5.1. Utilizing the torrential event database
Based on internationally recognized findings of the DOMODIS (Documentation of Mountain Disasters) project of the ICSU-CDR (International Council for Science, Committee on Disaster Reduction (formerly ICSU-SC IDNDR)) and the IAG (International Association of Geomorphologists), Hübl et al. (2002) published a recommendation of documentation sheets for natural hazards in Alpine regions. Together with various expert groups (e.g., Natural Hazards Carinthia), this recommendation has been adopted in the Austrian settings and a digital event cadastre was designed and developed as a standardized tool for recording and collecting event information on torrent and avalanche processes. This cadastre is maintained by the Centre for Natural Hazards Information of the Torrent and Avalanche Control (WLV). Although it is an internal database, it has been made available for scientific research (Naturgefahren 2021).
From the 191 columns in the "event” table of the WLV event register, only the information required for this study has been selected. The identified columns include the following information for each entry: “id”, “process type id”, “process type”, “state id”, “state”, “district id”, “district”, “community id”, “community”, “event date”, “event time”, “event time maxo”, “event year”, “event month”, “event day”, “event duration”, “event duration maxo”, “trigger id”, “trigger”, “trigger phenomenon”, “precipitation date”, “precipitation time”, “precipitation begin maxo”, “precipitation duration”, “precipitation duration text”, “precipitation duration maxo”, “precipitation amount”, “precipitation amount maxo” (BMLRT 2019). Within the selected columns, some seem to be similar, e.g., the two columns “trigger” and “trigger phenomenon”. The difference is explained by the following example. A torrential event can be triggered by a thunderstorm, therefore an entry in the column “trigger” is created. But during the storm, hail was the main cause, so this information is listed in the column “trigger phenomenon” (BMLRT 2019). Thus, it is possible to differentiate the triggers in greater detail and with higher accuracy.
All data entries are provided with a quality index by the WLV, which is based on the “MAXO-code”. The variables in this code refer to the following abbreviations: m = measured value or saved value, a = assumption, x = can be determined at a later point in time and o = cannot be determined (Andrecs and Hagen 2011). This information has been used to determine if the weather information in the event register from the WLV or the weather data from ZAMG is more accurate. In the case of “m”, associated with WLV entries, the values have been used in the following analysis and the ZAMG database has been applied subject to “a”, “x” and “o” marking the values. New entries in the WLV event register contain information on precipitation and triggering (BMLRT 2019).
The selected two tables (event and event location) are linked with the help of the column “id”. This allows the exact localization of each event on a map. To avoid false interpretations, all events that are not explicitly assigned as landslides and bedload transports are excluded. As a result, only the following torrential processes are included in the subsequent analysis: “debris flow-like transport”, “debris flow”, “earth flow”, “slope debris”, “mountain creep”, “rotational slide”, “translational slide”. In addition, the data for “undefined slides” and “bedload transport” have been used. Since there were still too many entries, the different process types have been grouped in the three classes “fluvial process”, “debris flow-like process”, and “sliding process" (Rimböck et al. 2013) by applying the triangle diagram developed by Phillips and Davies (1991).
Finally, events with no entry in the “event date” column have been removed. Since the used weather data from ZAMG started in the year 1980, all data before this year have been deleted also from the torrential event database.
5.2. Utilizing the meteorological database
ZAMG's network of stations comprises approximately 260 weather stations which are listed on their website including names and coordinates. Within ArcGIS Pro (Version 2.6.2), all Austrian weather stations have been visualized. To regionalize the local weather information of the individual meteorological stations, Thiessen polygons have been applied. This is still one of the most common methods to approximate precipitation over an area, where limited weather information is available (Lü et al. 2021). Compared to the inverse distance weighting method (IDW) and the Kriging method, the Thiessen polygons focus on geometric values only and ignore topographic influence (Zhou et al. 2020). Since the available weather data does not provide any respective details, it is reasonable to apply the Thiessen polygons for first approximations, despite their limitations (Lü et al. 2021). Consequently, each weather station is assigned to one Thiessen polygon. The resulting map is shown in figure 1a (ZAMG 2021b). When two polygons meet, the edges will be a perpendicular bisector to the connecting line between two weather stations. The still open polygons along the national border of Austria are closed by using the Austrian border line. Otherwise, these polygons would extend to infinity, which has to be avoided (Yamada 2016). Since the WLV database contains the digital torrential data until October 2019 only, the additional entries of the ZAMG database are deleted (ZAMG 2020, BMLRT 2019).
5.3. Merging the torrential event database, the meteorological database and the lithological database
In the next step, the torrential event and meteorological databases have been merged. As mentioned, the WLV table does not provide information on the precipitation date for all entries. To be able to carry out an analysis and to connect the precipitation value and type of precipitation with the ZAMG data, the following rules are applied:
For events with no event time available, the precipitation date of each entry has been checked. If no precipitation date is given, the WLV event date is used, otherwise, the ZAMG precipitation date is considered more accurate than the event date of the ELV database. For the values where only the WLV date of the event is known, the ZAMG daily precipitation value is used, as it is not known when the event happened throughout the day. If an event has a known event time it can be connected to the period it best represents.
WLV events with entries in the column “event time” can be similarly merged with the ZAMG data. No precipitation data or time means that only the event time can be used. If there is a ZAMG precipitation date, but no WLV precipitation time available, then the precipitation date is still considered to be more correct than the event time. That rule must be applied if the precipitation date is different from the event date and therefore the daily precipitation or daily precipitation type is applied. If the ZAMG precipitation date and the WLV event date are the same, then the event time is used since the time value is considered to be more accurate. If an event has a known precipitation time, it can be linked to the best fitting ZAMG value.
Now the WLV-ZAMG database can be merged on the base of the polygons and utilised within a GIS system. Only 79 weather stations in Austria can distinguish between different precipitation types, so for this analysis, these weather stations are used as shown in figure 1b (ZAMG 2020).
Since the lithology (figure 2) provides some basic information on the near-surface and thus, it is important on the effects of hail on torrential processes. This information has also been imported into ArcGIS Pro and adapted based on the UBA template (UBA 2004). The available information has been grouped into the following main lithologies: Bohemian Massif, Tertiary Basin, Quaternary, Penninic Nappes, Helvetic Nappes, Northern East Alps, Greywacke Zone, Central East Alps, and Southern East Alps (GBA 2015).
To evaluate whether there was any precipitation in the 5 days before an event and to verify if hail can be seen as a direct or indirect trigger for landslides and/or bedload transport, an analysis of the antecedent precipitation is carried out. In many regions, antecedent precipitation is an important aspect in the triggering of landslides, e.g., by increasing the soil moisture and therefore, reducing soil strength. Since no detailed information on antecedent periods for landslide-triggering events in Austria are available, a common length for an antecedent period for Central European regions has been applied from literature (Jemec and Komac 2012, Rahardjo et al. 2019, Kuradusenge et al. 2020). Based on these studies, a threshold of 5 days has been defined and implemented in this analysis (BMLRT 2019).
To add the missing precipitation duration and precipitation amount values in the WLV database, ZAMG precipitation values are used. It has to be noted, however, that due to the selection process of the meteorological ZAMG data, some ZAMG data do not increase the accuracy of the database. The limitation of the ZAMG data does not always allow a proper merging, 24-hour values cannot be divided into 1-hour values, because it is not clear when the precipitation in question occurred in the 24 hours. There would be misjudgements in this investigation and consequently, the analysis of the precipitation duration and the precipitation amount have not been further processed in this study (ZAMG 2020, BMLRT 2019).
If snowmelt was given as a trigger phenomenon, the entry was deleted. The assumption herein is, that the temperature increase - and not the precipitation on the date of occurrence - is the real reason for this event. If rain within the column “trigger” and snowmelt within “trigger phenomenon” was given, this dataset was not deleted. It can be assumed that the rain has influenced the melt and not only the temperature rise is the cause of this torrential event. If flooding was entered within “trigger”, the respective entries were deleted.
To avoid double entries, the ID is checked for every entry in the two tables and compared if there are any clones unintended created through the various work steps. Since entries with time information for precipitation are more accurate, they are superimposed over those without time information, if the rest of the data has the same information.
If heavy precipitation triggers many torrential processes in the same catchment, they are individually recorded in the WLV database. This might influence the significance and introduce a bias in the results of this study. To reduce the influence of major event entries within the same catchment these have to be compared with each other and merged into one event.
The finally resulting database was used for the respective analysis.