COMPARISON OF DATA FROM TWO

Environmental Monitoring and Assessment (2005) 100: 235–248

© Springer 2005

COMPARISON OF DATA FROM TWO VEGETATION MONITORING METHODS IN SEMI-NATURAL GRASSLANDS A. LISA M. CARLSSON1 , JENNY BERGFUR1,2 and PER MILBERG1∗ 1 Department of Biology-IFM, Linköping University, Linköping, Sweden; 2 Department of Environmental Assessment, SLU, Uppsala, Sweden (∗ author for correspondence, e-mail: [email protected])

(Received 18 March 2003; accepted 15 December 2003)

Abstract. Two vegetation-monitoring methods were compared: subplot frequency analysis (SF) and visual estimation of percentage cover (VE). Two independent observers collected data from two semi-natural, species-rich grasslands on three different occasions during the growth-season. During the last data collection period, survey times were also recorded. The two different data sets from the two methods were compared using partial Redundancy Analyses. The purpose of the comparison was to identify the method that explains most of the relevant variation in biodiversity-monitoring (interand intra-site variation), and the variation irrelevant when evaluating data (systematic inter-observer variation and variation due to phenological changes). Compared with VE data, more variation in SF data could be explained by spatial variables, while less variation depended on the observer and time of year surveyed. SF also found more species per plot but took on average five times longer to complete than VE. In conclusion, the different methods are suitable for different purposes: SF is more suitable for purposes demanding high accuracy and high precision, such as long-term biodiversitymonitoring when the identification of small changes has high priority, while VE might be more suitable for a one-time mapping of a large area. Keywords: accuracy, frequency analysis, ordination, permanent plot, precision, Sweden, vegetation monitoring, visual estimation

1. Introduction Data on semi-natural grassland vegetation are collected for the purpose of research, management planning and biodiversity monitoring. However, time series data from semi-natural grasslands are often considered unreliable (Stampfli, 1991). In part, this uncertainty is the results of data series coming from different observers and being collected during different periods of the year. Another factor potentially influencing the usefulness of time series data is the choice of method, which only rarely has been optimized for the specific purpose of the survey. For the above reasons, survey data can contain an unknown amount of variation that can hide true changes, but also have a systematic bias that can lead to inappropriate conclusions. When evaluating the suitability of a survey method, it is useful to think in terms of precision, i.e. the repeatability of the method, and accuracy, i.e. how well the method describes ‘reality’. When evaluating time series data, accuracy is much less important than precision (Gotfryd and Hansell, 1985). Another issue is what aspect

236

A. L. M. CARLSSON ET AL.

of reality we are trying to describe? Some have considered plant biomass to be a ‘golden standard’ (i.e. the best method currently available to describe an attribute). However, as biomass determination requires destructive sampling, it is not useful when monitoring permanent plots (Bråkenhielm and Qinghong, 1995). Instead, the various types of methods used to describe species abundance in vegetation can be grouped into one of two categories: (i) visual estimation of cover (most often as percentage of total area) and (ii) presence/absence, i.e. recording the frequency of species at various points or smaller sampling units (Kent and Coker, 1992). Most comparisons of survey methods have used some kind of ‘golden standard’ in their comparisons (Kirby et al., 1986; Floyd and Anderson, 1987; Stampfli, 1991; Leps and Hadincová, 1992; Bråkenhielm and Qinghong, 1995). Bråkenhielm and Qinghong (1995) used photographs as a ‘golden standard’; however, photographs tend to show the largest and the most visual plants of the vegetation cover, and hide the smaller or more low-grown ones. Some studies have used summarized data from different investigators as the ‘golden standard’ (Kirby et al., 1986; Floyd and Anderson, 1987), assuming that this is a better approximation of ‘reality’. In the present study, we focused on precision rather than accuracy when comparing the two methods, and we concentrated on how the variation in data can be attributed to different sources. These sources were either considered as relevant for monitoring or as irrelevant. In our study, relevant factors were defined as those related to spatial patterns in nature, i.e. differences between plots and sites, while irrelevant sources of variation were phenological changes during a growth season and consistent inter-observer differences. Our aim was to compare the usefulness of a subplot frequency analysis approach (SF) with that of visual estimates (VE) for monitoring semi-natural vegetation in permanent plots. More precisely, our aim was to: • compare the methods by identifying the variation in the different data sets caused by systematic differences between investigators, periods of investigation, plots and sites; • highlight the species that are hard to identify/estimate cover of/find in the field, and compare the different methods in this regard; • identify the method that detects most taxa; • estimate the time needed for fieldwork. 2. Materials and Methods 2.1. S TUDY SITES The two pastures studied are species-rich, semi-natural grasslands situated in the county of Östergötland, southern Sweden. They are both parts of larger nature reserves, which mainly consist of forest. Åsabackarna pasture (58◦ 17 N, 14◦ 55 E) is a calcareous grassland, partly wooded with Juniperus communis, Picea abies, Pinus sylvestris and Betula spp.

COMPARISON OF DATA FROM GRASSLAND VEGETATION MONITORING METHODS

237

The area has a complex geomorphology with ridges, hills and hollows made of both till and glaciofluvial material. The area is a pasture with xeric vegetation, and has been grazed by cattle for at least 200 yr. Solberga pastures (58◦ 21 N, 15◦ 11 E and 58◦ 21 N, 15◦ 12 E) are situated on glaciofluvial material. The river Svartån intersects the reserve, and on both sides of the river are ridges and hills covered with vegetation favoured by nutrient-poor soil. The pastures studied are partly wooded with Quercus robur. This grassland was historically used as a meadow, but nowadays cattle graze the area.

3. Data collection The fieldwork was conducted in 2002 and there were three inventory-periods: end of June, end of July and end of August. Together these periods represent the time span during which most vegetation monitoring work is conducted in southern Sweden. Eight (Åsabackarna) or seven (Solberga) plots were permanently marked according to the procedures used by the County Administrative Board, i.e. a 20 cm long underground iron bar (a metal detector was later used to locate the bar during field work). The plots were placed to represent the different species-rich parts of the pastures, a placement strategy that mimics that followed by the County Administrative Board in their monitoring work. A squared, 0.5 m2 wooden frame was used (0.7 m ∗ 0.7 m). With the help of a compass, its southern corner was placed exactly over the bar with the diagonal corner pointing north. The abundance of each taxon was estimated by two sampling methods, both described in Goldsmith and Harrison (1976): visual estimation of percentage cover (VE) and subplot-frequency analysis (SF). Nomenclature follows Karlsson (1998). The identification of the plants was, whenever possible, to species-level, except grasses and sedges which were noted as Poaceae, Luzula spp. and Carex spp. When estimating percentage cover (VE), the species in a plot were recorded at their exact percentage. As Tonteri (1990) argues, this way of estimating plant cover is more suitable than estimating cover in constant classes, minimizing the bias caused by discreteness of scale and the different sizes of coverage classes. Using subplot-frequency analysis (SF), 25 subplots within each main plot were used. For each taxon, presence/absence of rooted individual(s)/shoot(s) was noted for each subplot, giving a value of between 0 and 25 per (main) plot. In order to compare the two methods, two persons made the inventory independently. There were no discussions about species identifications, plant morphology or distribution during fieldwork. The two investigators had a similar educational background (a Swedish BSc in biology) and previous field experience in identifying grassland species. During the three inventory periods, the two observers used both methods. This means that every plot was examined 12 times.

238

A. L. M. CARLSSON ET AL.

To minimize the impact of a person’s familiarity with a plot, the two different examinations (SF and VE) were done with as much separation in time as possible. Inpractice, this meant that first all the plots were examined using SF, and thereafter by VE. The plots were also examined following a different order each time, and protocols from previous examination(s) were not available during fieldwork. In spite of these efforts to minimize the impact of familiarity, this might still have affected the data. The difference between the methods, persons, plots and surveying times may therefore be somewhat underestimated. At the third period of surveying, time spent per plot using the two different methods was recorded.

3.1. DATA ANALYSES 3.1.1. Explained and Unexplained Variation in the Data The two sets of data, originating from SF and VE, were analysed separately following the standard protocol (Table I). To partition explainable variation in the data sets, partial Redundancy Analyses (pRDA) were conducted with different combinations of independent variables and covariables. The software CANOCO 4.5 (ter Braak and Smilauer, 2002) and its default options, including centering by species, were used (Leps and Smilauer, 2003). To test for a statistical significance of the effects of the environmental variables for each pRDA, appropriate Monte-Carlo tests were done, using 9999 permutations.

3.1.2. Species Difficult to Survey Using the ordination scores from the first axis in the analysis evaluating systematic inter-observer differences (A in Table I), species deviating more than one standard deviation from the average ordination score among species were considered as ‘species difficult to survey’. Hence, these were the species for which systematic inter-observer differences were largest. To evaluate the performance of ‘species difficult to survey’ in the two methods, species-wise ordination scores from the VE data set were divided by those from the SF data set. Hence, this highlights to what extent each method identified the same taxa as ‘difficult to survey’.

3.1.3. Species Detection and Time Consumption for Fieldwork All the taxa in each plot were summarized, leading to 90 sums per method (15 plots ∗ 2 persons ∗ 3 periods). We used a paired t-test to evaluate if the two methods differed in their ability to detect taxa. This test was also used to compare the time taken for completing each method.

A B C D ABCD

Observer Period Site Plot Total explained variance

Investigated aspect

1, 2 1st, 2nd, 3rd Solberga, Åsabackarna Plot-ID (1–15) All variables above

Environmental variables

2 3 2 15 22

N

BCD ACD AB ABC –

Covariables

1.6 1.5 16.8 66.1 86.1

0.0001 0.0001 0.0001 0.0001 0.0001

8.06 3.96 17.87 26.28 26.15

Subplot frequency Explained p F variance (%)

2.4 2.4 9.0 62.7 76.5

0.0001 0.0004 0.0001 0.0001 0.0001

7.49 3.74 8.86 14.79 13.81

Visual estimation Explained p F variance (%)

TABLE I Outline of and results from the pRDAs used to partition the variation in vegetation data from two semi-natural grasslands in southern Sweden, collected by two independent observers with two different methods

COMPARISON OF DATA FROM GRASSLAND VEGETATION MONITORING METHODS

239

240

A. L. M. CARLSSON ET AL.

4. Results 4.1. E XPLAINED AND UNEXPLAINED VARIATION IN THE DATA The variation in the data that could be attributed to observers was low, less than three percent, irrespective of method (Table I). SF data contained less variation explainable by systematic inter-observer differences than VE. The variation in the data sets due to the time of year of the survey period was also relatively small, 1.5 and 2.4% for SF and VE, respectively (Table I). With both methods, the variation in the data depending on sites and plots were the largest (Table I). Comparing methods, SF gave a larger variation explainable by investigated site and plot than did VE. An important aspect of the pRDA results is to what extent the variation is explained by relevant variables. As seen in Table I, the relationship between variation explained by relevant factors (site and plot), and the variation explained by irrelevant factors (person and period), differs between the methods. Looking at all the factors investigated, SF identified a larger proportion of the variation, leaving only 13.9% as unexplained, while VE identified a smaller part of the variation, leaving 23.5% as unexplained (Table I). 4.2. S PECIES DIFFICULT TO SURVEY The species hard to survey, i.e. with large inter-observer differences, were to a large extent the same in both methods (Table II). The genera Ranunculus, Trifolium, Vicia, Taraxacum, Potentilla and Viola were all identified as problematic taxa, irrespective of method (all deviating > 1.5 SD from mean ordination score; Table II). This is also demonstrated in Table III, where many species in these genera have an ordination score ratio near 1, indicating a survey-identification problem not associated with either particular method. However, the survey difficulties of species having ratios far above or far below 1 are method-related. For example, Viola canina gave large variation with VE, but not with SF (Table III). Though small, the systematic inter-observer differences were highly significant (Table 1), and one observer consistently estimated higher plant covers and found higher frequencies of many species. This bias seemed slightly more pronounced with VE (Figure 1). 4.3. S PECIES DETECTION AND TIME CONSUMPTION FOR FIELDWORK Paired t-tests showed that SF and VE differed significantly in the number of species detected (p < 0.000001, df = 89, t = 12.937); average number of species detected was 23.4 and 20.3, respectively. However, the observer needed less time surveying a permanent plot using VE than SF (paired t-test, p < 0.000001, df = 29, t = 11.534); average time used was 10 and 56 min, respectively.

COMPARISON OF DATA FROM GRASSLAND VEGETATION MONITORING METHODS

241

TABLE II Species considered as ‘difficult to survey’, i.e. those with the largest and smallest ordination scores when analysing for systematic inter-observers differences (analyses A in Table I). Only taxa deviating > 1 SD from the average ordination scores are listed. ‘Expl. var.’ is the amount of variation in species abundance explained by observer Subplot frequency

Expl. var.%

Visual estimation

Expl. var.%

Ranunculus auricomus Trifolium medium Taraxacum sp. Potentilla tabernaemontani Taraxacum sect Erythrosperma Trifolium pratense Vicia cracca Daucus carota Viola riviniana Viola hirta Taraxacum sect Hamata Carex sp. Artemisia vulgaris Pimpinella saxifraga Artemisia campestris Ajuga pyramidalis Potentilla reptans Equisetum pratense Vicia tetrasperma Ranunculus polyanthemos Platanthera sp. Platanthera chlorantha Galium album Veronica verna Betula pubescens Betula sp. Erophila verna Juniperus communis Leontodon autumnalis Tragopogon pratensis Veronica arvensis Fragaria viridis Cerastium semidecandrum Lychnis viscaria Veronica serpyllifolia Centaurea jacea

21.97 10.60 9.69 8.91 8.75 6.21 4.79 4.65 4.65 4.55 4.48 4.25 3.86 3.86 3.55 3.54 3.31 3.31 3.17 3.05 2.81 2.81 2.66 2.54 2.27 2.27 2.27 2.27 2.27 2.27 2.25 2.14 2.10 2.04 1.85 1.85

Ranunculus auricomus Trifolium medium Trifolium pratense Poaceae Vicia tetrasperma Luzula sp. Taraxacum sect Erythrosperma Veronica arvensis Potentilla tabernaemontani Viola riviniana Viola canina Lathyrus linifolius Artemisia vulgaris Leucanthemum vulgare Vicia cracca Ajuga pyramidalis Juniperus communis Platanthera chlorantha Artemisia campestris Platanthera sp. Potentilla reptans Equisetum pratense Veronica serpyllifolia Lychnis viscaria Fragaria viridis Veronica verna Rumex acetosa Alchemilla sp. Aegopodium podagraria Hieracium pilosella Centaurea jacea

28.57 11.31 9.77 8.73 6.17 6.13 4.78 4.65 4.59 4.48 4.48 3.94 3.74 3.69 3.64 3.54 3.45 3.18 3.04 2.94 2.87 2.59 2.27 2.27 2.14 2.04 1.92 1.92 1.85 1.83 1.67

242

A. L. M. CARLSSON ET AL.

Figure 1. Species ordination scores from the pRDAs contrasting persons (analyses A in Table I) versus abundances. The species far to the right and far to the left explain a large part of the inter-person variation. Species with low abundance are not shown; the abundance of Poaceae in the lower figure was drastically higher than the other abundances.

COMPARISON OF DATA FROM GRASSLAND VEGETATION MONITORING METHODS

243

TABLE III The taxa with large inter-observer differences between the two survey methods. The numbers are species-wise ordination scores from the VE data set divided by those from the SF data set (i.e. numbers in Table II). Species only found with one method are excluded Species

VE/SF

Species

VE/SF

Viola canina Aegopodium podagraria Leucanthemum vulgare Rumex acetosa Poaceae Alchemilla spp. Lathyrus linifolius Luzula sp. Hieracium pilosella Veronica arvensis Vicia tetrasperma Trifolium pratense Juniperus communis Ranunculus auricomus Veronica serpyllifolia Platanthera chlorantha Lychnis viscaria Trifolium medium Platanthera sp. Ajuga pyramidalis Fragaria viridis

74.7 18.5 17.6 16.0 7.79 5.65 4.69 4.35 4.16 2.07 1.95 1.57 1.52 1.30 1.23 1.13 1.11 1.07 1.05 1.00 1.00

Artemisia vulgaris Viola riviniana Carex sp. Centaurea jacea Potentilla reptans Artemisia campestris Veronica verna Equisetum pratense Vicia cracca Taraxacum sect Erythrosperma Cerastium semidecandrum Potentilla tabernaemontani Betula pubescens Betula sp. Erophila verna Leontodon autumnalis Tragopogon pratensis Pimpinella saxifraga Ranunculus polyanthemos Taraxacum sect Hamata Viola hirta

0.969 0.963 0.932 0.903 0.867 0.856 0.803 0.782 0.760 0.546 0.533 0.515 0.493 0.493 0.493 0.493 0.493 0.425 0.367 0.250 0.218

5. Discussion 5.1. E XPLAINED AND UNEXPLAINED VARIATION IN THE DATA Systematic inter-observer differences explained only small amounts of variation, in both methods (Table I). This would indicate that differences between observers, at least with similar levels of experience and education, might not affect the data substantially, irrespective of method. Subplot frequency analysis (SF) generated even smaller variation explainable by inter-observer differences than visual estimation (VE). This can be related to many factors, but as discussed in Bråkenhielm and Qinghong (1995), fatigue and other human factors have a larger influence using VE than when using presence/absence. SF involves a more systematic search through

244

A. L. M. CARLSSON ET AL.

the plot, with a more homogeneous effort per subplot, leading to fewer opportunities to ‘speed up’ the work. It is also important to note that the recorded variation of abundant species in SF data is likely to decrease as frequency approaches its maximum. In contrast, inter-observer differences in cover estimates of abundant species are likely to increase with their abundance. For example, in the current study, the ubiquitous Poaceae contributed substantially to the VE solutions but hardly to the SF ones. Hence, the current analyses of VE data are likely to be substantially influence by a relatively small number of abundant species, while the SF data analyses would be affected by a larger number of species. Almost all previous attempts to evaluate inter-observer differences have done so using univariate response variable(s), e.g. number of species recorded, density of individual species, etc (Hope-Simpson, 1940; Smith, 1944; Sykes et al., 1983; Kirby et al., 1986; Kennedy and Addison, 1987; Tonteri, 1990). Such studies have often noted how poor vegetation data are and the need for using more than one observer, or setting high threshold for accepting a temporal change as real (e.g. Kennedy and Addison, 1987; Tonteri, 1990). Therefore, it is striking how relatively small the systematic inter-observer differences were in the current study. It is possible that the two observers in our study were unrepresentative by being very similar in experience and perception. However, as shown in Figure 1, one observer consistently recorded larger numbers, and there were also disagreements on the identity of some taxa (e.g. Trifolium pratense and T. medium), which should further inflate inter-observer differences. Therefore, it seems that multivariate analyses on vegetation data might be more robust than univariate ones comparing single species, species numbers, etc (see also Leps and Hadincova, 1992). The variation in the data sets due to the time of year of the survey periods was also relatively low (Table I). The reason might be that though June, July and August are all summer-months and the vegetation is growing substantially, the changes in species composition are relatively small. Again, SF gave a smaller variation than VE (Table I). This can be explained by the fact that as the vegetation grows, the percentage cover changes, but often the frequency of the plants remain the same. In other words, any vegetative expansion into new subplots, or seedling recruitment, might be a phenomenon of relatively modest importance, at least on the time scale involved in the present study. Not surprisingly, the variation in the data that depended on site and plot were the largest with both methods (Table I). It is also reassuring that a permanent plot approach seems worthwhile when compared with the obvious alternative of using random plots. The latter alternative would be much easier, but would also mean large amounts of random variation, something likely to obscure any subtle changes with time – changes that would be of interest in monitoring programs. A larger proportion of the total variation was unexplained with VE compared to that with SF (Table I). The less unexplained variation there is – the better the method, since known sources of variation are more easily dealt with. For example, even if the systematic inter-observer differences would be unknown in a specific

COMPARISON OF DATA FROM GRASSLAND VEGETATION MONITORING METHODS

245

study, knowing the general magnitude of variation normally attributed to this source will assist when evaluating the data. Finally, it is important to consider the potential sources for the remaining unexplained variation. In the analyses using observer, time period, site and plot (ABCD in Table I), residual variation should be attributed to interactions between these terms, to variation due to small differences in the exact placement of the plot and, probably most important, to non-systematic differences between and within observers. 5.2. S PECIES DIFFICULT TO SURVEY Most of the genera for which systematic inter-observer differences explained large parts of the variation (e.g. Ranunculus and Trifolium, Table II) are known to be problematic from earlier studies (Hope-Simpson, 1940; Kirby et al., 1986). Therefore, special attention to these species is vital when evaluating data; it is likely that the precision in the data will increase if observers doing field surveys on seminatural grasslands would get special training regarding these genera. Although Smith (1944) found that there is little profit in giving a general education to the observers, a specific, species-targeted training is likely to reduce the variation between observers. We distinguished three factors as contributing to the variation between observers: (1) incorrect identification, (2) differing cover estimations and (3) problems of finding the plants. (1) Incorrect identification of species is known to give survey data undue variation (e.g. Kennedy and Addison, 1987; Scott and Hallam, 2002; Brandon et al., 2003). This problem may in some cases be systematic according to observer: one observer consistently identifies plants as belonging to one species, while another observer considers the same plant as being another species. This problem should not depend on the size or spatial distribution of the plants in the plots, only on how much species resembles each other and on the skill of the observers. In the present study, the most conspicuous cases were Trifolium pratense vs. T. medium and Ranunculus spp. (Figure 1, Table II). Incorrect identifications may occur irrespective of method, and is likely to contribute substantially to variation in the data as species can vary from high abundance to zero and vice versa. Unfortunately, if it is not controlled for, it is likely that this ‘random’ variation will be attributed to a temporal change during analysis. (2) Estimating cover. The problem of estimating cover for particular taxa is only applicable to VE. In the present study, this problem concerned those species that were either small, had a high abundance or were winding plants (personal observation). The species with high abundance are likely to contribute most to the variation simply by their large cover estimates, allowing for numerically larger discrepancies between observers. (3) Finding the plant. The problem of finding the plant mainly concerns small species, especially if they are regularly distributed over the plot. This problem

246

A. L. M. CARLSSON ET AL.

should have a larger impact using the SF method. In every subplot, one may or may not find the species, and the ‘random’ variation this brings can be substantial when using the SF method. Using the VE method, the mistakes concerning these types of ‘hidden’ species are presumably not fewer, but they do not give the same variation in the data, as the cover value may differ between 0 and 5%, hardly more. By extension, this may lead to substantial, systematic differences between observers. It is difficult to know which of these three factors had the largest impact in our analyses. However, when looking at the ratio of the explained variation from the two methods (Table III), it is possible to discern a few patterns. Fairly abundant taxa such as Leucanthemum vulgare had a high inter-observer variation using VE, but almost none using SF (Table III). The reason for this should be the problem of estimating cover, since the leaves of this species are easy to find and do not resemble other species. One of the observers generally noted higher cover of this species than the other (Figure 1). In contrast, the reason for the low ratio of Pimpinella saxifraga in Table III should be the problem of finding the plant; the species had a larger variation using SF than using VE (Table III). Pimpinella saxifraga was highly abundant in both reserves, not as flowering individuals but as very small leaves often reduced in size by grazing. These leaves can be difficult to find when growing in a dense, grazed sward. In our study, it is obvious that the observers differed in how often they recorded P. saxifraga (Figure 1). In conclusion, the systematic bias between the observers (Figure 1) tells us that, irrespective of method, observer ‘personality’ will likely be manifested in the data. This bias is slightly larger using VE, which is probably due to the problem of estimating cover. 5.3. S PECIES DETECTION AND TIME CONSUMPTION FOR FIELDWORK The fact that the observers found, on average, 3.1 more taxa in every plot with SF than with VE, suggests that whenever the purpose of the inventory demands that as many species as possible are found, SF is the most appropriate survey method. The difference in time taken to complete each method was on average 46 min. The reason for this very large difference was probably the species-rich vegetation type, its small-scale heterogeneity, and the fact that grazing prevented most plants from flowering. This required a very thorough examination of every subplot, making the SF method quite time consuming. In contrast to our results, Bråkenhielm and Qinghong (1995) surveyed forest vegetation and found no difference in the time taken between a method of visual estimation and a subplot frequency method. Obviously, these contradictory observations show that time consumption depends very much on the complexity of the vegetation type. 5.4. C ONCLUDING REMARKS Our results showed that time consumption using SF was substantially higher than when using VE, which might lead to the recommendation to use VE in preference

COMPARISON OF DATA FROM GRASSLAND VEGETATION MONITORING METHODS

247

to SF from the point of view of survey economy or efficiency. The economic evaluation, however, is not quite as simple as all that. First, as noted by Hope-Simpson (1940), Bråkenhielm and Qinghong (1995), and Jalonen et al. (1998), using a method consisting of subjective estimations, the survey work must be controlled and often supplemented to obtain reliable data. For instance, VE is likely to require more time than SF for other tasks than just the time taken directly for surveying, e.g. targeted training, supervision, and post-collection data adjustments. However, if the quicker VE method leads to more plots being surveyed at a site, the statistical power will increase, compensating partly for the loss in precision. The choice of method should also be related to the purpose of the study. If VE, giving higher irrelevant variation with a smaller proportion of explainable total variation, is used for monitoring biodiversity, the consequences may be pastures being managed inadequately and biodiversity evaluations being made on deceitful data. In this respect, SF is the most suitable survey method for the monitoring of semi-natural grasslands. On the other hand, if detailed monitoring is not the main aim, VE is more eminently suitable because of the potentially larger number of plots that can be included, thereby resulting in a total sample that better captures the within-site heterogeneity - a feature which is often so prevalent in species-rich wooded semi-natural grassland in Sweden.

Acknowledgements We would like to thank Dan Nilsson and Ulrika Carlsson at the County Administrative Board of Östergötland for various inputs. The Swedish Environmental Protection Board provided financial support.

References Bråkenhielm, S. and Qinghong, L.: 1995, ‘Comparison of field methods in vegetation monitoring’, Water, Air, Soil Poll. 79, 75–87. Brandon, A., Spyreas, G., Molano-Flores, B., Carroll, C. and Ellis, J.: 2003, ‘Can volunteers provide reliable data for forest vegetation surveys?’, Natural Areas J. 23, 254–261. Floyd, D. A. and Anderson, J. E.: 1987, ‘A comparison of three methods for estimating plant cover’, J. Ecol. 75, 221–228. Goldsmith, F. B. and Harrison, C. M.: 1976, ‘Description and Analysis of Vegetation’, in S. B. Chapman (ed.), Methods in Plant Ecology. Blackwell Scientific Publications, Oxford, UK. pp. 85–156. Gotfryd, A. and Hansell, R. I. C.: 1985, ‘The impact of observer bias on multivariate analyses of vegetation structure’, Oikos 45, 223–234. Hope-Simpson, J. F.: 1940, ‘On the errors in the ordinary use of subjective frequency estimations in grassland’, J. Ecol. 28, 193–209. Jalonen, J., Vanha-Majamaa, I. and Tonteri, T.: 1998, ‘Optimal sample and plot size for inventory of field and ground layer vegetation in mature Myrtillus-type boreal spruce forest’, Ann. Bot. Fennici 35, 191–196.

248

A. L. M. CARLSSON ET AL.

Karlsson, T.: 1998, ‘The vascular plants of Sweden – a checklist’, Sven. Bot. Tidskr. 91, 241–560 (In Swedish with English summary). Kennedy, K. A. and Addison, P. A.: 1987, ‘Some considerations for the use of visual estimates of plant cover in biomonitoring’, J. Ecol. 75, 151–157. Kent, M. and Coker, P.: 1992, Vegetation Description and Analysis. A Practical Approach, John Wiley and Sons, Chichester, UK. Kirby, K. J., Bines, T., Burn, A., Mackintosh, J., Pitkin, P. and Smith, I.: 1986, ‘Seasonal and observer differences in vascular plant records from British woodlands’, J. Ecol. 74, 123–131. Leps, J. and Hadincova, V.: 1992, ‘How reliable are our vegetation analyses?’, J. Veg. Sci. 3, 119–124. Leps, J. and Smilauer, P.: 2003, Multivariate Analysis of Ecological Data Using CANOCO, Cambridge University Press, Cambridge, UK. Milberg, P., Rydgård, M. and Stenström, A.: 2003, ‘Evaluation of vegetation changes in permanent plots using ordination methods’, Sven. Bot. Tidskr. 97, 107–116 (In Swedish with English summary). Scott, W. A. and Hallam, C. J.: 2002, ‘Assessing species misidentification rates through quality assurance of vegetation monitoring’, Plant Ecol. 165, 101–115. Smith, A. D.: 1944, ‘A study of the reliability of range vegetation estimates’, Ecology 25, 441–448. Stampfli, A.: 1991, ‘Accurate determination of vegetational change in meadows by successive point quadrat analysis’, Vegetatio 96, 185–194. Sykes, J. M., Horrill, A. D. and Mountford, M. D.: 1983, ‘Use of visual cover assessments as quantitative estimators of some British woodland taxa’, J. Ecol. 71, 437–450. Ter Braak, C. J. F. and Smilauer, P.: 2002, CANOCO Reference Manual and User’s Guide to Canoco for Windows: Software for Canonical Community Ordination (version 4.5), Microcomputer Power, Ithaca, New York, USA. Tonteri, T.: 1990, ‘Inter-observer variation in forest vegetation cover assessments’, Silva Fennica 24, 189–196.