Paregiani

“ Automated Unsupervised Geomorphometric Classification of Earth Surface for Landslide Susceptibility Assessment ” Alessandro Paregiani and Maria Ioannilli International Conference on Computational Science and Its Applications ICCSA 2008 June 30th - July 3rd, 2008 Perugia, Italy "Geographical Analysis, Urban Modeling, Spatial Statistics" University of Rome “Tor Vergata”

Outline 4. Experimented Classification Methods 1. Landslide Hazard vs. Landslide Susceptibility 2. Purpose of the Work 3. Approaches to Landslide Susceptibility Analysis 6. Integrated Classification Method 5. Comparison of Intermediate Results 7. Correlation Analysis between Geomorphometric Classes and Types of Landslide 8. Conclusions

Landslides constitute one of the major hazards that cause losses in lives and property. To assess landslide occurrences is a complex analysis, involving multitude of factors and need to be studied systematically in order to evaluate the hazard. There are no universally accepted forecasting methods of "natural hazard" and in particular of landslide hazard. Landslide Hazard R = P x ( V x E ) (UNESCO)

Purpose of the work The definition of an automated method of terrain morphological classification in order to establish the correlation degree between topographic forms of the territory and landslide phenomena, by using a Landslide Inventory and a DEM as input A Landslide Inventory and a DEM Input Data Specific Objectives Identification of the most suitable measures to describe terrain topographic forms and to distinguish among geomorphically different landscapes (geometric signatures) Identification of a classification method, in order to obtain the best segmentation of terrain surface related to landslide phenomena Moving from the current literature: Building up the morphological parameters Classification by using different methods Evaluation of the goodness of each classification, by considering as factors the physical meaning of classes and the statistical correlation degree between classes and landslide phenomena Experiment end evaluation of a new integrated classification method Technical approach

Approaches to Landslide Susceptibility Analysis Indirect methodologies Geomorphometric Approach high degree of subjectivity this method doesn’t consider the relationships between instability factors Limits Algorithm Heuristic Approach Statistical Approach Method

The “science of quantitative land-surface analysis” It draws upon mathematical, statistical and image-processing techniques to quantify the shape of the earth at various spatial scales The quantitative analysis of a territory, and in particular of its shape, eliminates the limitations of the qualitative topographic information It stems from the need to establish a reliable numerical model in order to describe the earth shape. The quantitative characterization of topographical shape is a multidisciplinary technique applicable at any scale of analysis The geometric signature is an analytic tool of numerical land-surface classification The signature was defined as “a set of measurements sufficient to identify unambiguously an object or a set of objects” [Enzmann, 1966] Natural surface processes create different forms. The geometric signature abstracts those forms and expresses them numerically. Geomorphometric Approach Approaches to Landslide Susceptibility Analysis It considers combinations of instability factors, by introducing the “geometric signature” It eliminates the subjectivity of heuristic approach

Experimented Classification Methods Applying State-of-the-Art Classification Methods Supervised Classification: types of topography are recognized starting from selected “training samples” Unsupervised Classification: unconstrained by pre-set conditions, and allow the input data to determine “optimal” categories Parameters Authors Parameters - Single – Cell Topological Parameters “ Context” Parameters (extended neighborhood) Evans (1981) Pike – Iwahashi (2006) Pike (1971)   Nested - Means Divided Parameters Clustering  • Mean • S.D. • Variation coefficient • Symmetry • Slope gradient • Texture • Convexity - -  • Slope gradient • Aspect • Plan curvature • Profile curvature - Method   • Mean • S.D. • Variation Coefficient • Symmetry • Slope Gradient • Texture • Convexity  • Slope Gradient • Aspect • Plan Curvature • Profile Curvature Types of

It depends on the input data type the scale of analysis the desired output data quality and spatial resolution the availability of analytic and information tools Preliminary Processing The choice of a terrain-unit of analysis Input data analysis and preparation 30x30m DEM, computed by interpolating the altitude-points extracted from contour lines (10m interval) of the Technical Regional Cartography of Lazio Landslide Inventory of Tevere River Basin Authority (PAI), differentiating seven types of phenomena; the number of events totally registered is 351 with a total area of 19.35 square kilometres as a portion of land surface which contains a set of ground conditions which differ from the adjacent units across definable boundaries

Considered Parameters Z = A x2y2 + B x2y + C xy2 + D x2 + E y2 + F xy + G x + H y + I A, B, C ecc. are calculated using this polynomial and 9 elevation values as input data, as shown: (the reference system has its origin-point in the central cell): A = [(Z1 + Z3 + Z7 + Z9) /4 - (Z2 + Z4 + Z6 + Z8) /2 + Z5] /L4 B = [(Z1 + Z3 - Z7 - Z9) /4 - (Z2 - Z8) /2] /L3 C = [(-Z1 + Z3 - Z7 + Z9) /4 + (Z4 - Z6)] /2] /L3 D = [(Z4 + Z6) /2 - Z5] /L2 E = [(Z2 + Z8) /2 - Z5] /L2 F = (-Z1 + Z3 + Z7 - Z9) /4L2 G = (-Z4 +Z6) /2L H = (Z2 - Z8) /2LI = Z5 Curvature = -2 (D + E) * 100 (dz/dx) = [(a + 2d + g) - (c + 2f + i)] / (8 * L) (dz/dy) = [(a + 2b + c) - (g + 2h + i)] / (8 * L) Computational procedure to calculate “curvature”: Computational procedure to calculate “local convexity”: Focalmean(DEM) Computational procedure to calculate “texture”: Focalmedian(DEM) DEM – Focalmedian(DEM) “ slope gradient” “ section curvature” “ plan curvature” “ aspect” “ local convexity” “ texture” Computational procedure to calculate “slope gradient” in “e” cell i h g f e d c b a

Considered Parameters: Slope Gradient (first threshold) Local Convexity (second threshold) Surface Texture (third threshold) Classification of earth topography from DEMs by a nested-means algorithm and a three-part geometric signature Experiment 1: Nested-Means Multivariate Analysis (Pike – Iwahashi) TOPOGRAPHY: Continuous random surface Independent of any spatial orderliness imposed by geomorphic processes

Nested-Means Multivariate Analysis (Pike - Iwahashi) The classification underline a remarkable distinction among mountainside surfaces in four different classes characterized by increasing values of elevation and slope gradient

Statistical Multivariate Analysis Cluster method Maximum internal homogeneity and minimum external homogeneity Statistical Mean of parameter distributions and Covariance among parameter distributions

Considered Parameters: Slope gradient (SIMG) Local Convexity (CONVEX) Surface Texture (PITPEAK) Experiment 2: Statistical Multivariate Analysis (Pike - Iwahashi)

Experiment 3: Statistical Multivariate Analysis (Evans) Considered Parameters: Slope gradient Aspect Plan Curvature Profile Curvature normalized as follow: A remarkable distinction among terrain elements originated by hydrological and wind erosive activities, such as torrential (Class 1) and fluvial (Class 6) riverbeds and ridges (Class 8), with a topological continuity between Classes 1 and 6

Experiment 4: Statistical Multivariate Analysis (Pike) Considered Parameters (statistically derived): Mean Standard Deviation Variation Coefficient Symmetry Not-derived Parameters: Elevation Slope Gradient Curvature

Comparison of Intermediate Results

Integrated Classification Method grid-cell based analysis homogeneity between input data (30x30m cells) selection of relevant classes by a conditional function

Integrated Classification Method A particular of the classification discriminating mountainside surfaces Overlapping of the three classes representing ridges, fluvial and torrential riverbeds A particular of the new integrated classification that considers both mountainside surfaces and hydrological factors

Correlation Analysis Geomorphometric Classes/Landslides Integrated Classification Method K: type of landslide J: geomorphometric class M: number of k-type landslides in class j N: total number of k-type landslides

Conclusions Identification of the most suitable parameters to describe terrain topographic forms related to landslide susceptibility Slope gradient constitutes the main parameter in discriminating different classes with a clear physical meaning related to landslide susceptibility analysis Single – Cell Topological Parameters discriminate local physical terrain features “ Context” Parameters discriminate global physical terrain features Identification of a new classification method, in order to obtain the best segmentation of terrain surface related to the landslide phenomena The method working by the nested-means algorithm allows to identify global features Local features, such as fluvial and torrential riverbeds, have been identified by using the statistical multivariate method The goodness of each classification has been evaluated by considering as factors the physical meaning of classes and the statistical correlation degree between classes and landslide phenomena The results of this evaluation show that the integration of both classification methods allows to correctly classify the territory and to establish correlation degrees between geomorphometric classes and landslide phenomena This method could represent a useful tool in territorial-scale landslide susceptibility analysis. In fact, the application of this repeatable and reliable procedure may return the best results in a short time and with low economic resources.

Paregiani

More Related Content

What's hot

What's hot (20)

Similar to Paregiani

Similar to Paregiani (20)

More from Beniamino Murgante

More from Beniamino Murgante (20)

Recently uploaded

Recently uploaded (20)

Paregiani

Editor's Notes