Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Doctorat → Allemagne → Water erosion prediction by stochastic and empirical models in the Mediterranean : A case study in Northern Sicily (Italy)

University of Tübingen (2012)

Water erosion prediction by stochastic and empirical models in the Mediterranean : A case study in Northern Sicily (Italy)

Angileri, Silvia Eleonora

Titre : Water erosion prediction by stochastic and empirical models in the Mediterranean : A case study in Northern Sicily (Italy)

Prognose der Bodenerosion im Mittelmeerraum anhand statistischer und empirischer Modelle : Eine Fallstudie in Nordsizilien (Italien)

Auteur : Angileri, Silvia Eleonora

Université de soutenance : University of Tübingen

Grade : Doktors der Naturwissenschaften (Dr. rer. nat.) 2012

Résumé
The present thesis aimed to explore the methodological advantages as well as limitations in applying different modelling approaches to predict water soil erosion in Mediterranean environments. The research was accomplished in the central northern part of Sicily (Italy), considering this region to be representative of Mediterranean environmental conditions. In this region soil degradation problems, due to water erosion are becoming more and more serious. Consequently, defining models being able to predict erosion susceptibility and to discriminate environmental factors causing erosion is important to protect soil resources. The prediction of the spatial distribution of soil erosion processes was carried out by means of GIS tools and multivariate statistical analysis. A stochastic gradient boosting model (TreeNet) was proposed to classify erosion and mass wasting processes and to define the functional relationship between spatial data sets of driving factors and response variables. The TreeNet method allowed identifying a susceptibility model that accurately fits the relationship between a set of several attributes and the activity of different erosion processes with a high resistance to over-training. Moreover, a better understanding of the prediction model was provided by the evaluation of the relative overall importance of the predictive variables in the tree construction. In order to estimate the overall prediction skill of the model, the ROC (Receiver Operating Characteristic) curves for each of the predicted process were constructed. Results illustrated an outstanding and excellent performance of the TreeNet method to predict bank and gully erosion, respectively. Sheet and rill erosion and mass wasting phenomena prediction attested to acceptable and poor performance of the model. The erosion susceptibility model was exploited to regionalize the information in areas characterized by the same geo-environmental conditions. Among erosion processes, gully susceptibility was most intensely investigated due to their high contribution to soil loss in the Mediterranean. A GIS layer containing 260 ephemeral and permanent gullies was constructed by field surveys and interpretation of high detailed aerial images and a set of 27 environmental attributes was selected as explanatory variables. The statistical analysis was defined on the scale of grid cells and slope units. The functional relationships between gully occurrence and spatial variability of the controlling factors was explored by carrying out forward stepwise logistic regression analysis that allowed to calculate the probability of mapping units hosting gullies. Results of validation showed acceptable to excellent accuracy of the predictive models, illustrating a more stable performance of susceptibility models defined on cell scale. Finally, further logistic regression analysis was carried out to generate a cell- and a slope-unit based gully erosion susceptibility map, both demonstrating an excellent fitting precision. Furthermore, a procedure to evaluate the impact of anthropogenic activity on soil erosion dynamics by means of empirical methods was proposed. In cultivated catchments, man-induced elements influencing runoff processes are mainly linked to alteration of original terrain morphology and to the consequently spatial soil redistribution pattern. In order to simulate the impact of anthropogenic elements on soil loss, data related to the characteristics of these rural elements and to their spatial distribution in the basin were collected and included in soil erosion modelling procedures. The interplay between the RUSLE (Revised Universal Soil Loss Equation) and the USPED (Unit Stream Power Erosion Deposition) models allowed to define the spatial distribution of man-induced impacts on soil erosion processes. In the study area farmer activities play an important role in modifying the natural flow-path, on both field and basin scale. Unpaved roads resulted the main cause of important transformation mechanisms in the agricultural landscape. These linear features influence the drainage patterns and consequently soil erosion dynamics. The results of this study confirmed the reliability of the adopted methods that are objective, reproducible and able to be exploited to produce accurate erosion susceptibility maps : A useful instrument for land management and planning. In addition, the research demonstrated that spatial occurrence of erosion processes is strongly influenced by human pressure modifying the natural flow path of water, underlining the necessity to more specifically include this factor in erosion prediction modelling.

Mots clés  : Bodenerosion , Geoinformationssystem , Fernerkundung , Anthropogener Einfluss , Regressionsanalyse , Bodenschutz , Mittelmeerraum — Soil erosion , Geographic information systems , Remote sensing , RUSLE , USPED , Human impact, road , Artificial channels , DEM , Erosion scenario

Présentation )

Version intégrale (10,2 Mb)

Page publiée le 30 décembre 2015, mise à jour le 11 janvier 2019