Landscape Analysis of Soil and Crop Data Using Regression

Geostatistical modeling of soil and crop variability requires that large numbers of samples be collected across the landscape, often at close intervals. Sampling costs are high, and intense sampling is generally impractical in a management context. Another problem with geostatistical analysis is the assumption of stationarity, which treats the data as a random field with some degree of correlation amongst neighboring sample points. The assumption applies poorly to soil properties and crop growth, which do not only vary in random fashion. There is also a systematic component of landscape variability. It is this systematic component, caused by trends in parent materials, hydrologic functioning, and soil development across the landscape, that is of greatest interest when soil or crop attributes are assessed on a site-specific basis. A better method of mapping variability of soils and crops would require less input data, and focus on systematic variability. These concerns can be addressed through a regression approach to spatial modeling, using data collected from the land surface (e.g., topographic or remotely sensed data) to predict soil or crop attributes. The concept is not entirely new; soil scientists have traditionally used topographic information and aerial photographs in several ways. For example, hills lope partitioning (Ruhe & Walker, 1968) is one approach to explain topographic variation in soils or crops (Brubaker et aI., 1993; Stone et aI., 1985). Also, aerial photography has been used as an aid in soil mapping (Soil Survey Staff, 1951). Milfred and Kiefer (1976) showed that landscape patterns of loess thickness could be discerned using aerial photographs taken during crop growth. Other studies have applied regression analysis (or shown correlation) between surface-collected data and soil or crop parameters. Terrain modeling (Moore et aI. 1991) has been usedc to map variation of soil properties within small drainage basins (Moore et aI. 1993) where runoff is an important hydrologic component. Phototone data obtained by scanning aerial photographs (taken during bare soil conditions) has been related to soil organic matter (Robert, 1993; Zheng & Schreier, 1988), and