Statistical analysis of soil data from the McMurdo Dry Valleys, Antarctica
Degree GrantorUniversity of Canterbury
Degree LevelPostgraduate Certificate
Degree NamePostgraduate Certificate in Antarctic Studies
Pedological data from over a decade of field trips to the McMurdo Dry valleys has been collated into a dataset. This data includes site observations (location, topographical position, estimated glacial history and soil age), morphological observations (full pit profile description), soil taxonomic classification, surface observations (weathering characteristics of boulders), chemical measurements of the major anions (Cl, S04, N04) and cations (Ma, Mg, Ca, K), electrical conductivity and pH. An exploratory statistical analysis was performed on this dataset to determine which analyses were appropriate given the format of the dataset, and its quality and quantity. Box plots were used to study the variability of variables according to different groupings. Multivariate analyses including a factor analysis, discriminant analysis, cluster analysis and machine learning algorithms were all applied. Geostatistical analyses investigated the spatial dependence of some of the observations. Most of the variability analyses indicated little differences in the ranges of soil properties between groups (weather stage, eco-climatic zone, taxonomic class, geological age). Where there were differences some trends were obvious and others were unexpected. The multivariate analyses did separate the pits and observations into groups that seem reasonably sensible. Little spatial dependence was found. It is concluded that the Bockheim dataset is sufficiently comprehensive for statistical analyses. The next stage in this work requires pedological input to refine those analyses that either have results of interest or have the potential to provide information of interest.