Landscape diversity and local temperature, but not climate, affect arthropod predation among habitat types

Arthropod predators are important for ecosystem functioning by providing top-down regulation of insect herbivores. As predator communities and activity are influenced by biotic and abiotic factors on different spatial scales, the strength of top-down regulation (‘arthropod predation’) is also likely to vary. Understanding the combined effects of potential drivers on arthropod predation is urgently needed with regard to anthropogenic climate and land-use change. In a large-scale study, we recorded arthropod predation rates using artificial caterpillars on 113 plots of open herbaceous vegetation embedded in contrasting habitat types (forest, grassland, arable field, settlement) along climate and land-use gradients in Bavaria, Germany. As potential drivers we included habitat characteristics (habitat type, plant species richness, local mean temperature and mean relative humidity during artificial caterpillar exposure), landscape diversity (0.5–3.0-km, six scales), climate (multi-annual mean temperature, ‘MAT’) and interactive effects of habitat type with other drivers. We observed no substantial differences in arthropod predation rates between the studied habitat types, related to plant species richness and across the Bavarian-wide climatic gradient, but predation was limited when local mean temperatures were low and tended to decrease towards higher relative humidity. Arthropod predation rates increased towards more diverse landscapes at a 2-km scale. Interactive effects of habitat type with local weather conditions, plant species richness, landscape diversity and MAT were not observed. We conclude that landscape diversity favours high arthropod predation rates in open herbaceous vegetation independent of the dominant habitat in the vicinity. This finding may be harnessed to improve top-down control of herbivores, e.g. agricultural pests, but further research is needed for more specific recommendations on landscape management. The absence of MAT effects suggests that high predation rates may occur independent of moderate increases of MAT in the near future.

spatial scales, the strength of top-down regulation ('arthropod predation') is also likely to vary.

23
Understanding the combined effects of potential drivers on arthropod predation is urgently needed with 24 regard to anthropogenic climate and land-use change. In a large-scale study, we recorded arthropod 25 predation rates using artificial caterpillars on 113 plots of open herbaceous vegetation embedded in 26 contrasting habitat types (forest, grassland, arable field, settlement) along climate and land-use gradients in 27 Bavaria, Germany. As potential drivers we included habitat characteristics (habitat type, plant species 28 richness, local mean temperature and mean relative humidity during artificial caterpillar exposure), 29 landscape diversity (0.5-3.0-km, six scales), climate (multi-annual mean temperature, 'MAT') and 30 interactive effects of habitat type with other drivers. We observed no substantial differences in arthropod 31 predation rates between the studied habitat types, related to plant species richness and across the Bavarian-32 wide climatic gradient, but predation was limited when local mean temperatures were low and tended to 33 decrease towards higher relative humidity. Arthropod predation rates increased towards more diverse

122
Plant species richness per plot was derived between May and July 2019 from plant species records in seven 123 subplots (10 m 2 total sampling area). Further details and a species list are provided in Fricke et al. (29).

124
Local weather conditions during caterpillar exposure were derived from thermologgers (ibutton, type 125 DS1923). Those were attached north-facing to a wooden pole, at 1.1 m above ground and roughly 0.15 m 126 below a wooden roof, which prevented direct solar radiation. One thermologger was established per plot.

127
We extracted mean temperature and mean relative humidity (in the following referred to as 'local mean 128 temperature' and 'mean relative humidity') during the study-site specific exposure period of the artificial 129 caterpillars from hourly measurements of the thermologgers. around the centre point of the plots at six spatial scales (0.5-3.0 km, in 500-m steps). At 2-km scale, low 136 landscape diversity equated a dominance of forest or arable land, and the land-cover proportions of semi-137 natural habitat and water were below 7.5% and 10.2%, respectively.

141
Prior to data analysis, data exclusion criteria were applied to standardize data. We excluded artificial 142 caterpillars exposed to attack for more than 54 hours (exceeding 48 ± 6 h limit), 'released' later than 25 th 143 May, and recovered incomplete with a loss of more than 20% (<16 artificial caterpillars per plot). In total,

144
we achieved standardized data on 113 plots. Artificial caterpillars from 58 of these plots (51%) were

148
Arthropod predation rate data were analysed with binomial generalized mixed effect models to cope with 149 proportional data (derived from absence-presence data) using the R-package 'glmmTMB' (31) with R 150 version 4.0.3 (32). Region was included as a random term to account for the nested study design and was 151 retrieved throughout the model selection process (33). Due to zero-inflation (complete absence of attack 152 from 17% of plots), confirmed using the R-package 'DHARMa' (34), we added a zero-inflation term. We 153 did not account for exposure duration of the artificial caterpillars in the models, since data were standardized 154 by exposure duration (48 ± 6 h limit) and similar exposure durations of 48.2 ± 1.7 h (mean ± SD) were 155 realized among plots. richness, local mean temperature and mean relative humidity (during artificial caterpillar exposure),

158
landscape diversity and MAT. Candidate predictors were z-transformed prior to analysis, while presented 159 models contain untransformed predictor variables.

160
To parametrize the zero-inflation term, we considered factors which might explain absence of attack on 161 plot level, e.g. arthropod activity limited by low temperatures (19). Besides, we visually screened the 162 candidate predictors for accumulation of absence-of-attack events (predation rate = 0) at the extremes of 163 the predictor ranges. Local mean temperature was the only candidate predictor in which absence of attack 164 marks was frequently observed at the lower range on a per plot basis. Therefore, local mean temperature 165 was included as a single candidate predictor in the zero-inflation term. Additionally, we run a separate 166 analysis on presence-absence of attack on plot level (data extracted from predation rate data; predation rate 167 > 0 replaced by 1) to investigate how the probability of attack on plot level was affected through local mean 168 temperature using binomial generalized linear mixed effect models including region as random term (see 169 S1 Table).

170
When analysing the data, we first conducted multi-model averaging to identify the most relevant predictors 171 and spatial scales. Models with all possible predictor combinations were created separately for each spatial 172 scale (0.5-3.0 km, six scales). Akaike weights were computed using the dredge-function from the 'MuMin' 173 R-package (35). Achieved Akaike weights (w i ) were summed per predictor and spatial scale, whereby high 174 summed Akaike weights (Σw i ; range: 0 (low) -1 (high)) indicate a high relative importance of a predictor,

175
corresponding to high cumulative probability that a predictor occurs in the best model at the respective 176 spatial scale (36).

177
In a second step, we analysed potential interactive effects of habitat type with plant species richness, 178 weather conditions during artificial caterpillar exposure (local mean temperature, mean relative humidity),

179
landscape diversity and MAT. Therefore, we added single interaction terms (e.g. local habitat type * plant 180 species richness) to the best model at the most relevant spatial scale derived from multi-model averaging.

181
Model selection was done based on Akaike's information criterion corrected for small sample size (AICc).

182
Thereby, models with lower AICc were considered better, and models with ∆AICc < 2 were considered 183 equal and the more parsimonious model was chosen.

184
Pearson correlations between continuous candidate predictor variables were rather low ≤ 0.33 (S2 Table) 185 with two exceptions. MAT was positively correlated with local mean temperature (Pearson's r = 0.59) and 186 negatively correlated with mean relative humidity (Pearson's r = -0.51). However, all variance inflation 187 factors (VIF) fell below the commonly applied threshold for collinearity of variance inflation factor ˃10 188 (30, see S3 Table), unless interactions with the only categorical variable habitat type were included (S4 189  Table), which commonly inflates VIF; the latter were calculated using the R-package 'performance' (37). 199 probability when local mean temperatures were above 7°C (Fig 1, S1 Table). On plots with arthropod attack, 200 on average 26% (mean; ± 19% SD) of the artificial caterpillars were attacked per plot within 2-d exposure; 201 across all plots, the average predation rate was 21% (mean± 20% SD).

205
Due to landscape diversity as landscape parameter, models at intermediate scales (1.5, 2.0 or 2.5-km) -

206
particularly at 2-km scale -were more important for explaining arthropod predation rates than models at 207 smaller (0.5 km, 1.0 km) or larger scales (3.0 km), as shown by sum of Akaike weights (Σw i , Fig 2A). The 208 relative importance of candidate predictors for explaining arthropod predation rates revealed a similar 209 pattern across all spatial scales, with high relative importance of landscape diversity and local mean 210 temperature as zero-inflation term, intermediate relative importance of mean relative humidity, and low 211 relative importance of MAT, plant species richness, local mean temperature (as fixed effect) and habitat 212 type (Figs 2B, 3). Thus, landscape diversity and -as a zero-inflation term -local mean temperature have a 213 high probability to appear in the best fitting model across spatial scales (Fig 2B), with the most substantial 214 contribution in models including landscape diversity at the intermediate 2-km scale (Fig 2A, see also S3 215 Table). 219 plant species richness, Temp or RH: local mean temperature or mean relative humidity during artificial caterpillar 220 exposure, zi: included as zero-inflation term) and filled blue symbols to regional factors (LandDiv: landscape diversity, mean ± SD, forests 0.20 ± 0.20, grasslands 0.22 ± 0.20, arable fields 0.21 ± 0.20, settlements 0.21 ± 0.20),

225
and across the observed range of plant species richness ( Fig 3B) and local mean temperature (Fig 3C), while 226 higher relative humidity tended to decrease arthropod predation rates (Fig 3D)(S3 Table). Local mean 227 temperature as zero-inflation term equals a higher probability of arthropod attack at plot level with higher 228 local mean temperatures (Fig 1). Particularly at 2-km scale (Fig 2A), arthropod predation rates increased 229 towards diverse landscapes ( Fig 3E). Higher maximum predation rates and more frequently high predation 230 rates were observed in more diverse landscapes than landscapes dominated by a single land cover type (Fig   231  3E, e.g. compare landscape diversity < 0.69 and ≥ 0.69, landscape diversity value of 0.69 equals an effective 232 number of two land-cover types). MAT did not substantially affect arthropod predation rates (Fig 3F). We 233 observed no interaction effects of any predictor on arthropod predation rates depending on habitat type (S4 234  Table).

In this study, we assessed drivers of arthropod predation in open herbaceous vegetation in typical habitat
242 types of the temperate region. Arthropod predation rates in different habitat types were similar and 243 responded similarly to both local and regional drivers. Towards diverse landscapes, particularly at 2-km 244 scale, arthropod predation rates increased, whereas they tended to decrease towards higher mean relative 245 humidity and were frequently absent from plots with low local mean temperatures. Plant species richness 246 and MAT did not substantially affect arthropod predation rates.

247
The observed average arthropod predation rate of 21% (in 2 days) in May was in the same order of

277
The absence of habitat type effects but increasing predation rates towards higher landscape diversity does 278 not mean that directly adjoining habitat type is less important to arthropod predation than general landscape

283
Local weather conditions during artificial caterpillar exposure shaped arthropod predation. In our study,

284
higher local mean temperatures made it more likely to observe predation (predation rates > 0), but did not 285 substantially increase predation rates. This seems to be in contrast to observations from pitfall trap catches,

286
where numbers of many ground-active arthropod species in the catches increased with temperature (44),

287
which similarly could have increased the likelihood of encounter with an artificial caterpillar. However, as 288 we did not study predation rates as time-series but on different plots, natural enemy communities possibly 289 differed between plots and entailed arthropod species with different temperature preferences (19) and 14 290 sensitivities (see 44), which can explain the absence of a clear temperature relationship in our study.

291
Furthermore, local mean temperatures measured 1-m above ground possibly reflected the conditions 292 experienced by a predator differently depending on the effects of vegetation structure on microclimate and 293 the daily activity pattern. Thus, local weather conditions may influence predation rates but this effect might 294 be masked in our study, possibly through differences in natural enemy communities among plots and a 295 discrepancy between the measured and experienced temperature by ground-active arthropods. However,

296
we more frequently observed the absence of attack (predation rates = 0) at low local mean temperatures.

297
Accordingly, temperature thresholds may apply more broadly to arthropod predation, at least in spring.

298
Both, because emergence after overwintering is temperature-dependent (45)  MAT, which may suggest that other factors impact natural enemy communities more strongly than MAT.

312
However, even if natural enemy communities change along the MAT gradient, this may not have led to 313 differences in predation rates. This is supported by the observation that several independent studies using 314 artificial caterpillars in temperate regions reported predation rates in the same order of magnitude (see above) -which likely encompasses large differences in natural enemy communities -, but also by the 316 marked relevance of key predators for predation functions, e.g. compared to natural enemy richness (12).

317
Key predators can be, for instance, particular voracious predator species (47)  pool, but our data suggests that this is not a ubiquitous or dominant pattern.

329
Our results provide insights into herbivore regulation through arthropod predators, but are limited by the 330 method of artificial caterpillars as sentinel prey.