Conference Agenda

Traffic on the Czech rail network is often interrupted by falling trees as 30 % of the rail network is closer than 50 m to forests. Tree falls on railway tracks and overhead power lines cause considerable damage. In order to help the national rail infrastructure administrator (Správa železnic, SZ) deal with these incidents, a web-map application called Stromynazeleznici (i.e., trees on railway tracks) has been developed. It provides a forecast of tree-fall hazard on a 3-hour basis for the following two days. The model incorporates data from weather forecasts (Aladin model) and a tree-fall susceptibility layer which delimits the locations where falling trees are capable of crossing railway tracks.

The tree-fall susceptibility layer is prepared from the raster of a normalized digital surface model. One-meter cells contain information about the absolute height of the surface above the relief model. All non-vegetated areas (all types of buildings, tall objects, bridges, masts, etc.) and areas with low vegetation that do not pose a hazard are filtered out. Impact zone buffers are defined for the remaining vegetation areas according to the actual height of the vegetation. The final output is a proportion of the length of railway lines per unit section which are threatened by falling trees.

Stromynazeleznici contains tree fall evidence for recording, presenting, and exporting incidents. Data can be entered either via a web form or through a mobile application for the Android system. The forecast is based on a regression model programmed in R (server solution Project R). A multivariate logistic regression was chosen as the most suitable approach to construct the model according to cross-validation results and practical requirements. The following meteorological elements and characteristics of the rail infrastructure surroundings were selected as explanatory variables in the logistic regression: maximum daily wind gust, soil saturation index, snow index, the occurrence of thunderstorms, the season, the range of altitudes in the vicinity of the rail track, the median height of trees along the railway tracks, and the length of the rail track section with trees along the rail track.

Meteorological data are sent four times a day via an SFTP server by the Czech Hydrometeorological Institute. The hazard level of tree falls is calculated for the "hectolines" (i.e., 100-meter segments) of the railway track. These are then aggregated into three levels of administrative units defined by SZ. The hazard level is calculated for three-hour intervals, covering a 45-hour forecast period – resulting in 15 time slots for each hectoline. The forecast is updated four times a day as new meteorological data become available.

The data is stored in a database and presented in the form of graphs, tables, and an interactive map. The tree-fall hazard level is represented by a five-level colour scale for individual administrative units. When zooming in, the risk is shown in relation to the hectolines. A timeline is located at the bottom of the screen, allowing users to switch between different time slots or aggregated time windows.

Transportation network is a vital part of moder-day society. It allows for the mobility of people and goods across large distances. When natural disasters hit transportation networks the results are often a number of closed parts. As a results, certain roads or rail tracks may be even destroyed, but the majority of them are usually only closed for traffic and can be reopened after a relatively short period of time. Functioning transportation network is among the primary environments securing economic growth. Therefore, its robustness and resilience have to be maintained. Data about these incidents which can affect the transportation network performance is important for designing relevant security measures.
In Czechia, data on all problems in road transport (including traffic collisions, planned maintenance) is being gathered by numerous organizations and provided in an online system of traffic information (JSDI). The main aim of this system is to offer an overview on actual situation on roads. Among other features, it offers an automatic data interface. Records are sent in real-time using the HTTP POST protocol in XML format. The JSDI database has not been planned as a source of this kind of information. Therefore, all information regarding natural hazards and their impacts had to be data-mined from text descriptions which is among the attributes. We developed a full-text filter that determines whether a disruption has occurred and, if so, what type. In the application, we distinguish disruptions caused by flooding, landslides, rock falls, falling trees, and snow.
Similarly, also data for railways are available, albeit from a different data source. The state-owned company, Správa železnic (SZ), which is responsible for the majority of rail tracks in Czechia, collects information on all problems that affected rail network. In addition, there is a database of the fire brigade unit, which deals with the consequences of these incidents.

CDV stores this data from all these sources for further analyses in order to study and evaluate the impacts of natural processes on transportation infrastructure. For this purpose, we created a spatial database which includes data for roads as of 1997 and railways (as of 2002). The spatial database, called RUPOK, is automatically updated.

For road network, the majority of complete road blockages were caused by fallen trees (64%), followed by snowing (31%). Flooding and landsliding (including rockfalls) caused 4% incident (1% respectively), but with considerable higher impacts on infrastructure.
For railways, the situation is similar as for roads. The majority of railway track blockages were caused by fallen trees (90%), followed by snowing (6%) and flooding (4%). The least common were landslides and rockfall incidents with less than 2% share. It is important to mention, however, that incidents may overlap in part, as snowing can also cause tree fall.
Data on incidents and certain elementary statistics is presented via a webmap application. The core is a MariaDB database with the Spatial extension, which allows for the management of spatial data. The application is programmed using PHP, jQuery, and the Google Maps API.

Floods are one of the most relevant natural hazards Worldwide and in Switzerland, causing significant socio-economic damage every year. Despite the recent progress in assessing flood hazards and risks, predicting rivers' responses to flooding and anticipating their consequences remains challenging. This is particularly true in forested mountain rivers, where floods are much more than extreme discharges, as they trigger geomorphological changes, such as bank erosion and channel widening, leading to significant sediment erosion and transport while recruiting and mobilizing trees and large pieces of wood. However, flood hazard and risk analysis rarely quantify or fully consider these cascade processes.
During large flood events, entrained and transported instream wood (i.e., large wood, which includes trunks, logs, branches and root wads) may accumulate at particularly vulnerable locations such as bridges, culverts, and other hydraulic structures, enhancing flooding impacts. However, unlike flow and sediment monitoring, the monitoring of wood in rivers is scarce, with a generalized lack of data, monitoring stations or standard metrics to quantify the instream wood regime. Therefore, monitoring large wood transport during flood events is critical for improving flood hazard assessment and infrastructure management.
The work presented here summarizes several research projects aiming at designing a monitoring framework for wood transport in rivers and identifying critical bridges in terms of wood trapping.
Our research combines fieldwork, remote sensing, drone surveys, and in-situ sensor networks to track wood movement during flood conditions, ranging from large floods to more frequent, seasonal floods, and to assess the factors influencing wood mobilization and deposition. We propose a monitoring framework that combines stationary or drone-mounted cameras with a novel machine-learning algorithm to automatically detect wood transport.
The research also focuses on identifying variables related to river morphology, surrounding forest, bridge geometry and characteristics that control wood trapping. These variables are then used to train a machine-learning decision tree and random forest that classify wood-prone bridges.
The results revealed that wood transport during floods is highly episodic, occurring predominantly during the rising limb and peak discharge, and is influenced not just by the river characteristics and flood magnitude but by other factors, such as the wood availability, flood hydrograph shape, sequence of floods, and the presence of obstacles and human structures.

The presence and number of bridge piers, their shape and the channel energy (in terms of stream power) were particularly important for identifying bridges prone to trapping large wood.
This study provides a more comprehensive understanding of wood transport during floods. More importantly, the monitoring framework using cameras and the model to identify critical infrastructures can be easily replicated at other geographical locations with varying features and characteristics. Integrating these methods into flood hazard assessment will improve the analysis of potential risk and guide the design of more resilient infrastructure to mitigate the effects of large wood accumulations during extreme events.