Estimating effects of arable land-use intensity on farmland birds using joint species modeling.

Declines in European farmland birds over past decades have been attributed to the combined effects of agricultural intensification and abandonment. Consequently, aspirations to stop declines should focus attention to reversing these changes through voluntary or policy driven interventions. The design of such interventions should ideally be informed by scientific knowledge of which aspects of the transformation of agricultural landscapes have contributed to the farmland bird declines. Declines may be associated with loss of natural habitats or the intensification and homogenization of land-use management on production land, and furthermore, these changes may interact. Here we applied an orthogonal design exploiting spatial variation in land use in a major agricultural region of Sweden to seek evidence for benefits to farmland birds of reversing some of the intensifications on and among arable fields and whether effects are modified by the availability of semi-natural habitats (pastures and field borders) in the landscape. We accounted for the potentially confounding effect of interactions between species by using a joint species distribution model explicitly controlling for additional variation and covariation among species. We found that interventions aimed specifically at land in production could provide benefits to farmland birds. Landscapes with a higher proportion leys or fallows and/or with a more diverse set of crops held higher abundances of most farmland birds. However, effects were only apparent in landscapes with low availability of semi-natural habitats, and were sometimes even negative in landscapes with high amounts of such habitats, demonstrating context dependence. Even if we found little evidence of interactions between species, the joint modeling approach provided several benefits. It allowed information to be shared between species making analyses robust to uncertainty due to low abundances and provided direct information about the mean and variability in effects of studied predictors among species. We also found that care needs to be taken regarding prior and distributional assumptions as the importance of species interactions might otherwise be overstated. We conclude that this approach is well suited for evaluating agricultural policies by providing evidence for or against certain interventions or to be linked to policy scenarios of land-use change. This article is protected by copyright. All rights reserved.