Integrating forest soils information across scales: spatial prediction of soil properties under Australian forests.

Abstract Quantitative spatial predictions of forest soil properties and land qualities can be produced using field measurements of soil layer and site properties as response variables and the more readily available spatial environmental explanatory variables derived from digital elevation models (DEM), terrain analysis, digital climatic surfaces, and geophysical and multi-spectral remote sensing. The two sets of measurements have contrasting scales, or geometric support, and this raises a number of methodological issues. Use of environmental correlation for predicting soil distribution is most effective when an a priori pedogenic model is proposed for a region. This model is used to design the sampling strategy and evaluate the adequacy of available soil data. Geographic positioning systems (GPS) enable accurate location of field samples and correct registration with environmental coverages (e.g. DEM, remote sensing, etc.). Field measurement should focus on forest soil processes or land qualities that affect forest productivity and management such as soil physical and chemical fertility. The final modelling phase extends the limited point information to the landscape level by developing statistical models. A major challenge in environmental correlation modelling is to generate spatial measures of the critical environment factors affecting soil development. A second major consideration is the varying scales of process, observation and prediction. Robust prediction requires consideration of at least three levels of organisation. First is the level in question (e.g. catena); second is the level below which provides an insight into mechanisms (e.g. pedon); and third is the level above which provides context and significance (e.g. watershed). Predictive relationships developed at one level may not be useful for prediction at more than one level removed. Digital elevation models provide the basis for spatial representations of topography, climate and, thus, the majority of available environmental coverages. The resolution of these surfaces is often coarser than that of actual soil measurement creating an inequality in scale. Causal processes controlling soil formation may not have clear surface expression in ancient landscapes or where endogenous soil forming processes dominate; thus limiting the use of land-surface morphometry. Geophysical remote sensing is currently the only means of obtaining comprehensive, subsurface geological information. Results from two quantitative forest soil surveys at different scales in eastern Australia are presented. The Bago–Maragle study covers a 50,000 ha area of sub-alpine eucalypt forest of southern New South Wales. The second survey is at a finer scale, covering two native eucalypt catchments in a granitic landscape in south-eastern NSW. Statistical models are presented with the soil and site properties as the response variables and the various spatial environmental coverages as the explanatory variables.