Bias in vegetation databases? A comparison of stratified-random and preferential sampling

Aim Vegetation plots collected since the early 20th century and stored in large vegetation databases are an important source of ecological information. These databases are used for analyses of vegetation diversity and estimation of vegetation parameters, however such analyses can be biased due to preferential sampling of the original data. In contrast, modern vegetation survey increasingly uses stratified-random instead of preferential sampling. To explore how these two sampling schemes affect vegetation analyses, we compare parameters of vegetation diversity based on preferentially sampled plots from a large vegetation database with those based on stratified-random sampling. Location Moravian Karst and Silesia, Czech Republic. Methods We compared two parallel analyses of forest vegetation, one based on preferentially sampled plots taken from a national vegetation database and the other on plots sampled in the field according to a stratified-random design. We repeated this comparison for two different regions in the Czech Republic. We focussed on vegetation properties commonly analysed using data from large vegetation databases, including alpha (within-plot) diversity, cover and participation of different species groups, such as endangered and alien species within plots, total species richness of data sets, beta diversity and ordination patterns. Results The preferentially sampled data sets obtained from the database contained more endangered species and had higher beta diversity, whereas estimates of alpha diversity and representation of alien species were not consistently different between preferentially and stratified-randomly sampled data sets. In ordinations, plots from the preferential samples tended to be more common at margins of plot scatters. Conclusions Vegetation data stored in large databases are influenced by researcher subjectivity in plot positioning, but we demonstrated that not all of their properties necessarily differ from data sets obtained by stratified-random sampling. This indicates the value of vegetation databases for use in biodiversity studies; however, some analyses based on these databases are clearly biased and their results must be interpreted with caution.