How does vegetation sampling in different parts of the growing season influence classification results and analyses of beta diversity?

Abstract

Questions Several studies have demonstrated that data sets that combine vegetation plots recorded in different parts of the growing season might contain significant seasonal variability. The effects of this variability might confound some data analyses, such as vegetation classifications or beta diversity estimates, but the magnitude of these effects is unknown. Here, we try to quantify how strong these effects are, depending on the range of seasonal variation within the data set. Location Southern Moravia, Czech Republic. Methods We used two data sets of permanent plots (Forests and Dry Grasslands from the Czech Republic, each recorded in spring, summer and autumn) to analyse the similarity of partitions in hierarchical classifications with (1) different parameter settings (transformations of cover data and the beta parameter of the Beta flexible clustering method), and (2) different proportions of plots recorded in different parts of the growing season (added non‐hierarchical k‐means classification). Results Single‐season classifications based on the summer records were mostly robust to various cover data transformations and Beta settings, whereas spring and autumn records showed high variability in the resulting partitions. The comparisons of partitions based on the same parameter settings, but using two‐ or three‐season data sets, revealed considerable discrepancies. In the analyses comparing summer records with seasonally heterogeneous data sets, the similarity of partitions rapidly declined after the substitution of a few plots recorded in different parts of the growing season, and non‐hierarchical clustering showed higher partition similarity than hierarchical clustering alone in the Dry Grasslands. Compared to single‐season data sets, we found higher beta diversity when combining spring and summer plots in both Forest and Dry Grassland data sets. Conclusions The sampling date might strongly affect the results of classifications of temperate deciduous forests and dry grasslands. Therefore, for classification, we highly recommend using only summer‐recorded plots. These plots are most robust with respect to various classification settings, correspond approximately to the phenological optimum of these vegetation types and are the most represented in vegetation databases from temperate regions. When the summer‐recorded plots are less represented, we suggest careful seasonal stratification and the inclusion of information concerning the seasonal ratio of analysed data sets into each study.

Emergent wetland vegetation data recording: Does an optimal period exist?

“…Vymazalová et al. (, ) have statistically proven that using a database composed of relevés made in different seasons can distort the results of the classification, so they suggested that the sampling date should be observed as an important factor for resampling large phytocoenological data sets. It would be good for practical applications in the further investigation of emergent wetland vegetations to establish the magnitude of the sampling date effects relative to the impact of other choices commonly made in the analysis of vegetation data, including the size of the vegetation plots, type of cover–abundance transformation and classification method, etc.…”

“…Seasonal variability in species composition and cover is one of the features typical for plant communities (Kennedy & Addison, ; Kirby et al., ). Neglecting seasonal changes may have serious effects on the classification and description of syntaxonomic units (Vymazalová, Tichý, & Axmanová, ). During the development of phytocoenology, potential solutions for eliminating this problem were given that were mainly related to data collection.…”

“…(), Vymazalová et al. (, ) defined the optimal sampling time of vegetation in a different way, resorting to the extensive use of statistical analysis and simulations. Their approach partly coincides with Ellenberg's () proposal because it involves only one sampling during the entire growing season.…”

“…Their approach partly coincides with Ellenberg's () proposal because it involves only one sampling during the entire growing season. So far, optimal sampling time has been proposed only for certain types of terrestrial vegetation, namely the forests and dry grasslands of Central Europe (Vymazalová et al., ).…”

Optimal transformation of species cover for vegetation classification

“…For lower classes of the Braun-Blanquet scale, this conversion provides slightly higher values than most of the published conversion tables (Tüxen & Ellenberg, 1937;Braun-Blanquet, 1964;van der Maarel, 1979van der Maarel, , 2007McNellie et al, 2019), meaning that species with low cover values will have a slightly higher importance in the classification than in the case of other conversions. It is important to note that the differences between various conversions are usually smaller than the error of visual cover estimation (Sykes et al, 1983;Kennedy & Addison, 1987;Lepš & Hadincová, 1992;Klimeš, 2003) or phenological changes in species cover over the growing season (Vymazalová et al, 2014).…”