Biases in estimation of insect herbivory from herbarium specimens

Effect of sampling protocol on leaf area lost to insects

We analysed 248 samples collected from 17 species of woody plants native to the study region. Among these, 85 samples were collected using the protocol developed for ecological research, and 163 samples were collected as herbarium specimens. Each herbarium specimen, on average, contained four-fold fewer leaves than a sample collected by ecological methods (13.5 and 50.3 leaves, respectively).

Measurements of leaf area lost to insects were performed by M.V.K., who was aware of research hypothesis and sample origin, and by J. Rikus, who was blinded to these factors. The measurements by both persons yielded the same results (average difference ± SE: 0.52 ± 0.48%; P > 0.38), indicating that the results by M.V.K. used in subsequent analyses were not affected by confirmation bias.

The average losses of woody plant foliage (Supplementary Data S1) were significantly lower for herbarium specimens (4.87%) than for ecological samples (7.96%), although the differences between these two types of samples varied with the plant species (Table 1, Fig. 1). Collectors generally prefer branches with low levels of herbivory, and when insect damage increases in nature, the difference between herbarium specimens and ecological samples becomes greater (see e.g. Sorbus aucuparia and Tilia cordata on Fig. 1). Individual collectors significantly differed in their attitudes to leaf damage by insects (F_{14, 79} = 1.93, P = 0.04; Fig. 2) while collecting herbarium specimens, with their choices ranging from careful selection of branches with nearly undamaged leaves to taking almost no account of the extent of insect herbivory. As a result, the actual levels of herbivory and the damage in herbarium specimens varied independently from each other (regression analysis: F_{1, 15} = 0.07, P = 0.80; Fig. 3). Additional analysis showed that exclusion of Sorbus aucuparia, the species with the extreme differences between the levels of herbivory measured from two types of samples, did not change our main result: the correlation between the levels of herbivory in ecological samples and in herbarium specimens remained not significant (data not shown).

Preference of press-dried plant specimens by herbarium curators

The 17 herbarium curators reported the use of the following criteria for selecting herbarium specimens for accession: good representation of morphological characters (13 respondents); high quality of preservation and preparation, including less overlapping plant parts (9 respondents); large size, in terms of the amount of material available for a study (7 respondents); low damage by insects or fungi (7 respondents); presence of additional identifiable organisms, e.g. lichens and leafminers (5 respondents); high aesthetic quality/beauty (3 respondents); and high variation in leaf size (1 respondent). Interestingly, two respondents reported a preference for specimens with leaf mines (i.e. with damage by identifiable insects) but avoidance of specimens with high damage by chewing insects (i.e. damage that cannot be associated with a particular insect species).

Two respondents refused to select between paired images, because (as they argued) all or almost all plant individuals used as test objects do not deserve accession into herbarium collections. The remaining 15 respondents (Supplementary Data S2) clearly preferred less damaged specimens (which were selected 211 times from 315 offers; the difference from the random selection: χ² = 18.71, d.f. = 1, P < 0.0001). Consequently, the selected plant specimens had only 60% of the average foliar damage found in the rejected plant specimens (F_{1, 28} = 41.54, P < 0.0001; Fig. 4). The selection was independent of the position (left or right) of the more damaged plant specimen within a pair (left or right; F_{1, 14.9} = 0.29, P = 0.60).

The respondents significantly differed in their preferences towards less damaged plant specimens (χ² = 10.15, d.f. = 1, P = 0.0007). The percentage of less damaged specimens selected by individual respondents from 21 pairs of images varied from 38.1 to 90.5%.

The probability of selection of less damaged specimen also varied among pairs of plant specimens (χ² = 41.63, d.f. = 1, P < 0.0001). The percentage of less damaged specimens selected by 15 respondents from a pair of images varied from 0 to 100%, and in 10 pairs of images this preference was statistically significant (P < 0.05): in eight pairs, the respondents consistently selected the less damaged specimen, but in two pairs, they clearly preferred the more damaged specimen. The differences in the absolute levels of herbivory between the paired specimens did not explain the detected preferences (F_{1, 13.6} = 0.09, P = 0.77).

Discussion

Collection of multiple specimens of the same species in the same locality is often performed when inventorying vegetation of a certain region. Collection of duplicates is also recommended by some manuals³⁰. Therefore selection among several specimens collected in the course of field excursions is a part of herbarium practice³¹, and our experiments mimic this practice in as many details as possible.

The results of our two experiments consistently demonstrated that measurements of losses of woody plant foliage from herbarium specimens strongly underestimate the levels of background insect herbivory observed in nature, thus confirming our first hypothesis. This finding is not surprising, because many textbooks, both old and recent^13,15,16,17, advise the collection of undamaged plants and plant parts for herbaria. The likely underestimation of actual herbivory has been mentioned in previous studies exploring herbarium specimens^9,10,24, but we now have provided the first quantitative estimate: the combined preferences of the specimen collectors and the herbarium curators decrease the level of background foliar damage of herbarium specimens, on average, to about a half the level observed in nature at the time of sampling.

The overall underestimation of herbivory due to the non-random sampling of herbarium specimens would not create a major problem if the differences in herbivory between samples obtained from the same site by two different methods (botanical and ecological) remained constant. However, this is not the case, because, as shown in our study, the increase in levels of leaf damage by insects in nature was not followed by proportional increase in herbivory measured from herbarium specimens. In other words, collectors dampen the differences in herbivory between more and less damaged plants; consequently, analysis of herbarium specimens does not allow a distinction between situations when plant damage in nature is low and of no concern to the collector (see, for example, Rubus idaeus in Fig. 1) and when the damage in nature is rather high and forces the collector to search for specimens showing the least damage (see, for example, S. aucuparia in Fig. 1).

Moreover, one plant species collected by us (Rubus saxatilis) showed the opposite pattern: the leaves of herbarium specimens demonstrated 2.5-fold higher losses to insects when compared with the ecological samples (Fig. 1). We explain this exceptional pattern by different handling of the runners (the long creeping shoots) of this species. The collectors chose only vertical (fruiting) stems with 3–4 leaves and discarded the long (up to 2 m length) runners, presumably because they did not fit the size of the herbarium sheets. By contrast, the ecological samples included these long runners. The leaves of runners are younger than the leaves of vertical stems, and these young leaves suffered less damage due to their shorter exposure to herbivores. Consequently, exclusion of runners from the herbarium specimens caused an overestimation of the plant-specific leaf damage. This example suggests that collector’s preference may be influenced by plant species traits.

Thus, not only the magnitude but even the direction of the effects of non-random sampling of herbarium specimens on measures of herbivory may change with the plant species (Fig. 1). Consistently, we found that regression analysis, contrary to our second hypothesis, cannot be used to estimate field herbivory from the herbarium specimens, even before the specimens underwent additional selection by the curators for inclusion in the herbaria. The inclusion step then further affects the level of herbivory observed in herbarium specimens by deciding their fate during accession and de-accession procedures.

Our conclusion regarding impossibility to deduce the levels of insect herbivory occurring in nature from herbarium specimens is based on the analysis of the variation in herbivory among plant species, which were collected at the same time. However, a space-for-time substitution is commonly used as an alternative to long-term studies in global change research³². Consequently, our results suggest that in the years with high herbivory the botanists are more selective against damaged plant species than in the years with low herbivory. The effects of this presumed among-year variation in the strength of selection for less damaged plant specimens on long-term herbivory data extracted from herbaria remain to be studied.

An earlier study¹⁰ reported non-significant variation in herbivory between the samples of different collectors. However, the cited study involved plants that had been sampled at different sites and on different dates, and this spatio-temporal variation was likely to prevent detection of the preferences of individual collectors with respect to plant damage. Furthermore, the subsequent accession/de-accession preferences of curators were also not taken into account. By contrast, in both our experiments, the collectors simultaneously sampled the same plant species at the same site and the curators evaluated the same set of plant images. This study design revealed a significant variation in attitudes of both collectors and curators towards plant damage by insects.

Our first experiment involved botany students who were qualified for sampling of plants for herbarium collections, whereas the second experiment involved experienced botanists. In our opinion, it is unlikely that more experienced collectors are less selective in respect of collected plant specimens than the botany students. The differences in herbivory between herbaria and ecological samples would rather increase with the increase in collector’s experience.

The strength of the curator’s impact on the fate of damaged plant specimens would depend on the policies of a particular herbarium and on the individual preferences of the curator, as well as on the type of plant damage by insects. In particular, the presence of leaf miners, even in high numbers, facilitated the preference of a plant specimen by some curators, whereas other curators considered leaf mines to be ordinary damage that diminished the value of the plant specimen. Similarly, branch infestation by scale insects documented from herbarium specimens³³ may also overestimate the infestation occurring in nature, because some curators reported a positive attitude towards additional, identifiable organisms associated with herbarium specimens. A disproportionately large percentage of specimens deposited in individual herbaria have been collected by a very few individuals, termed mega-collectors¹⁹, and most herbaria have only a few curators. Therefore, the combined impacts of mega-collectors and curators, with their associated preferences and idiosyncrasies, may end up shaping the patterns of collection bias in each individual herbarium in a truly unpredictable manner. This variation can be neglected only when analysing very large samples taken at random from a number of different herbaria.

To conclude, cross-validation with field data demonstrates the need for correcting for underestimates (or overestimates) of herbivory that arise from the use of herbarium collections due to variations associated with the levels of herbivory, the plant species traits, and the directions and strengths of preferences for less or more damaged plant specimens by both the collectors and curators. In terms of overall losses of leaf area to insects, herbaria may act as distorting mirrors and further studies are needed before they could serve as reliable sources of information about historical levels of insect herbivory.

Source: Ecology - nature.com