Well known indicator groups do not predict the decline of insects

Musters, C.J.M.; Honkoop, H.P.; Snoo, G.R. de

doi:doi:10.1016/j.ecolind.2023.111458

n decision making for insect conservation, one depends largely on knowledge of the relationship between changes in environmental factors and abundance of a very limited number of species. The species we have knowledge on cannot be regarded as a representative sample of all insects. How accurately do changes in the abundance of these species predict the changes in other species? To answer this question, we studied 373 insect species belonging to the Apidae (bees), Lepidoptera (butterflies), Orthoptera (grasshoppers), Ephemeroptera (mayflies), Trichoptera (caddisflies), Odonata (dragonflies), and Plecoptera (stoneflies), with known population trends and attributes in the Netherlands. The 78 attributes included morphological and demographic trait values, as well as habitat requirements of species. We trained Random Forests (RFs) with random samples and with taxonomic groups to predict the decline of the species based on their attributes. Then we used the trained RFs to predict the...Show moren decision making for insect conservation, one depends largely on knowledge of the relationship between changes in environmental factors and abundance of a very limited number of species. The species we have knowledge on cannot be regarded as a representative sample of all insects. How accurately do changes in the abundance of these species predict the changes in other species? To answer this question, we studied 373 insect species belonging to the Apidae (bees), Lepidoptera (butterflies), Orthoptera (grasshoppers), Ephemeroptera (mayflies), Trichoptera (caddisflies), Odonata (dragonflies), and Plecoptera (stoneflies), with known population trends and attributes in the Netherlands. The 78 attributes included morphological and demographic trait values, as well as habitat requirements of species. We trained Random Forests (RFs) with random samples and with taxonomic groups to predict the decline of the species based on their attributes. Then we used the trained RFs to predict the decline of the species outside the training groups and checked the accuracy of the predictions. The results showed that accuracy of the predictions of the RFs trained by the random samples increased from 0 to 0.20 (maximum 0.40, on a scale of 0 to 1) with sample size increasing from 10 to 90% of the insects. Moreover, we found that the accuracy of the predictions by the RFs trained with the taxonomic groups were zero in case of butterflies and grasshoppers, and low in other groups (maximum 0.37, in case of bees predicting terrestrial insects). Accuracy depended significantly on the size of the taxonomic group. Large over- or underestimation of number of declining species occurred in all cases. Further, we found that the taxonomic groups had few attributes important for predicting in common. The attribute ‘Active dispersion’ had the highest importance when all insects were used for training the RF. Using ‘indicator groups’ for predicting the decline of insects has a high risk of over- or underestimating the actual number of declining species and should therefore be advised against unless the indicator group is sure to be representative.Show less

Leiden University Scholarly Publications

View statistics

Documents

In Collections

Well known indicator groups do not predict the decline of insects