On the spatial pattern of soil nutrients in desert ecosystems

Ecology, March, 1996 by William H. Schlesinger, Jane A. Raikes, Anne E. Hartley, Anne F. Cross

INTRODUCTION

Several years ago, we postulated that various processes, especially overgrazing, that create spatial heterogeneity in the soils of arid and semiarid grasslands may lead to the invasion of these communities by desert shrubs (Schlesinger et al. 1990). Shrubs may further localize soil fertility under their canopy, leading to the development of "islands of fertility," which characterize shrub-desert and steppe ecosystems worldwide (Crawford and Gosz 1982, Noy-Meir 1985). Our hypothesis suggested that the accumulation of nutrients under desert shrubs is an autogenic process that may promote the persistence of shrubs in the community and the desertification of grasslands that are invaded by shrubs (Reynolds et al., in press).

Implicit in this view was the assumption that the distribution of soil resources in grasslands was relatively uniform. Hook et al. (1991) found a fine-scale pattern of soil nitrogen in a semiarid grassland, suggesting that the changes in soils that accompany the transition from grassland to shrubland may be associated with a change in the scale of soil heterogeneity, rather than with the initial development of heterogeneity per se (cf. Tongway and Ludwig 1994). Thus, the main objective of this paper is to compare the scale of soil heterogeneity in arid and semiarid grasslands to that in desert shrublands of the southwestern United States.

In this paper, we use geostatistics to describe the spatial variation in soil nutrient distribution in different ecosystems by the calculation of a semi-variogram [ILLUSTRATION FOR FIGURE 1 OMITTED], which shows the average variance found in comparisons of samples taken at increasing distance from one another, the lag interval. For randomly distributed data, one would expect little change in the semi-variance ([Gamma]) encountered with increasing distance (i.e., the total sample variance is found at all scales of sampling), and the semi-variogram is essentially flat (Rossi et el. 1992, Robertson and Gross 1994; [ILLUSTRATION FOR FIGURE 1 OMITTED], curve a). For patterned data, the semi-variogram first rises from comparisons of neighboring samples that are similar and autocorrelated and then levels off at the sill, indicating the distance beyond which samples are independent ([ILLUSTRATION FOR FIGURE 1 OMITTED], curve b). A spherical model is often used to fit this form of semi-variogram (Webster 1985, Issaks and Srivastava 1989). Statistics from the spherical model indicate the range over which samples show spatial autocorrelation ([A.sub.0] in [ILLUSTRATION FOR FIGURE 1 OMITTED]), an index of the scale of spatial pattern in the community involved. Variance that exists at a scale finer than the field sampling is found at 0 lag distance and is known as nugget variance ([C.sub.0]). A high nugget variance indicates that most variance occurs over short distances, and a high ratio of nugget variance ([C.sub.0]) to sill variance ([C.sub.0] C) is an indication of a random pattern among the data (Trangmar et al. 1985).

We examined soil nutrient distributions in 11 sites representing arid and semiarid habitats of the southwestern United States (Table 1). We sampled grassland and shrubland sites at the Sevilleta National Wildlife Refuge (n = 1 each) and the Jornada Experimental Range (n = 2 each) in the Chihuahuan Desert of New Mexico, where various desert shrubs, including Larrea tridentata, have invaded perennial grasslands, dominated by Bouteloua eriopoda, during the last century (Buffington and Herbel 1965). We compare the distribution of soil nutrients in these shrubland communities to that in two sites dominated by Larrea tridentata in the Mojave Desert of California, where desert shrublands have existed in conditions of extreme drought for millennia (Thorne 1986) and some individual Larrea tridentata appear [greater than] 10 000 yr old (Vasek 1980). We also compare the distribution of soil nutrients in the Bouteloua eriopoda communities of the Chihuahuan Desert to that in the short-grass steppe, dominated by Bouteloua gracilis, at the site sampled by Hook et al. (1991) in the Central Plains Experimental Range of northeastern Colorado.

We hypothesized that the distribution of soil nutrients in desert shrublands would show spatial autocorrelation up to the average size of the dominant individuals. This pattern should be seen for biologically essential elements, especially those that are limiting in desert ecosystems (e.g., N), but spatial autocorrelation [TABULAR DATA FOR TABLE 1 OMITTED] could be seen for nonlimiting, mobile elements (e.g., Na, Li, and Cl) as well, if physical processes (e.g., wind erosion) are important to the development of shrub islands. For grassland ecosystems, in which grass clumps are typically [less than]20 cm diameter, our hypothesis was that most of the variance in soil nutrients would be found at a distance less than our smallest lag, 20 cm, and we expected that the semi-variogram would show a high nugget-to-sill ratio, indicating a random pattern, for all elements.