Show simple item record  

dc.contributor.authorSchicker, Renee Deborah
dc.contributor.authorMoon, Vicki G.
dc.date.accessioned2012-06-25T03:14:21Z
dc.date.available2012-06-25T03:14:21Z
dc.date.copyright2012-08
dc.date.issued2012
dc.identifier.citationSchicker, R. & Moon, V. (2012). Comparison of bivariate and multivariate statistical approaches in landslide susceptibility mapping at a regional scale. Geomorphology, 161-162, 40-57.en_NZ
dc.identifier.urihttps://hdl.handle.net/10289/6425
dc.description.abstractLandslide susceptibility assessment was undertaken for the Waikato Region, New Zealand. Landslide inventory data were extracted from a pre-existing database that included few landslides in the region (1.4% of area), and is limited in terms of completeness of record and location uncertainty. This database is in contrast to those normally used for research, which are derived for the research project and are complete and accurate, but is representative of those that may exist within government bodies. This paper applies statistical methods to derive a meaningful predictive map for planning purposes from such a relatively poorly defined database. Susceptibility maps for both logistic regression and weights of evidence were derived and evaluated using success, prediction, and ROC curves. Both statistical methods gave models with fair predictive capacity for validation samples from the original database with areas under ROC curves (AUC) of 0.71 to 0.75. An independent set of landslide data compiled from observations made in Google Earth showed lower overall prediction quality, with the logistic regression method giving the best prediction (AUC = 0.71). For this regional assessment, categorical data proved a major constraint on the application of logistic regression as the area considered has complex geology and geomorphology. As a result, the large number of categories required led to a complex and unwieldy statistical model, whereas division into fewer categories meant that real variability in the area could not be adequately represented. This limited the result to a model with two continuous variables, slope and mean monthly rainfall. The incomplete record in the database proved of little concern for the logistic regression method as the model was able to generalise landslide locations from the known sites well, giving a similar AUC value for the original and independent data; the same was not true for the weights of evidence method which was not successful at predicting landslides outside those in the original data.en_NZ
dc.language.isoen
dc.publisherElsevieren_NZ
dc.relation.ispartofGeomorphology
dc.relation.urihttp://www.sciencedirect.com/science/article/pii/S0169555X12001626en_NZ
dc.subjectlandslide susceptibilityen_NZ
dc.subjectGISen_NZ
dc.subjectlogistic regressionen_NZ
dc.subjectweights of evidenceen_NZ
dc.titleComparison of bivariate and multivariate statistical approaches in landslide susceptibility mapping at a regional scaleen_NZ
dc.typeJournal Articleen_NZ
dc.identifier.doi10.1016/j.geomorph.2012.03.036en_NZ
dc.relation.isPartOfGeomorphologyen_NZ
pubs.begin-page40en_NZ
pubs.elements-id37635
pubs.end-page57en_NZ
pubs.volume161-162en_NZ
uow.identifier.article-noCen_NZ


Files in this item

FilesSizeFormatView

There are no files associated with this item.

This item appears in the following Collection(s)

Show simple item record