PREDICTING INSECT DISTRIBUTIONS FROM CLIMATE AND HABITAT DATA

Authors: Ulrichs, Christian; Hopper, Keith

Submitted to: BioControl
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/8/2007
Publication Date: 11/27/2007
Citation: Ulrichs, C., Hopper, K.R. 2008. Predicting insect distributions from climate and habitat data. BioControl. 10.1007/s10526-007-9143-8, 1-14.

Interpretive Summary: Knowing the effects of climate and habitat on insect distributions would help target the search for predators and parasites to introduce to control pests and increase establishment and efficacy of such predators and parasites. Therefore we used a new approach, using the actual geographical distributions of insects to develop statistical models for the effects of climate and habitat on distribution. We tested this approach using six insect pests found in the United States: European corn borer, Russian wheat aphid, corn earworm, Colorado potato beetle, imported fire ant, and plum curculio. The models correctly predicted presence for at least 93% of the observations on each insect species. The models correctly predicted absence for 59% to 77% of the observations on each species except the corn earworm, where absence was correctly predicted for only 21% of observations. Predictions of insect absence were more difficult because absence data were less abundant and perhaps less reliable. Our approach offers potential for the analysis of existing data to produce useful predictions about establishment of insects. However, the quality of the predictions depends heavily on the quality of data, and in particular, more data are needed from locations where insects are sampled but not found.

Technical Abstract: Knowing the effects of climate and habitat on pest and natural enemy distribution would help target the search for natural enemies, increase establishment of intentional introductions, and improve risk assessment for accidental introductions. Existing methods used to predict insect distributions either involve subjective comparisons of climate or require data concerning insect responses to climate. Furthermore, current methods don not include effects of habitat. We took a new approach, using the actual geographical distributions of insects to develop statistical models for the effects of climate and habitat on distribution. We tested this approach using six insect pests in the US: Ostrinia nubilalis, Diuraphis noxia, Helicoverpa zea, Leptinotarsa decemlineata, Solenopsis invicta, and Conotrachelus nenuphar. By randomly separating the data into model-building and test sets, we were able to estimate prediction accuracy. For each species, a unique combination of predictor variables was identified. The models correctly predicted presence for more than 93% of the data for each insect species. The models correctly predicted absence for 59% to 77% of the data for five of the six species. Absence predictions about H. zea were poor because distribution data for this species were limited and inaccurate. Our approach offers potential for the analysis of existing data to produce predictions about insect establishment. However, accurate prediction depends heavily on data quality, and in particular, more data are needed from locations where insects are sampled but not found.