Local and landscape drivers of aquatic-to-terrestrial subsidies in riparian ecosystems: a worldwide meta-analysis

Cross-boundary fluxes of organisms and matter, termed “subsidies”, are now recognized to be reciprocal and of roughly equal importance for both aquatic and terrestrial systems, even if terrestrial input to aquatic ecosystems has received most attention. The magnitude of aquatic to terrestrial subsidies is well documented, but the drivers behind these subsidies and their utilization by terrestrial consumers are characteristically local scale studies, limiting the inferences that can be drawn for broader geographic scales. We therefore built and analyzed a database of stable isotope data extracted from 21 studies worldwide, to identify both landscape and local scale variables that may affect the diet of terrestrial predators in riparian ecosystems. Our meta-analysis revealed a greater magnitude of aquatic-to-terrestrial subsidies (> 50%) than previously reported, albeit with large geographic and inter-annual variations. We demonstrated a large effect of landscape-scale factors on aquatic-to-terrestrial subsidies, particularly anthropogenic land use and tree cover. Local human population was the only relevant factor at the local scale. We also found that studies on landscape-scale and anthropogenic land use effects on aquatic-to-terrestrial subsidies are currently strongly under-represented in the ecological literature. Such studies are needed to improve our understanding of how land use and environmental change might influence future patterns of biodiversity and ecosystem function.

In this study, we conducted a worldwide meta-analysis of studies that have assessed 62 aquatic-to-terrestrial subsidies using stable isotopes. We quantified the effects of 63 ecosystem size, stream morphology and land use on aquatic subsidies to terrestrial 64 predators. First, we estimated the overall proportion of aquatic subsidies in the diet of 65 several groups of terrestrial predators, and tested whether the proportion of these prey 66 was significantly higher than that of terrestrial prey. We hypothesised that the proportion 67 of aquatic subsidies varied between taxonomic groups of predators, hydrological system 68 type (hydro-ecoregion) and year. Next, we assessed the relative importance of biotic and 69 abiotic variables at local-and landscape-scales (100 m buffers and catchments, 70 respectively) for the proportion of aquatic subsidies in the diet of spider and carabid 71 beetle predators. We hypothesised that landscape-scale variables related to anthropogenic 72 land use would be of at least equal importance in explaining predators' diets as 73 commonly-assessed local-scale variables. 74

Methods 75
Our meta-analysis focused on the use of aquatic subsidies by terrestrial predators. We 76 restricted the subsidies to aquatic organisms actively crossing the boundary between 77 aquatic and terrestrial ecosystems (i.e. macro-invertebrates). All predators consuming 78 aquatic macro-invertebrates were included. In order to get a more accurate estimation of 79 the proportion of aquatic subsidies in the diet of predators, we restricted our meta-80 analysis to studies using stable isotopes, which integrate the use of prey types over a 81 longer period of time than do stomach content analyses (Tieszen et al. 1983). 82

Data retrieval 83
We searched the Web of Science and Google Scholar for studies focusing on riparian 84 habitats and using stable isotopes as a tool to infer the contribution of aquatic prey to the 85 diet of terrestrial predators. The keywords used were "aquatic subsidies" AND "stable 86 isotope" AND "diet", which gave 69 results. From these 69 articles we refined the 87 selection in several steps. First, a selection was made based on words in the title and a 88 second one on words in the abstract. We then screened the bibliography of the selected 89 studies to find new references and iterated this search procedure until we did not find any 90 new documents. This procedure reduced the 69 papers to 47. At last, a selection of 91 studies was based on the number of sampling sites and replicates in the different studies, 92 i.e. we kept studies with at least two sampling sites or studies with repeated 93 measurements in time and studies including sampling of two predator species. 94 As studies using experimental manipulation of subsidies (and using stable isotopes) were 95 very rare, descriptive studies were also included. Studies on predators' diet based on 96 means and standard errors were available we used a modified version of the JAGS 120 models used by Parnell et al. (2013) to include standard error of the consumer isotope 121 values as a prior of the model. Source aggregation (terrestrial vs aquatic) was made a 122 priori as the number of sources included in models was variable between studies, which 123 is problematic for a posteriori aggregations if one wants to compare diets (Stock et al. 124 2018). We chose not to give any prior to the proportion of aquatic preys in diet 125 (generalist diets) which means that all possible combinations of proportions of aquatic 126 and terrestrial preys were likely a priori (Stock et al. 2018). 127

Predictors 128
The catchment draining to each sampling location was delineated using QGIS 2.18.18 129

Statistical analysis 149
We used the proportion of aquatic subsidies in the diet minus 0.5 as an effect-size to test 150 for differences between proportion of aquatic and terrestrial subsidies in the diet of the 151 terrestrial predators. Freshwater ecoregion, sampling year and taxonomic group of the 152 predators were included in the model as fixed factors. We used the metafor package 153 (Viechtbauer 2010) with restricted maximum-likelihood estimator to test the effect-size. 154 The selection of landscape and local variables best explaining the proportion of aquatic 155 subsidies in predators' diet was done using partial least square regression (PLS) on mean 156 % of aquatic subsidies in the diet per sampling site. Given the low number of studies 157 available for some groups (Table 1)  Results 175

Dataset description 176
The final dataset resulted in 21 studies representing 159 sampling sites and 400 diets 177 (Table 1). This corresponds to almost half of the studies initially selected. Twenty-six 178 studies could not be used, mainly because they did not report data in a suitable format 179 and quality for analysis of diet partitioning. Among these 21 studies, two were not used 180 in the PLS because we could not locate the sampling sites with enough accuracy. Spiders 181 and carabid beetles were the two most studied groups whose diets were estimated in 182 51.3% and 41.6% of the studies, respectively. The studies were mainly located in the 183 northern hemisphere with cold or temperate climates ( Fig. 1 and Table 1). 184 Study site locations were strongly biased toward small forested catchments with very low 185 human population density and urbanization extent and located mainly in the northern 186 hemisphere ( Fig. 2 and 3). Conversely, a few studies were also located in rivers with very 187 large catchments or/and high human population. 188 Predator's reliance on aquatic subsidies 189 The contribution of aquatic subsidies was significantly higher than 50% (effect size = In the PLS regression model for carabid beetles (two components: R 2 = 0.112 and R 2 = 207 0.041), percent tree cover, forests, and water bodies at the landscape scale were the most 208 important variables for low proportion of aquatic prey. The percentage of non-forested 209 natural habitats, urban areas and agriculture at the landscape scale and the river width of 210 the local scale were most important variables for high proportion of aquatic prey ( fig. 6). 211

Discussion 212
Our study extends recent findings that demonstrate high levels of aquatic-to-terrestrial 213 pattern of high aquatic subsidies use by terrestrial predators, we also documented 221 significant inter-annual and geographic variations in these subsidies, largely driven by 222 hydrologic cycles and ecoregion, respectively. 223 We found the diet of riparian predators to be highly dependent on aquatic subsidies (> 224 50%, overall effect size = 0.07). Since we re-computed diet partitioning from raw data to 225 reduce mixing-model and discrimination-factor biases (Bond and Diamond 2011), our 226 estimate is likely the most robust to date. This suggests that, in general, the proportion of 227 aquatic subsidies in predator diets may be even higher than the 40% reported in Bartels et 228 al.'s (2012) meta-analysis. We could not find any significant effect of predator taxonomic 229 group, which might be due to the small number of studies dealing with groups other than 230 carabid beetles and spiders. Given the wide geographic spread of our analysis and the 231 pattern of high proportion of aquatic-derived carbon across the study sites, it seems likely 232 that most predator taxa in riparian systems rely on these subsidies for more than 50% of 233 their diet. 234 Perhaps unsurprisingly, we also found significant temporal (inter-annual) and spatial 235 Studies are also needed on southern hemisphere streams. 300 Our study is the first worldwide meta-analyses to use exclusively stable isotope studies in 301 order to better integrate the temporal component of terrestrial predator diets. We 302 demonstrated a high reliance (more than 50%) of terrestrial predators on aquatic subsidies 303 across broad geographic regions, despite large geographic and inter-annual variations. 304 We further demonstrated a large effect of anthropogenic land use at the catchment scale 305 across geographic regions. Linking these two key findings suggests that more attention to 306 broad-scale landscape patterns is warranted to improve our understanding of how these 307 cross-boundary energy flows affects biodiversity and ecosystem function of tightly 308 coupled aquatic-terrestrial systems. 309

311
We would like to thank Karlstad University for funding this study within the framework 312 of its strong research groups program. We particularly want to thank authors that sent us 313 raw data: B.K. Jackson