Giacomo D’Amato Avanzi

Shallow landslides susceptibility assessment in different environments

ABSTRACT The spatial distribution of shallow landslides is strongly influenced by different climatic conditions and environmental settings. This makes difficult the implementation of an exhaustive monitoring technique for correctly assessing the landslide susceptibility in different environmental contexts. In this work, a unique methodological strategy, based on the statistical implementation of the generalized additive model (GAM), was performed. This method was used to investigate the shallow landslide predisposition of four sites with different geological, geomorphological and land-use characteristics: the Rio Frate and the Versa catchments (Southern Lombardy) and the Vernazza and the Pogliaschina catchments (Eastern Liguria). A good predictive overall accuracy was evaluated computing by the area under the ROC curve (AUROC), with values ranging from 0.76 to 0.82 and estimating the mean accuracy of the model (0.70–0.75). The method showed a high flexibility, which led to a good identification of the most significant predisposing factors for shallow landslide occurrence in the different investigated areas. In particular, detailed susceptibility maps were obtained, allowing to identify the shallow landslide prone areas. This methodology combined with the use of the rainfall thresholds for triggering shallow landslides may provide an innovative tool useful for the improvement of spatial planning and early warning systems.

Fragility of Territory and Infrastructures Resulting from Rainstorms in Northern Tuscany (Italy)

In mountainous urbanized areas, shallow landslides cause significant, often unpredictable, hazard conditions. Landslides may involve and destroy infrastructures, kill people. Source areas are often located along the road network, representing a recurring situation also during not heavy rainstorms in Northern Tuscany. The landslides are generally first time debris slide-flows, occurring in peculiar environments: colluvium/debris slope cover (0.5–2 m thick), semipermeable-impermeable bedrock, hollow shaped slope, high slope gradient. Despite little size, they cause damage and deaths owing to their high velocity and erosion power. The source area along the road network is generally associated to the lack of efficient drainage systems. In fact, the concentration of uncontrolled runoff downslope creates an “unnatural” increase in pore pressure that the only rainfall should not have produced. This generates reaching and exceeding of the local critical rainfall threshold, making landslide hazard assessment more difficult. As example, in October 2010, almost 60 % of the landslides source area in the Massa area was located along the road network.

Assessing shallow landslide susceptibility by using the SHALSTAB model in Eastern Liguria (Italy)

On 25 October 2011 a heavy rainstorm hit Eastern Liguria (Vara Valley and Cinque Terre) and North-western Tuscany (Magra Valley), causing floods and hundreds of shallow landslides.This study aims at assessing the shallow landslide susceptibility using the physically based model SHALSTAB (Shallow Landslide Stability Model) in the Pogliaschina Torrent basin (Vara Valley). The susceptibility map elaborated with SHALSTAB was compared with the landslide inventory map, which confirmed the good performance of this model for the study area.The implementation of the SHALSTAB model provided a preliminary shallow landslide susceptibility map of the Pogliaschina T. basin and quite promising results on the shallow landslide spatial prediction.

Developing and testing a data-driven methodology for shallow landslide susceptibility assessment: preliminary results

In this work a data-driven methodology for shallow landslide susceptibility assessment is presented. The procedure is based on the Generalized Additive Model (Hastie and Tibshirani, 1990) and it is developed to be applied in different contexts, using terrain attributes, land use and lithological data. The application of the method in three different contexts in Italy shows the good forecasting capability of the model. The implementation of this method allows for building landslide susceptibility maps, which are a fundamental basis in hazard and risk assessment.

Preliminary analysis of the November 10, 2014 rainstorm and related landslides in the lower Lavagna valley (eastern Liguria)

(*) Università di Genova Dipartimento di Scienze della Terra, dell’Ambiente e della Vita Corso Europa, 26 16132 Genova, Italy (**) Università di Pisa Dipartimento di Scienze della Terra Via S. Maria, 53 56126 Pisa, Italy Corresponding author: cevasco@dipteris.unige.it PRELIMINARY ANALYSIS OF THE NOVEMBER 10, 2014 RAINSTORM AND RELATED LANDSLIDES IN THE LOWER LAVAGNA VALLEY (EASTERN LIGURIA)

A study on the 10 November 2014 intense rainfall and the related landslides in the lower Lavagna valley (eastern Liguria)

On 10 November 2011 some small coastal basins of the Tigullio gulf and the middle Lavagna and Sturla valleys (eastern Liguria), were hit by an intense rainfall. Flash floods involved the city of Chiavari and shallow landslides were triggered on the slopes, causing severe damage to settlements and road network. Some buildings were destroyed and two casualties occurred at Leivi, in the lower Lavagna valley.This paper synthesizes the results of preliminary surveys carried out few days after the event in the area most affected by shallow landslides. The investigations, which are still in progress, have contributed to understand the characteristics of both triggering rainfall and landslide source areas. These achievements will be helpful in studying rainfall thresholds, shallow landslide modeling, landslide susceptibility and risk assessment for the study area.

Comparison of statistical methods and multi-time validation for the determination of the shallow landslide rainfall thresholds

Shallow landslides are unforeseeable phenomena often resulting in critical conditions in terms of people’s safety and damage. The main purpose of this paper is the comparison of different statistical methods used to determine the rainfall thresholds for the shallow landslide occurrence. Rainfall data over a 46-year period were collected for one rain gauge located in a test area of northwest Italy (Riviera Spezzina; RS). In the RS, intense rainfalls often induce shallow landslides causing damage and sometimes casualties. The rainfall events occurred in the 1967–2006 period were classified as events inducing shallow landslides (SLEs1967–2006) and events that did not trigger shallow landslides (NSLEs1967–2006). Thresholds for various percentiles of SLEs1967–2006 were computed by identifying the lower limit above which shallow landslides occurred. Another set of thresholds, corresponding to different probabilities of occurrence, was determined using SLEs1967–2006 and NSLEs1967–2006. The least-squares linear fit (LSF) and the quantile regression (QR) techniques were employed in the former approach, while the logistic regression (LR) was applied in the latter. The thresholds were validated with the same data used for their definition and with the data recorded in the 2008–2014 period. Contingency tables were created and contingencies and skill scores were computed. The 10% probability threshold obtained using the LR method is characterized by the best values of at least two skill scores for both periods considered; therefore, it may be considered the “best” threshold for the RS. The results of this work can help the choice of the best statistical method to determine the shallow landslide rainfall thresholds.

Statistical Methods for the Assessment of Rainfall Thresholds for Triggering Shallow Landslides: A Case Study

La Spezia Province (880 km2; Liguria, northwestern Italy) is frequently hit by intense rainfalls, which often cause shallow landslides and damage to population and environment. In this regard, the Provincial Administration of La Spezia and the Earth Sciences Department, University of Pisa, promoted a study to define the rainfall thresholds for shallow landslides occurrence. In fact, on 25 October 2011 a very intense rainfall hit two parts of the provincial territory (Cinque Terre-Riviera area and Vara Valley) causing at least 3500 shallow landslides. This event was analyzed together with other 134 shallow landslide events occurred from 2008 to 2014. The rainfall conditions of these events were determined using an algorithm implemented by the CNR-IRPI of Perugia. The rainfall thresholds at different exceedance probability levels of landslide were defined using two statistical techniques: least-squares linear fit (LSF) and Quantile Regression (QR). The results highlight that the LSF thresholds seems to be the best performing from a statistical point of view and, consequently, the “best” for the study area.

Analysis of the Predisposing Factors for Different Landslide Types Using the Generalized Additive Model

In this paper, a semi-parametric nonlinear regression technique, known as Generalized Additive Model (GAM), was implemented for the landslide susceptibility assessment in the Gravegnola catchment (Northern Apennines, Eastern Liguria, Italy), which was affected by more than 500 shallow landslides on the 25 October 2011 intense rainfall event. Twelve explanatory variables derived from DEM with 5-m resolution, river network, land use and geological maps were considered to investigate their influence on landslide type occurrence. The predictive performance of different combinations of explanatory variables has been evaluated through a cross-validation technique and ROC curve analysis. Different susceptibility maps for each landslide type were finally produced and the results were compared. The preliminary results show the higher ability of GAM than a single regression technique in selecting the most influent predisposing factors on the basis of the type of movement involved in landsliding.

Remarks on the Role of Landslide Inventories in the Statistical Methods Used for the Landslide Susceptibility Assessment

Open image in new window In this work, the role of different landslide inventories in susceptibility assessment was evaluated using a non linear regression technique, namely the generalized additive model (GAM).The investigation was carried out in three study areas: the Versa catchment (Oltrepo Pavese, Southern Lombardy, Italy), the Vernazza catchment (Cinque Terre, Eastern Liguria, Italy) and the Pogliaschina catchment (Vara Valley, Eastern Liguria, Italy). Two landslide inventories related to the 2009 and 2013 rainfall events were taken into account in the Versa catchment, whereas two landslide inventories (referred to the same 2011 rainfall event) which differ for methods of detection and criteria adopted for the landslide mapping were considered in the Vernazza and Pogliaschina catchments. The predictive performance of GAM for each landslide inventory was evaluated. The results related to different inventories were compared. The results show that the predictive capability of the model and the landslide susceptibility are significantly influenced by the type of landslide inventory. Thus, the work highlights that a standard criterion for preparing inventories should be adopted in order to produce landslide susceptibility assessment as reliable as possible.