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

Accueil du site → Doctorat → Australie → 2003 → Geographical issues in predictive vegetation modelling : error and uncertainty in GIS data, methods and models

Australian National University (2003)

Geographical issues in predictive vegetation modelling : error and uncertainty in GIS data, methods and models

Van Niel, Kimberly

Titre : Geographical issues in predictive vegetation modelling : error and uncertainty in GIS data, methods and models

Auteur : Van Niel, Kimberly

Université de soutenance : Australian National University

Grade : Doctor of Philosophy (PhD) 2003

Résumé
Predictive vegetation modelling makes extensive use of GIS data methods and models, yet error in each may affect the integrity of vegetation modelling outcomes. While error has increasingly been studied in GIS, it has not been extended to inte1face applications such as vegetation modelling. In this study, a framework was developed for considering error and uncertainty within GIS data, methods, and models, and how they affect the process and outcomes of predictive vegetation modelling. As a basis for the study, predictive vegetation models were developed for 17 tree species in a temperate Eucalypt forest in New South Wales. GIS method error was examined in relation to stochastic analyses, particularly focusing on error propagation. GIS data error was examined based on an error analysis of the DEM and the propagation of that error in the development of the predictive environmnental variables, and the ultimate effect of the propagated error on model development and results. To examine error due to GIS models, the individual species predictions, and associated errors and uncertainties, are used to demonstrate the use of individual species data for the development of traditional outputs, such as maps and GIS layers, that are flexible and data rich. The results of this thesis have a number of implications for predictive vegetation modelling. The analysis of error in GIS methods showed that selection of the pseudo-random number generator and the method of spatial autocorrelation development in stochastic analyses of propagation of uncertainty can threaten the integrity of results. The examination of GIS data error showed that some environmental variables are more sensitive to error in the DEM than others. The results contradict the prevailing, and until now untested, theory that indirect variables would be less prone to sensitivity due to their quantitative proximity to the original DEM data set. Also, error in environmental variables may have a much larger effect on the process and results of predictive vegetation modelling than had previously been suggested. In this analysis, propagated GIS data error has an extensive effect on model development, species-environment relationship interpretation, statistical fit, and spatial predictions. These results show that selection of predictive variables should consider GIS data error and uncertainty and that this type of uncertainty needs to be reported for model outputs, for both interpretations and predictions. Finally, the study into GIS model error shows that it is feasible to develop models for species, retaining a stronger theoretical link to current vegetation theory, while still providing traditional output, which can incorporate methods of tracking and reporting error and uncertainty information developed throughout the modelling process. The thesis concludes with a discussion of how this framework could be expanded to develop a deeper understanding of.error and uncertainty during data development, selection, and modelling, which can ultimately lead to reducing and controlling error in the modelling process.

Présentation

Version intégrale (81 Mb)

Page publiée le 27 janvier 2021