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

Keynote III - Geospatial Analytics in Risk Management

00:00

Formale Metadaten

Titel
Keynote III - Geospatial Analytics in Risk Management
Untertitel
Geospatial Analytics in Risk Management
Serientitel
Teil
65
Anzahl der Teile
193
Autor
Lizenz
CC-Namensnennung 3.0 Deutschland:
Sie dürfen das Werk bzw. den Inhalt zu jedem legalen Zweck nutzen, verändern und in unveränderter oder veränderter Form vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen.
Identifikatoren
Herausgeber
Erscheinungsjahr
Sprache

Inhaltliche Metadaten

Fachgebiet
Genre
Abstract
At the Joint Research Centre (JRC), scientists involved in maritime situational awareness are confronted with a growing volume of data. Every day millions of ship positions from terrestrial and satellite receivers are gathered globally and in real-time, as well as optical and radar Earth Observation images, leading to a significant variety of data. To support the researchers, policy makers and operational authorities in their activities a analysis platform with WebGIS functionality has been developed with the aim of turning data into valuable information and demonstrating pre-operational tools for maritime awareness. The platform is mostly based on FOSS software and consists of a front-end visualization tool and a back-end analysis engine. Fusion algorithms provide the ability to integrate data from multiple sources on the fly. A series of tools provide predictive analysis, activity mapping, anomaly detection, and cross disciplinary information, to support maritime security and safety and to improve marine knowledge. The web application is developed using open source programming languages (e.g. Javascript, Python), frameworks (e.g. Django, Geoserver), and interchange data format (JSON) to enable researchers to seamlessly integrate ad hoc algorithms developed in scientific languages (e.g. R, Matlab). A case study will be presented, showing examples of how the WebGIS architecture can provide visualisation and analysis tools to support decision makers and scientific and operational actors in the fields of fisheries science, maritime spatial planning, and maritime surveillance.
78
Vorschaubild
51:51
154
Vorschaubild
35:04
EichtheorieElementargeometrieDreiVollständiger VerbandInnerer PunktAnalysisZeitzoneNuklearer RaumLokales MinimumEreignishorizontVarianzGeradeNatürliche ZahlGesetz <Physik>SpywareNeuronales NetzAnalysisCASE <Informatik>EinfügungsdämpfungInformationMultiplikationsoperatorNatürliche ZahlInstantiierungGüte der AnpassungKategorie <Mathematik>KatastrophentheorieEreignishorizontGeradeGarbentheorieDifferenteGewicht <Ausgleichsrechnung>Schnitt <Mathematik>URLMeterGebäude <Mathematik>PunktProdukt <Mathematik>GraphfärbungNetzwerkverwaltungCybersexAnalytische MengeTermTypentheorieParametrische ErregungKreisflächeInhalt <Mathematik>Ordnung <Mathematik>DoS-AttackeBitPerspektiveWort <Informatik>Virtuelle MaschineAggregatzustandNuklearer RaumFakultät <Mathematik>Interrupt <Informatik>ParametersystemLeistung <Physik>CodierungSoftwaretestQuick-SortCodeSchwach besetzte MatrixPhysikalisches SystemPhysikalismusFrequenzMakrobefehlSoundverarbeitungVerschiebungsoperatorTeilbarkeitOffice-PaketMAPLesezeichen <Internet>Faktor <Algebra>Bildgebendes VerfahrenFlächeninhaltInteraktives FernsehenWurzel <Mathematik>ComputerarchitekturComputeranimation
TypentheorieMetropolitan area networkEreignishorizontVierEinfügungsdämpfungPersonal Area NetworkSpezielle unitäre GruppeElementargeometrieHasard <Digitaltechnik>VorgehensmodellKontrollstrukturARM <Computerarchitektur>DISMANatürliche ZahlKlasse <Mathematik>Computerunterstützte ÜbersetzungPhysikalisches SystemInformationDatentypCase-ModdingDreiNeunzehnDatenmodellVerhandlungs-InformationssystemSystemprogrammierungKlassische PhysikAnalysisGerade ZahlSummierbarkeitBitrateFaktorenanalyseEndliche ModelltheorieMapping <Computergraphik>Computerunterstützte ÜbersetzungURLMAPSichtenkonzeptCASE <Informatik>FlächeninhaltMechanismus-Design-TheorieGeradeProzess <Informatik>PaarvergleichKartesische KoordinatenAeroelastizitätEreignishorizontPunktSchnitt <Mathematik>SoftwaretestNatürliche ZahlBitHasard <Digitaltechnik>KreisflächeZellularer AutomatRelativitätstheorieMeterNetzwerkverwaltungEin-AusgabeInformationAnalysisBitrateMultiplikationsoperatorKlassische PhysikTermMailing-ListeAnalytische MengeUnrundheitBeobachtungsstudieDesign by ContractGradientSummierbarkeitKatastrophentheorieTelekommunikationSkalarproduktClientEinflussgrößeZahlzeichenCybersexDatenfeldAutomatische HandlungsplanungEinfache GenauigkeitGewicht <Ausgleichsrechnung>Puffer <Netzplantechnik>SystemidentifikationKlasse <Mathematik>EinfügungsdämpfungComputeranimation
DatenmodellComputerunterstützte ÜbersetzungBitrateNatürliche ZahlHasard <Digitaltechnik>KontrollstrukturSummierbarkeitFünfRuhmasseMessage-PassingSoziale SoftwareNetzwerkverwaltungEuler-WinkelNichtlineares ZuordnungsproblemMehragentensystemCOMTLSEinfügungsdämpfungSchätzungSatellitensystemSpezialrechnerAnalysisDatenflussWellenlehreGradientInformationsspeicherungTypentheorieBinärdatenKreisbewegungProzessautomationMAPFlächeninhaltVerhandlungs-InformationssystemCloud ComputingTaupunktVorgehensmodellMathematikNotepad-ComputerAnalytische MengeNatürliche SpracheSichtenkonzeptEreignishorizontInformationKartesische KoordinatenInstantiierungCASE <Informatik>StabAnalysisBilderkennungBildschirmmaskeSatellitensystemElektronischer ProgrammführerPunktInformationsspeicherungTermMultiplikationsoperatorKatastrophentheorieMaßerweiterungGraphfärbungApp <Programm>MereologieMessage-PassingExpertensystemDichte <Physik>TopologieMapping <Computergraphik>Web-SeiteMAPEDV-BeratungSprachsyntheseQuadratzahlFlächeninhaltAttributierte GrammatikGamecontrollerArithmetisches MittelNetzwerkverwaltungPhysikalisches SystemWort <Informatik>StellenringSummierbarkeitStatistische HypotheseTwitter <Softwareplattform>Dienst <Informatik>HypermediaBildgebendes VerfahrenDoS-AttackeGebäude <Mathematik>Gesetz <Physik>Total <Mathematik>Natürliche SpracheMetropolitan area networkZahlenbereichWeb SiteRechenwerkNatürliche ZahlZeitzoneProzess <Informatik>URLSkalarproduktDifferenteCodierungSoftwareFastringElementargeometrieZweiPolygonComputerunterstützte ÜbersetzungRechter WinkelComputeranimation
SimulationNeunzehnDreiInformationAnalysisComputeranimation
Transkript: English(automatisch erzeugt)
We start with our next keynote speaker. Mr. Thomas Zewek came from Munich. He's from Munich, RE, a reinsurance,
one of the biggest reinsurances in the world. I think he also studied at least two years in Bonn. So it's kind of coming back to his home, and he will talk about geospatial analytics and risk management.
Thanks a lot, Till. Thanks for the invitation. And thank you very much for the fantastic conference so far and for the fantastic location.
It's really great. So thanks for that. Geospatial analytics and risk management. I want to talk a little bit about some aspects of this topic from our perspective, from the perspective of the reinsurance industry. And to start, a few words about what reinsurance or insurance
is or does, actually. So I think everyone more or less knows what an insurance is. It's pretty straightforward. Everyone has one, probably. And insurance ensures private people or companies
against different risks, against different dangers. And the reinsurance, to make it simple, ensures the insurers. So what do we insure? What different types of things we have? And what everybody probably knows or comes in mind is property, what we call property insurance.
So that is if you insure your building against storm damage, earthquake damage, flood damage, for instance. But it's not only the building itself. Even more important often is the content. So the building might have a certain value. But of course, if you think of a semiconductor factory,
for instance, then the content, the machines within this building, their value is much, much higher. And another thing which is also really interesting and important is what we call business interruption. So this is if, for instance, a street gets damaged,
that might not be a big damage simply to repair the street. But the time it costs to repair the street, which might cause that a certain company cannot deliver their goods to the port nearby, and so they cannot make business for a week or two, this damage might be much, much higher than the property
damage itself. Other lines of business, as we call them, are casualty, for instance, marine. So this is marinas, for instance, if you think of a big cargo boat. But it's not only the boat itself. What is mostly of much more value
is the cargo on this boat. And that can, nowadays, sum up until $500 million, for instance. These ships can transport about 20,000 containers nowadays. And if such a ship sinks, then it's really getting into a huge damage.
Other things would be engineering. So this is while a bridge or a big building is built. Agro, so this is crop failure, for instance. So weather risks are insured here. Cyber risks are something new. So there are a lot of different lines of business which play a role in insurance and reinsurance,
and not only what we think of in the first term of a building. Oh, that's interesting. So we have here some, what is it called, things which are not displaying on your computer.
OK. Another difference, something which I want to mention, is what is also playing a role in our terms is the difference between a single risk and a portfolio risk. We call it faculty or treaty.
A faculty risk is one single risk. You know exactly where it is. It is a big risk where you have the exact location, a nuclear plant, for instance. And you can really do a detailed risk analysis for this one single risk. A treaty on the other side, that's a huge portfolio of risks. So that would be all houses of aliens in Bavaria
insured against storm, for instance. Then you have thousands or 10,000 of single risks. And you do not analyze each and every single risk, but you look at it as a whole. And you often don't even have the information where your risk exactly is located.
You might only get the information, OK, in this postcode zone, we have 10 millions. In this postcode zone, we have 5 million, et cetera, et cetera. Of course, this makes it difficult if you want to model this risk against flooding, for instance. Because in flood, you should have an exact location
because it might make a difference if it's exactly at the river or 100 meters away. So you have to think about ways how you can clever handle this lack of information, actually, to finally come to a good risk management.
And something special, which I wanted to mention, because it's also interesting from a natural catastrophe point of view, is what we call trigger products or parametric insurance. In this case, it works a little bit different
than what we normally know from insurances. Normally, you have a damage. You send this to your insurance company, and the insurance company hopefully pays for it. In the trigger products, the insurance company does not pay for the damage. It pays when a certain event occurs.
So when a certain trigger or a certain physical parameter, as we call it, is triggered by an event. So for instance, if the speed of a storm exceeds a certain, whatever, 200 miles per hour, then the insured gets the insurance payment. The advantage is that as an insured person
or as an insured state or nation, I don't have to wait until all the damages are categorized, but I immediately get my insurance payment as soon as the event is triggered.
Sorry, I got a cold a little bit, so I have to drink every now and then. So talking about risks and disasters in reinsurance or in the insurance business, we mainly make a difference between net cut and man-made cut, so natural catastrophes and man-made catastrophes.
And in the natural catastrophes, we have some main sections that we distinct, meteorological events that would be all kinds of storms, more or less, geophysical events, that's earthquakes, tsunamis, but also volcanic eruptions, for instance, hydrological events that would be mainly floods.
And then we have other climatological events, as we call them, for instance, wildfire. On this map, you get a little overview over the events of 2015. We have what we call a net cut service,
so that's a few guys in our company who collect all the events, natural catastrophe events over the year, and try to collect all information about these events, like what was the loss, what was the insured loss, how many people were,
how many people, how many fatalities do we have, et cetera. And here you can see those events identified by their color in terms of the type of events and identified by the size of the circle in order to display the size of the event.
And I want to pick out a few just to point out on some things which are important in this case. For instance, in last April, we
had this earthquake in Nepal. 9,000 people dead. Overall loss is $4.8 billion, so pretty high. Insured loss is only $210 million. And if you compare this to the winter storm event in the USA and Canada, for instance,
then we have only a few fatalities there. We have pretty high overall losses, but not as high as in Nepal, for instance, but much, much higher insured losses. And this is, of course, something which is for us pretty important or which makes a big difference
for the insurance company. What is a big catastrophe in the human terms, and which is a big event, might not necessarily be a big event for the insurance companies and the other way around. Something which is not that big, if you think of it in terms of catastrophes, might be a huge damage for the insurance companies.
The man-made catastrophes, I don't have such nice pictures for them, unfortunately. So just a few bullet points here. Fire and explosion would be a major reason for man-made cuts. Terrorism, of course, cyber attacks, just to name a few.
And here's a little overview about the events in 2015. And you can see how high the insured losses have been, what was the costliest event, and in comparison to that, the most costliest event ever.
To point out the man-made cuts event in 2015, I will come to this later again. It was the explosion in the port of Tianjin and I don't know how many people of you even have heard of it or remember it,
because it was not that big in the newspapers. It was a little bit, but not that big. But it was a huge damage for the insurance industry. Also, it was really located pretty much on one place. The damage is still not sure how much it really is, but it's about $3 billion US.
So that's a really big damage. So to come to our topics, more or less, what can we do with geospatial analytics in risk management or risk and disaster management? What is our role? What do we do in our company? Where does the geospatial aspect come into all of that?
And I just listed a few fields where we do geospatial analytics. Risk assessment is one of them. That's a qualitative approach. So that's really about getting an idea where your risks are,
where your exposure is. That's what we call the exposure, so all the single or the aggregated information about what we ensure. Where is your exposure and where is it in relation to the hazards, of course? So that you get an idea where it is at all.
And you would be surprised how many people in the insurance industry don't have any clue where their exposure is. So it's really surprising. Risk modeling and pricing, that would be then the more quantitative approach. So there, it's really about getting a price.
I mean, finally, you need to know how much do I have to earn over a certain time. You should not need to pay more than you earn, because then you get into trouble. So over a longer time, it has to pay off.
So you need to model your risk. You need to know, OK, in this area, every 50 years, I have a total damage. And then you can more or less calculate how much you should get for this risk. Accumulation control, that's a pretty important issue. And that's more or less where I'm also working in the company.
You need to avoid that all of your insured exposure is in one place. Because if you have everything in New Orleans, and then you have Katrina, then you are in trouble, then you are bankrupt. So you need to know where your exposure is, and you have to make sure that not everything is
concentrated in one single point. And finally, disaster management, that is more or less when it all happened, to see as fast as possible what happened, what was damaged, and how much to estimate, more or less,
how much it will cost the company. OK, risk assessment, net cut. One example here is natural hazard maps. It's pretty simple, actually. I mean, it's simply a map, a worldwide map, which shows the hazard in a classified way. In this case, it's earthquake rates or earthquake classes
from yellow to red. Pretty simple. And displayed on top of it in the green dots is the exposure. And as you can imagine, as people who have to do with cartography and with maps, the darker green the dot is, the larger the dot is,
the larger is the total sum insured. So pretty easy, but this is something which a lot of people in the insurance companies don't know. They haven't ever seen their exposure. And it's really surprising, because if you see it like this, you immediately get an idea where your exposure is
and what kind of problems you might run into. Another example, and this is not really catastrophe, it's more damage, but I wanted to integrate it because there were a few interesting steps in it from the geospatial point of view.
What was it all about? It was about subsea cable and pipeline damage. So when you have an offshore wind park or if you want to connect an island or two continents by a subsea cable, you somehow put it on the ground of the sea. And if you have a damage there, then it gets expensive.
So why is there damage? One of the few most reasons are damage by anchoring or ground net fishing. And what we did was we created a risk map. Pretty simple one, I have to say. But we created a risk map displaying this risk
in the area for ground net fishing and anchoring damage. So we took information from the AIS, the Automatic Identification System, which every ship above a certain size has, displaying its location, its status, its speed, et cetera.
And we, well, I would say we crunched this data. We did a lot of editing there and analysis, also by integrating some Hadoop mechanisms to, well, 40 million is not that big data, but already a little bit. So there was also a test case, a POC,
to see what we can do with big data in GIS. And finally, we came up with a risk map displaying the most affected area. So it was only for a certain study area in the North and Baltic Sea.
We came up with this risk map displaying areas of higher danger for subsea cables and pipelines. Again, that is a pretty easy thing. It's not really rocket science. But it is an important information for the people who do the contracts.
That's what we call the underwriters. So it was an important information for the underwriters in their risk assessment and as well as in their communications with the clients, with the people who they ensure, because they could easily show them, OK, this cable there, we have certain problems in these areas.
So you should have some measures against that. Or we need to have a higher price or whatever. And we built an application around this risk map where the underwriter can simply draw in his new planned pipeline or subsea cable and get something
like you can see here, this buffered line on the map indicating the risk of the subsea cable. Another example here for a man-made cat would be conflagration analysis. So if we want to ensure or reinsure
residential buildings or single private houses against fire and explosion, we also need to know what is in the circle of 200 meters to see the total risk which might be there in case
of an explosion or fire. And I mean, this is something really classical. It's a classical GIS analysis. It's simply a buffer and to summarize on this area. And that's it. So again, no rocket science, but an important input for the risk assessment of this exposure.
OK, I will get a little bit faster with this because it's more or less as what you already saw with the hazard maps.
In this case, I'm talking about the rate maps. It's more or less the same. But in this case, it's really about the price. So the hazard maps, it's really about risk assessment. So getting an idea, a feeling, a qualitative approach for the risk. In this case, there would be some rate behind it. So there would be some information.
OK, we will have total damage here every 100 years. So you have to take this in this price. Same with man-made cat. In this case, would be terrorism exposure. So we have information where is a higher and where is a lower terrorism risk. Accumulation control, I mentioned
that at the beginning. Pretty important. Well, you can have your portfolio either this way or either this way. And if you display it on a map, you see it immediately. If you display it in an Excel sheet, you don't have a clue at all. So those two examples are exactly the same portfolio
in terms of total sum insured, in terms of number of risks, in terms of whatever additional attributes might be there. But from the location point, it's a big difference. And as I mentioned, if you have a Katrina and you have a portfolio like that one on the right side,
then you might be in trouble. Disaster management, it's about getting an idea or an information where the damage happened and how big it is. And one thing we did there or we tried there was by a master thesis to get a damage footprint
out of social media data. And my colleague, Florian Uslander, he identified or he did the thesis and he tried to find out for a certain event, can he also get this event information
out of Twitter messages, out of tweets? Can he identify that there happened something? Can he identify the damage? And can he also spatially identify it? So can he identify the region?
And while it's just a few images here, you can see some density maps on the left side. And you can see the tweets on the lower right side. And this is about an event, the earthquake Napa event two years ago,
which had happened in the middle of the night, local time. And the interesting thing is it happened about 3.20 a.m. and people fall out of bed and the first thing they do is send a tweet. So it's quite surprising. So actually it works to a certain extent at least.
Here in this map you can see on the green shaded colors that would be the event footprint as we call it. So that would be what we got from the USGS and the red and blue squares or dots are indicating what we got from the tweets
and do the cold spot and hot spot analysis. So the red spots would be what we would have identified. And it's not so bad actually, but of course it's not really exact. It's not an exact hit and you have certain problems there.
Tweets are often not geo-referenced, especially if you, for instance, if you go to Germany then you suddenly see that there are no geo-referenced tweets anymore. You might have problems that in areas where there is no one living, then of course you don't get information. Although there might be something
which is interesting for you as an insurer. If you think of industries for instance and the earthquake happens in the middle of the night, you won't get any tweets there. But you might have damage. So still a lot to do but an interesting approach. Another example, I already mentioned it,
the Port of Tianjin explosion. In this case we tried something with satellite image analysis. So just a few facts about this explosion. The Port of Tianjin, no one ever heard of it probably, is one of the largest ports worldwide. And there happened an explosion at a chemical storage site
and in the first place no one knew what exactly it was about and if it, well not so, not such a big thing. And after a few days, slowly there came the surprising information that this is really a big thing for the insurance industry, a lot of damage.
And the question was how do we get the information, what really happened and how much is damaged? Because the access to the area is restricted of course. It's restricted for health reasons of course, but perhaps in this case also for political reasons.
So it's really difficult to get information. And what the colleagues did together with the GAF in Munich was they analyzed the satellite image simply by manual analysis. So nothing fancy here again, simply by manual analysis.
They looked at the image and tried to find out how high the damage is on the one side and then of course on the other side, what was there? What was in this area? Was this a container area? Was this an area where cars were located, et cetera?
And from this information, they slowly got an idea about what the damage is and what exactly happened. And of course this is also something where we see a certain use case for Copernicus information, for Copernicus data, especially from the EMS,
from the emergency mapping service, where we get information in this case from a certain flood event near Skopje, which might be of use for this damage analysis. Okay, so to sum it all up,
some challenges for the future, for the insurance industry, and that's also something then which in the second step concerns us as a geospatial department. The values of the buildings or of the, yeah, whatever it is, the values increase more and more
and they are more and more concentrated in certain areas. So we need more and more risk transparency to really get a good idea about what we are actually insuring and where it is. The costs of natural catastrophes get higher
and they are more frequent and we will have probably more natural catastrophes or at least extreme events due to the climate change. And we have to get prepared for that. And what are some requirements for geospatial analytics?
And they partly come out of that what I just mentioned, but partly they are general requirements and I would say they are not only general requirements in our company, but they are general requirements in the business at all. What we need is geospatial tools and apps for the business people.
We don't need any tools and apps for us as experts. We know how to deal with them, but we need them for the people who have no clue what geospatial is. And it's really important for us in our department, but probably for everyone who wants to do consulting
and GIS and geospatial stuff. It's really important to speak the business language, to know what the people do. What is their business? How do they earn the money? And then to understand what we could do with our geospatial knowledge and our geospatial tools to help them.
Those tools we have on those software and those systems, they have to seamlessly integrate into the workflows and processes. So it's pretty seldom that you have an application where you have a map and all this nice stuff we are talking about here so often. Normally, you have a system which does a certain business thing,
I would call it like this. So in our case, for instance, you try to price the risk. You do all the wording, so the treaty stuff and all that kind of things. And within the system, you need some geo information.
You need some geo coding. You need some point and polygon analysis to put it that way. And the user, he doesn't care and he doesn't know. So our systems and our tools have to integrate in this often already existing systems.
And the third point which I think that's more important is that we get more and more data, more detailed data, more up-to-date data. So in this case, it's really about that we have to handle more data. I mean, just put down the term big data here. It's not necessarily big data, but it's more data
and performance is an important issue for us. Okay, I think that was it. A few ideas, a few informations about what we do in risk assessment or risk analysis in the insurance business. Thanks a lot.