them later on the later slides. And we tested the statistical learning methods in order to deduce these resilience indicators from OpenStreetMap.
So first of all before I go into the methods part I want to introduce the region of Stuttgart because I'm
not sure if all of you know it or know where it is and what dimension it has or it covers. We chose it because it has a very high OSM coverage and a very high contribution index compared internationally and offered a sound data source there.
So it's in Germany in the southwest in the federal state of Baden-Württemberg. And Baden-Württemberg consists of four federal districts, one of which is the federal district of Stuttgart which also holds the capital of the federal state Stuttgart here in the southwest, the economical center.
And as you can see at this map that the economical center or basis of this district is around Stuttgart.
In Stuttgart there are many sub-centers, Halbron, Ludwigsburg and so on with a high economical capacity whereas the eastern part and central part of the district is really rather rural area, small municipalities, small villages. But some global players, hidden champions inside some of these municipalities and this has some
influence as you will see later on maps where we compare the performance of the predictors. The data pre-processing, we already heard some of this before but we had a little different approach because
we could in terms of computational power in this example because the coverage on the data amount was lower. First we downloaded the database from Diofabric and pre-processed the data so throughout some information
of the database that was not very important like authors and so on and change information. Then we imported it into post GIS database with OSM2PGSQL with the full data set which was not clipped, not yet clipped
at the boundaries of the municipalities but we wanted to have statistics for each municipality so we clipped the geometries with QGIS. And built them in the post GIS database again the statistic tables for all polygons and lines and points so
we had statistics for each municipality for area length and counts and we normalized it by population in the municipalities. This table then could be or these tables, there were a few, could be connected with a package to R statistics where we performed basically these methods that you already heard from Daniel.
We decided Random Forest after testing all of them because it produced just the best outputs in comparison and deduced the indicators. What was the training data or the four indicators we've selected to be predicted by the model?
The four indicators were migration balance, demographic age structure, business tax revenue and unemployment representing these four dimensions.
You can see on the left of resilience dimensions, socioeconomic all of them and on the right you see the sources of these indicators. So they are tested and applied, as you know the business tax revenue data sets that usually each year
is produced, others like migration balance might not be available for each community on an annual basis and so on. So you already can see at this selection that the conventional method would be rather complicated in contrast to an optimized dynamic, maybe dynamic method that we did or developed within this study.
So let me show you some results. I don't have to explain the tags anymore because Marco already did. It's basically a key value pair from OpenStreetMap.
The tables here show the results of the different dimensions or indicators we measured and the number MSE, the increase of the mean squared error shows you the importance of the specific tag for the prediction of the model output.
So barrier liftgate for example has the highest impact on the model quality at the end, on the model prediction. If you leave it away, the mean squared error would increase accordingly, double as for example if you would leave it away, Deutsche Post, German Post.
We were surprised a bit by looking at the tags that were selected by the model that were of high importance. Some of them are obvious regarding the dimensions, some of them are not. We really wonder but it's basically the same surprise as we have with the world risk index.
That's the result or the funny thing with random forest models. You don't know what comes out. So if we have a look at for example tax revenues, it's very clear, industrial buildings, private access and so on, land use, industrial has a high impact as everyone would expect.
But when you look at unemployment, for example the situation is completely different, what has occurred, what has an orchid to do with unemployment you might ask.
But anyway, we didn't understand yet fully what the reasons behind these obviously logical connections are. But if you take a look at the maps, you really see that the results are quite convincing. On the left side, I will show you all the four indicators within the next slides. On the left
side you see the statistical data, the conventional mapping and on the right side the random forest output model. And yeah, feeling of belonging which we measured with the migration balance. So how much people migrate in to the municipality. You
can see there's a really good coverage or a really good similarity in the southwest, in the central areas, the economically strong areas. Whereas we have a bit of problem in the prediction here around this northeastern part. We didn't understand why at
the beginning but then we looked a bit into detail of the municipalities and we found out that here in Bad Meggentheim, we have the large company, I don't want to name it, I don't know if that is okay.
So it's a big producer of screws for industry and this economical asset pulls many people into the region but the infrastructure around it doesn't really reflect this economical strength of the region.
So the spatial attributes not always, obviously, can predict these special situations and this difference here or this specific issue about these, as I named them before, these hidden champions that are
located somewhere in rural regions, pops up in the other results as well a few times. Not as much here with the age structure but still one class left and here with the tax revenue, the prediction is rather good.
So another, the last indicator we mapped is the unemployment which also matched very well as you can see.
So it brings me already to the conclusions. This method we applied, of course, it's just a first step but already showed that there are really hidden attributes in OpenStreetMap regarding risk, regarding resilience.
Socioeconomic topics or social topics where the database was not made for but it's still hidden there and it's a quality of OpenStreetMap. We have found or proven that the method we applied is very simple. It
doesn't need a big computational power. You can easily do it with a good computer. Of course, you need some patience for the modeling but a few days should do it. It's just a first step. A lot of more research has to be done definitely and it might not
be sufficient to just use such methods and just use OpenStreetMap in order to have a dynamically produced indicator. It might be also useful to have other methods include in order to improve the quality of the prediction.
For example, remote sensing data that is also dynamically available could be linked with these methods. Thank you for your attention. That was it and I'm free for questions. Thank you. Another great presentation. Are there questions?
In the case of Daniel's presentation with the vulnerability, I can see more instinctive connection to physical features with resilience.
I think it's quite extraordinary that you get that sort of close mapping based on what is mostly mapping of physical features. On the face of it, it wouldn't have much to do with unemployment or migratory trends and then because some of this is just black box passing through,
the interesting thing here is how much the physical features that are being mapped in an area are a reflection of sort of certain issues like unemployment and stuff like that, because we don't map these statistics in any way.
And then the follow-up question to that would be how do you get any feel from this, from what isn't being mapped that could be mapped for these factors? Because everybody has their priorities when they do OSM mapping, they focus on certain things,
but are there things coming out of this that you think should be mapped that aren't? To be honest, both questions are tough questions because we're not that far yet. We produced these maps rather recently and we're still very surprised and the
point is we still don't know exactly why this is producing a good output or that good output, although we cannot really relate the tags to the topic or the spatial structures to the social sphere.
I just have to say where we will get more into detail and have a deeper look at it to uncover it. In Calgary maybe next year, we will show some new results explaining exactly these open questions and hopefully also why or what else could be integrated to produce more results.
Daniel, you have an idea for that question or an answer to that question? I join you basically in your answer, but interesting was looking especially at the age structure in our region.
It was what you would have expected somehow that you have an older age and more rural and looking at indicators, it was connected to farmland, hiking, and even the quite local, where is it?
It was mentioned when I really had to laugh because it's really a club to go hiking on the Swabian alp for older people and that the tier was somehow really unexpected, but also somehow for me proof of the model that such features
because they have houses to go to walk around and hiking, farmland. So that was similar like the tax revenue so you could understand how it is related. And it's also like the migration balance is such a softer topic and therefore it's harder to understand the connections again.
I have one question. In contrast to Daniel, you have applied basically the same method but on a very local area,
starting from the hypothesis that OSM is very well mapped because this has been proven that in Stuttgart it is very well mapped. My question is did you try also to run this in another country or another region of Germany when this is not the truth?
And if yes, can you provide some argument on how sensible it is, the completeness? Because one of the main issues of OSM is the uneven spatial coverage. So that has been studied a lot and of course this is a fact.
So again kind of same questions as to Daniel before. So how sensible is this to the completeness of the data? And again I have to say that's exactly the point where we have to re-enter now after this conference and that's what we're planning to do to just apply this model now to other regions with different spatial distribution and attributes,
especially the coverage of OpenStreetMap. To do this on the region of Stuttgart, the reason behind it was exactly that we had some output from the international perspective of the World Risk Index that was definitely of a bad prediction quality because of missing data and this data coverage difference
and we didn't have it in Stuttgart. That's exactly the reason why we did it there. But many regions in Europe are well covered so we definitely have a chance to apply it elsewhere
and then compare this in order to get or make some sensitivity tests of the model and improve the quality at the end or find out what really fosters the quality of prediction. And I think it's also going to be quite interesting. I really want to see in this term if we now take the same model for a different federal state,
first of all, does it work the same way and then if we move to another country, of course, to see are there different texts. So in order to develop a European indicator set basically for socioeconomic, would we need to have like models with validation for each
or for what resolution for each country, for each federal state or could we make them robust enough with using one model over more countries or does the cultural differences reflect upon the text related to a certain age structure.
So I think I'm really looking forward to see what's coming up there. Thank you. Any other input from the audience? Okay, so just to close the session and before the coffee break I would like just to make some final remark. Again, coming back to the keynote given yesterday by our GEO president where
she basically presented OSM data as the complement to the open source software. We are actually speaking about these two different sides of the same thing and I would like to thank Maria and Andrea to put actually a track
on, I mean a small session on OpenStreetMap in this academic track. I think this should be the usual and should be actually kept in the future editions of actually any academic track, especially on Phosphor-G. I've seen that as mentioned in the first presentation and as really shown by the second and the third, OSM is really strictly connected to open source.
So whenever you want to have some tools to process or to extract something from OSM you need actually to use Phosphor-G. So we are different communities but basically we are the same community. Just to close, again I will invite everyone to attend the academic track at State of the Map conference that will take place in Heidelberg, Germany, Baden-Württemberg.
In three or four weeks, so 21st to 23rd of September. Again, there will be kind of a similar conference to this one, so a general track and then one day of academic track where we will have similar presentations to this one. So you can see really the potential of OSM also for scientific academic applications.
So thank you to the speakers and to the audience for this very successful session and wish you all a nice coffee break and nice afternoon and evening. Thank you.