Towards a system-level causative knowledge of pollinator communities

Abstract

Pollination plays a central role in both crop production and maintaining biodiversity. However, habitat loss, pesticides, invasive species and larger environmental fluctuations are contributing to a dramatic decline of pollinators worldwide. Different management solutions require knowledge of how ecological communities will respond following interventions. Yet, anticipating the response of these systems to interventions remains extremely challenging due to the unpredictable nature of ecological communities, whose nonlinear behaviour depends on the specific details of species interactions and the various unknown or unmeasured confounding factors. Here, we propose that this knowledge can be derived by following a probabilistic systems analysis rooted on non-parametric causal inference. The main outcome of this analysis is to estimate the extent to which a hypothesized cause can increase or decrease the probability that a given effect happens without making assumptions about the form of the cause–effect relationship. We discuss a road map for how this analysis can be accomplished with the aim of increasing our system-level causative knowledge of natural communities. This article is part of the theme issue ‘Natural processes influencing pollinator health: from chemistry to landscapes’.

Protection promotes energetically efficient structures in marine communities

“…S1). Importantly, this causal graph satisfies the conditions to measure genuine direct causation between MPAs and community structure (33, 46) (Notes and Table S2).…”

“…Lastly, to formally study the cause-effect relationship between MPAs and community structure under the context of our studied internal and external variables, we follow a nonparametric causal-inference approach using do-calculus (33). This approach allows us to translate (whenever possible) interventional conditional distributions P ( Y = y | do ( X = x )) into observational conditional distributions P ( Y = y | X = x ) (46). Note that applying the do ( X = x ) operator to observations involves the removal of all incoming effects from the intervened variable in a causal graph (33).…”

“…Once we empirically corroborated our causal graph, we calculated the direct average causal effects ( ACE s) following the rules of do-calculus. This implies working with a modified version of the causal graph ( P ( Y | do ( X )) = P m ( Y | X )), where all the incoming links to the node X have been removed (46). We used product decomposition to simplify conditional independencies as much as possible, and to use the observed probability distributions (i.e., P ( Y | X )) whenever P m = P .…”

“…Causal inference. We applied a nonparametric probabilistic approach to calculate cause-effect relationships rooted on do-calculus (33,46). First, we tested our causal graph by confirming all the possible d-separations emerging from the directed acyclic graph (DAG) depicted in Figure S1 (also test with 99% confidence level (see Table S2 in SI).…”

Natural processes influencing pollinator health

“…Given the range and variety of these drivers, many of which are covered in this issue, the experimental work to investigate this is simply too great [ 76 ]. By contrast, Saavedra et al [ 77 ] provide a statistical approach that could enable us to understand causal drivers of, for example, pollinator richness, based on observational rather than experimental data. Given the wealth of observational data in the scientific literature, and the relative ease with which it can be collected (as opposed to the cost and complexity of ecological experiments), application of probabilistic systems analysis rooted in nonparametric causal inference holds out real hope for the scientific community to take apart the complex relationships between pollinator communities and ecological and environmental drivers.…”