Spatial patterns of vegetation, soils, and microtopography from terrestrial laser scanning on two semiarid hillslopes of contrasting lithology


  • Ciaran J. Harman,

    Corresponding author
    1. Department of Geography and Environmental Engineering, Johns Hopkins University, Baltimore, Maryland, USA
    2. Department of Hydrology and Water Resources, University of Arizona, Tucson, Arizona, USA
    Search for more papers by this author
  • Kathleen A. Lohse,

    1. School of Natural Resources and the Environment, University of Arizona, Tucson, Arizona, USA
    2. Department of Biological Sciences, Idaho State University, Pocatello, Idaho, USA
    Search for more papers by this author
  • Peter A. Troch,

    1. Department of Hydrology and Water Resources, University of Arizona, Tucson, Arizona, USA
    Search for more papers by this author
  • Murugesu Sivapalan

    1. Department of Geography, University of Illinois at Urbana-Champaign, Urbana, Illinois, USA
    2. Department of Civil and Environmental Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois, USA
    Search for more papers by this author


Shrublands in semiarid regions are heterogeneous landscapes consisting of infertile bare areas separated by nutrient rich vegetated areas known as resource islands. Spatial patterns in these landscapes are structured by feedbacks driven by the transport of water and nutrient resources from the intershrub space to areas below shrubs, and the retention of these resources to locally drive productivity and tight biogeochemical cycles. Most understanding of plant-soil feedbacks is based predominantly on studies of low topographic gradient landscapes, and it is unclear whether the patterns of association between soils and vegetation, and the autogenic processes that create them, also occur on more steeply sloping terrain. Here we analyze the spatial patterns of soils, vegetation, and microtopography on hillslopes of contrasting lithology (one granite at 16°, one schist at 27°) in the Sonoran desert foothills of the Catalina Mountains. We also describe a method of extracting vegetation density from terrestrial laser scanning point cloud data at 5 cm × 5 cm scales and find that it correlates well with soil organic carbon measurements. Vegetation was associated with microtopographic mounds (relative to the mean slope) extending 0.3 m downslope and 1.8 m (schist) and 0.9 m (granite) upslope on the study hillslopes. Soils below the shrub canopies exhibited 2–3 times more soil organic matter and 2–4 times higher hydraulic conductivity than the interspaces. Soils enriched with organic matter were found to extend at least two canopy radii downslope of woody shrubs, but not upslope. These plumes were clearest in the lower gradient granite site where vegetation mounds created distinct patterns of microtopographic convergence and divergence. At the steeper schist site, microtopography appeared to have a weaker control on topographic flow accumulation. Collectively, our findings suggest that the spatial structure of association between soils and microtopography and vegetation on these slopes exhibit many of the features observed in lower gradient areas. However, microtopography and soils are more asymmetric along the downslope axis of the hillslopes than lower gradient areas and vary with lithology. Alluvial and colluvial processes are likely more important in shaping vegetation and soil dynamics on hillslopes, and these factors need further consideration in scaling results to the landscape level.

1 Introduction

Feedbacks between soils, biota, and hydrology are believed to be an important control on dryland ecosystems through their control on hydrologic partitioning [Huxman et al., 2005], runoff generation, erosion [Wainwright et al., 2000], and nutrient transport [Schlesinger et al., 1999; Brazier et al., 2007]. The spatial patterns of vegetation play a key role in modifying the spatial patterns of water, carbon, and nutrient resources in semiarid areas [Noy-Meir, 1973; Ludwig and Tongway, 1995; Schlesinger and Pilmanis, 1998; Reid et al., 1999; Ludwig et al., 2005] and are an indicator of desert ecosystem health (and conversely, desertification) [Kéfi et al., 2007]. On stable geomorphic surfaces, the areas around individual shrubs, patches, or groves of plants develop higher concentrations of nutrients [Schlesinger et al., 1996; Schlesinger and Pilmanis, 1998], soil organic matter [Gonzalez-Polo and Austin, 2009], soil moisture [Bhark and Small, 2003], and microbial biomass and activities [Bolton Jr. et al., 1993] compared to surrounding nonvegetated areas.

These spatial patterns are believed to arise from feedbacks between vegetation, soils, and hydrologic processes across scales [Turnbull et al., 2008; Reynolds et al., 2004; Caylor et al., 2009; Ludwig et al., 2005]. In part, these feedbacks are generated by the alteration of surface soil properties in the vicinity of the plant in ways that help retain resources transported by overland flow [Puigdefabregas et al., 1999]. Vegetated patches have been found to have higher roughness [Muñoz Robles et al., 2011] and a greater concentration of macropores [Dunne et al., 1991], and are often raised relative to the surrounding area [Buis et al., 2010]. Surface and root litter generated by the plants is incorporated into the patches' soil organic matter pool, where it increases soil aggregation [Stavi et al., 2010]. These soils often have higher infiltration rates [Dunne et al., 1991; Dunkerley, 2002; Thompson et al., 2010] and higher water retention capacity [Greene, 1992; Wilcox et al., 2003], though opposite trends have been observed [Caldwell et al., 2008]. Water, sediments, and nutrients transported from interspaces by overland flow create pulses of productivity and reinforce the vegetation pattern [Tongway and Ludwig, 1997; Reynolds et al., 2004; Ludwig et al., 2005; Austin et al., 2009].

In this way the landscape becomes self-organized into a mosaic of patches and interspaces [Virginia, 1983; Schlesinger et al., 1996], where interspaces act as a source of resources deposited below canopy (including atmospheric depositions of dust and precipitation) and the vegetated areas are a sink, enriched in these resources and carbon from litter, and protected by the vegetation and altered soil properties [Puigdefabregas, 2005]. This patchiness allows a larger biomass to grow in semiarid regions than uniform vegetation cover could sustain [Aguiar and Sala, 1999]. Several studies have shown that the boundary distinguishing altered soils from the surrounding area is not sharp, but rather, there is a broad transition that produces a “zone of influence” on the order of 2–4 times the aerial extent of the canopy [Schlesinger and Pilmanis, 1998; Dunkerley, 2000a, 2002; Madsen et al., 2008; Bedford and Small, 2008; Caldwell et al., 2008].

However, this conceptual understanding of plant-soil feedbacks is based predominantly on studies performed in low topographic gradient landscapes. It is unclear whether these patterns and self-reinforcing processes hold on more steeply sloping terrain. Bedford and Small [2008] studied spatial patterns of soil properties on a range of landforms at the Sevilleta National Wildlife Refuge in New Mexico and found no difference in infiltration capacity between canopy and interspace areas on steeper bedrock-controlled hillslopes with slopes around 18–24°. They suggested that the downslope transport of sediments erased the alteration of soil properties observed under vegetation on more stable landscapes. Puigdefabregas et al. [1999] found that when compared to the more stable alluvial fan surfaces, the steep upper hillslope at the Rambla Honda site in Spain showed much lower levels of spatial differentiation in texture, conductivity, and runoff ratio between bare ground and plant clumps.

The recent ability of field-portable high-resolution Terrestrial Laser Survey (TLS) equipment makes it possible to simultaneously observe vegetation and microtopographic structure at high resolution. For example, this technology has facilitated a number of recent studies into the relationships between fine-scale microtopography and erosion following fire [Soulard et al., 2012; Sankey et al., 2012a; Eitel et al., 2011; Sankey et al., 2011]. Combining this data with soil sampling and analysis provides a detailed picture of spatial patterns in vegetation, the soil surface, and the distribution of soil nutrients and hydraulic properties. However, it is challenging to extract the vegetation structure from the TLS point cloud data.

Here we develop a method of analyzing point cloud data and use it to analyze the spatial patterns of vegetation, microtopography, and soils on two steep bedrock-controlled slopes of contrasting lithology in semiarid southern Arizona. The sites studied have contrasting lithology and are both moderate to steeply sloped, while other state factors, such as climate, potential biota, and age [Jenny, 1941], are similar.

We use the vegetation and soil sampling data to test the following hypotheses. First, that there are significant differences between soils in the canopy and interspace areas in terms of organic matter content, bulk density, coarse fraction, and hydraulic conductivity on these steep bedrock-controlled hillslopes. Second, that the variations in soil organic matter content are tightly associated with the vegetation density measured with the TLS at the scale of a hillslope transect of observations. Finally, we test whether a clear “zone of influence” of biotically altered soils exists around individual shrubs, as measured by the surface soil organic matter content, or whether this zone is absent as Puigdefabregas et al. [1999] and Bedford and Small [2008] observed at their sites.

2 Study Sites and Methods

2.1 Study Sites

Study sites were chosen on north-facing hillslopes in two small watersheds in the foothills of the Santa Catalina Mountains, Arizona, where strong geologic contrasts create a natural laboratory for examining the role of geology on soil-vegetation-hydrology feedbacks. The watersheds are part of the Jemez-Santa Catalina Critical Zone Observatory (JRB-SCM-CZO, [Chorovera et al., 2011], are located within 500 m of one another, and have the same history of climate and land use. The sites are at around 1050 m elevation, with a mean annual precipitation of around 490 mm (at the nearby Oracle gauge). The climate is characterized by hot summers and mild winters with two distinct rain seasons. Intense, localized summertime convective rainfall (the North American Monsoon) accounts for more than half of the total precipitation. Mean temperatures range from 26°C in July to 7°C in January.

One of the hillslopes is underlain by Tertiary-aged granites, and the other by Paleozoic aged metamorphic schist [Richard et al., 2000]. The study area on the granite slope has an average gradient of 16°, while the schist is steeper at around 27°. The granite hillslope is an approximately 100 m long side slope of a valley incised into a low-gradient upland that extends over a larger area. Due to an outcrop of rock at the top of the study area, overland flow contributions from the upslope area are directed away from the study hillslope. The schist site is a 100 m side slope below a sharp interfluve of only a few meters width. The different parent materials weather to produce very different soils: the granitic rocks yield coarse quartz and feldspar crystals that resist further weathering, while the foliated schist weathers to produce channery soils (containing a significant fraction of thin, flat coarse fragments) with a fine matrix. Soils on the foothill slopes are shallow entisols of around 30–50 cm thickness. The schist site is mapped as the Romero series (an Ustic Torriorthent), while the granite site is not mapped but closely resembles the Oracle complex (an Ustic Haplargid). The granite hillslope consists of areas of soil up to 30 m across between fins of exposed bedrock oriented downslope. The schist site has a more uniform mantle of soil, with only occasional, isolated areas of highly weathered bedrock exposed on the surface. Study locations were selected in planar upper midslope areas (so that they were not dominated by depositional processes) that were largely free of exposed bedrock.

Vegetation can be broadly classified as Upper Sonoran Desert [Whittaker and Niering, 1965, 1975], dominated by a mosaic of bare soil and bunch grass interspersed with cactus, small woody shrubs, and velvet mesquite (Prosopis velutina). Each site contained a Mesquite tree in the central portion of the site. The sites are open to grazing, but grazing is light due to the steep terrain and proximity of more productive and accessible valley bottoms. Other biogeomorphic activity such as animal burrowing is much more evident, particularly at the granite site.

Our initial visual impression of the sites suggested that there was a high level of spatial segregation in lighter and darker colored soils (assumed to indicate enrichment in organic matter), particularly associated with the woody shrub Calliandra eriophylla and Gutierrezia sarothrae and with the mesquite. Isolated perennial bunch grasses appeared to have less association with these enriched surface soils. The spatial distribution of the soils appeared to be enriched downslope of the canopies. The field work described below was conducted to quantify this apparent pattern.

2.2 Terrestrial Laser Scanning

The terrestrial laser scans were carried out using a Leica C10 ScanStation between December 2011 and January 2012. At each site nine to eleven low or medium point density scans of 360° views were taken to obtain a usable point cloud coverage of an area approximately 10 m across the slopes and 20 m downslope. The ScanStation was mounted on a standard tripod, around 5 feet above ground level. This instrument uses a 532 nm laser and has a ±1σ positional accuracy of 6 mm at up to 50 m range. Scans were all conducted at locations within or on the perimeter of the study area and chosen to minimize shadowing effects. Four circular targets were used at each site to register scans together with a positional accuracy of around 2 mm. Scan point cloud data were processed using Leica Cyclone [Leica Geosystems, 2009] to obtain a unified cloud.

A digital terrain model (DTM) was obtained by taking the lowest return in 5 cm by 5 cm pixels in a grid. The grid was aligned along the direction of steepest descent of a plane fitted through the raw data. This set of lowest return points was then filtered using a 5-by-5 pixel moving-average window to identify lowest returns more than 5 cm above the local mean, which were excluded under the assumption that they represented points in the vegetation, not the ground. The resulting point set was then used to linearly interpolate a 5 cm by 5 cm gridded terrain model. Areas beneath Sotol (Dasylirion wheeleri) were completely shielded from observation (and most likely overland flow) by a dense apron of dead leaves and were excluded from the DTM. The DTM was analyzed using GRASS [GRASS Development Team, 2012] to obtain rasters of upslope accumulated area (using the multiple flow direction algorithm in r.watershed) and local slope. Note that the study areas are midslope, and the calculated accumulated areas did not include the contributions from upslope of the site.

2.3 Extracting Vegetation Density From the Point Cloud

Point cloud density cannot be used to measure vegetation density, although the TLS point cloud is denser in regions where vegetation is dense. Objects close to the center of the scanner are measured at a higher density than objects further away, which creates a spatial bias. The method we use to map vegetation structure at high spatial resolution must be able to filter this bias.

A robust method for extracting the vegetation was developed to estimate vegetation density based on a box-counting approach. Box counting is often used to give a measure of the “volume” of a fractal object when observed at a specific resolution [Liebovitch and Toth, 1989]. Vegetation density was first estimated in 5 cm cubic voxels. A column of 5 cm high cubic volumes (voxels) was constructed above each 5 cm by 5 cm pixel in the DTM. Each voxel was subdivided into 125 cubic subvoxels of 1 cm3. Subvoxels were identified as “occupied” if any returns were recorded within the voxel, and “unoccupied” otherwise. The number of occupied subvoxels was taken to represent the volume occupied by vegetation within the voxel in cm3. Biovolume (per unit area) was determined by summing the biovolume in a column of voxels and dividing by the voxel footprint (25 cm2). The values obtained by this method are necessarily dependent on the resolution of the voxels used and were chosen to reflect the accuracy of the TLS returns and scan alignment.

This approach was checked by calculating the redundancy of returns in each 5×5 cm2pixel. The “return redundancy” is the ratio of the number of returns in the column above a pixel relative to the minimum number that would be needed to register the given biovolume. For instance, if the biovolume of a 5×5 cm pixel is 10 cm3/cm2, a minimum of 250 returns would be needed to register at least one return in 250 of the 1 cm3 subvoxels above that pixel. If 750 returns were registered, the redundancy ratio would be 3, since 3 times as many returns were registered. A high value of the redundancy ratio suggests that we can have a higher confidence that the returns represent the surface of a real object, rather than noise. The result (not shown) showed that the redundancy ratio is on average greater than 2. Thus, the point density generally included many redundant returns, which gives us confidence that the this analysis is giving a realistic representation of the vegetation density.

2.4 Vegetation Mapping and Validation of the TLS

To better understand the vegetation patterns of the site, and to provide some validation of the biovolume method to ensure that the values were realistic, we conducted vegetation transect surveys.

Vegetation species were mapped in four transects at each site, and using a grid of point observations in the field. At each site, two transects were taken downslope and two across the slope. Transects were conducted in March 2010. The transects cover only a 12 m by 14 m section of the upslope part of the schist site, while the surveys extend across the length of the entire granite site. One of the transects in each direction was sited to intersect a mesquite tree in each site, and the other was laid in parallel part-way across the study area. Along each transect, the species (or genus, if species could not be determined), position, and canopy dimensions of each plant crossing the transect tape were recorded. Because high-accuracy GPS equipment was not available to georeference the TLS scans, the position of the vegetation transects and soil sampling relative to the scans was estimated by matching the position of individual plants. The biovolume voxels corresponding to each plant in the transect were summed in each dimension, and the distribution of the results (in cm3/cm2, or simply cm) was compared to the recorded height and canopy width and length. To characterize the site, ground cover and canopy cover were also recorded in an approximate 1 m grid. A larger area was mapped at the granite site (414 point observations, 66 m of transects) than the schist site (254 point observations, 53 m of transects). In both cases, only perennial vegetation was recorded.

2.5 Soil Sampling and Analysis

Soils were sampled on three occasions to examine whether significant differences could be found between canopy and interspace soils, to examine the association with aboveground vegetation density, and to examine the extent of the “zone of influence” of individual plants or clumps of plants. The first set of samples was used to determine whether there were significant differences below and between canopy. Eight vegetated patches were randomly chosen, and a pair of samples were taken from below the canopy and from the nearest bare interspace in any direction. The largest individual in each patch at the granite site was a mesquite (Prosopis velutina) in two cases, Calliandra eriophyllain four cases, Artemisia ludoviciana in one case, and Bouteloua curtipendula once. At the schist it was Prosopis velutina in four cases and Gutierrezia sarothrae in the other four. At all locations, coarse surface litter was collected from the sampling point and a 57 mm diameter corer containing three 3 cm sampling rings was used to extract intact samples from 0 to 3 cm and 3 to 6 cm depths for analysis of hydraulic properties and bulk density [Grossman and Reinsch, 2002]. Cores that contained large rocks or had been disturbed during extraction were discarded and new ones collected. A further set of nonintact (bulk) cores were then collected to provide samples of 0–3 cm, 3–9 cm, and (where possible) 10–19 cm for analysis of their composition.

The bulk soil samples were sieved to 2 mm, and the coarse fraction weighed. Roots that remained on the 2 mm sieve were collected and dried, and the <0.5 mm fraction, 0.5–2 mm fraction, and >2 mm fractions were weighed. Subsamples of the soil were dried in an oven at 60°C for 24 h, weighed and then combusted in a furnace at 350 C for 24 h to determine soil organic matter (SOM) content from loss on ignition (LOI) [Nelson and Sommers, 1996; Schulte and Kaufmann, 1991]. To test the error in the LOI measurements, a subset of soils, also dried at 60°C, were ground and sent to the University of California, Davis Stable Isotope Facility, for total carbon analysis on an elemental analyzer (Elementar Vario EL Cube or Micro Cube elemental analyzer, Elementar Analysensysteme GmbH, Hanau, Germany). LOI SOM was correlated with the elemental analyzer C data after both were log-transformed to normalize the distributions. Analysis of covariance (ANCOVA) using MATLAB [MATLAB, 2012] showed that the slope of the relationship was not significantly different from 1 (p=0.48, N=52) for the schist site but was different for the granite (p=0.001, N=32) due to a tendency for LOI to overestimate when C was small (<1%). This may be due to the presence of residual water in the samples prior to combustion. When the three smallest samples (out of 32) were excluded, the slope was not distinct from 1 (p=0.06, N=29). Root mean square error was 15% and 16% for the granite and schist sites, respectively, which is comparable to errors observed in other studies [Konen et al., 2002; Howard and Howard, 1990].

Soil texture was characterized at each site using a laser diffraction particle analyzer (Beckman Coulter LS12 320, Brea, CA) at the Center for Environmental Physics and Mineralogy (CEPM) at the University of Arizona. The analysis was conducted on 12 composite samples, representing soils from three depths, below and between canopy, at the two sites. To ensure that observations were representative, the composite samples were obtained by combining soils from three of the sampling locations.

Saturated hydraulic conductivity of the intact cores was estimated using falling head permeametry in a Reynolds tank [Reynolds et al., 2002] following methods similar to Chief et al. [2008]. Samples were saturated from below over a 12 h period prior to being analyzed.

The second set of soil samples aimed to quantify the relationship between soil organic matter and vegetation density derived from the TLS and consisted of a representative set of samples taken along a transect. Samples of the top 5 cm of soil were collected at 1 m intervals along a 30 m tape running upslope through the mesquite tree at each site. These samples were sieved and analyzed for LOI as above. Sample locations were estimated on the DTM.

The final set of soil samples were collected from 10 transects at each site to quantify the spatial distribution of the SOM under individual canopies. Five transects at each site were oriented downslope below woody shrubs and Velvet mesquite (Prosopis velutina). Samples were taken directly below the center of the canopy, at three points upslope, and three points downslope. Another five transects were taken along unvegetated flow lines (protorills) immediately adjacent to the vegetated transects. These flow lines were identified using the TLS data to estimate regions of high accumulated upslope area. Surface samples were taken, sieved to 2 mm, and combusted as before to obtain SOM estimates. The samples were separated by a distance approximately equal to the canopy radius, so that the samples were more closely spaced in small shrubs and wider under the trees. The shrubs were selected by walking around the sites and sampling the first woody shrub found that did not have another shrub within two canopy radii downslope and three upslope. This was done to ensure that the spatial structure around an individual shrub could be examined without overlap with the “zone of influence” of plants upslope or downslope. Some bias may be introduced by this sampling, as only a portion of shrubs at each site met this criteria. The soils around the shrubs selected for sampling may receive less protection than those in a composite patch created by several shrubs. At the granite site, Calliandra eriophylla and Artemisia ludoviciana were sampled, while at the schist site, the shrubs sampled were Acacia gregii and Gutierrezia sarothrae.

3 Results

3.1 Vegetation

The summed biovolumes approached the dimensions of the plant at some points but were generally much lower (Figure 1). This is as we would expect, given that the measured dimensions represent the maximum extent of the canopy, while the biovolume reflects its porous nature. Small shrubs had an average biovolume that was 6% of their total height, while mesquite were only 1.4%. Plants close to the ground tended to have values that were a higher percentage of their observed height (e.g., 13% on average for like C. eriophylla and 30% for Artemisia ludoviciana). Despite the fact that Dasylirion wheeleri do actually fill much of the volume attributed to them, their measured biovolume was only 6% of their actual height on average. This highlights a shortcoming of the method, since the internal parts of the plant do not generate points in the point cloud and are therefore not measured. However, those internal parts are part of the trunk of the plant and are therefore not associated with a patch of soil. Thus, they do not affect our results, and we are satisfied that the remaining biovolume results are realistic.

Figure 1.

Comparison of the volumes of plants estimated from the transect surveys and the biovolume estimated from the TLS. Transect surveys give volume of box around the whole plant and are therefore naturally much larger than the biovolume estimates, which are based on 5 cm cubic voxels.

Transect data and on-site observations clearly showed that the spatial structure of the vegetation at the two sites was composed of a mosaic of trees, large shrubs, small shrubs, bunch grasses, and bare interspaces (Figures 2-4). Average aboveground biovolume was larger in the granite site, with an mean of 135 cm3/cm2 and a standard deviation of 177 cm3/cm2. At the schist, mean biovolume was 88.9 cm3/cm2 and the standard deviation was 118.0 cm3/cm2. A larger number of species were observed at the schist site, even though a slightly larger area was surveyed at the granite site. Table 1 lists the species observed at each site.

Figure 2.

TLS data aligned along three of the eight transects used to map vegetation at the (top) granite site and (bottom) schist site. Images show the raw TLS point returns in a 50–100 cm wide swath extracted from the point cloud, with a 5 cm wide swath highlighted in green to illustrate the density of this data. The blue line shows the ground topography approximated by the lowest return in 5 cm by 5 cm pixels along that swath. Below each transect is the ground topography (solid line) exaggerated around the mean slope (dashed line)—scale of the vertical exaggeration is as shown by the 50 cm scale. Vegetation key is given in Table 1.

Figure 3.

Spatial data extracted from the TLS point cloud (left to right): hillshade of the ground surface topography, upslope accumulated area, and local slope. Gaps occur where Sotol bushes obscure the ground completely. (top) Schist site and (bottom) granite site. Dotted lines indicate the location of the sampling transect used to examine the relationship between soil organic matter and TLS data.

Figure 4.

Spatial data extracted from the TLS point cloud (left to right): total biovolume less than 1 m from the ground, total biovolume above 1 m, and vegetation mapped along the transect. Line width and color indicates the vegetation type overlapping the transect tape. (top) Schist site and (bottom) granite site. Dots on the middle plot indicate soil sampling locations for the transect samples.

Table 1. Species Observed in Vegetation Transects at Each Sitea
Scientific NameSymbol (Common Name)
  1. a

    Note the greater diversity of shrub species at the schist site.

  2. b

    Observed at both sites.

  3. c

    Observed at schist only.

  4. d

    Observed at granite only.

  5. e

    Observed at granite transects only but present near the surveyed area at the schist site.

Woody Species
Acacia gregiib (Cat-claw)ACGR
Artemisia ludovicianab (Silver wormwood)ARLU
Muhlenbergia porterib (Bush muhly)MUPO
Prosopis velutinab (Velvet mesquite)PRVE
Calliandra eriophyllab (Fairy duster)CAER
Gutierrezia sarothraee (Broom snakeweed)dGUSA
Fouquieria splendense (Ocotillo)FOSP
Condalia warnockiicCOWA
Acacia neovernicosacACNE
Celtis pallidacCEPA
Lycium berlandiericLYBE
Mirabilis bigeloviicMIBI
Gutierrezia sarothraecGUSA
Ziziphus obtusifoliacZIOB
Perennial Grasses
Bouteloua curtipendulabBOCU
Bouteloua eriopodabBOER
Eragrostis lehmannianabERLE
Digitaria californicadDICA
Aristida speciescARIS
Dasylirion wheelerib (Sotol)DAWH
Opuntia spp.b (Prickly pear and Cholla)OPSP

In addition to the greater species diversity at the schist site, there were differences in the spatial cover of vegetation. Ground cover of perennial grasses was much greater at the schist site (23%) than the granite (4%), while at the granite site, C. eriophylla and A. ludoviciana were more common ground covers. A larger fraction of the granite site was bare ground (54% versus 43%). Litter from the woody vegetation covered slightly less than a third of the ground (granite 31%, schist 27%) at both sites. The remaining ground at each site was covered by dead standing vegetation, a small fraction of exposed bedrock and points where D. wheeleri obscured the ground.

Although a larger fraction of the ground at the granite site was bare, a larger fraction of the canopy cover at the granite site was woody vegetation (54% versus 40%). Shrubs like F. splendens, C. eriophylla, and A. ludoviciana made up 40% of the canopy at the granite site, compared to only 22% at the schist, despite the higher diversity there. Mesquite tree cover was only 4% at the granite site but was 15% of the schist. The remaining cover was Opuntia sp. (granite 4%, schist 2%).

3.2 Microtopography and Vegetation

Topographic features shown in Figure 2 and in transect in Figure 3 revealed structures at a range of scales with a strong association with vegetation. Vegetation was associated with local topographic highs relative to the mean slope (Figure 2). Mounds rarely if ever represented actual local topographic “peaks” but rather represented a local minimum in gradient upslope (approaching horizontal in extreme cases) and maximum in gradient downslope. These mounds broadly scaled with the size of the vegetation, though bunch grasses and sotol were observed to have strong microtopographic associations with sediments accumulated upslope, armored, and deflated areas to either side, and a steep toe-slope. Areas below some large shrubs had a concave profile. The granite site had a more well-defined rill-interrill structure, whereas the schist site is more dispersed, with no well-defined rill structure.

To quantify the connection between topography and vegetation, we examined the correlation between total biovolume V and the elevation residual Δz(xx), defined as Δz=z(xx)−[z(x)−SΔx]. In other words, Δz(xx) is the difference between the actual elevation z at a point Δx downslope of the point x and the elevation predicted by extrapolating the mean hillslope gradient S from z(x). The correlation coefficient for different values of Δx was determined (Figure 5) and showed that larger values of V were associated with larger downslope increments in elevation (more negative Δz) at a distance of Δx=0.3 m downslope, and reduced increments (more positive Δz) upslope at Δx=−1.8 m in the schist site and at Δx=−0.9 m in the granite site.

Figure 5.

Relationship between biovolume V less than 1 m above the ground and the elevation residuals Δz at a distance Δx downslope (where Δz is difference Δz=z(xx)−[z(x)−SΔx] and S is the mean hillslope gradient). (top) Larger local values of biovolume are correlated with a larger downslope increments at a distance Δx=0.3 m downslope and reduced upslope increments upslope at Δx=−0.9 m in the granite site and Δx=−1.8 m upslope in the schist site. (middle and bottom) At these distances, the differential increments (vertical axis, in centimeters) closely follows a three parameter function (dashed line) of V4 (given in cm3/cm2).

Models predicting the average deviations from the mean slope are given below:

display math(1)
display math(2)
display math(3)
display math(4)

The reported Pearson correlation coefficients in Figure (4) cannot be regarded as a true estimate of the correlation coefficient associated with these relationships, since the individual observations of Δz and V are both autocorrelated in space. There is a significant amount of scatter in the relationships too, indicated by the low values of the computed correlation coefficient. However, what is important here is not that V is a mediocre predictor of the topography but that the microtopography shows a clear spatial organization that is asymmetric upslope versus downslope and is similar in character—though different in magnitude—between hillslopes with very different lithology, soils, and overall gradient.

3.3 Soils, Microtopography, and Vegetation

3.3.1 Canopy-Intercanopy Samples

Particle size analysis indicated that differences in texture between canopy and bare areas were negligible: in all cases differences were less than 3 percentage points, and in more than half were less than 1%. The granite site soils generally consisted of loamy sand at the surface, and sandy loam below 10 cm, whereas the schist site was mostly loam (Table 2). Clay content more than doubled between the 0–3 cm and 10–19 cm samples at the granite site and increased by just over 50% at the schist site. Silt content increased with depth in all cases and sand content declined.

Table 2. Soil Texture Measured by Laser Diffraction of Composites of a Representative Set of Samples Obtained for Each Class Were Analyzed
Veg0–3 cmLoamy Sand4.8212.0883.1
 3–9 cmLoamy Sand6.2612.3981.35
 10–19 cmSandy Loam9.7414.6675.6
Bare0–3 cmLoamy Sand4.6911.4483.88
 3–9 cmLoamy Sand7.651379.35
 10–19 cmSandy Loam9.7214.1476.15
Veg0–3 cmLoam7.8941.5250.6
 3–9 cmLoam9.1749.0541.78
 10–19 cmSilt Loam11.9851.236.83
Bare0–3 cmSandy Loam7.2440.1452.63
 3–9 cmLoam9.3347.1543.53
 10–19 cmLoam11.1849.0539.78

Figure 6 shows the results of the soil composition analysis on the eight pairs of canopy-intercanopy sampling locations and indicates the results of pairwise t tests comparing the means of samples stratified by sample depth. At both sites, the bulk density, coarse fraction, and organic matter contents of the soils were significantly different below canopy compared to between (particularly near the surface, and particularly in the Schist site, where p values exceeded 0.01), despite the high variability both below canopy and in the interspace. Soil bulk density was generally reduced near the surface 3 cm below vegetation (1.1±0.18 g/cm2 versus 1.6±0.27 g/cm2 in the granite and 0.89±0.25 g/cm2 versus 1.3±0.21 g/cm2 in the schist), but differences between vegetated and bare patches were eliminated below 3 cm in the schist site and below 9 cm in the granite site. Coarse fraction tended to be significantly reduced in the surface soil below the vegetation (11±4.2% versus 24±9.5% in the granite and 13±7.4% versus 28±7.7% in the schist). Soil organic matter at the schist site tended to be higher on average, reaching an average of 7.8±4.0% in the top 3 cm under vegetation versus 3.4±3.4% between, while at the granite organic matter was 4.0±1.8% below vegetation and only 1.3±0.33% in the interspaces. This content declined rapidly with depth, and below 10 cm at both sites, the content at the vegetated patches was not distinct.

Figure 6.

Comparison of the soil composition below canopy and in bare interspaces. Bar shows average values, and whiskers are standard deviation. Shade represents the significance of the differences between canopy and interspace samples grouped by depth in a pairwise t test. White bars indicate that canopy and interspace soils are not different at that depth at p<0.05, grey bars are significant, and black bars are highly significantly different (p<0.01).

Soil hydraulic conductivity was significantly higher below the canopy than in the bare interspaces. At the granite site, the values averaged 12 mm/h in the interspace and 40 mm/h below canopy (a difference significant at p=0.01 in Welch's t test), while at the schist site, the values were lower at 11 mm/h (between canopy) and 28 mm/h (below canopy, p=0.01). The conductivity was strongly correlated with soil organic matter content. Linear regression relationships were found predicting the log of conductivity from the log of SOM with R2=0.48 in the granite site and 0.72 in the schist (Figure 7).

Figure 7.

Hydraulic conductivity measured on intact cores as a function of soil organic matter content.

3.3.2 Long Transect Samples

A strong spatial association of surface soil organic matter and vegetation was clearly identified in the transect data (Figure 8). A significant correlation was found between the log-transformed soil organic matter and the biovolume obtained from the TLS data at the sampling points (schist R2=0.65, p=0.0003; granite R2=0.65, p=0.0043). Using these relationships, the following predictive model was constructed:

display math(5)
display math(6)

where V is the TLS-derived biovolume. To obtain these relationships, only points with a nonzero biovolume were included, and both variables were log-transformed. The mean organic matter in the areas with zero biovolume is used as the lower bound on the predicted SOM. It is interesting to note that the exponents of the two models are quite similar to each other, although the coefficients are not.

Figure 8.

(left) Vegetation biovolume (V) and soil organic matter (SOM) in the upper 5 cm along 30 m upslope transects in the granite and schist sites. (right) Good power-law relationships were found between biovolume observed by the TLS and soil organic matter (with a lower bound on the relationship given by the average SOM observed in bare sites, where V=0).

3.3.3 Shrub Transect Samples

At the scale of individual shrubs, the spatial distribution of the organic matter-enriched soils was visibly asymmetric in the granite site, but only subtly so in the schist (Figures 8 and 9). At many points it was possible to observe plumes of soil organic matter extending downslope from the areas beneath the canopies of deciduous vegetation. Photos of this phenomenon under C. eriophylla on the granite site (taken with a near-infrared camera to highlight green leaves in red) are shown in Figure 9. Surface soil organic matter in one plume was sampled using the same protocol as the transects and showed a decline in organic matter downslope (along points labeled P1 to P4 in the left photo) from values similar to those within the vegetation patch (at C) to those immediately outside in the adjoining flow line (at F).

Figure 9.

Near-infrared photographs showing typical patterns of soil organic matter redistribution around Calliandra eriophylla at the granite site. NIR channel is highlighted in red, while visible channels have been desaturated and contrast enhanced to highlight pattern. (left) Values of SOM (m/m%) observed at the points indicated (standard deviation in parentheses where replicates were taken). Blue lines show approximate overland flow paths. Letters are referred to in the text.

Figure 10 presents the average soil organic matter at points measured along the transects, with the horizontal coordinates for each transect normalized to units of canopy radius (positive downslope). At the granite site, elevated soil organic matter content extended at least 2 canopy radii downslope from the edge of each canopy. The difference in SOM content from the adjacent flow line points was all significant (p<0.05) starting at the upslope edge of the canopy (−1 radius in Figure 10) but were not significant upslope of the canopy. The downslope edge of the canopy is enriched in SOM by 76% compared to the upslope edge (averaging 3.6% versus 6.3%, a difference that is significant in a two-tailed t test with p<0.05), and the two points sampled downslope of the canopies were 36% higher than the two upslope (3.9% versus 2.9%, p<0.05).

Figure 10.

Transects of soil organic matter below vegetation canopies. Distance downslope is measured in canopy radii away from the canopy center. (top) Raw data and (bottom) mean values (error bars are ±1 standard error in the mean). SOM in the granite site shows a clear asymmetry, with a plume of SOM extending downslope. The pattern is more subtle and variable in the schist site, though unvegetated areas downslope of the canopy are enriched in SOM compared to upslope. The areas along flow lines have consistently low soil organic matter.

The results of the schist site were more variable, though the data support the existence of elevated SOM extending downslope. The upslope and downslope edges of the canopies were not significantly different from each other (p=0.05). Areas beyond the canopy were different with a marginal significance (p=0.08), with the downslope area (2–3 canopy radii) enriched in SOM by 42% compared to the points upslope (5.9% versus 4.2%).

4 Discussion and Conclusions

In this work we used vegetation surveys, soil sampling, and high-resolution topography obtained from TLS to analyze the spatial patterns of soils, vegetation, and microtopography on two steep hillslopes of contrasting lithology in the Sonoran desert. We also developed a method for processing TLS point cloud data into high-resolution maps of biovolume, a measure of vegetation density. We found that surface soil organic matter strongly correlated with this biovolume and exhibited spatial patterns suggesting slope-dependent controls on its spatial distribution.

4.1 Spatial Patterns of Association

Many of the spatial patterns commonly found in lower gradient areas occurred at the study sites, though with some important differences. Vegetation was patchy and there are considerable bare areas with no vegetation, consistent with other studies of Sonoran desert vegetation [Whittaker and Niering, 1965, 1975; Virginia, 1983; McAuliffe, 1994; Schade and Hobbie, 2005]. However, the sites exhibited a diverse assemblage of shrubs, trees, and grass, rather than being dominated by a single species (such as Larrea tridentata or Prosopis velutina) [Virginia, 1983; Potts et al., 2010]. Microbial crusts were not observed at the sites. The low stability of coarse-textured soils on steep slopes has been found to reduce the incidence of biological soil crusts [Belnap, 2003; Belnap et al., 2005].

Despite the steep slopes, vegetation was associated with microtopographic deviations from the mean slope, but the microtopography was considerably asymmetric along the downslope axis, as shown in Figures 2 and 4. The analysis shows that there were steeper and elongated areas downslope of individual shrubs and shallow sills upslope.

The paired canopy interspace sampling supported the hypothesis that there are significant differences between below-canopy and interspace areas. Soils below vegetation were enriched in organic matter, but the enrichment was only in the surface horizons (on the order of a few centimeters). Significant differences in bulk density between the vegetated and unvegetated areas can be attributed to the combined effects of differences in the coarse fraction and differences in the soil organic matter fraction. The saturated hydraulic conductivity was more than 2 times higher under the vegetation at the schist site, but more than 3 times higher in the granite, and in both cases correlated with the organic matter content. We acknowledge that the hydraulic conductivity values represent a relatively small sample of the conductivity of the soil matrix, measured at a small spatial scale (57 mm diameter, 30 mm deep cores). These cores may be too small to provide reliable estimates of the saturated hydraulic conductivity at scales relevant to runoff generation [Wainwright et al., 2000, 2002; Parsons et al., 2006]. While the data clearly show a highly significant difference between canopy and interspace (p values in both lithologies are close to 0.01) and correlations with organic matter content (again, p values less than 0.01 in both cases), soil infiltration properties in semiarid areas can be highly spatially variable, and infiltration can occur preferentially in localized areas of high infiltration capacity [Parsons et al., 1992; Abrahams et al., 1995]. Furthermore, runoff generation in semiarid areas is often controlled by surface saturation above a shallow low-conductivity layer [Puigdefabregas et al., 1999] rather than by the surface hydraulic conductivity per se.

The correlation between biovolume and soil organic matter also supports the spatial differentiation of surface soils around vegetation on these slopes. The strength of the relationship suggests that lidar-derived measurements of biovolume provide a good estimate the spatial patterns of soil organic matter on these hillslopes. The systematic deviations from the relationship created by the downslope plumes of organic matter (discussed below) are potentially important for the ecosystem dynamics but are a second-order effect. However, the study was performed using data mostly collected during two winter periods when canopy density is high and the litter on the ground has probably not been replenished since the previous fall. It is possible that seasonal variations in canopy density and in the quantity and freshness of the litter would lead to different relationships, especially around the drought deciduous mesquite.

Observations from the transects under shrubs showed that the “zone of influence” of vegetation on the hillslopes is asymmetric on the granite slopes, but the data were more equivocal on the schist. On the granite plumes of organic matter extended downslope from the canopies of drought deciduous trees and shrubs into bare areas (Figure 9 and 10). Concentrations of organic matter two canopy radii downslope of the shrubs were comparable to those within the canopy at the upslope edge. Note that the data presented in Figure 2 give the impression of individual shrubs producing discrete plumes. However, when the shrubs were selected for sampling, candidates were excluded that had shrubs within three canopy radii upslope. The data suggest that this would be a minimum spacing required on these slopes to observe individual plumes. It was observed in the field that many more of the shrubs on the slopes grow close together, and their associated plumes appeared to merge together.

At the schist site, spatial pattern of the “zone of influence” was no less clear but was not significantly asymmetric in terms of the concentration at the canopy edge, though organic matter was enriched downslope. The differences in soil composition between canopy and interspace were most significant in the schist site (where as Figure 6 shows, differences were significant at the p=0.01 level), but the effects of the microtopography on the patterns of microtopographic convercence and divergence (as seen in the TLS data, Figure 3) were less pronounced. The schist had a higher overall organic matter content, likely caused by a combination of the finer texture of the schist soils (whose greater surface area can stabilize more organic matter through organomineral complexation) and a higher overall vegetation cover.

These results contrast with those few studies that have directly addressed this type of landscape, which have tended to show reduced differentiation between vegetated patches and interspaces on steeper hillslopes [Puigdefabregas et al., 1999; Bedford and Small, 2008]. Patterns of both higher [Dunkerley, 2000b] and lower [Madsen et al., 2008] hydraulic conductivity have been found in association with vegetation patches and organic matter in semiarid areas. The lower (unsaturated) conductivity observed by Madsen et al. [2008] was associated with hydrophobicity, and differences disappeared when the samples were prewetted. The technique we used to measure hydraulic conductivity is insensitive to hydrophobicity, since samples were saturated. Nicolau et al. [1996] found that saturated hydraulic conductivity below shrubs on an erosional slope underlain by micaschist bedrock (similar to the schist site here) was higher than in bare interspaces, but that on a nearby relic depositional terrace surface, the relationship was inverted, with higher conductivity in the bare areas. It is possible that the differences in the patterns of vegetation and conductivity between steeper and lower gradient sites may be in part due to the preferential aeolian deposition of fines associated with vegetation [Ravi et al., 2007]. The minimal differences in silt size fraction between the canopy and bare areas at the sites studied here suggest that preferential aeolian deposition of fines was not as significant in these landscapes. On these steeper surfaces, the rate of deposition by fines may be outstripped by organic matter turnover and fluvial and colluvial transport processes.

4.2 Possible Explanations for the Spatial Patterns

Field observations of armoring in the interspaces and light, easily transportable material in the below-canopy plume areas suggest that these areas are bounded by areas that receive more frequent concentrated overland flow. The low concentrations of organic matter in the flow lines suggests that transport of particulate SOM is rapid once it leaves the plumes. The differences in the evidence of plumes at the two sites (Figure 9) and in the microtopographic structure (Figures 3 and 5) suggest that the processes controlling the spatial structure of the resource islands may differ between the granite and schist sites. Vegetation mounds are believed to be formed partially by erosional lowering of the surrounding crust and partially by redeposition below the canopy [Buis et al., 2010]. In the granite site, the higher infiltration capacity and stronger microtopography may limit the frequency of overland flow in the areas downslope of the shrubs. On the other hand, the less-pronounced microtopographic structure and higher overall gradient at the schist prevent the mounds under vegetation from having a strong effect on flow direction.

However, we cannot be sure that these patterns in fact represent a stable feature of the landscape, since the study here presents only a “snapshot” of the landscape as it is today. Recent studies into the effects of fire of resource islands and microtopography have emphasized the way disturbance can lead to vegetation microsites switching from being primarily a sink for elemental resources to being a source [Field et al., 2012; Sankey et al., 2012b, 2012a; Ravi et al., 2009]. The reversion to being a sink depends primarily on the trajectory of recovery. It is possible that the elevated organic matter under shrubs represent a relic of A-horizon soils formed under previous grassland vegetation and preserved only under established shrubs, as Abrahams et al. [1995] have argued is the case in Walnut Gulch, another semiarid rangeland site in the southwest US.

Alternatively, the patterns could be a stable feature maintained by annual inputs of litter from the drought deciduous shrubs and protected from overland flow erosion by the microtopography. Both of these explanations point to a significant role of downslope transport controlling the spatial structure of the of resource islands on these slopes but must be distinguished on the basis of further evidence. One way to do this could be to carefully track the carbon balance of below-canopy and interspace sites across periods of fresh litter fall at the onset of the driest periods of the year in spring and fall through the wetter periods in summer and winter.

4.3 Implications for Ecosystem Function

The observations suggest that the plumes evident at the granite site could be regarded as a third component to the spatial mosaic, distinct in some ways from both the below-canopy and interspace areas. Since they are located outside the canopy, these areas are subject to the same energy environment as the interspace and do not receive the protection from rainsplash and trampling that the canopy areas do. However, the elevated organic matter suggests that they do receive some protection from the transport processes that prevent organic matter from accumulating in the interspaces.

The function of these areas within the ecosystem may be quite distinct from the canopy and interspace areas. Labile soil carbon rather than water availability per se has been postulated as a principal limitation of microbial activities in semiarid ecosystems and the distribution of this carbon is determined largely by the distribution of the vegetation [Gonzalez-Polo and Austin, 2009]. Thus, the SOM and other nutrients in these plumes contributes resources of fixed carbon and nitrogen to parts of the landscape where they would otherwise be much more limited. Other studies have shown that soil respiration [Sponseller, 2007], nutrients [Schlesinger et al., 1996; Schlesinger and Pilmanis, 1998], microbial biomass and activity [Herman and Provencio, 1995], heterotrophs and nitrogen efficient guild members [Herman and Provencio, 1995] follow the distribution pattern predicted by the resource island hypothesis with higher activities and numbers under shrubs than between shrubs. These studies together suggest that the plumes might be associated with increasing activities of microbial populations in areas that would otherwise be carbon limited and relatively barren. In addition the higher infiltration rates associated with elevated soil organic matter may tend to increase the availability of water to microbial communities. This is an area that requires further study.

4.4 Future Work

The results presented here contrast a large body of work that has focused on spatial patterns of lower gradient areas, such as alluvial fans. The transport that creates these plumes is inverted from the more commonly observed redistribution of resources from the bare interspace to the vegetated patches. If slope-dependent transport processes (such as rainsplash) are significant, this also suggests that modeling based on fluvial transport alone is insufficient to capture carbon transport dynamics in these areas.

The patterns described in this paper have significant implications for the connection between ecohydrologic dynamics of water-limited ecosystems [Newman et al., 2006], the interactions between vegetation and erosion on hillslopes [Marston, 2010; Osterkamp et al., 2011], and the coupling of geomorphic processes and the spatiotemporal dynamics of desert microbial communities. The formation and coevolution of spatial structures on hillslopes are becoming important to hydrologic sciences (as they have been in other disciplines, notably ecology and geomorphology, for considerably longer) [Sivapalan, 2005]. It is hoped that this will provide a clearer landscape context for hydrologic sciences, provide a way to deal with the spatial heterogeneity that limits our ability to connect processes across scales, and help us understand the interactions with other related (and equally critical) landscape processes, such as biogeochemical cycles.

More work is required to understand the transport processes operating in the plume and below-canopy areas and to determine whether they are sufficient to maintain a dynamic steady state plume structure, which will require direct measurement of the fluxes involved. This will require more detailed spatial analysis and modeling at these sites and comparative analysis with data published in other studies.


This research was supported by the National Science Foundation EAR 0910666 (University of Arizona) and EAR 0911205 (University of Illinois at Urbana Champaign) and by the Critical Zone Observatory (NSF JRB-SCM CZO—EAR 0724958). Kathleen Lohse, now at Idaho State University, was supported by the National Science Foundation under award number EPS-0814387. Thanks to Steve DeLong, Biosphere 2, for use of the terrestrial laser scanner which was purchased with funds provided by the Philecology Foundation of Fort Worth, Texas. Thanks to Ty Ferre and Andrew Hinnell for assistance and advice in the soil hydraulic analysis. Thanks also to Brittney Bates, Patrick Broxton, Ingo Heidbüchel, Caitlin Orem, Allison Peterson, and Clare Stielstra for assistance in the field and with soil analyses. Particle size analysis was conducted by Mercer Meding through a grant from the Center for Environmental Physics and Mineralogy (CEPM) at the University of Arizona. Thanks to Sally Thompson for stimulating discussions and a review of an early draft. Thanks also to the constructive comments of the three anonymous reviewers.