Landslide hazard mapping at a basin scale using remote-sensing data and artificial neural networks

Sandro Moretti(1) , F. Catani(1) , A. Bartolomei(1) , M. Kukavicic(1) , V. Tofani(1) , P. Farina(1) , and F. Marks(1)

(1) Università di Firenze, Via La Pira 4, 50121 Firenze, Italy


A landslide hazard map of the entire Arno River basin (Central Italy) has been completed in the framework of a research project, sponsored by the Basin Autorithy of the River Arno and by the European Space Agency (ESA/SLAM Project).

The drainage basin of the Arno River is almost entirely situated within Tuscany in Central Italy. The river is 241 km long, while the catchment has an area of about 9131 km2 and a mean elevation of 353 m a.s.l. The basin is representative of the typical Mediterranean conditions, in terms of landslide type and environmental setting.

Conventional investigation methods, such as aerial-photograph interpretation and field surveys, have been improved by the use of different remote sensing techniques, employing both optical and synthetic aperture radar (SAR) images.

Differential SAR interferometry (DInSAR) and the Permanent Scatterers (PS) technique, implemented by using ERS and JERS data, have been applied in order to detect new slope movements and to better evaluate the state of activity and the boundaries of landslides mapped with independent techniques. About 350 SAR images, covering the period 1992-2002, have been interferometrically processed by means of the PS technique, detecting about 600,000 PS, corresponding to natural reflectors on the ground where it is possible to assess precisely the velocity history over the investigated period.

About 28,000 individual landslides were recognized and mapped through the integration of the different data sources. Most of the mapped landslides are earth slides / earth flows (75%), solifluctions and other shallow slow movements (17%) and flows (5%) while soil slips, and in general shallow landslides, seem to be of limited importance within the basin. The relationships between landslide characteristics and environmental factors have been preliminary assessed through statistical analysis. As expected, the results show a strong control of lithology and morphology on landslide distribution. The landslide frequency size distribution shows a typical scaling behavior, already underlined in other landslide inventories worldwide. This scaling law holds for areas greater than 104 m2 and shows a rollover for smaller landslides that can be ascribed either to the undersampling of small phenomena due to the survey scale or to some physical threshold concerning the minimum required mass for the development of rotational earth-slides, which are the dominant type of mass movement in the study area.

The assessment of landslide hazard in terms of probability of occurrence in a given time, based on direct and indirect observations of the state of activity of mapped landslides, has been extended to landslide-free areas through the application of statistical methods implemented in an artificial neural network (ANN). Unique condition units (UCU) were defined by the map overlay of landslide preparatory factors (lithology, land cover, slope gradient, slope curvature and upslope contributing area) and afterwards used to construct a series of model vectors for the training and testing of the ANN. Various different ANNs were selected throughout the basin, based on the prevailing physiographic settings until each UCU was assigned a degree of membership to a hazard class. The validation of the hazard map has been based on the comparison between predicted and observed mass movement, using two different methods: the first one relying upon the inventory map and directed field surveys, the second one using the information from the PS dataset. The following intersection of hazard values with vulnerability and exposure figures, obtained by the reclassification of digital vector mapping at 1:10,000 scale, has lead to the definition of risk for each terrain unit.

The final results of the research are now undergoing a process of integration and implementation within land planning and risk prevention policies of the Arno Basin Authority for the Hydrogeological Structure Plan. The proposed methodology represents one of the first examples in Europe of full integration of remote sensing data into land regulations concerning landslide hazard.


Workshop presentation


  Higher level                 Last modified: 07.10.03