Community interaction webs describe both direct and indirect interactions among species. Changes in direct interactions often become noticeable soon after a perturbation, but time lags in responses of many species may delay the appearance of indirect effects and lead to temporal and/or spatial variation in interaction webs. Accurately identifying these shifts in the field requires time-specific, spatially-differentiated interaction webs. We explore how variation in browsing affects interaction webs in a long-unburned chaparral shrubland near the central California coast. Most prior work in chaparral focused on rapid changes for <5 yrs after a wildfire that have been assumed to determine community patterns until the next fire. Here, we report results of the first 15 years of an on-going experiment monitoring how interaction webs in long-unburned chaparral (at least 100 yrs postfire) respond to experimental variation in browsing by deer and rabbits on dominant shrubs (Arctostaphylos pumila, Ceanothus cuneatus var. rigidus and Ericameria ericoides). We hypothesized that variation in browsing would directly affect foodplants, indirectly modify growth and survival of other shrubs and impact habitat needed by herbaceous plants. We found a dynamic web of plant-herbivore and plant-plant interactions that responded rapidly to changes in deer browsing on Ceanothus followed by indirect interactions that continued developing over several years, affecting shrubs, open space, herbaceous plants, and small mammals. Experimental variation in the intensity of deer browsing led to temporal and spatial changes in interactions that produced three different community interaction webs. With deer, community webs were complex, having numerous direct and indirect interactions. Removing deer simplified the community web, changed outcomes of interactions, and reduced open space and herbaceous plants densities. Finally, changes in Ceanothus morphology without deer allowed woodrats to browse these shrubs, with negative impacts on Ceanothus growth and survival. General field observations also showed that all three alternative interaction webs occur naturally at our fieldsite, varying across space and over time. Long-unburned chaparral communities browsed by deer maintain high biological diversity, but maintenance of this diversity involves many key direct and indirect biotic interactions.

We worked in a 242 ha reserve on the former U.S. Fort Ord Army base, now part of the University of California Natural Reserve System. The on-going herbivore exclusion experiment regulated deer and rabbit access to chaparral plants. The experiment had open plots as controls and two types of caged plots. Control plots were uncaged, with natural levels of browsing by deer and rabbits (+D+R). One cage type prevented deer from browsing in a plot but allowed access to rabbits (-D+R), and the other cage type excluded both deer and rabbits (-D-R). Deer were excluded by 2 m high fences, with 20 cm x 10 cm mesh. Rabbits were excluded by 1.5 cm mesh (aviary wire) on the lower 1 m of half of the exclosures. Each plot was 9 m2. Four experimental blocks, each with a control and one of each cage type, were randomly located at each of two randomly selected sites ~1 km apart (sites called F and N in this paper are designated F4, N1 and N2 in the FONR mapping system). In total, there were 24 plots among eight blocks over two sites. Baseline data collection began immediately after the cages were completed in August 1996. We censused and measured all shrubs in late summer or early autumn, annually from 1996-2000 and 2002-2007, and in 2011. We also recorded browsing intensity and shrub mortality during censuses as well as each spring and sporadically at other times of the year.

Observations variously related to individual-, species-, or plot-level response metrics. Univariate and multivariate response variables were modeled separately as they generally corresponded to different time periods. Unless otherwise noted, we assessed trends only over years 1997-2007, since we did not measure plants between 2007 and 2011.

We modeled individual-, species-, and plot-level response variables with generalized linear mixed models (GLMMs) and explored directional community composition changes among the three dominant shrub species with non-metric multidimensional scaling (NMDS) ordination. The initial experimental design established four blocks, each nested in two spatially distinct random sites; however, sites explained zero variation so were removed from model random structures. We modeled GLMMs in PROC GLIMMIX in SAS® 9.4 (SAS Institute Inc. 2013) and NMDS with package vegan 2.4-4 in R 3.4 (Moen et al. 1993, Oksanen et al. 2017, R Core Team 2017).