Automatic landform stratification and environmental correlation for modelling loess landscapes in North Otago, South Island, New Zealand

Abstract

The advent of quantitative soil-landscape modelling techniques has seen the mapping of pedological phenomena placed on a more scientific footing. In New Zealand, the spatial distribution of loess has a significant influence on soil properties, and recent regional-scale quantitative models have attempted to model the thickness and pattern of this important soil parent material. Here we apply two quantitative modelling techniques, automatic landform stratification and environmental correlation, to test and refine loess-landscape models for a higher-resolution study window in North Otago, South Island. Field validation demonstrated that previously developed coarse-scale models were moderately successful in predicting loess thickness, but these models required refinement. Automatic landform stratification based on conceptual models of loess distribution was a good predictor of primary loess in the field (85% success), but predicted poorly occurrence of colluvial and thin loess. Environmental correlation using nominal logistic regression with equal prior probabilities was a good predictor of primary loess in the field (70% success), but predicted poorly occurrence of colluvial and thin loess. Quantitative loess-landscape models need to be further refined using higher-resolution terrain data derived from LIDAR and photogrammetric surveys. In addition, morphometric similarities between different loess landscapes needs to be addressed and the historical and spatial contingency inherent in loess-landscape evolution incorporated in models for improved spatial prediction of loess parent materials.