Predicting the Phenology of Herbivorous Insects

Abstract

Models of herbivorous insect phenology can be used to make agriculture more sustainable and to better manage the effects of climate change on natural communities. The phenology of herbivorous insects depends on heat time, but exactly how it varies across populations and the causes of this variation are unclear. Here, with multilevel Bayesian models, we performed a comparative analysis of 601 published herbivorous insect phenology models. We found that variation in herbivorous insect phenology can be explained by variation in phylogenetic relatedness, adult body size, feeding site, host plant taxonomy, geographic location, and the approaches that researchers used for model parameterization. Contrary to previous analysis, we also found that the minimum temperature required for development varies across life stages in a way that could be adaptive. Our analysis demonstrates that by accounting for more information on the variation across insect populations and their environments, we can make better and more generalizable predictions of herbivorous insect phenology.

Keywords: degree days; integrative pest management; lower developmental threshold; multilevel Bayesian models; phylogenetic models; phytophagy; thermal requirements.

FIGURE 1

Variation across life stages in LT, the lower temperature threshold for development. The plot is based on model LTstage2 (see Table 2). Egg stage is the reference level and its posterior distribution was standardized by subtracting the mean value. The dashed line marks the mean egg‐stage effect. LT is significantly lower for growth stages and pupae than for egg, but does not significantly differ between growth stages and pupae (see Section 3).

FIGURE 2

Fixed effects on LT, the lower threshold for development, for the (a) egg and (b) growth stages. LT was measured in °C. Host breadth and body size were log‐transformed. Latitude, host breadth, and body size were centered on the mean and scaled by standard deviations. Error bars represent 95% high density intervals (HDI) for taxonomy models and 95% confidence interval (CI) for phylogenetic models. Effects for taxonomic models with the experimental host–plant family as a random effect (LTE1 and LTL1) are shown in blue, while those included nested effects for host–plant family and genus (LTE2 and LTL2) are in violet. Phylogenetic model effects are shown in green. Effect estimates with a 95% HDI or CI that does not span zero are marked with an asterisk. No effects in the phylogenetic models (PLTE and PLTL) met that criterion. The reference level for insect feeding site is “exophagous.” The reference level for measurement type is “field experiment.” The phylogenetic models do not include measurement type. Note that the phylogenetic model of the egg LT (PLTE) does not include any example of underground feeders.

FIGURE 3

Fixed effects on DD, the heat time required for development from egg to adult. DD was log‐transformed and measured in degree days. Model names are taken from Table 2. † indicates that the effect has a p value < 0.05 in phylogenetic models. Error bars represent 95% high density intervals (HDI) for taxonomy models and 95% confidence interval (CI) for phylogenetic models. Effects for the taxonomic model with the experimental host–plant family as a random effect (DD1) are shown in blue, while those including nested effects for host–plant family and genus (DD2) are in violet. Phylogenetic model (PDD) effects are shown in green. Effect estimates with a 95% HDI or CI that does not span zero are marked with an asterisk. The reference level for insect feeding site is “exophagous.” The reference level for measurement type is “field experiment.”