Minimizing insect mortality during grassland mowing: The potential of insect chasing devices

<b>Abstract</b>1. Early and frequent mowing is a threat to grassland arthropods. To date, measures to counteract these losses have largely been limited to extensification concepts, which are not always feasible on intensive grasslands.2. One option for insect-friendly mowing is the use of chasing devices, which are designed to remove arthropods from the grassland before mowing. Scientific research on their effectiveness is limited to one study on parasitic microhymenoptera. Studies on other groups are lacking.3. In a three-year experiment we have investigated the effectiveness of three devices, i.e. two mechanical flushing bars (truck tarpaulin, hanging down metal brackets) and a blowing device. We tested different driving speeds and distances of the flushing bar from the mower on seven abundant arthropod groups in grassland that differ in their escape strategies.4. We found the blowing device to be the most effective. At the maximum driving speed of 12 km/h it had an effect on almost all taxonomic groups studied except Araneae and Hymenoptera. At this speed, the metal bracket flushing bar was only effective on Diptera and Coleoptera. The truck tarpaulin flushing bar was only effective at a low driving speed of 5 km/h, albeit on four out of seven taxa.5. Chasing devices have a great potential to remove arthropods prior to mowing, thereby reducing mortality. Whilst a costly blowing device works at higher speed, inexpensive mechanical flushing bars have a limited effectiveness (metal brackets) or only work at low speed (truck tarpaulin).<b>Materials and Methods</b>The experiments took place in the years 2022 to 2024 on a study site of the Kleinhohenheim Agricultural Experiment Station of the University of Hohenheim, Stuttgart, Germany (48.734 °N, 9.204 °E). In each year, the study site was mown twice, once in June/July and once in September (Figure 1). The experimental design consists of five blocks, each separated into five plots. This design allows to repetitively sample five different treatments per block (von Berg <i>et al.</i>, 2024). More detailed information on the study sites and design can be found in von Berg <i>et al.</i> (2024)Here, we compared the direct effects of two flushing bars and a blowing device together with a conventional disc mower (Disco 320, without conditioner, working width 3 m, Power Take-Off (PTO) speed 850-1000 rpm, Claas Saulgau GmbH, Bad Saulgau, Germany) on the arthropod fauna. We tested the following treatments: (a) disc mower with flushing bar made of truck tarpaulin, (b) disc mower with hanging down metal brackets (c) disc mower with a blowing device and (d) a disc mower without flushing bar as control. The first two devices were self-made prototypes (instructions and exact measurements are publicly available: dryad link available after acceptance) while the blowing device was supplied by a manufacturer of mowing machines (leaf blower type F4052, capacity 240 m³/min, PTO 850 rpm, Fischer Maschinenbau GmbH &amp; Co. KG, Gemmrigheim, Germany) (Figure 2).The sampling took place twice a year from end of June to beginning of July 2022–2024 and end of September 2022–2024, depending on the weather (Table 1). The experimental plots were treated and sampled in similar time intervals. For detailed sampling description see von Berg <i>et al.</i> (2024)The devices were attached in either 1 m (truck tarpaulin in 2022, 2023), 5.5 m (truck tarpaulin and metal brackets) or 6.8 m (blowing device; in 2024) distance to the mower. Based on the first year’s results attempts have been made in subsequent years to continually optimise the design of the flushing bars, the driving speed and the distance between the flushing bar and the mower: Flushing bars were installed either directly to the mower (truck tarpaulin in 2022, 2023) or at the front of the tractor (2024) at a distance of 5.5 m and 6.8 m from the mower, for the mechanical flushing bars and the blowing device, respectively. They were installed so that the lower edge of the mechanical flushing bars and the blowing device were 15 cm and 30 cm above the ground, respectively. The driving speed of the tractor was 5 km/h (in 2022) and 12 km/h (in 2023 and 2024) with a cutting height of 7 cm (Table 1, Figure 2).Three randomly placed 1m² isolation squares (Mühlenberg, 1989) were set up per plot and completely vacuumed for a total of 5 min with an insect vacuum (ecoVAC Insect vacuum, ecoTech Umwelt-Messsysteme GmbH, Bonn, Germany). Further, we removed the mown grass from the isolation squares by hand and weighted the grass with a crane scale. In 2022, the whole content of each insect vacuum sample bag was transferred into vials filled with 80% denatured ethanol. From 2023 onwards samples were transferred to zip-lock plastic bags and stored at -18 °C until further processing (for detailed information see von Berg <i>et al.</i>, (2024))In the first year, we sampled 60 isolation squares (disc mower: 30 samples, dm mower with truck tarpaulin: 30 samples; driving speed 5 km/h; Table 1). In the second year, we sampled 60 isolation squares (disc mower: 30 samples, disc mower with truck tarpaulin: 30 samples; driving speed 12 km/h; Table 1), and in the third year, 120 isolation squares (disc mower: 30 samples; disc mower with truck tarpaulin: 30 samples; disc mower with metal brackets: 30 samples; disc mower with blowing device: 30 samples; driving speed 12 km/h; Table 1). All 240 samples were sorted for the following taxonomic groups: Araneae, Orthoptera, Cicadina, Heteroptera, Hymenoptera (parasitoid wasps and Anthophila), Coleoptera, and Diptera. All non-target taxa (e.g. Collembola, Aphididae, Formicidae) from the analysis. Only individuals that were not damaged or injured by the mower were considered for the subsequent data analysis (von Berg <i>et al.</i>, 2024). All sorted individuals were stored as voucher specimens in 80% denatured ethanol at the Universities of Hohenheim and Tübingen. All data have been deposited in and are available from Dryad repository (link available after acceptance).All data analyses were performed in RStudio (Posit team, 2023) using R 4.3.2 (R Core Team, 2023)following Santon <i>et al.</i> (2023) to conduct generalized linear mixed models (glmmTMB; (Brooks <i>et al.</i>, 2017)) using Template Model Builder (TMB; (Kristensen <i>et al.</i>, 2016)). Due to the different experimental setups, each year was analysed separately and samples with missing values for one or more of the predictors were excluded from the analyses. This resulted in a sample size of n = 30 for all treatments in all experiments, except for n = 29 in case of the truck tarpaulin flushing bar in 2022 and 2023. In addition, one outlier for Diptera (92 individuals per m²) was excluded from the analyses for the 2022 truck tarpaulin flushing bar (n = 28). Date was used as factor predictor, daytime and grass weight were included as numeric predictors and block and plot were used as random factors. All models were tested for interaction effects between date and treatment and were included where necessary, i.e. for Heteroptera, Coleoptera and Diptera in the 2024 experiment. Tukey post hoc tests using ‘multcomp’ (Hothorn <i>et al.</i>, 2008) for models without interactions or ‘emmeans’ (Lenth, 2023) for models with interactions were used to compare data from the various scaring devices in 2024. Pairwise comparisons (mean and 95% compatibility intervals), considering the model predictions, were used to determine the percentage differences between the individual treatments (Santon <i>et al.</i>, 2023).Referencesvon Berg, L., Frank, J., Betz, O., Steidle, J.L.M., Böttinger, S., Sann, M. (2024). Disc mower versus bar mower: evaluation of the direct effects of two common mowing techniques on the grassland arthropod fauna. <i>Journal of Applied Ecology</i>, <b>n/a</b>.Brooks, M.E., Kristensen, K., Benthem, K.J. van, Magnusson, A., Berg, C.W., Nielsen, A., Skaug, H.J., Mächler, M., Bolker, B.M. (2017). glmmTMB Balances Speed and Flexibility Among Packages for Zero-inflated Generalized Linear Mixed Modeling. <i>The R Journal</i>, <b>9</b>: 378–400.Hothorn, T., Bretz, F., Westfall, P. (2008). Simultaneous Inference in General Parametric Models. <i>Biometrical Journal</i>, <b>50</b>: 346–363.Kristensen, K., Nielsen, A., Berg, C.W., Skaug, H., Bell, B.M. (2016). TMB: Automatic Differentiation and Laplace Approximation. <i>Journal of Statistical Software</i>, <b>70</b>: 1–21.Lenth, R.V. (2023). emmeans: Estimated Marginal Means, aka Least-Square Means.Mühlenberg, M. (1989). <i>Freilandökologie</i>. 2nd ed. Quelle &amp; Meyer Verlag, Heidelberg.Posit team. (2023). RStudio: Integrated Development Environment for R.R Core Team. (2023). R: A Language and Environment for Statistical Computing.Santon, M., Korner-Nievergelt, F., Michiels, N.K., Anthes, N. (2023). A versatile workflow for linear modelling in R. <i>Frontiers in Ecology and Evolution</i>, <b>11</b>.<b>Description of the dataset</b>Numbers of individuals for the seven assessed arthropod groups (Araneae, Orthoptera, Cicadina, Heteroptera, Hymenoptera, Coleoptera and Diptera) recorded for the different treatments: disc mower (dm), disc mower with truck tarpaulin (dm), disc mower with metal brackets (dm_brackets) and disc mower with blowing device (dm_air) in three consecutive years (2022, 2023, 2024). Sampling was carried out using 1-m² isolation squares and an insect vacuum on the following sampling dates: 2022-07-04, 22.09.2022, 2023-06-15, 2023-09-19, 2024-07-08, 2024-09-25. The study site was divided into five blocks (B1-B5). Each block contains up to 5 sampled subsections (plot P1-P5). Per subsection and treatment three samples were taken at the same time. The time of sampling (daytime), the driving speed of the tractor (speed in km/h), the cutting height (in cm), the weight of the grass in each isolation square (grass weight in kg) as well as the distance between the flushing bar and the mower (in m) were recorded. Distance between the flushing bar and the mower is not applicable (N.A.) for the disc mower only (dm) treatment.

<b>摘要</b>
1. 早期高频刈割对草地节肢动物（grassland arthropods）存在显著威胁。截至目前，针对该类种群损失的应对措施大多局限于草地粗放经营理念，然而这类方案在高强度经营的草地中往往难以实施。
2. 适配昆虫友好型刈割的一种方案是使用驱离装置（chasing devices），这类装置旨在刈割前将节肢动物从草地中驱离。目前针对其有效性的科学研究仅局限于一项针对寄生性膜翅目小蜂（parasitic microhymenoptera）的研究，尚缺乏对其他类群的相关探索。
3. 本研究通过为期三年的野外试验，评估了三类装置的驱离效果：两类机械冲洗杆（mechanical flushing bars）——分别为卡车篷布型、悬挂式金属支架型，以及一台风力驱离装置（blowing device）。我们针对7类具有不同逃逸策略的优势草地节肢动物类群，测试了不同的拖拉机行驶速度，以及冲洗杆与刈割机（disc mower）之间的间距。
4. 结果显示，风力驱离装置的驱离效果最优。当拖拉机以最大行驶速度12 km/h运行时，除蜘蛛目（Araneae）和膜翅目（Hymenoptera）外，该装置对几乎所有受试类群均有效。在该速度下，金属支架型冲洗杆仅对双翅目（Diptera）和鞘翅目（Coleoptera）有效；而卡车篷布型冲洗杆仅在较低行驶速度5 km/h时生效，且仅对7类类群中的4类有效。
5. 驱离装置在刈割前驱离节肢动物、进而降低其死亡率方面具备巨大应用潜力。尽管成本较高的风力驱离装置可在较高行驶速度下工作，但低成本的机械冲洗杆要么驱离效果有限（金属支架型），要么仅能在低速下生效（卡车篷布型）。

<b>材料与方法</b>
本试验于2022年至2024年在德国斯图加特霍恩海姆大学克莱因霍恩海姆农业试验站的试验样地开展（北纬48.734°，东经9.204°）。每年试验样地进行两次刈割，分别在6月/7月和9月（图1）。试验设计包含5个区组，每个区组再划分为5个样地，该设计允许在每个区组内重复开展5种不同处理的试验（von Berg et al., 2024）。关于试验样地与试验设计的详细信息可参见von Berg等人（2024）的研究。

本研究比较了两类冲洗杆、一台风力驱离装置，搭配常规圆盘刈割机（disc mower，型号Disco 320，无压扁装置，作业幅宽3 m，动力输出轴（Power Take-Off, PTO）转速850~1000 rpm，产自德国绍高克劳阿斯公司（Claas Saulgau GmbH, Bad Saulgau, Germany)）对草地节肢动物群落的直接影响。我们设置了以下4种处理：(a) 搭载卡车篷布型冲洗杆的圆盘刈割机；(b) 搭载悬挂式金属支架型冲洗杆的圆盘刈割机；(c) 搭载风力驱离装置的圆盘刈割机；(d) 未搭载任何驱离装置的圆盘刈割机（对照组）。前两类装置为自制原型（制作说明与精确尺寸可在论文录用后公开获取：Dryad数据仓库（Dryad repository）链接），而风力驱离装置由一家刈割机械制造商提供（型号F4052型风机，风量240 m³/min，PTO转速850 rpm，产自德国盖明根菲舍尔机械制造有限公司（Fischer Maschinenbau GmbH & Co. KG, Gemmrigheim, Germany)）（图2）。

采样于2022年至2024年的6月末至7月初，以及9月末开展，具体时间依天气条件调整（表1）。试验样地的处理与采样均保持相近的时间间隔。详细的采样流程可参见von Berg等人（2024）的研究。

装置与刈割机的间距分别设置为：卡车篷布型冲洗杆1 m（2022、2023年）、5.5 m（2022年后的卡车篷布型与金属支架型），以及风力驱离装置6.8 m（2024年）。基于第一年的试验结果，后续年份中我们持续优化了冲洗杆的设计、拖拉机行驶速度，以及冲洗杆与刈割机的间距：机械冲洗杆要么直接安装在刈割机上（2022、2023年的卡车篷布型），要么安装在拖拉机前部，与刈割机的间距分别为5.5 m（机械冲洗杆）与6.8 m（风力驱离装置）。安装时，机械冲洗杆的下边缘距地面15 cm，风力驱离装置的下边缘距地面30 cm。拖拉机行驶速度在2022年为5 km/h，2023与2024年为12 km/h，刈割高度均为7 cm（表1，图2）。

每个样地设置3个随机布设的1 m²隔离样方（Mühlenberg, 1989），使用昆虫真空采样器（insect vacuum，ecoVAC型，产自德国波恩生态环境测量系统有限公司（ecoTech Umwelt-Messsysteme GmbH, Bonn, Germany)）对样方进行总计5分钟的真空采样。此外，我们手工收集隔离样方内的刈割牧草，并用吊秤称重。2022年，我们将昆虫采样器收集袋中的全部样品转移至装有80%变性乙醇的样本瓶中；2023年起，样品被转移至密封塑料袋中，并置于-18 ℃环境下保存以待后续处理（详细信息参见von Berg等人，2024）。

第一年我们共采样60个隔离样方（圆盘刈割机组：30个样本，搭载卡车篷布型冲洗杆的刈割机组：30个样本，行驶速度5 km/h；表1）。第二年采样60个隔离样方（圆盘刈割机组：30个样本，搭载卡车篷布型冲洗杆的刈割机组：30个样本，行驶速度12 km/h；表1）。第三年采样120个隔离样方（圆盘刈割机组：30个样本，搭载卡车篷布型冲洗杆的刈割机组：30个样本，搭载金属支架型冲洗杆的刈割机组：30个样本，搭载风力驱离装置的刈割机组：30个样本，行驶速度12 km/h；表1）。全部240份样品均针对以下7个分类类群进行分拣：蜘蛛目（Araneae）、直翅目（Orthoptera）、蝉亚目（Cicadina）、异翅亚目（Heteroptera）、膜翅目（Hymenoptera，包括寄生蜂与显花昆虫Anthophila）、鞘翅目（Coleoptera）以及双翅目（Diptera）。分析中排除了所有非靶标类群（如弹尾纲Collembola、蚜科Aphididae、蚁科Formicidae）。仅统计未被刈割机损伤的个体用于后续数据分析（von Berg et al., 2024）。所有分拣后的个体均作为凭证标本（voucher specimens）保存于80%变性乙醇中，存放于霍恩海姆大学与蒂宾根大学。所有数据已提交至Dryad数据仓库（Dryad repository）（论文录用后可公开获取链接）。

所有数据分析均在RStudio（Posit团队，2023）中完成，使用R 4.3.2（R核心团队，2023），并遵循Santon等人（2023）的流程，使用Template Model Builder（TMB）构建广义线性混合模型（generalized linear mixed models, glmmTMB; Brooks et al., 2017）。由于试验设置存在差异，我们每年单独进行分析，并剔除了存在一个或多个预测变量缺失值的样本。最终所有试验处理的样本量均为n=30，仅2022与2023年的卡车篷布型冲洗杆处理组样本量为n=29。此外，2022年卡车篷布型冲洗杆处理组的双翅目数据中存在1个异常值（每平方米92头个体），该异常值被剔除，最终该组样本量为n=28。采样日期作为分类预测变量，采样时段与牧草重量作为数值预测变量，区组与样地作为随机效应变量。所有模型均测试了采样日期与处理之间的交互效应，并在必要时将其纳入模型，例如2024年试验中的异翅亚目、鞘翅目与双翅目数据。针对无交互效应的模型，我们使用‘multcomp’包（Hothorn et al., 2008）进行Tukey事后检验；针对存在交互效应的模型，我们使用‘emmeans’包（Lenth, 2023）进行比较。基于模型预测结果，我们通过两两比较（均值与95%置信区间）确定了各处理组之间的百分比差异（Santon et al., 2023）。

<b>参考文献</b>
von Berg, L., Frank, J., Betz, O., Steidle, J.L.M., Böttinger, S., Sann, M. (2024). 圆盘刈割机与杆式刈割机：两种常见刈割技术对草地节肢动物群落直接影响的评估。<i>Journal of Applied Ecology</i>, <b>n/a</b>.
Brooks, M.E., Kristensen, K., Benthem, K.J. van, Magnusson, A., Berg, C.W., Nielsen, A., Skaug, H.J., Mächler, M., Bolker, B.M. (2017). glmmTMB：兼顾零膨胀广义线性混合模型构建速度与灵活性的R包。<i>The R Journal</i>, <b>9</b>: 378–400.
Hothorn, T., Bretz, F., Westfall, P. (2008). 广义参数模型的同时推断。<i>Biometrical Journal</i>, <b>50</b>: 346–363.
Kristensen, K., Nielsen, A., Berg, C.W., Skaug, H., Bell, B.M. (2016). TMB：自动微分与拉普拉斯近似工具。<i>Journal of Statistical Software</i>, <b>70</b>: 1–21.
Lenth, R.V. (2023). emmeans：估计边际均值，又称最小二乘均值。
Mühlenberg, M. (1989). <i>Freilandökologie</i>. 第2版. Quelle & Meyer Verlag, 海德堡.
Posit团队. (2023). RStudio：R语言集成开发环境。
R Core Team. (2023). R：统计计算语言与环境。
Santon, M., Korner-Nievergelt, F., Michiels, N.K., Anthes, N. (2023). R语言线性建模的通用工作流。<i>Frontiers in Ecology and Evolution</i>, <b>11</b>.

<b>数据集说明</b>
本数据集包含7类受试节肢动物类群（蜘蛛目、直翅目、蝉亚目、异翅亚目、膜翅目、鞘翅目、双翅目）的个体数量统计结果，这些数据来自4种不同处理的试验：圆盘刈割机（dm）、搭载卡车篷布型冲洗杆的圆盘刈割机（dm）、搭载金属支架型冲洗杆的圆盘刈割机（dm_brackets）以及搭载风力驱离装置的圆盘刈割机（dm_air），试验周期为连续三年（2022、2023、2024年）。采样采用1 m²隔离样方结合昆虫真空采样器的方式，采样日期分别为：2022-07-04、2022-09-22、2023-06-15、2023-09-19、2024-07-08、2024-09-25。试验样地被划分为5个区组（B1~B5），每个区组包含最多5个采样子区域（样地P1~P5）。每个子区域与处理组合下同时采集3份样本。记录的变量包括：采样时间（时段）、拖拉机行驶速度（km/h）、刈割高度（cm）、每个隔离样方内的牧草重量（kg），以及冲洗杆与刈割机之间的间距（m）。仅未搭载任何驱离装置的圆盘刈割机（dm）处理组无需记录冲洗杆与刈割机的间距（N.A.）。