Research Archive

Quantitative soil-landscape modelling in North Otago, South Island, New Zealand

Research Archive

Quantitative soil-landscape modelling in North Otago, South Island, New Zealand

Hughes, Matthew W.
Type: Thesis
Permalink: https://hdl.handle.net/10182/2500
Originally published: 2003 by Lincoln University
Keywords: digital elevation model; global positioning system (GPS); Geographic Information System (GIS); discriminant function analysis; digital terrain attributes; soil mapping unit; soil taxonomic unit; qSLM 1; qSLM 2; A horizon percent C and N
Fields of Research: 080110 Simulation and Modelling; 050305 Soil Physics; 0406 Physical Geography and Environmental Geoscience

Abstract:

Digital Elevation Model (DEM), Global Positioning System (GPS) and Geographic Information System (GIS) technologies can be used to improve soil survey method and soil resource information systems, and to explicitly define soil-landscape models (SLMs) in a quantitative manner. Soil data can be more realistically modelled and represented in a semi continuous manner rather than using chloropleth maps. The aim of the study was to develop quantitative SLMs for a part of the North Otago downlands, with three objectives: 1) to replicate mapping rules used for a 1:50,000 soil map of North Otago (qSLM 1), 2) to map soil taxonomic units (STUs) based on a new dataset from a randomised-stratified sample of soil profiles observed from auger borings (qSLM 2) and 3) to investigate relationships between A horizon percent organic C and N and parent materials, terrain attributes and microclimate. Discriminant Function Analysis (DFA) used training data from 1) a randomised subsample (n=100) of each Soil Mapping Unit (SMU) of the 1:50,000 map for qSLM 1 and 2) 145 field sample points allocated to predefined STUs for qSLM 2. Digital terrain attributes derived from a 25 m DEM (elevation, aspect, slope, profile and plan curvature, wetness and stream power indices) were used as variates in the DFA. Slope, profile curvature and stream power index were the most useful terrain attributes for discrimination between SMUs in qSLM 1 and success of DFA-derived classification functions ranged 39-56% among lithological units. Stream power index, slope, upslope area and plan and profile curvature were the most useful terrain attributes for qSLM 2, with overall classification success 51 %. The mapping procedure for qSLM 1 and qSLM 2 consisted of allocating 25 m grid cells to SMU or soil series, respectively, on the basis of discriminant function scores. For a given grid cell of the map, discriminant function scores were calculated from the terrain attribute values generated by intersecting that grid cell with the terrain attribute grids. The grid cell was assigned to the SMU (qSLM 1) or STU (qSLM 2) whose associated discriminant function produced the highest score. Correspondence between predicted SMUs of qSLM 1 and the original 1:50,000 map is 39% averaged across all SMUs (range 8-93%). When considered with respect to success in predicting soil taxa at the 145 field sites the 1:50,000 map performed appreciably better (54%) than qSLM 1 (39%). Some STUs were predicted successfully enough by qSLM 1 to justify using it to extrapolate to unmapped areas of similar geology, physiography and loess deposition regime. qSLM 2 had no independent data set, and comparison with the original 1:50,000 soil map was not considered appropriate because the two mapping approaches operate at much different resolutions of landform. Both qSLM 1 and qSLM 2 produced maps with much smaller grain size (characteristic delineation) than the original 1:50,000 soil map. The patterns mirrored landform elements and appeared realistic. It was concluded that qSLM 1 rules defining the spatial distribution of some soils could serve as the basis of further modelling and extrapolation. qSLM 2 is based on more objective data, yet requires further field data for testing its validity. DEMs, GPS and GIS show potential for further detailed modelling of soil types and properties in North Otago. A horizon percent organic C and N values (n=175) were highly correlated (R2=0.94). There were significant differences in mean C and N values between parent materials (P<0.001) but classification by parent material explained only 24% and 29% variance in C and N values, respectively. Digital terrain attributes explained only 6% and 5% of variance in C and N values, respectively. Lowest summer rainfall microclimate data explained only 4% and 5% of C and N variance, respectively. The poor correlation between %C or %N and terrain attributes precluded a SLM being developed on the basis of those attributes. A SLM based on parent material alone could have been produced but, generally, variances of %C and %N within parent materials were sufficiently high to make this of little value.

Show full item record

Files in this item

[PDF]
Name: hughes_mapplsc.pdf
Size: 30.40Mb
Format: PDF
Description: thesis
[PDF]
Name: hughes_mst_permis ...
Size: 140.5Kb
Format: PDF
Description: permission (admin. only)

This item appears in the following Collection(s)

Search Research Archive


Browse

My Account

Statistics