Skip to main content
Dryad logo

Large ecosystem-scale effects of restoration fail to mitigate impacts of land-use legacies

Citation

Turley, Nash; Brudvig, Lars (2021), Large ecosystem-scale effects of restoration fail to mitigate impacts of land-use legacies, Dryad, Dataset, https://doi.org/10.5061/dryad.crjdfn339

Abstract

Ecological restoration is a global priority, with potential to reverse biodiversity declines and promote ecosystem functioning. Yet, successful restoration is challenged by lingering legacies of past land-use activities, which are pervasive on lands available for restoration. Although legacies can persist for centuries following cessation of human land uses such as agriculture, we currently lack understanding of how land-use legacies affect entire ecosystems, how they influence restoration outcomes, or whether restoration can mitigate legacy effects. Using a large-scale experiment, we evaluated how restoration by tree thinning and land-use legacies from prior cultivation and subsequent conversion to pine plantations affect fire-suppressed longleaf pine savannas. We evaluated 45 ecological properties across four categories: 1) abiotic attributes, 2) organism abundances, 3) species diversity, and 4) species interactions. The effects of restoration and land-use legacies were pervasive, shaping all categories of properties, with restoration effects roughly twice the magnitude of legacy effects. Restoration effects were of comparable magnitude in savannas with and without a history of intensive human land use; however, restoration did not mitigate numerous legacy effects present prior to restoration. As a result, savannas with a history of intensive human land use supported altered properties, especially related to soils, even after restoration. The signature of past human land-use activities can be remarkably persistent in the face of intensive restoration, influencing the outcome of restoration across diverse ecological properties. Understanding and mitigating land-use legacies will maximize the potential to restore degraded ecosystems.

Methods

This research took place within a large-scale experiment at the Department of Energy Savannah River Site (SRS), a National Environmental Research Park in South Carolina, USA (33.20°N, 81.40°W) (Fig. 1). The sandy uplands at SRS historically supported open canopy longleaf pine savannas, which were largely converted to corn, cotton, and other crops between 1865 and 1950. All agriculture was abandoned in 1951 when the United States government acquired SRS and these fields were subsequently converted to plantations of longleaf (Pinus palustris), loblolly (P. taeda), and slash pine (P. elliotii).

            The experiment included 126 1-ha (100×100 m) plots grouped into 27 blocks. Each block was centered around the boundary between a longleaf pine savanna with no known history of agriculture and a former agricultural field supporting a mature pine plantation (P. palustris where possible). We determined land-use history using historical aerial photography and confirmed no differences in soil types on plots with and without agricultural history. Blocks included at least four and as many as 10 plots, depending on the sizes of areas with and without agricultural history, with half of the plots located within the savanna lacking agricultural history and half of the plots within the post-agricultural pine plantation. Due to a history of fire suppression, all plots supported closed canopy woodland at the onset of the experiment. To restore open canopy conditions, we randomly assigned a restoration thinning treatment in 2011 to half of the plots with and half of the plots without agricultural history, using logging equipment to remove trees from plots. The thinning treatment reduced tree density from an average of 650 trees/ha to 10 trees/ha . One or more prescribed surface fires were subsequently conducted within each block. In sum, this resulted in a 2 × 2 factorial manipulation of agricultural/plantation history and restoration thinning.

            Between 2012 and 2017 we quantified abiotic conditions, the abundance of individuals, the diversity of species, and species interactions within the experimental plots (. Abiotic variables included temperature and light, the percentage ground cover by leaf litter, depth of the O horizon (leaf litter and duff), percent canopy closure (which influences understory light availability), soil water holding capacity, the percentage of ground area burned during prescribed fires, soil compaction, percent soil moisture, soil pH, soil organic matter, and soil phosphorus. Abundance variables included the summed captures of three rodent species using live traps, the number of observations of three individual rodent species using remote trail cameras, counts of grasshoppers, the total number of individuals established from seed addition of 12 herbaceous understory plants, numbers of fire ant (Solenopsis invicta) mounds, numbers of pyramid ant (Dorymyrmex bureni) mounds, counts of bees, and floral cover. Species diversity variables included the richness of grasshoppers, rodents, soil bacteria, soil fungi, bees, vascular plants naturally occurring in plots, and the richness of 12 herbaceous understory plants added through seed addition. Measures of species interactions included rates of herbivory on four understory herbs, rates of pollination to sentinel black mustard (Brassica nigra) plants, the effects of root competition on four understory herbs, and rates of seed removal, as a measure of granivory, on the seeds of six understory herbs and one tree species (Quercus nigra). We excluded any data collected within 10 m of the land-use boundary, to avoid potential influence of edge effects. 

            We used meta-analysis techniques to evaluate responses of the 45 response variables to land-use history and restoration. We first calculated standardized effect sizes, using absolute values of the difference between treatments because directionality does not have a consistent interpretation across the variables we measured. This allowed us to focus on the magnitude of responses across variables, though we do consider directionality during interpretation of individual variable responses. For each variable, we then averaged subsamples to get a single value for each of the four treatment combinations (plot types) within each block, providing 7-27 replicates per variable. We did this because some variables had multiple measurements per plot and some blocks had multiple replicate plots of each treatment combination. We then calculated the mean and standard deviation across blocks for each variable in each of the four treatment combinations: unrestored post-agricultural/plantation, unrestored non-post-agricultural, restored post-agricultural/plantation, restored non-post-agricultural. We used these values to calculate Hedge's g, a measure of standardized effect size, for the differences between each pair of treatment combinations by using the escalc function in the metafor package in R. For example, to quantify how restoration affects the ecological properties within post-agricultural/plantation areas (“Restoration +Ag. history”) we calculated the effect size of the difference between unrestored post-agricultural/planation plots and restored post-agricultural/plantation plots. This was repeated for each treatment combination, resulting in four total effect sizes: the effect of agricultural and planation history in unrestored plots (Ag. history - Restoration), the effect of agricultural and planation history in restored plots (Ag. history + Restoration), the effect of restoration in post-agricultural/planation plots (Restoration + Ag. history), and the effect of restoration in plots without agricultural/planation history (Restoration - Ag. history) .

We next fit meta-analysis models with the rma function in the metafor package to statistically test our four questions, using the effect sizes, variances, and sample size we obtained from the above Hedge’s g calculations. The models included ‘dataset’ as a random effect to account for potential non-independence of variables collected from the same project.  We then ran post-hoc tests using the glht function in the multcomp package to obtain p-values for the contrasts among each of the four effect sizes. 

Usage Notes

20200807 Remnant project meta analysis data.csv is the main dataset with raw data from the field experiment. The other files are outputs from the R code that are used for the analyses and making figures. 

Funding

Department of Agriculture and Department of Energy, Award: DE-EM0003622

Department of Agriculture and Department of Energy, Award: DE-EM0003622