Creating and testing a geometric soil-landscape model in dry steeplands using a very low sampling density

Abstract

The aim of this study was to determine whether a predictive geometric soil-landscape model, potentially applicable to 400 000 ha of seasonally dry greywacke steeplands in New Zealand, could be created for 29 soil properties, using a very low soil sampling density. We postulated that in these deeply dissected steeplands which have relatively uniform geology and slope form, landscape geometry (through its effects on microclimate), rather than vegetation, geology, or slope form will control the soil pattern. To create and test the soil-landscape model we sampled the 26 000 ha Benmore Range, South Canterbury, New Zealand, in a formally stratified way so that trends of soil carbon, soil nutrients, and profile characteristics could be established for predominant slopes, at various altitudes and aspects. We used a factorial sampling system (3 land systems × 3 altitudes × 4 aspects × 2 slope positions), giving 72 sampling sites in total, and a sampling density of one site per 360 ha. Altitude and aspect had significant (P < 0.05) effects on many topsoil characteristics, particularly those likely to be related to soil moisture status, leaching, and weathering (e.g. topsoil pH, carbon, nitrogen, and phosphate retention). For most soil properties the effect of slope position was not significant (P > 0.05). The soil-landscape model was tested by comparing predicted and actual soil properties at a further 22 sites. Soil properties that were laboratory-determined were generally satisfactorily predicted by the model, but properties based on several measurements (e.g. nutrient amounts in units of kg/ha) were less satisfactorily predicted, presumably because they incorporate more measurement error. Trends of soil properties that showed strong altitude and aspect relationships were effectively illustrated using 360° ‘radar diagrams’. We conclude that for dry steeplands of uniform geology, with simple and repeated landforms at the output scale being used, a geometric soil-landscape model based on a very low sampling density successfully predicts soil properties on dominant landscape units. The methodology has application to national resource inventories.