Abstract
Wild bee populations are threatened by current agricultural practices in many parts of the world, which may put pollination services and crop yields at risk. Loss of pollination services can potentially be predicted by models that link bee abundances with landscape‐scale land‐use, but there is little knowledge on the degree to which these statistical models are transferable across time and space. This study assesses the transferability of models for wild bee abundance in a mass‐flowering crop across space (from one region to another) and across time (from one year to another). The models used existing data on bumblebee and solitary bee abundance in winter oilseed rape fields, together with high‐resolution land‐use crop‐cover and semi‐natural habitats data, from studies conducted in five different regions located in four countries (Sweden, Germany, Netherlands and the UK), in three different years (2011, 2012, 2013). We developed a hierarchical model combining all studies and evaluated the transferability using cross‐validation. We found that both the landscape‐scale cover of mass‐flowering crops and permanent semi‐natural habitats, including grasslands and forests, are important drivers of wild bee abundance in all regions. However, while the negative effect of increasing mass‐flowering crops on the density of the pollinators is consistent between studies, the direction of the effect of semi‐natural habitat is variable between studies. The transferability of these statistical models is limited, especially across regions, but also across time. Our study demonstrates the limits of using statistical models in conjunction with widely available land‐use crop‐cover classes for extrapolating pollinator density across years and regions, likely in part because input variables such as cover of semi‐natural habitats poorly capture variability in pollinator resources between regions and years.
Highlights
Pollination by wild animals is a key ecosystem service that is highly important for 35% of the world’s crops (Klein et al 2007), and wild insects are especially important in supporting yields (Garibaldi et al 2013, Rader et al 2016, Dainese et al 2019)
We focused on two land-use cover variables: percentage of winter oilseed rape and percentage of semi-natural habitats, which included forest and permanent, non-intensively managed grasslands (Supporting information)
To assess whether the cover of mass-flowering crops and semi-natural habitats in the landscape show consistent effects on wild bee abundance across space and time, we analyzed the relationship between wild bees and the explanatory variables oilseed rape and semi-natural habitat by building a hierarchical model that contains all region–study–year combinations, and obtaining a parsimonious random effects structure (the results of model selection for support of random slopes (Supporting information)
Summary
Pollination by wild animals is a key ecosystem service that is highly important for 35% of the world’s crops (Klein et al 2007), and wild insects are especially important in supporting yields (Garibaldi et al 2013, Rader et al 2016, Dainese et al 2019). To support wild pollinators and reduce the risk for economic vulnerability induced by low pollination levels (Gallai et al 2009), it is recommended to change the landscape-scale land use to ensure the availability of nesting and overwintering habitats and pollen and nectar resources supplied by wild and cultivated flowering plants (Carvell et al 2006, Smith et al 2014, IPBES 2016). Linking landscape-scale land-use to availability of resources for pollinators across space and time (Baude et al 2016), or more often proxies such as cover of habitat assumed to be rich in resources for pollinators, is a key aspect of applied ecological research on pollinators. The application of models for decisionmaking often requires generating predictions for locations and time periods that are distinct from those for which the available empirical data was recorded (i.e. extrapolation). Despite the critical role of model transferability, i.e. how well models generalize to new contexts, is poorly studied
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
