Disentangling effects of climate and land use on biodiversity and ecosystem services – a multi-scale experimental design

Climate and land-use change are key drivers of environmental degradation in the Anthropocene, but too little is known about their interactive effects on biodiversity and ecosystem services. Long-term data on biodiversity trends are currently lacking. Furthermore, previous ecological studies have rarely considered climate and land use in a joint design, did not achieve variable independence or lost statistical power by not covering the full range of environmental gradients. Here, we introduce a multi-scale space-for-time study design to disentangle effects of climate and land use on biodiversity and ecosystem services. The site selection approach coupled extensive GIS-based exploration and correlation heatmaps with a crossed and nested design covering regional, landscape and local scales. Its implementation in Bavaria (Germany) resulted in a set of study plots that maximizes the potential range and independence of environmental variables at different spatial scales. Stratifying the state of Bavaria into five climate zones and three prevailing land-use types, i.e. near-natural, agriculture and urban, resulted in 60 study regions covering a mean annual temperature gradient of 5.6–9.8 °C and a spatial extent of 380×360 km. Within these regions, we nested 180 study plots located in contrasting local land-use types, i.e. forests, grasslands, arable land or settlement (local climate gradient 4.5–10 °C). This approach achieved low correlations between climate and land-use (proportional cover) at the regional and landscape scale with |r ≤0.33| and |r ≤0.29|, respectively. Furthermore, using correlation heatmaps for local plot selection reduced potentially confounding relationships between landscape composition and configuration for plots located in forests, arable land and settlements. The suggested design expands upon previous research in covering a significant range of environmental gradients and including a diversity of dominant land-use types at different scales within different climatic contexts. It allows independent assessment of the relative contribution of multi-scale climate and land use on biodiversity and ecosystem services. Understanding potential interdependencies among global change drivers is essential to develop effective restoration and mitigation strategies against biodiversity decline, especially in expectation of future climatic changes. Importantly, this study also provides a baseline for long-term ecological monitoring programs.


Introduction
Human actions are threatening the interdependent yet fragile balance of the biosphere, 79 with far-reaching consequences for the diversity of plants (Brummitt et al., 2015) and animals 80 (Dirzo et al., 2014). As biodiversity contributes a wealth of ecological services, cascading  Wagner, 2020). 89 Yet the full cross-taxon magnitude of declines and the relative contributions of man-made 90 drivers remain poorly understood. 91 One of the greatest threats to biodiversity is land-use change, the transformation of 92 terrestrial ecosystems for infrastructure, human settlements and the production of crops, 93 animals and timber (Newbold et al., 2015). Landscape simplification, urbanization,

112
Climate is another major driver of biodiversity. Long-term data on species 113 distributions along latitudinal and elevational climatic gradients demonstrate significant 114 poleward and upward shifts of species' ranges driven by global warming (Parmesan, 2006). In   Understanding the independent and combined impact of land-use and climate change on 125 biodiversity, community composition and ecosystem services is needed to predict future 126 changes and allow for management strategies to mitigate further losses. However, less than 127 7 10% of available studies analyse combinations of those drivers (Rillig et al., 2019). Land-use 128 change may also feedback to the atmosphere and alter regional climate including water 129 availability by precipitation ( Dale, 1997;Laux et al., 2017;Williams & Newbold, 2020), 130 resulting in correlated land-use and climate gradients that make it difficult to disentangle 131 individual effects (Peters et al., 2019). Furthermore, long-term data on climate, land use and 132 biodiversity are currently lacking, recently established monitoring schemes will not deliver 133 sufficient data in the near future and time-series analysis may be prone to biases (Didham et  Here, we report on a novel protocol (Fig. 2) for a comprehensive study design that programs and long-term ecosystem monitoring but will also allow for predictions of potential 159 interactive impacts of climate and land use in a space-for-time approach.  Step 1 -Selection of study regions based on climate and land-use zones 184 At the regional scale, a stratified sampling approach ensured complete coverage of   Step 2 -Create heatmaps to reduce correlations among landscape variables 214 Within each of the 60 study regions, we aimed to investigate the impact of local land  Step 3 -Selection of local study plots 256 Within each quadrant, we aimed to establish local study plots of 0.5 ha size within 257 contrasting land-use types (Fig. 2). Although four local, dominant land-use types had been 258 identified during the heatmap process (forest, grassland, arable land or settlement), not all 259 were present in each quadrant. Therefore, we focused on three out of four land-use types per 260 quadrant by considering availability (if only three types present) or regional dominance (three 261 types with highest proportional cover) and contrast (whenever proportional cover of two land-262 use types was similar). We then used the respective heatmaps to preferentially place study 263 plots in grid cells that had a low predicted correlation values for the specific land-use type.  Assessing efficiency of study design 282 We assessed the efficiency of our stratified selection and heatmap approach by a)

298
Implementation of the experimental design 299 Our design and selection process (Fig. 2) allowed us to minimize the potential 300 correlations between climate, land use and landscape metrics at multiple scales and resulted in 301 an approximately even distribution of 60 study regions (quadrants) across Bavaria (Fig. 3).

302
These regions covered a climate gradient of 5.6-9.8 °C (8.2 ± 0.8 °C, mean ± SD) and 614-303 1820 mm of annual precipitation amounts (939 ± 263 mm). Across all quadrants, the cover of 304 our dominant regional land-use types (i.e. landscape composition) ranged from 0.8 to 97.1%   For each study region, the heatmap procedure yielded four heatmaps for the local 318 land-use types forest, grassland, arable land and settlement, which were used to identify 319 potential study plots within dominant local land-use types (Fig. 6B-D). After ground-truthing 320 of sites and gaining permission of landowners, three final plots were chosen per quadrant 321 (Fig. 6E), yielding 179 out of 180 expected study plots (Fig. 6A). One study plot was 322 discarded as landowner permission was denied. Forest (n = 55) was the most selected local 323 land-use type, followed by grassland (n = 46), arable land (n = 43) and settlement (n = 35).

340
Compared to potential correlations based on random selection of study plots, the 341 heatmap approach resulted in lower correlations between landscape composition and 342 configuration (in 1-km radius around study plots) for plots located in forest, arable land and 343 settlements (blue line, Fig. 7A, C, D). Only for grassland, the final correlation was positive 344 and higher than predicted (blue line, Fig. 7B). Taking all study plots independent of the local    First, the crossed and nested design at the regional scale resulted in relatively weak 381 correlations between climate and land use (proportional cover of forest, near-natural and 382 urban area). The design also decoupled regional climate and land-use effects from the 383 influence of small-scale land use due to the selection of three out of four dominant local land-384 use types (forest, grassland, arable land or settlements) within our 60 study regions.

385
Regarding landscape composition and configuration in a 1-km radius around study plots, the 386 heatmap approach lowered correlations compared to average potential correlations for 387 specific local land-use types (blue lines, Fig. 7), but these benefits were not that substantial in