Climate, foliage chemistry and herbivory

Climate and geology, not single soil nutrients, set the stage for insect herbivory in tropical montane rainforest. Hierarchical Bayesian models traced their influence through soil chemistry and foliage chemistry to herbivory itself, splitting direct from indirect pathways.

The evidence comes from 25 sites in the Australian Wet Tropics, chosen across temperature, precipitation and geology gradients. 3 widespread rainforest tree species anchored the comparison. Trees of different species responded differently to the same resources, which is the practical warning of the paper.

Architecture

The models are hierarchical, specified in JAGS and processed in R. Structured pathway coefficients carry the network: climate and geology to soil nutrients, soil to foliage chemistry, chemistry to herbivory. The model estimates direct and mediated paths together, with random effects for site nesting. Posterior analysis reads each pathway’s strength with its uncertainty.

The decision that was hard

Cascades invite a pile of pairwise regressions: soil against foliage, foliage against herbivory, each defensible alone. Pairwise results cannot separate a direct effect from a route through a mediator, and that separation was the question. We committed to one structured network model with explicit pathway coefficients. It cost model complexity and harder fitting, and it is the only design that answers what drives what.

What was measured

Climate and geology showed an overarching influence over soil chemistry, foliar nitrogen and insect herbivory, directly and indirectly. Each pathway carries a posterior estimate of its strength and its uncertainty. The 3 tree species responded to the same resources in different ways.

What did not work

The expected workhorse, individual soil nutrients, did not deliver. Once site geology was accounted for, single nutrients showed equivocal influence on foliage chemistry. A simple nutrient-to-leaf story would have been publishable and wrong. The honest conclusion names specific limiting factors, not convenient proxies of resource availability.

Role

I designed the pathway models, structured the multivariate field dataset, fitted and interpreted the Bayesian models and wrote the manuscript.

What this taught me about evaluation