Daily rainfall data spatialization for the analysis of shallow landslide triggering conditions

Shallow landslides are mass movements capable of causing severe damage to infrastructures and loss of lives. Similarly to other weather-driven geological processes, the spatial analysis of either hazard or susceptibility to shallow landslides by means of data-driven methods often involve two types of factors. Stable factors, such as geomorphological variables, represent the predisposing conditions for landslide occurrence. These factors, almost constant over time, are predominantly used in susceptibility assessments to identify areas potentially prone to landslides, irrespective of meteorological conditions. On the other hand, triggering factors, which are typically associated with highly dynamic variables, are also usually analyzed as they influence frequency and magnitude of landslide phenomena, hence being essential for hazard mapping. Heavy rainfalls may be regarded as the main triggering factor for shallow landsliding.

Rainfall is not a regular phenomenon and it is characterized by high spatial variability, particularly in mountainous regions, hence evaluating its spatial-temporal distribution represents a quite hard task, despite the availability of long-term meteorological stations records.

The primary objective of this study is to evaluate different interpolation methods for spatializing daily rainfall data to support shallow landslide hazard mapping. The study focuses on the Alpi Apuane region located in northern Tuscany (Italy), characterized by complex topography rising sharply few kilometers near the Ligurian sea coast. Daily precipitation data, collected over nearly seventy years, were obtained from various meteorological networks operating within the study area.

Different spatialization methods were selected to facilitate automated computation of the available large station dataset, such as the Inverse distance weighted interpolation, as well as different kriging methods, including the use of elevation data as a secondary variable for precipitation mapping.

The performance of the different methods was assessed for a set of significant precipitation days and involving an iterative process for random validation subsets selection.

Considering that landslides often occur in inaccessible areas and are generally poorly reported, their occurrence dates in landslide inventories are either frequently missing or uncertain.

In order to mitigate this issue, an inventory of shallow landslides was created for the study area through the visual interpretation of a multitemporal set of orthorectified aerial photographs. The available images used for landslide mapping span the period from 1954 to 2021. The acquisition of these aerial images was not temporally constant, the intervals between the acquisition range from 24 to 2 years,  with an average value of 6 years. The last two decades (2003-2021) instead are characterized by a regular acquisition of aerial images of about 3 years. For each landslide, the triggering period was defined by the time interval between two consecutive image acquisition dates t(n) and t(n+1), the latter representing the oldest image where the landslide was recognized. The pre-landslide period was defined to correspond to the time interval preceding t(n), i.e. the youngest image acquired before landslide triggering. The computed daily precipitation maps were used for the analysis of intense rainfall events occurred during both pre-landslide and triggering periods, enabling the assessment of triggering daily precipitation associated to the landslide areas of the multitemporal inventory.