Paving the way for improved insect metabarcoding

Abstract Metabarcoding is becoming an increasing popular method for broadscale insect monitoring. However, to complement or replace traditional insect monitoring approaches, the reliability of metabarcoding has to be confirmed. Therefore, we have evaluated the accuracy of species identifications of metabarcoding against the standard of morphology with binary classification in a confusion matrix to measure model performance. Within 12 German nature protected areas using a transect of five traps, metabarcoding found a total of 15,107 OTUs. Identifying 3096 individuals of three selected taxa by morphology, we found in total 151 species, compared with 130 species by metabarcoding. Species numbers for single traps differed substantially between methods, with a significant correspondence only found for Syrphidae (Diptera). Congruence at the species level was low, with sensitivity and precision below 50%, and even lower for Red List taxa. The match increased when aggregating traps and sites across Germany, or when comparing genera instead of species. Abundance curves strongly differed for species identified by both methods. So far, metabarcoding is lacking the necessary accuracy at the species level. We discuss possible causes for these inconsistencies and make suggestions for improvement for metabarcoding identification.