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

Accueil du site → Master → Canada → Bayesian networks and GIS techniques for modelling the causality, intensity and extent of land degradation in drylands

Trent University (2008)

Bayesian networks and GIS techniques for modelling the causality, intensity and extent of land degradation in drylands

Ahmed, Oumer

Titre : Bayesian networks and GIS techniques for modelling the causality, intensity and extent of land degradation in drylands

Auteur : Ahmed, Oumer

Université de soutenance : Trent University (Canada),

Grade : Master of Science (MS) 2008

Résumé
In this thesis, a new probabilistic approach to assess land degradation and its causes in dry lands is introduced. The suitability of Bayesian Networks for modelling the causality of land degradation intensity and extent through the integration of driving forces, pressures, states impacts and responses (DPSIR) is evaluated. In an attempt to describe the relationships between bio-physical states of degradation to their social, economic and demographic causes, the proposed DPSIR framework offers a new probabilistic approach to the establishment of the major root causes of the states of degradation in a study area, resulting in a practical Bayesian network modelling application and implementation to land degradation data. A Bayesian network model has been constructed and tested using DPSIR indicators of land degradation in El Alegre watershed, San Luis Potosi, Mexico, using data obtained from measurements recorded in field forms and questionnaires applied during interviews with farmers and herders and local experts and officials. These data were used as input to the model developed using Netica™ software. The Bayesian network model was developed by linking indicators of Drivers and Pressures to State indicators based on their presumed cause-effect relationships. These relationships were derived from expert knowledge and available combination of data sources in the study area. Values (intensity or extent) of status were assigned to each degradation indicator based on all combinations of the status (intensities or extents) of each of its identified causes. The final built model enables the visualization of the causality, intensity, and extent (of coverage over the area) of each indicator (drivers, pressures and states) within the model and to identify the most probable causes (drivers and pressures) of each of the state Indicators of land degradation from the sensitivity analysis of the model. This determines the most influencing causes for each indicator of the state of degradation. The causal relationships predicted by the model were validated independently through a confusion matrix and the Cohen’s Kappa technique using local farmers’ perceptions of the causes for a given type of degradation collected from interviews in the field through questionnaires. The results showed that the agreement between farmer perceptions of causes for each degradation state and the predicted causes by the model was good, but modest. This modest agreement was attributed, to a large extent, to the degree of subjectivity involved in interpreting vague farmer responses in the questionnaires. However the causality model proved empirically accurate according to the knowledge of local experts. Finally, using GIS the results of the present states of degradation of such dry lands, and their causes (drivers and pressures) were mapped coding each degradation indicator in an ad-hoc map legend, including their intensity, spatial extent and most influencing causes.

Présentation (Amicus)

Version intégrale

Page publiée le 28 janvier 2012, mise à jour le 7 février 2018