Defining rainfall thresholds for debris flows in catchments with short monitoring periods and rare debris-flow events.

In mountainous regions, debris flows represent a significant hazard, causing extensive damage and casualties each year. Among the various triggering factors, rainfall is the primary driver of debris flows in catchments with high sediment availability. Determining critical rainfall thresholds for debris-flow initiation is therefore essential for improving early warning systems and mitigating associated risks. This study focuses on:
a) Defining rainfall thresholds for debris-flow initiation using monitoring data collected over a relatively short period (three years), even in the absence of a large number of observed debris-flow events;
b) Gaining deeper insights into catchment dynamics by not only differentiating between debris-flow and no debris-flow conditions but also identifying rainfall thresholds that correspond to increased water levels and sediment transport within the stream;
c) Achieving these goals through the implementation of a monitoring station that is simple, cost-effective, and easy to install and operate.
The study area is the Blè catchment, a drainage basin covering 2.9 km², located in Val Camonica in the Central Italian Alps. Monitoring activities began in 2021, supported by funding from Regione Lombardia. The catchment is monitored through a network of seven stations distributed along the debris-flow channel. One station was installed by the University of Bologna, while the remaining six form the monitoring and early warning system developed by Hortus Srl. These stations are equipped with a variety of sensors, including rain gauges, radar-level sensors, geophones, and cameras, enabling comprehensive observation of debris-flow dynamics.
A crucial aspect in determining rainfall thresholds is the identification of individual rainfall events. In this study, we applied time windows of varying durations to separate consecutive events and analysed how rainfall thresholds change as the duration of these time windows varies. Using images from the cameras, we associated each rainfall event with the corresponding catchment response. Alongside the recorded debris-flow events (one in August 2021 and one in October 2022), we also considered events characterized by increased runoff in the stream, both without evident sediment transport and with evident sediment transport. The classical approach to defining rainfall thresholds was adopted, utilizing the duration and average intensity of rainfall in their logarithmic form. Rainfall threshold determination was performed using Linear Discriminant Analysis (LDA), a method that identifies threshold values by maximizing differences between categories of catchment responses while minimizing variability within each category. Two distinct thresholds were defined: an upper threshold for debris-flow initiation and a lower threshold to distinguish events characterized by increased runoff with sediment transport. These thresholds were determined for each of the five monitoring stations equipped with rain gauges. Additionally, we analysed how the average rainfall intensity and duration varied between rain gauges installed in close proximity within the same small catchment. Our approach revealed particularly valuable for areas with short monitoring periods and infrequent debris-flow occurrences.