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

Accueil du site → Doctorat → Royaume-Uni → 2005 → Modelling soil properties at the landscape scale in a desertification context

King’s College London (2005)

Modelling soil properties at the landscape scale in a desertification context

Fonseca Ines Lopes da

Titre : Modelling soil properties at the landscape scale in a desertification context

Auteur : Fonseca Ines Lopes da

Université de soutenance : King’s College London

Grade : Doctor of Philosophy (PhD) 2005

Résumé
Despite recent advances in the development of spatially distributed hydrological modelling there remain significant challenges to obtaining data about the spatial distribution of soil properties at scales appropriate for these data to be used in landscape-scale hydrological studies. It is suggested that soil-landscape modelling may contribute to efficient spatial derivation of soil properties at the landscape-scale and thus to promoting a large scale perspective in which it is possible to more fully understand the impact of soil variability upon the hydrological response of catchments. This thesis (1) examines different pattern recognition methodologies of landscape modelling for deriving spatially distributed soils data from routinely available sources, (2) assesses which technique(s) best predict(s) soil depth, texture, rock fragment content and bulk density and, (3) presents the results of a modelling exercise to understand better the importance of the spatially structured variability of these soil properties in mediating hydrological response at the catchment scale. This is achieved using a combination of fieldwork in the Marina Baixa (SE Spain), soil laboratory analysis and hydrological modelling. It was found that structural pattern recognition techniques are better at producing landscape functional units to predict soil properties than are clustering techniques. Nevertheless, issues such as data normalisation in clustering are shown to be less important than the choice of algorithm, especially when choosing between discrete and continuous algorithms. Also, the finer landscape variability produced by the application of methods that produce continuous boundaries can only be fully exploited if soil sampling is concomitant with the level of landscape variation. The resulting soil property patterns are used as distributed data input in the PATTERN LUE hydrological model and comparisons with lumped and random data inputs illustrate how the spatial variation of soil properties, the structure of soil patches, topography and type of rainfall event strongly influence runoff. The relative importance of each of these factors depends on a complex interaction between rainfall intensity/magnitude, in situ soil characteristics and the soil spatial structure. Thus, for accurate estimation of runoff for mitigation of further degradation of soils in semi-arid areas it is essential that not only soils be adequately represented but also the flow routing algorithm and rainfall data be realistic. Thus, efforts should also concentrate on developing tools for better describing the land surface and the partitioning of flow.

Présentation

Version intégrale (115 Mb)

Page publiée le 9 avril 2019