Challenges With Inferring How Land-Use Affects Terrestrial Biodiversity: Study Design, Time, Space and Synthesis

Abstract

Land use has already reshaped local biodiversity on Earth, with effects expected to increase as human populations continue to grow in both numbers and prosperity. An accurate depiction of the state of biodiversity on our planet, combined with identifying the mechanisms driving local biodiversity change, underpins our ability to predict how different societal priorities and actions will influence biodiversity trajectories. Quantitative syntheses provide a fundamental tool by taking information from multiple sources to identify generalisable patterns. However, syntheses, by definition, combine data sources that have fundamentally different purposes, contexts, designs and sources of error and bias; they may thus provide contradictory results, not because of the biological phenomena of interest, but due instead to combining diverse data. While much attention has been focussed on the use of space-for-time substitution methods to estimate the impact of land-use change on terrestrial biodiversity, we show that the most common study designs all face challenges—either conceptual or logistical—that may lead to faulty inferences and ultimately mislead quantitative syntheses. We outline these study designs along with their advantages and difficulties, and how quantitative syntheses can combine the strengths of each class of design.

Introduction

Land-use change and intensification are among the most widespread pressures facing terrestrial biodiversity worldwide. Only 39% of land has never been converted to human use; around 265,000 km² of unaltered landscapes is lost each year, while around 290,000 km² is abandoned and reverting to secondary vegetation (Hurtt et al., 2017). Loss of habitat to human use is the most commonly identified direct driver of global extinction risk to terrestrial species (e.g. Brummitt et al., 2015; Estrada et al., 2017). As well as global loss, local biodiversity loss is also of concern, as it can harm ecosystem function and jeopardise delivery of ecosystem services (Cardinale et al., 2012; Díaz et al., 2006; Hooper et al., 2012). Although there is consensus that global biodiversity is declining (McGill et al., 2015), there is disagreement about the average trend in local biodiversity (Dornelas et al., 2014; Gonzalez et al., 2016; McGill et al., 2015; Newbold et al., 2015; Vellend et al., 2013). Given that sustainable development depends on using natural resources sustainably (Sustainable Development Goals; Griggs et al., 2013), there is a pressing need to synthesise available evidence on how local biodiversity is changing in the face of land-use change and intensification.

One complication facing such syntheses is that many ecological aspects of the study system are likely to influence how the intensity of a given change in anthropogenic pressure affects different measures of site-level biodiversity. Observed effects can depend on—among other things—the taxonomic or functional group studied (e.g. Gibson et al., 2011; Lawton et al., 1998; Murphy and Romanuk, 2014), how long the landscape has experienced strong human impacts (Balmford, 1996), the habitat stratum sampled (Dumbrell and Hill, 2005) and the location of the study (e.g. Gibson et al., 2011). Our main focus here, however, is on a further possible source of disagreement that has recently become prominent (França et al., 2016; Gonzalez et al., 2016; Vellend et al., 2017)—the spatiotemporal design of the study. Some studies directly observe biodiversity change over time (by repeated sampling), others make spatial comparisons to infer temporal changes (space-for-time substitution), while others combine spatial and temporal comparisons. Global syntheses of time series data of assemblage diversity show no overall temporal trend (Vellend et al., 2013; Dornelas et al., 2014; but see Gonzalez et al., 2016), whereas global meta-analyses of spatial comparisons between sites varying in land use suggest a net decline in diversity (Gibson et al., 2011; Murphy and Romanuk, 2014; Newbold et al., 2015), and sampling both before and after a disturbance at both control and impact (i.e. undisturbed vs disturbed) sites can yield even larger estimates of decline (França et al., 2016).

Another aspect of study design that can affect inferences about land-use impacts on biodiversity is the experimental approach used, from manipulative experiments to correlational ‘quasi-’ or mensurative experiments (Block et al., 2001; McGarigal and Cushman, 2002). In the former, land-use change treatments are assigned by the researchers in order to minimise effects of confounding variables (e.g. randomly), while the latter approach makes use of externally imposed variation in land use. Manipulative experiments are rare, especially at a large enough spatial and temporal scale to fully capture the real-world effects of land-use change (see Ewers et al., 2011 for a notable exception), because land-use change is seldom within a researcher's control. Though much more common, quasi-experiments face the risk that the land-use variation incorporated as a predictor can also reflect other factors that may confound inferences (McGarigal and Cushman, 2002) and can rarely disentangle effects of different land-use pressures that are inherently linked (e.g. habitat loss vs fragmentation effects; Ewers and Didham, 2006).

Synthetic analyses of trends in local terrestrial biodiversity may not only be confounded by ecological differences, but also the study designs can influence the trends implied by different data sets. Recent extensive, thorough and careful syntheses of temporal and spatial comparisons disagree whether average assemblage-level diversity is declining (Newbold et al., 2015; Vellend et al., 2017). This disagreement makes it harder for global and regional assessments, such as those of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (Díaz et al., 2015), to deliver clear messages about human impacts on biodiversity. To clarify how methodology can influence our capacity to detect responses of local terrestrial biodiversity to land-use change, we explicitly consider the advantages and disadvantages of the study designs most commonly used to achieve this objective. We highlight when each study design is particularly likely to mislead and suggest a new synthetic modelling framework in which all relevant designs can be incorporated. We also address the broader question of whether syntheses of published data can provide an accurate global picture of how terrestrial biodiversity is responding to human impacts.