1. Introduction
Each year, natural or man-made disasters damage and destroy our environment, goods and infrastructures, cause fatalities, pose severe implications to our economy and, consequently, lower our quality of life. According to the European Environmental Agency (EEA) [
1], economic losses up to 520 billion Euros occurred in the 32 EEA member states only related to weather and climate extremes between 1980 and 2020. While the majority of losses and damages are related to single weather and climate disasters (compare widespread river floods, winter storms or the heatwave in 2003 causing most fatalities in this period), WMO [
2] and Munich Re [
3] also report that the total number of weather- and climate-related disasters as well as their associated damages in general have increased over the past 50 years. Regarding the increase in loss and damage numbers, it has to be noted that socio-economic or demographic factors, increased vulnerability and exposed assets, as well as improved reporting or counting issues, may also have to be considered (e.g., compare [
4,
5,
6]). Climate change, nevertheless, is one driver in this development, and a warmer global climate system will lead to intensified extreme climate conditions [
7]. It is, therefore, not surprising that disasters and their impacts are ranked as top risks for our societies and economic systems (compare World Economic Forum [
8]).
As a consequence, local, national, European and international developments, standards and regulations have evolved and have been agreed upon to support a resilient society (compare [
9,
10]). Besides others, EU Decision No 1313/2013/EU on a Union Civil Protection Mechanism (UCPM,
https://eur-lex.europa.eu/eli/dec/2013/1313/oj, accessed on 18 April 2022) calls participating states to periodically develop national risk assessments, which should address all relevant issues such as the EU Floods Directive, EU Solidarity Fund or EU strategy on adaptation to climate change. At the international scale, the UNDRR Sendai Framework for Disaster Risk Reduction [
11] defined four priorities and set associated global targets to reduce losses and damages from disasters by 2030. Independent of framework and scope, disaster risk management (DRM) and disaster risk reduction (DRR) activities focus on measures to identify, monitor, assess and govern risks. For their success, they heavily rely on data and observations. However, even data-rich countries, such as Austria, are still lacking a consistent event-based loss and damage database that covers multiple hazards. Although there are high-quality databases, they are often not intercomparable. Amongst others, this is due to various national and federal mandates, diversity of intentions and various standards. International systems such as EM-DAT [
12] or NatCatService [
13] also integrate and indicate Austrian information, but this information is often insufficient for robust multi hazard assessments at the local/regional level or not publicly available [
14,
15].
In this study, we assess the feasibility and benefit to build a consistent Austrian multi-hazard loss and damage database by using existing data and building on established national and international standards and technologies. The resulting demonstrator should support the national risk assessment, the Sendai Framework Monitor (
https://sendaimonitor.undrr.org/, accessed on 22 April 2022) and add value to the federal provinces’ disaster management duties.
Section 2 briefly depicts the framework of the study.
Section 3 focuses on the data,
Section 4 on the data harmonization process and event identification.
Section 5 shows the results and
Section 6 discusses these results and derives recommendations.
3. Data
3.1. Data Description
We collected a variety of different datasets in order to develop, test and apply concepts for transforming, harmonizing and merging data on natural event-based disasters from heterogeneous sources. The collected datasets differ, amongst others, in (i) the level of detail of spatial and temporal resolution, (ii) temporal coverage, (iii) types of hazards covered, (iv) documentation focus (events, processes, damages, losses, etc.) and (iv) the vocabulary used for hazards and damaged elements. Regarding data providers, we focused on those with internal quality check procedures, existing metadata and long-term availability. In the following, the selected datasets are described in detail.
The GIS-based data management system GEORIOS of the Geological Survey of Austria is one of the main Austrian information sources on mass movements (see, e.g., [
17]). Documented events are point-located. Date and time of the events are recorded in text format, with varying levels of detail (from to-the-hour to to-the-year) and certainty. Mass movement processes are classified in three different levels of hierarchy. Any damage caused (e.g., human loss, property damage) is documented as free text, which, however, does not contain any monetary loss volumes. For the period 2005 to 2018, the dataset contains about 6400 mass movement events in Lower Austria and Styria.
The torrent and avalanche register (WLK) of the Austrian Forest Engineering Service in Torrent and Avalanche Control records events concerning flood and sediment disasters, snow avalanches, landslides and rockfalls (see, e.g., [
18]). Documented events are point-located. Their date, time and duration are recorded in varying levels of detail. The MAXO code (M = measured, A = assumed/estimated, X = still unclear, O = not determinable) informs about the reliability of single data entries. Besides very detailed event documentations and process descriptions, the dataset contains loss and damage information for about 30% of the recorded events. This information may include human losses, damaged property and reproduction costs of damaged torrent and avalanche barriers. For the period 2005 to 2018, the dataset on Lower Austria and Styria contains almost 1900 hydrological and mass movement events and about 170 snow avalanches.
Since 2013, the Austrian Federal Water Engineering Administration has been documenting flood events in their flood database HWFDB (see, e.g., [
19]). The level of detail of the documentation varies depending on the extent of the event. Events are located by means of points and/or lines. Date, time and duration are recorded in varying degrees of detail (from to-the-minute to to-the-day) and reliability, the latter indicated by the MAXO code. Documented information of damages includes rough estimates of the total direct economic loss and reproduction costs of damaged flood protection structures of the Austrian Federal Water Engineering Administration. For the period 2005 to 2018, the dataset contains 183 flood events in Lower Austria and Styria.
The Austrian meteorological and geophysical service collects data on damaging extreme weather events based on media coverage in their VIOLA database. VIOLA includes information on (i) short-term events such as heavy rain, hail and windstorms, (ii) damage-causing events of longer duration such as droughts, continuous rain or periods of heat and cold, and (iii) events indirectly caused by extreme weather such as floods due to continuous rainfall, debris flows due to heavy rainfall, or avalanches due to intense snowfall. The spatial resolution ranges from municipality to state level. Information about start and end dates include uncertainty spans. The classification of the damage-causing process or hazard consists of two hierarchy levels. In addition to information on meteorological variables, VIOLA also contains information on the type and extent of damage to property, persons and animals—in some cases including rough monetary loss estimates—as well as information on the emergency forces involved. For the period 2005 to 2018, the dataset encompasses about 1200 events for Lower Austria and Styria.
Since June 2007, the Lower Austrian Fire Brigade Association has been recording its operations. Recorded information comprises, amongst others, the start and end date of an operation, point locations with varying degrees of precision and the cause or type of operation (e.g., required pumping out, thunderstorm, flood, storm damage, etc.). For the period 2005 to 2018, the dataset contains about 34,500 operations in Lower Austria.
In Austria, the Provincial Administrations are in charge of processing compensation payments for non-insurable extraordinary losses in the event of natural disasters, of which parts are refunded by the national disaster fund. For this purpose, they are collecting and assessing the reproduction costs of damages in the property of natural and legal persons, municipalities and states. Different departments, depending on the type of property, collect the data. In total, four different datasets were provided. The spatial resolution of provided datasets refers to the municipal level. The temporal resolution of the date of loss ranges from to-the-day to to-the-year. Besides the cause of loss, the datasets include information about the damaged object and the amount of granted compensation. For the period 2005 to 2018, the provided datasets contain about 103,800 loss entries.
Table 1 summarizes all included datasets with their relevant features for the CESARE system.
3.2. Data Management and Protection
Acknowledging the role and sensitivity of loss and damage data, we developed a data management plan and data exchange protocol in order to meet all providers’ requirements as well as data protection regulations. In general, two legal frameworks were considered in detail: ZAMG as maintainer of the database and its basic legal mandate as well as the general data protection regulation (GDPR). ZAMG’s basic tasks are defined within the Forschungsorganisationsgesetz (FOG), article 22, which covers the collection, storage, analysis and management of meteorological as well as geophysical data for Austria. Moreover, supporting the national crises management and mandated international DRM and DRR organizations, ZAMG is also allowed to hold relevant information to govern natural and man-made disasters.
Referring to GDPR, we based all our data management activities on §§ 2d Abs 2 lit c und 2f FOG, § 7 DSG (Art. 89 DSGVO), Art. 6 Abs. 1 lit e DSGVO and especially valued the principle of data minimization. The provided data are not forwarded to third parties or published, and only anonymized data are applied. Concerning data security, all data processing is performed exclusively within the European Union.
5. Results
In the following, we will exemplarily highlight our procedure for one selected disastrous event in Styria, then summarize our main findings including the presentation of the CESARE web interface.
In the second week of September 2014, a prominent low-pressure vortex with a core over Slovenia caused high precipitation totals, especially in the peripheral mountains of western Styria and in parts of eastern Styria. The situation was intensified by local thunderstorms and heavy rainfall events, especially on 13 September 2014. These meteorological conditions led to mass movements as well as to floods at numerous watercourses in Styria on 13–14 September, with the Sulm and Saggau rivers and their tributaries being most severely affected.
The database GEORIOS reports 54 point-located mass movement events in Styria for the considered period. The textual damage description does not contain references to human losses but mentions multiple damaged roads and several affected buildings.
VIOLA also reports several mass movement processes in various municipalities, which may coincide with some of the mass movements documented in GEORIOS. Firefighters had to turn out, but there is no information available about the operation costs. Textual damage descriptions refer to affected buildings, roads and public infrastructure. In addition, human loss in the form of about ten evacuees is documented.
The HWFDB dataset documents 17 line-located flood events in the considered period, causing damage to flood protection infrastructure in the amount of EUR 2.4 million.
The WLK dataset reports another nine point-located flood events in the considered period. There is no human loss or loss in protection property (torrent and avalanche protection structures) reported in the WLK dataset with respect to these nine flood events.
The documentations of the Provincial Administration on extraordinary losses in the event of natural disasters include 1174 loss records for the considered period that report non-insured losses (reproduction costs) of a total amount of EUR 9.3 million due to flooding and mass movements. EUR 7.9 million stem from the data set on losses in the property of natural and legal persons and another EUR 1.4 million from the data set on losses in the property of the Styrian state. The affected property includes buildings and facilities, transport and pipe infrastructure as well as agricultural and forest goods.
Figure 5 summarizes the harmonized information extracted from the available datasets about the composite event with the detailed disaster records given above. The harmonized total documented loss of property sums up to EUR 11.7 million and stems from three complementary datasets.
Figure 5 also shows how this total loss splits to single municipalities. Four different types of property were affected: buildings and facilities (49% of total loss), transport and pipe infrastructure (7%), agriculture and forestry (23%) and protective infrastructure (20%). In addition, the dataset VIOLA reports on human loss in the extent of about ten evacuees. The datasets WLK, GEORIOS and HWFDB provide additional information about the point or line location of single events or processes.
In total, we analyzed over 140,000 event and loss descriptions from the various available data sources. Applying our proposed harmonization procedure together with the event-defining algorithm, we derived 63,972 events (52,579 composites, as shown in
Figure 6) with a total direct property loss of EUR 1125 million (in current prices), total (emergency) operation costs of EUR 21 million (in current prices) and 20,000 directly affected people (dead, injured or evacuated) between 2005 and 2018 in the considered Austrian provinces. Although we do not claim for any completeness, as still important data sources such as from insurance companies are missing, the harmonized numbers give the most robust and detailed picture on loss and damage information for the considered hazards currently available for the study regions. For the same period and the selected hazards, EM-DAT features 16 database entries in total for entire Austria with an undefined number of events on the provincial scale, damages of EUR 2500 million (only uninsured damages taken into account) and about 4000 people affected. The DRMKC Risk Data Hub provides relative loss numbers between 0–1‰ GDP for buildings and critical services for Lower Austria and Styria for a comparable period with no information about the number of the included events. Regarding affected people, the Risk Data Hub shows 0–1 people/100,000 population. However, even considering national databases, we could show the importance of considering all sources of information to obtain the correct picture. A systematic omission of single data sources may result in significant underestimation of loss and damage accountings and a wrong geographical representation.
All our provided loss numbers are corrected or quality-checked for multiple counting and overlapping of different sources. Based on our decision tree (
Figure 3), we derived a prioritization of data sources depending on the affected element considered. In case of overlapping human loss information, our findings suggest prioritizing VIOLA over WLK and WLK over GEORIOS. Although we do not consider VIOLA information to be more reliable than the information in the WLK and GEORIOS datasets, the records in VIOLA usually refer to larger spatial units and may thus include additional human loss information outside the specifically overlapping locations, which are not recorded in the other two datasets. WLK is prioritized over GEORIOS, since in GEORIOS information on the specific number of people affected is often lacking. Information on property loss can be found in the documentations of the Provincial Administrations, who collect loss data in the course of processing compensation payments, and in the VIOLA, WLK and HWFDB datasets. Some of the information is complementary, but often, there is a significant amount of potential overlap. Due to the large risk of multiple counting, information from the documentations of the Provincial Administrations is used as the main source on property loss. It includes losses that are either assessed by experts and damage assessment commissions or proved by means of invoices. Hence, the information is regarded as more reliable than the media-based property loss information in VIOLA or the rough estimates on total losses in the WLK and HWFDB datasets. Information on losses in protection infrastructure, on the other hand, is taken from the WLK and HWFDB datasets, as it largely complements the information on property loss included in the documentations of the Provincial Administrations. Information on the monetary extent of (emergency) operation costs can only be found in the documentations of the Provincial Administrations. There are no overlaps with other currently available datasets.
To access all our results, we implemented a web-GIS portal that enables the temporal as well as geographical aggregation, filtering, statistical analysis and visualization of all integrated data. Data can be filtered by date, type and affected element. In addition, we implemented a list pre-selects single events based on frequently asked disasters. The service itself consists of two modules. A map view shows loss and damage sums per administrative division as well as the point-locations of damages, if available (see
Figure 6), and offers overlays of additional data for forensics (e.g., derived weather indices or EO data, compare
Section 5).
The second module consists of three dashboards that display more in-depth information on how loss and damage sums split over assets, hazards and administrative divisions (one dashboard for monetary losses and one for people affected; compare
Figure 7 and
Figure 8). The third dashboard plots the losses and damages against Sendai indicators for global targets A, B and C and shows a temporal evolution of the respective damages and losses (compare
Figure 9).
Concerning our investigated hazards, our harmonized test data, for instance, show that the majority (about 48%) of losses and damages are due to floods, followed by mass movements and storms. Roads are the most affected property type, followed by buildings and facilities, agricultural and forestry areas. Between 2005 and 2018, the total damage numbers show no trend (keeping in mind that a sample size of 14 years only allows for limited validity in trend analysis)
Besides our efforts to harmonize existing loss and damage data, we also assessed the usability and feasibility of contributing data layers such as EO data and, if suitable, added such information to selected events. Furthermore, we elaborated on the topic of estimating near miss events, which are essential to derive robust statistics on possible future damages, based on weather data.
Figure 10 shows the potential of EO-based event identification. Using the Sen2Cube.at Sentinel 2 data, we could clearly identify spatio-temporal changes semantic concepts that can be referred to single disasters—in this case, a debris flow event. Overall, our analysis yields that especially for damage events with a larger spatial and temporal extent (landslides/debris flows or storm damage in forests), the EO Sentinel-2 time series can help to spatially delineate the events and to analyze their duration which also includes potential recovery. In such cases, (open and free) remote sensing data can provide important complementary information to enrich the database with spatial information or duration of events. For flood events, there are often limitations in optical data regarding cloud cover. Analyses based on Sentinel-1 radar data can be helpful here. A potential extension could be the coupling of the damage database with results of the new Copernicus Global Flood Monitoring (GFM) component, which will provide, in the near future, a continuous monitoring of floods worldwide by immediately processing and analyzing all incoming Sentinel-1 Synthetic Aperture Radar (SAR) satellites. For events with a smaller spatial extent or events that cannot be seen from “above” (e.g., flooded basements) or temporal very limited events (e.g., flash floods), remote sensing (within the evaluated specifications) cannot provide meaningful additional information.
Referring to our HTPs analysis for mass movements in South-Eastern Styria, a list of dates when, from a merely meteorological point of view, a potentially damaging event could have occurred was calculated, related to the total number of days between 2005 and 2018 and averaged over the municipality. This way, areas at potential risk could be derived (see
Figure 11). The darker the shades of red, the greater the proportion of these events. Our results indicated the most potentially affected communities in the northeast of the target region. However, when additionally considering the actual events that occurred in the area (black points) over that time period, the communities with the highest proportions do not match the highest number of actual recorded events. This may be due to already taken mitigation measures or other factors relevant to the occurrence of gravitational mass movements, such as vegetation or terrain characteristics, that are not taken into account in this approach. Further investigations would be needed to assess the contributions of all relevant components and thereafter the real impact of meteorological triggers to loss and damage events.
6. Discussion
Based on the needs and requirements of DRM-relevant governmental stakeholders in Austria as well as national and international regulations and recommendations, we developed and successfully implemented a demonstrator (CESARE system) for an event-based loss and damage accounting and forensics from local to national scale. The demonstrator’s objective was to obtain an as-complete-as-possible picture about the occurrence, frequency and impacts of selected hazards and their composites. Our proposed system builds on existing loss and damage data from various sources and applies a harmonization procedure in order to make them comparable. This way, the primary sources and already established monitoring processes remain unchanged, and a synergetic usage is enabled. Although we integrated many relevant Austrian hazard databases, our system does not claim completeness as our objective was to demonstrate the feasibility of the database itself. Nevertheless, our harmonized test data already exemplarily feature the advantages of a local/regional scope database compared to already existing international repositories or the usage of single national databases. Furthermore, finer-resolved and quality-checked data allow for more robustness concerning geographical representation as well as derived statistics for hazard occurrence, frequency and related impacts. Although a quantitative comparison was outside the scope of this paper, the GAR 2013 report showed that moving from global to national databases increased global loss estimates by 50% [
27]. Similarly, Llasat et al. [
28] highlighted the significant benefit for a common/harmonized flood database focused on the Mediterranean region. We could also show that only considering selected existing national databases may likely result in systematic underestimation of losses and damages due to their specific intentions.
For future application, the CESARE system is conceived in such a way that it allows for the extension of further hazards as well as datasets with minimal technical effort. With our web-GIS portal including a map and dashboard services, the resulting dataset can easily be accessed, aggregated, filtered and analyzed as well as visualized. With such functionalities, typical annual reports, such as the national risk analysis or the Sendai Monitor reporting, are supported. Furthermore, by comparison of the major data sources, accounting biases can be assessed. As harmonized data are built on international standards, the data can, if required, easily be integrated and interlinked with the EU recommended Risk Data Hub by DRMKC and therein hosted data. With new EU taxonomy regulation (
https://eur-lex.europa.eu/eli/reg/2020/852/oj?locale=de, accessed on 28 April 2022), loss and damage data as basis for risk analysis will, in addition, become more important and more valuable.
Besides the advantage of merging data and events, the retrospective harmonization features limitations that need to be considered when interpreting the resulting “harmonized” information.
The risk of overlapping or multiple event entries: The original datasets partly show differences in how they define and classify events as well as uncertainties and inaccuracies with respect to date and location of an event. These differences, uncertainties and inaccuracies complicate the identification of coinciding events across different sources and lead to the risk of overlapping or double counted events in the “harmonized” dataset. The validity of the “harmonized” dataset in terms of the number of events is therefore limited. By contrast, the risk of multiple loss counts in the “harmonized” dataset is low, since the harmonization procedure requires selecting the most reliable and/or comprehensive source in case of potentially overlapping loss data. What is still not implemented, but will be followed up, is the indication how different data sources contribute to harmonized loss and damage numbers.
Imprecise differentiation in source data: Uncertainties in the original datasets due to imprecise differentiation with respect to the type of hazard or the type of affected element are automatically transferred to the “harmonized” dataset.
Differing loss definitions: Especially the documentations of the Provincial Administrations, who collect loss data in the course of processing compensation payments, may differ in what they define as “eligible” loss—i.e., loss entitled to receive allowance—and hence, in the extent of loss documentation. Depending on the state, motor vehicles are for instance eligible for compensation or not. Full harmonization might be difficult, especially if the respective affected object is not reported as stand-alone category, but part of a broader category of affected elements. Wherever retrospective harmonization is not possible, definitional differences are reported in the metadata.
Algorithm for event and composite ID: The algorithm for merging events across different sources and assigning IDs still shows potential for further optimization. In the current version, the grouping is based on the (start) date, geographical proximity and matching of types of natural hazards. Meteorological data, on the other hand, has not yet been integrated into the algorithm. Their inclusion has only been tested in a semi-automated process with a high share of manual work. Especially for larger events that extend over several days, the additional consideration of meteorological data provides noticeable improvements in the resulting event definitions but is quite time-consuming when not fully automated. The future goal is therefore to incorporate meteorological data into the automated algorithm and to refine it further.
Regarding contributing information from EO data (Sentinel 2 in our case), we could demonstrate the overall potential for event identification as well as disaster assessment. However, the applicability depends on the regarded hazard type and the field of the investigation and can be hampered by cloudy conditions, which normally occur along with severe weather conditions. Nevertheless, the functionality itself was well appreciated by the CESARE stakeholders, which indicates that a facilitated access to more EO based disaster related information should be fostered. Concerning our near miss investigations, we can conclude that our meteorological proxy data may give an indication but should be considered with care and only be taken into account in combination with other relevant components.
From a practical point of view, we have seen that a fully automated event identification and data harmonization process is not feasible. Therefore, recommending our semi-automated process, a certain knowledge and experience is needed for a successful implementation as well as for an informed data interpretation. Given also the enormous time resources needed for building relevant data provider networks, the operational maintenance of a quality based national loss and damage database can only be assured by a dedicated and mandated group of experts (also reinforcing [
9]). Besides the obvious administrative and policy implications, the relation with the national weather service also allows the usage of relevant data not only for retrospective analysis, but also for impact-oriented warnings and, therefore, for near-real-time applications.
Future implementations should focus on avalanches, forest fires and earthquakes, as there already exist disciplinary data repositories. Furthermore, the integration of insurance data as well as loss and damage information concerning federal assets would help to complement the CESARE data hub and enable more reliable assessments for single hazards such as storms. Other possibilities for improving the CESARE data hub, for instance, include the provision of inflation-adjusted and normalized loss data. The normalization of loss data addresses changes in wealth and assets over time and space and hence goes beyond mere inflation adjustment by additionally considering changes in asset values or—in case of human losses—changes in inhabitants. Including the feature of normalization in the CESARE data hub would thus improve the temporal and spatial comparability of the documented loss data.