Methods to describe the botanical composition of vegetation in grassland research

In plant ecology, scales based on intervals into which the assessed values fall are often used, resulting in so-called interval-censored data (Onofri et al., 2019). The seven-grade scale of Braun-Blanquet (1964) is used very frequently, which has a variation range of 20% (class 2) resp. 25% (class 3, 4 and 5). This scale is a so-called abundance/dominance-scale, and is used to estimate the species occurrence taking also into consideration the plant density for low cover values (Voigtländer and Voss, 1979). Some ordinal scales offer unequal intervals, with tighter intervals for low cover/yield proportion and wider intervals for high values (e.g., van der Maarel, 1979; Gauch, 1982; Dietl, 1995). These scales are rather similar to logarithmic scales (Figure 8), which reflect the fact that by means of a visual estimation, it is easier to judge small differences for low values (e.g., between 3 and 5%), whereas it is almost impossible to properly catch the same difference for high values (e.g., between 45 and 47%). In case of parameters with their total value being 100% (i.e., top cover and yield proportion), however, high values can be quite reliably estimated by subtracting first from 100 the less abundant species, which are easier to assess (Traxler, 1997). Logarithmic scales better reflect the human perception than percent scales and avoid claims of excessive accuracy. However, there is some disadvantageous loss of information caused i) by the unknown estimation ability of the single observer (well trained and skilled observers may still catch small differences at high values better than average observers), ii) by the impossibility of an iterative adaptation of the estimate, which in total should account for 100% when adding all estimated values and iii) by the non-additive properties of these scores, which do not allow for the computation of meaningful sums or means. In such cases, the mid-point of the corresponding interval of the score is often imputed (i.e., a percent value of 62.5 is assumed instead of a score corresponding to an interval from 50 to 75%). It is straightforward that the sum of the assumed values of all species cannot reproduce the sum of the real values of the species. They are just a proxy for the real values and, in case of broad intervals, it is unlikely to properly catch differences between experimental units or changes over time being smaller than the interval width. Nevertheless, logarithmic scales are well suited in vegetation ecology for performing ordination procedures and multivariate statistics (Gusmeroli, 2012). There are additional scales like that of Pfadenhauer et al. (1986) using 8 categories, Londo (1976) with 12 categories, the Londoscale modified by Zacharias (1996) using 20 categories or those of Schmidt (1974, cited in Pfadenhauer et al., 1986), Bornkamm and Hennig (1982), Wilmanns (1989) and Dierschke (1994). Censored data can be analyzed by means of the body of techniques known as “survival analysis” (Onofri et al., 2019).

Figure 8

In most cases, the most suitable time for vegetation assessments on meadows is right before performing the first cut, as most species can be more easily found at this time. At the same time, the vegetation stage of the dominating species should be evaluated as well. As the quantitative parameters of the botanical composition change along the phenological development of species, this information allows for an evaluation of the comparability of data obtained from the same plot in different growing seasons.

In extensive, late-cut grassland, the species determination is made easier by the advanced phenological stage of most plants, allowing to use diagnostic characters of generative organs (e.g., flowers or fruits). In intensively managed grassland, instead, the determination must be mainly based on vegetative organs. If the aim of the survey is an exhaustive description of the botanical diversity, the obtained species list should ideally be controlled and completed during the following regrowths, as some species might easily be missed in the first growth (e.g., fall dandelion) or are developing later. For the regrowths, it is advisable to estimate at least the weight-% of the species groups of grasses, forbs and legumes to document changes between the different growths.

The aspired level of botanical detail depends on the aim of the survey. Whilst for ecological studies, aiming at characterizing the botanical diversity, a complete list of species (and subspecies) is a quite straightforward requirement, other less time-consuming target levels of botanical detail may be reasonable for other aims.

The dry-weight-rank method (Mannetje and Haydock, 1963; Jones and Hargreaves, 1979; Tothill et al., 1992) averages the results of a certain number of assessments in small quadrats. In each quadrat, only the rank of the three most abundant species in terms of yield proportion is recorded and standard values are assigned to the ranks. For instance, this procedure was found to be acceptably accurate for a study aiming at the quantification of community-weighted means of functional traits (Lavorel et al., 2008).

For agronomic studies, the assessment at the species group level of grasses (including also graminoids such as sedges and rushes), legumes and forbs allows classifying the sward into types according, for example, to the Swiss system (Daccord et al., 2007): rich in grasses (> 70% grasses), balanced (between 50% and 70% grasses), rich in forbs (> 50% forbs, legumes < 50%) and rich in legumes (legumes > 50%). This kind of classification can be quickly learnt and applied by practitioners with reasonable precision even after a short training (Peratoner et al., 2018) and allows in turn obtaining some hints (Table 2) about yield potential (Troxler and Thomet, 1988), forage quality and its stability along the advance of the phenological stage at the first cut of meadows (Nußbaum et al., 1999; Daccord et al., 2007; Peratoner et al., 2016a; 2016b), suitability for silage conservation (Weißbach et al., 1977; Frame and Laidlaw, 2004), drying speed and risk of crumbling losses at hay making (Höhn, 1988; Frame and Laidlaw, 2004) and risk of contamination with soil (Resch et al., 2014).

Especially for investigations with a focus on plant sociology and species diversity, the proper selection of a representative survey area is an important precondition. In managed grassland, such areas should be as homogenous as possible. Bohner and Sobotik (2000) estimated the ideal area size to be about 100 m² to catch all diagnostic relevant species, of which some are growing sparsely. A similar size (50–100 m²) is suggested by Müller-Dombois and Ellenberg (1974) for dry grassland, but 10–25 m²for productive meadows and 5–10 m² for fertilized pastures may suffice according to the same authors. The size of the optimal survey area can also be determined by compiling a chart showing the relationship between species number and the size of the investigated area. Within a frame of a certain size (e.g., 0.5 × 0.5 m), the number of species is recorded and then this frame is turned over as often, as there is no more increase of the cumulated number of species. The species number is then plotted against the area size to infer the minimum size required. The point of inflection of the curve, representing the size at which the increase in species number declines, can be considered a good compromise between the required effort and detection of most species (Whalley and Hardy, 2000), whilst the area size at which no further species increase is attained represents the minimum area necessary to retrieve all species.

As in field experiments the size of the survey area is limited by the plot size, only the central, homogenous part should be used to avoid boundary effects. Considering this requirement, and given a certain plot area, a circular form of the plots allows to reduce the boundary effects to a minimum, but it bears the disadvantage of being less easily combinable with other plots within a field design. Furthermore, the boundaries of a circular form are more difficult to delimit on the field and to manage (i.e., fertilization or harvest) than quadrats or rectangles. Quadrats have a smaller boundary proportion than any other rectangle, whilst a rectangular form presents an increasing boundary proportion with increasing length of the longer side. However, in case of assessments to be performed without entering the plots, in order to avoid trampling the vegetation, rectangles allow a better visual and/or physical access to the plot area (Traxler, 1997). Transects are an extension of the rectangular form and are appropriate whenever changes of the vegetation are to be detected or described along a predefined path encompassing environmental gradients (Whalley and Hardy, 2000). Some methods to assess cover and frequency (i.e., point quadrats, linear analysis) are performed along transects, which can be also positioned within plots.

Both the number of recordings and the size of the survey areas should be adapted to the site homogeneity and to the used investigation method. Heterogeneous grasslands, which often occur in large grazed plots, poses particular challenges to describe the vegetation. In such cases, as it is not possible to assess the total area, it is advisable to repeat the assessments on a sufficient number of small subareas (pseudo-replicates) adequately reflecting the reality on average. Random location is appropriated if the heterogeneity is not extreme and the method chosen allows for a large number of observations without excessive increase of the time effort. In this case, however, caution is required to avoid unconscious bias in the choice of the subareas (Whalley and Hardy, 2000). Fixed grid sampling (and, in general, planning in advance the position within the plots of the areas to be sampled) overcomes this problem. It also allows better detection of temporal changes by means of repeated measurements, if the assessments are performed over time at the same position. Stratified random sampling is appropriated in case of clear patterns: in this case, the number of subareas for each occurring sward type is determined proportionally to its total area within the plot and then the single sampling areas are assigned randomly within the respective type (Whalley and Hardy, 2000).

The number of replicates necessary to detect significant differences between treatments is subject to statistics theory and can be estimated given the aspired width of the confidence interval and precision (smallest relevant difference between expected and observed mean) (Traxler, 1997).

Methods to describe grassland vegetation

Different methods have been developed and used in the previous century all over the world to describe grassland vegetation (Hanson, 1934; Braun-Blanquet, 1951; Johnston, 1957; Schechtner, 1958). Numerous authors have given attention to reliability, replicability and comparability of different methods (Goodall, 1952; Tüxen, 1972; Greig-Smith, 1983; Everson et al., 1990; Lepš and Hadincová, 1992; Grant, 1993; Traxler, 1997). The most frequently used methods are described in the next paragraphs, with the description mainly focusing on the operative issues. This allows clustering the methods, for the sake of clarity, in just few categories.

The main advantage of this method is the independency from technical devices and the low time expenditure. Voigtländer and Voss (1979) as well as Dethier et al. (1993) assert that, under favorable conditions, visual estimation methods are competitive with objective methods in terms of accuracy. However, accuracy of estimate depends on several factors (Traxler, 1997):

Furthermore, the subjective condition, training level, experience and routine of the observer play an important role (Peratoner et al., 2018). Observations at the end of a long working day are for instance less precise than those at the beginning of a day. Moreover, previous knowledge of the botanical composition of the surveyed plot or of the plant community under survey allows the observer a targeted search for species that are expected to be found. In the ideal case, the investigated areas should always be surveyed by the same person (Vittoz and Guisan, 2007). Even though some systematic, subjective error of estimate occurs, sufficient and reliable comparisons between different treatments can be obtained by means of this procedure. Visual estimations are mainly suitable for determining cover and yield proportion.

Following the method by Klapp/Stählin (Klapp, 1930; cited in Voigtländer and Voss, 1979), the yield proportion of particular species related to the entire above ground biomass (= 100%) is estimated and this can be done quickly by experienced observers. Usually, a list of all occurring species is compiled and then the yield proportion of species groups (grasses, including sedges and rushes, herbs and legumes) are estimated. The proportion of these groups is then allotted between the single species, typically starting with poorly represented species, which are easier to estimate. According to Traxler (1997), visual estimation of cover is the most used method in vegetation ecology for reasons of quick realization and sufficient accuracy. At AREC Raumberg-Gumpenstein, a modified version of the Braun-Blanquet method (1951) was established by Schechtner (1958). In contrast to five-part to nine-part ordinal scales, which are often used in plant sociology and botany, a percentage-scale is used to estimate the cover of single species (CS). This method needs a lot of experience, routine and is quite time consuming. In practice, initially, all occurring species at the observed site are recorded and allotted to the groups of tall grasses, medium grasses, bottom grasses, graminoids (sedges and rushes), legumes and forbs. Then, the cover of all single species is estimated as area-%, whereas the following symbols are used to determine very low cover values:

++ = rare (= 0.66 area-%), + = very rare (= 0.33 area-%) and r for species with one or two individuals occurring in the survey area

Following the determination of all single species, the total area proportion within the plant groups as well as the overall vegetation cover (CV) is calculated (= sum of group cover resp. sum of species cover). In tall and dense grassland stands, the total cover can amount to 130–140 area%, whereas in short and sparse grassland stands, this value can even fall under 100%. Validation of the results can be done in different ways, for example, as a comparison with other observation areas to decide whether the total cover is too high or too low. In addition, the cover of single species or species groups can be compared.

In field experiments, treatments are usually replicated several times, mostly providing a different total cover of vegetation (CV). This must be considered, when the mean cover of a plant species is calculated. For this reason, the cover proportion of species can be standardized (CSZ) as follows:

CSZ = standardized cover proportion of a species

CS = cover proportion of a species (area-%)

CV = total cover of vegetation (area-%)

The example in Table 3 shows a considerable difference of approximately 4 area-% between the uncorrected (CS) and the standardized (CSZ) mean value.

The use of wooden or metal frames provides a precise definition of the area to be surveyed. However, one critical point is the decision whether plants that are partly outside the frame have to be included in the observation or not. Usually, rooting within the frame is the relevant criterion for this decision. Nevertheless, the decision remains partly subjective for plants growing at the frame’s margins. Moreover, the size of the frame is relevant in this respect: the smaller the frame, the higher is the proportion of plants at the frame’s margins and the higher is the risk of wrong decisions (Müller-Dombois and Ellenberg, 1974). In grassland science frames with a size of 0.5 × 0.5 m or 1.0 × 1.0 m are frequently used (Whalley and Hardy, 2000). Such frames can be used for the estimation of cover, plant density or frequency.

For frequency measurements, frames are typically subdivided into 100 quadrats and in each of them the occurrence of every single species is recorded. The number of quadrats, in which a single species occurs, is representing its frequency. Circular frames with a size of 0.1 m²can be used for a random selection of observation spots by throwing the frames repeatedly followed by recording the species occurring within the frame.

This method is known as the point intercept method, point quadrats or point method (Levy and Madden, 1933; Goodall, 1952; Goodall, 1953). With this method, the occurrence of a species is not recorded within a certain area, but at a defined point. Usually, wire pins are vertically lowered downwards at the intersection points of a grid frame or through regularly spaced holes (guide channels) in a frame (Figure 9). The contact point between the pin and the vegetation identifies the species to be recorded. If leaf area index (LAI) is the targeted parameter, an angle of 32.5° of the pins to the horizontal minimizes the error due to the underestimation of the foliage area of species with erect habit of foliage (Wilson, 1960).

Figure 9

In France and Italy, a modified point intercept method is frequently used for surveys on pastures. For this so-called linear analysis, according to Daget and Poissonet (1971), a metal bar or a bayonet is stuck into the soil along a measuring tape at regular distances (Ostermann, 1991). Again, the contact between the bar (or bayonet edge) and a plant defines the species that is recorded.

The point intercept method can be used for assessing the top cover if only the first contact is recorded, but also for frequency measurements, provided that all contacts per observation point are recorded. When all contacts per point, including the repeated contacts of the same species at the same point are recorded, specific proportion can be derived and seen as a relative expression of aboveground biomass (Goodall, 1952; Ostermann, 1991). This type of recordings can also be done with the help of telescopic tubes, in which the function of wires is replaced by cross-hairs (Traxler, 1997). The point intercept method offers better objectivity and accuracy for investigations of vegetation dynamics than visual estimate (Stampfli, 1991; Vittoz and Guisan, 2007). However, there is some disadvantage due to the high time effort and in case of windy conditions and tall vegetation, which can impede the measurements or even make them impossible. There is also the need of a high number of replicates to catch rare species (Traxler, 1997). As an acceptable compromise, species that are not intercepted but occur in the surveyed area can be added to the species list with a frequency of less than 1% (Peratoner, 2003; Pittarello et al., 2018).

Theoretically, an exact determination both on species level and species group level, can also be carried out by extremely time-consuming manual separation of the harvested yield. Therefore, manual separation in grassland is mostly done with defined subsamples for the species groups of grasses, herbs and legumes to search for their functional traits and their impact on specific parameters like crude protein or carbohydrate fractions (Pötsch and Resch, 2007; Weichselbaum, 2015; Gabauer, 2018).

These methods, including both proximal and remote sensing, imply the availability of instruments to capture, process and record information related to the vegetation under survey. Their main advantages are the ability to collect large amount of data in short time, thus reducing labor and time effort and, in case of remote sensing, the possibility to collect data also from remote or hardly accessible locations, potentially covering every location in time and space (Wachendorf et al., 2018). Great advance has been achieved in the last two decades in estimating biomass and forage quality mainly, but they have proved to be suitable also in providing valuable information about other parameters related to the botanical composition of the sward, as long as i) there is only one target species, or ii) the botanical composition is simple, or iii) an information at a community level fulfils the aim of the survey. For instance, photography and digital image analysis (DIA) have been shown to provide reliable information about cover and yield proportion of legumes in binary clover-grass mixtures (Himstedt et al., 2010) and to allow the automatic detection of a broad-leaved weed (Rumex obtusifolius) in grassland (Gebhardt et al., 2006; Gebhardt and Kühbauch, 2007). Strong emphasis is currently placed on spectrometry, a non-invasive method that allows analyzing biological systems in terms of components, structures and molecular functions. Vegetation and its growth dynamics can be described at a community level by means of various indices derived from field-spectrometric measurements and partly related also to vegetation traits (i.e., vegetation mapping, species richness) at a community level (Govender et al., 2007; Psomas et al., 2011; Hollberg and Schellberg, 2017). Spectrometric techniques can also be used for detecting weeds (Glenn et al., 2005) or, as expected in future, for estimating the optimal time of harvesting in combination with satellite data (Schaumberger and Schellberg, 2015). However, compared to the manual and visual methods, spectrometry requires both costly equipment and advanced knowledge in data processing as well.

It must not be forgotten that data describing the botanical composition of the vegetation can be used to derive further parameters useful to summarize and synthesize the information gained by the assessments. They allow to quantify diversity and compute community-weighted means to characterize the vegetation from an ecological and agronomical point of view.