The use of diversity indices to assess the diversity of vegetation in managed boreal forests

The use of diversity indices to assess the diversity of vegetation in managed boreal forests

Forest Ecology and Management 112 (1998) 121±137 The use of diversity indices to assess the diversity of vegetation in managed boreal forests Sari Pi...

326KB Sizes 1 Downloads 22 Views

Forest Ecology and Management 112 (1998) 121±137

The use of diversity indices to assess the diversity of vegetation in managed boreal forests Sari PitkaÈnen* University Of Joensuu, Faculty of Forestry, P.O. Box 111, FIN-80101 Joensuu, Finland Received 23 May 1997; accepted 5 May 1998

Abstract A classi®cation of biodiversity in managed forests is presented. The classi®cation is based on the abundances of ground vegetation species, the classes being described using stand variables and diversity indices. Four alpha diversity indices, three measures of evenness, and three beta diversity indices are calculated. Discriminant analysis is used to determine the stand variables that best describe the classes. A total of 14 forest classes are established based on the results. The signi®cant variables in the classi®cation, in order of importance, are: the number of coniferous and broadleaved tree species, prescribed burning, site fertility, topography, mean diameter of trees, dominance of the tree layer by spruce, number of canopy layers, soil type, drainage and arti®cial regeneration. # 1998 Elsevier Science B.V. All rights reserved. Keywords: Alpha diversity; Assessment; Beta diversity; Classi®cation; Discriminant analysis; Managed forests; Species turnover

1. Introduction Forest management is one reason for the impoverished biodiversity in the forests of Finland. The total area of old forests has decreased markedly (Kaila et al., 1994), which has in turn reduced the scope for many species to live in forest habitats. Thus, about half of Finland's threatened invertebrate species, for example, are denizens of the forests (Kaila et al., 1994). Further, since management smooths natural multimodal forest stand structure, managed forests are often dominated by generalist species. On the other hand, it has been shown on a number of occasions that lowintensity forest management can lead to increases in *Corresponding author. Tel.: +358-13-2513633; fax: +358-132514444; e-mail: [email protected]

biological diversity (Attiwill, 1994a, b; Rescia et al., 1994; Larsen, 1995 Butter®eld, 1995; Halpern and Spies, 1995). Thus, the developing of systems of sustainable forest management and an objective method for assessing the resulting diversity is a matter of some urgency. Most studies of biodiversity have concentrated on developing indices to measure alpha or beta diversity, where alpha diversity refers to species richness within an area (Whittaker, 1972) and beta diversity to the rate of species turnover along a complex environmental gradient (Whittaker, 1972; Wilson and Shmida, 1984). At the landscape level biodiversity can be described in terms of gamma diversity; the number of species of different communities in an area and the level of differentiation of the communities (Whittaker, 1972).

0378-1127/98/$ ± see front matter # 1998 Elsevier Science B.V. All rights reserved. PII: S0378-1127(98)00319-3

122

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137

The ®rst heterogeneity index, developed by Simpson as early as 1949 (Whittaker, 1972), described the probability of two randomly selected individuals belonging to the same species. Many versions of Simpson's index of diversity have been developed since then (Baev and Penev, 1995). Another frequently used diversity index is that created by Shannon and Wiener in the same year (Whittaker, 1972), while Kempton and Taylor (1976) later developed a new alpha diversity index that they called Q-statistics. This index was based on quartiles of the species abundance distributions. Besides the different alpha diversity indices, several measures of evenness have been developed. Evenness has been considered to be a fundamental factor on any site containing more than one species (Molinari, 1989). Evenness indices describe the equality of species abundance in the community. The ®rst evenness indices, developed by Hill (1973); made it possible to compare quantitative results obtained at different sites. Later, Alatalo (1981) and Molinari (1989) proposed measures of evenness based on the work of Hill. Molinari (1989) developed his measure in order to avoid two disadvantages inherent in the F index of Alatalo, which, he claimed, overestimates evenness and has a non-linear relationship to it. The third popular index of evenness, developed by Pielou (Peet, 1974), relates the observed diversity to the maximum value the index could have in the given community. Measures of beta diversity were developed much later, the ®rst being introduced by Whittaker in the late 1960s (Wilson and Shmida, 1984). Later, Wilson and Shmida (1984) developed a beta diversity index based on that of Whittaker, and Wilson and Mohler (1983) developed one based on gradient length and species turnover along the gradient. Further, Oksanen and Tonteri (1995) developed an application of the Gaussian response function to measure compositional turnover along gradients. Meanwhile, ékland (1990a) developed a new method to determine beta diversity based on species importance in detrended correspondence analysis (DCA) ordination of sample plots. Ecological studies concerning biodiversity have on many occasions concentrated on alpha diversity alone. Species richness (Zobel et al., 1993; Tonteri, 1994; Rey Benayas, 1995) or various alpha diversity indices (Zobel et al., 1993; Rescia et al., 1994; ToÂthmeÂreÂsz, 1995), and also species presence (Nieppola, 1992),

have been used to describe alpha diversity and its changes. Beta diversity has been described in terms of half-change units describing species turnover along a gradient (Whittaker, 1972, Wilson and Mohler, 1983) ékland (1990a); in turn, used ordination of sample plots to determine beta diversity in terms of the standard deviation of species turnover. The present work is part of an effort to develop methods of assessing forest biodiversity, which refers here to species richness, evenness and species turnover in the ground vegetation. The ®rst aim of the work was to relate alpha and beta diversity measures to forest stand structure as described with classes based on ordination and classi®cation of the ground vegetation, calculating the different diversity indices for these classes. The second aim was to analyze forest stand classes according to various stand characteristics in order to determine which variables best describe the different classes and their biodiversity. 2. Material and methods The data used here were gathered by the Finnish Forest Research Institute in the Lieksa-Nurmes area of northern Karelia, eastern Finland, in the course of the 7th National Forest Inventory (NFI) of Finland (Fig. 1). They apply to 166 circular plots at forest sites on mineral soils. Detailed measurements were made of stand and site characteristics on these plots (Hotanen and Nousiainen, 1990): site fertility, dominant tree species, numbers of coniferous and deciduous tree species, topography, soil type, age of the stand, mean diameter of trees, number of canopy layers, crown cover, damage (from insects, wind, etc.), basal area, soil preparation and manuring, drainage, whether the stand was established by arti®cial or natural regeneration and different management measures applied to the stand. Also, some trees at the plot were measured as sample trees involving accurate determinations of tree species, diameter at breast height, crown layer, diameter at six metres and height of the tree. In addition, the vegetation of the sample plots was assessed in eight quadrats of 1 m2 each (Hotanen and Nousiainen, 1990). The abundances of the ground vegetation species were recorded as cover percentages. A total of 163 species were found on the plots.

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137

123

Fig. 1. The forest area used as a source of data (darkened) and boundary between the southern and middle boreal vegetation zones (Hotanen and Nousiainen, 1990).

PitkaÈnen (1997) used these data to develop a new classi®cation of forest sites. This classi®cation forms a basis for the present work. The vegetation data were ®rst subjected to ordination, to identify the most important ecological gradients that affect the species composition and abundances on the plots (Fig. 2). Four ordination methods were used; DCA and three variants of multi-dimensional scaling (MDS). DCA ordination is simple to use and results in interpretable axes which can be used as a robust measure of the compositional turnover of species (ékland, 1990b), but it has two serious shortcomings: it suffers from the tongue effect, caused by its tendency to minimize variation in the higher axes, and it assumes that the species response curves are unimodal and symmetric, which is not the case (Minchin, 1987). MDS ordination was developed to solve the problem of ordination caused by the non-linearity of ¯oristic similarity measures (ékland, 1990b), and is more capable of ®nding the global optimum for ordination than DCA.

The disadvantage, however, is that the gradients have an arbitrary scaling (ékland, 1990b). Correlations between environmental variables (tree and site characteristics) and the ordination, the eigenvalues of the axes (DCA) and the minimum stress values for the axes (MDS) were calculated for each set of four ordinations. The highest correlations and the lowest minimum stress values were achieved with global non-metric multidimensional scaling (GNMDS). DCA ordination was almost as good, but the GNMDS solution was accepted as an ordination in view of the shortcomings of DCA (PitkaÈnen, 1997). The placement of the plots was similar in each ordination, those with similar species composition and similar stand characteristics being situated close to each other in the ordination space. Also, the two main gradients were the same regardless of the ordination method, the most important gradient being site fertility and the second main gradient the successional stage of the stand (PitkaÈnen, 1997).

124

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137

Fig. 2. The calculation procedure.

In addition, the data were classi®ed by two-way indicator species analysis (TWINSPAN), using the ordination as a reference when evaluating the relevance of the classi®cation (PitkaÈnen, 1997). The classi®cation was based on the vegetation data as well (Fig. 2). TWINSPAN is a hierarchical clustering method which uses differential species to characterise and separate classes (Hill, 1979; ékland, 1990b). After the classes had been formed, the stand characteristics were considered and, if necessary, the classi®cation was altered, i.e. sample plots were classi®ed at

a higher level, to obtain classes with uniform stand variables (Fig. 2) (for further details, see PitkaÈnen, 1997). As a result, two classes were combined (classes 21 and 22 into *) and two classes (classes 23 and 24) were omitted on account of heterogeneity in stand variables (Fig. 3). Thus, a total of 21 forest classes were left for further analysis (Fig. 3 and PitkaÈnen, 1997). Biodiversity indices for these classes are calculated in the present work (Fig. 2). Ten diversity indices and species richness were calculated for the 166 plots to describe the diversity

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137

Fig. 3.

TWINSPAN

125

classification of the data. The classification levels are indicated by broken lines.

of the ground vegetation in the data (Table 1). The diversity of the classes was described in terms of the mean, minimum and maximum values for each index (Peet, 1974) calculated from the classi®ed plots. The average number of species was considered an appropriate measure for comparing areas with equal sample sizes (Hill, 1973; Peet, 1974). Species richness was calculated as the number of species present and recorded for six vegetation groups: shrubs, dwarf shrubs, grasses, ferns, mosses and lichens. The dominance of the vegetation groups over each other was then examined within each class (Huston, 1994). In addition, the percentage of the cover accounted for by the two most abundant species (Minchin, 1991) was calculated for each plot in each class. Wilson and Shmida (1984) proposed that their own measure and that of Whittaker were the most suitable among the beta diversity indices available. This is partly because they are easy to calculate and interpret (Wilson and Shmida, 1984). The measure of Cody was also considered appropriate (Baev and Penev, 1995),

although it is not entirely independent of alpha diversity as are the measures of Whittaker and Wilson and Shmida (Wilson and Shmida, 1984). These measures are all independent of sampling, however, and are not affected by the particular location of the samples along the gradient (Wilson and Shmida, 1984). The various alpha and beta diversity indices were calculated with BIODIV software (Baev and Penev, 1995). Another way to calculate beta diversity was proposed by ékland (1990a), who used DCA ordination of the plots to determine beta diversity in the data. The DCA ordination used in the present case is that of PitkaÈnen (1997). The calculation of beta diversity is based on DCA since this scales the axes in units of average standard deviation of species turnover, S.D. (ékland, 1990a). Both main axes were ®rst divided into segments of 0.25 S.D. each. ékland (1990a) suggests that only segments containing 10 plots or more should be used for further analysis, but due to the low number of plots used here, segments with ®ve plots or more were accepted. As a result, 13 segments

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137

126 Table 1 Diversity indices evaluated Index

Reference

Formula

Comments

Peet, 1974 Hill, 1973 Kempton and Taylor, 1976 Hill, 1973

P H0 ˆ ÿ i† ÿP i p2 i ln…p ÿ1 N2 ˆ i pi Q ˆ …S=2† log…R P 2 =R1 † N1 ˆ exp‰ÿ i pi ln…pi †Š

Sensitive to rare species Not affect by rare species Discriminates well between sites Expresses diversity on a uniform scale

Peet, 1974

J 0 ˆ ‰ÿ

Alatalo's F

Alatalo, 1981

Fˆ(N2ÿ1)/(N1ÿ-1)

Molinari's G

Molinari, 1989

Beta diversity indices Whittaker

Gˆ[(arcsin F)/908]F, when F>H0.5; GˆF3, otherwise

Wilson and Shmida, 1984

ÿ1 W ˆ s=

Cody Wilson and Shmida

Wilson and Shmida, 1984 Wilson and Shmida, 1984

C ˆ ‰g…H† ‡ l…H†Š=2  T ˆ ‰g…H† ‡ l…H†Š=2

Alpha diversity indices Shannon's H0 Simpson's reciprocal Q-statistics Hill's N1 Evenness indices Pielou's J0

P i

pi ln…pi †Š=ln S

Total number of species in the community should be known Little affected by sampling bias, values depend only on evenness Independent of species richness Relates and diversities to overall diversity S Define ecotones between community types

R1ˆ25% quartile of the species abundance distribution. R2ˆ75% quartile of the species abundance distribution. pi is the proportion of species i in the community; S is the total number of species; is the average number of species in the samples; g(H) is the number of species gained along the habitat gradient H. l(H)is the number of species lost along the habitat gradient H.

were assigned to the ®rst axis and 11 to the second. This omitted the plots situated at the ends of the axes. Importance values Iijk for each species i in segment j along axis k were then calculated from Iijk ˆ

cij ln…1 ‡ dij † ln 101

(1)

where cij is the ratio between number of plots containing species i in segment j and number of plots in segment j and P percentage cover dij ˆ i n n being the number of plots containing species i. These importance values were ordinated and the length of the ®rst DCA axis was taken as an estimate of beta diversity (ékland, 1990a). The ordination was performed for four sets of Iijk values on two axes: (1) all species (163 species), (2) trees (12 species), (3) ®eld layer (85 species) and (4) ground layer (66 species). Thus, a total of eight estimates of beta diversity were obtained. Discriminant analysis, a technique for classifying individuals into two or more classes on the basis of measured variables (A®® and Clark, 1990), was used

to ®nd the stand characteristics that best distinguished between pre-de®ned TWINSPAN classes (Fig. 2). The method was implemented using the SAS statistical software, version 6.0, assuming multinormality of the variables. The equality of the group covariance matrices was tested (A®® and Clark, 1990). The goodness of the classi®cation was tested by crossvalidation, which produces almost unbiased estimates with low variance (A®® and Clark, 1990). The variables were selected using the stepwise method based on Wilks' lambda. Discriminant analyses were performed on the sixth, ®fth, fourth and third levels of the TWINSPAN classi®cation (Fig. 2 and 3) with all possible stand variables to examine which of the levels would best distinguish the classes and which would then be signi®cant, i.e. have the most typical values in each class. All the stand and site variables measured were used in the analysis. The ability of the various diversity indices and of species richness to distinguish between the forest classes was tested with Duncan's multiple range test, which is based on studentized range distribution (Fig. 2). Duncan's test was chosen because the diversity indices were normally distributed, except reciprocal of Simpson and Hill's N1, which were

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137

normalized with logarithmic transformation. The test identi®es classes that have signi®cant differences in the mean values of their diversity indices as calculated from the plots belonging to each class. In addition, correlations were calculated between the various diversity indices and between these and the two main environmental gradients in order to describe the relationships between the diversity indices and to connect the diversities with the gradient structure of the vegetation and the environment. Kendall's nonparametric t was selected for this purpose, since it takes only the ranks of the variables into account. 3. Results 3.1. Discriminant analysis On the sixth level of TWINSPAN classi®cation 85.78% of the plots were classi®ed correctly, i.e. they were allocated to the same class as in the vegetation analysis, but only 59.72% were classi®ed correctly by cross-validation (Table 2). The signi®cant variables at the 95% signi®cance level were the numbers of coniferous and broad-leaved tree species, number of canopy layers, site fertility (herb-rich, mesic forest sites), soil type (silt, loam, till), topography (hollow or shaded hillside), mean diameter of the trees, and dominance of the tree layer by spruce (Picea abies). The forest management variables were prescribed burning, drainage and arti®cial regeneration. On the ®fth level of TWINSPAN classi®cation, 88.74% of the calibration data were classi®ed correctly (Table 2) and the cross-validation result improved a little, 64.66% of the plots being classi®ed correctly. The variables describing the classes were almost the same as those on the sixth level of classi®cation. Table 2 Results of discriminant analysis on different levels of the TWINSPAN classification Level 6 5 4 (13 classes) 4 (14 classes)

Calibration data

Cross-validation

Correct, %

Error %

Correct, %

Error %

85.78 88.74 85.83 86.29

14.22 11.26 14.17 13.71

59.72 64.66 77.50 73.91

40.28 35.34 22.50 26.09

127

The fourth level of TWINSPAN classi®cation was the most exact. The classi®cation result for the calibration data was slightly poorer (85.83%) than on the ®fth level, but cross-validation produced by far the best result, 77.50% of the plots being classi®ed correctly (Table 2). It was noticed at this point, however, that one of the classes should be divided into two on account of clear differences in site type and the age of the stand, i.e. the classi®cation should be made at the ®fth level (class * in Fig. 4 should be divided into classes 13 and 14). This division improved the classi®cation of the calibration data somewhat (86.29%), but weakened the outcome of cross-validation, 73.91% of the data being classi®ed correctly (Table 2). The signi®cant variables in the discriminant analysis were the following: numbers of coniferous and broad-leaved tree species, site fertility (herb-rich forest site, mesic forest site), topography (hollow or shaded hillside), mean diameter of the trees, dominance of the tree layer by spruce, number of canopy layers, soil type (silt loam, till), prescribed burning, drainage and arti®cial regeneration. As a result, a total of 14 classes out of the original 21 were left to describe the forest area (Fig. 4 and Table 3). On the third level of TWINSPAN classi®cation there was too much variability in the values for the stand variables for correct plot classi®cation. According to the alpha diversity measures, the classes with the highest and lowest diversity were classi®ed with 100% correctness by cross-validation (Table 4). Classes 10 and 14 were among the poorest classi®cation results (Table 4), as they did not differ clearly from the other classes in terms of stand variables. 3.2. Correlation analysis Pielou's J0 , Q-statistics and Simpson's reciprocal had signi®cant correlations with every alpha diversity index and measure of evenness (p<0.001), whereas Alatalo's F and Molinari's G were correlated only with each other and with Pielou's J0 , Q-statistics and Simpson' reciprocal. The other alpha diversity indices correlated signi®cantly with each other and with species richness. All the correlations except those between Q-statistics and F and G were positive. The highest correlations (r >69) were found between

128

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137

Fig. 4. The final classification result and its classes.

Hill's N1, Shannon's H0 , Simpson's reciprocal and Pielou's J0 . Two major groups can be distinguished; indices that are connected with the evenness of the vegetation (Alatalo's F, Molinari's G, Pielou's J0 , Qstatistics and Simpson's reciprocal) and indices connected with species richness (Hill's N1, Pielou's J0 , Qstatistics, Shannon's H0 and Simpson's reciprocal). Pielou's J0 , Q-statistics and Simpson's reciprocal seem to describe both. Correlations were also calculated between the GNMDS ordination and the alpha diversity indices, yielding signi®cant relations (p<0.001) for the ®rst axis with Hill's N1, Pielou's J0 , Shannon's H0 , Simpson's reciprocal, Q-statistics and species richness, and for the second axis with Shannon's H0 , Pielou's J0 , Hill's N1 and Simpson's reciprocal. Neither Alatalo's F nor Molinari's G was signi®cantly correlated with either of the axes. It can be concluded that fertility (axis 1) is more important for diversity than the stage in the succession (axis 2). 3.3. Species diversity within the classes The calculated alpha diversity indices pointed to very low species diversity (Table 5 ), in addition to

which no rare species were detected in the ground vegetation. As a whole, the plots with the highest species richness were located in the classes representing young forests with a low crown cover, classes 3, 5, 7, 12 and 13. The site type in these classes varied from herb-rich to dry forest sites, the former having a higher species richness. Class 3, for example, had a mean of 39.5 species, a minimum of 36 and a maximum of 43. The vegetation within the most fertile classes was also even, with no speci®c species or functional group dominant. The most common functional groups were grasses and mosses, while shrubs and lichens were less frequent. In the classes representing dry forest sites, the mosses and lichens, especially Cladonia spp., were more frequent and the number of grasses decreased. Class 14, representing one of the older stands, had almost as high a species richness as the younger forests in the area, the mean value being 30.5. This was a mature spruce forest on a fertile site where no particular species was dominant in the ground vegetation. The other classes with mature tree stands were dominated by moss and lichen species (Dicranum spp., Sphagnum spp. and Cladonia spp.). The same classes, e.g. classes 3, 13 and 14, were identi®ed as more diverse in terms of the various alpha

Dryish Not measured

Dryish Plain, sunny hillside 8.2 (5.7±11.1)

3

1

2

Fine sandy till or Fine sandy till or Fine sandy till or sandy till sandy till sandy till

No

No

14.0 (5.5±22.5), shelterwood Yes

Herb-rich Not measured

2

1±10 1

Class 3

Fine sandy till or sandy till

2

No

Dry Plain, sunny hillside 18.6 (4.7±50)

1

61±180 2

Class 4

Fine sandy till or sandy till

2

No

13.1 (4.2±29.8)

Dry Hollow

0

11±60 1

Class 5

Fine sandy till or sandy till

2

No

Dryish Plain, sunny hillside, hill op 11.6 (4.0±29.0)

1

1±60 1

Class 6

1

No

Dryish Plain, shadowed hillside 10.2 (6.9±13.7)

0

11±30 1

Class 7

Fine sandy till or sandy till Treatment Prescribed burning N1 6.32 (4.72±8.35) 7.80 (4.70±11.05) 16.51 (15.76±17.26) 5.07 (2.72±8.02) 9.29 (6.17±11.86) 6.73 (2.83±10.92) 8.81 (7.12±10.04) Simpson's reciprocal 4.3 (3.7±5.0) 21% 5.1 (2.4±8.5) 19% 10.65 (9.8±11.5) 27% 3.43 (1.6±5.7) 19% 6.07 (3.5±8.4) 21% 4.12 (1.7±7.9) 15% 5.85 (4.2±7.2) 23% Chamaedaphne Calluna vulgaris, Salix phylicifolia, Differential species Empetrum nigrum Chamaedaphne Agrostis tenuis, Vaccinium calyculata, Peltigera apthosa, Polytrichum commune, calyculata Lycopodium uliginosum, Polytrichum strictum, Hylocomium Cladonia gracilis V. myrtillus, annotinum, Dicranum bergerii splendens, Pohlia Rubus saxatilis, Calluna vulgaris, nutans, Cladonia Cladonia Dryopteris cenotea carthusiana arbuscula, Hylocomium splendens

Dominance by spruce Number of canopy layers Soil type

20.3 (5.6±33.7)

2

0

Mean diameter, cm

71±140 2

1±60 1

Age, a Number of conifer tree species Number of broad-leaved tree species Site fertility Topography

Class 2

Class 1

Variable

Table 3 Forest vegetation classes

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137 129

3 Silt loam

2

Fine sandy till or sandy till

Treatment N1 Reciprocal of Simpson Differential species

Partly drainaged 4.59 (2.64±8.93) 3.58 (2.0-6.3) 21% Lutzula pilosa, Geranium sylvaticum, Melica nutans, Pohlia nutans, Hylocomium splendens

Mesic, herb-rich Plain, hill top, Sunny hillside 25.7 (3.1±64.8) Yes

Dryish Plain, sunny Hillside 26.0 (3.8±4.7.3) No

7.39 (6.21±10.99) 4.86 (3.2±7.5) 19% Pinus sylvestris

3

0

Mean diameter, cm Dominance by spruce Number of canopy layers Soil type

81±180 2

61±160 1

Age, a Number of conifer tree species Number of broad leaved tree species Site fertility Topography

Class 9

Class 8

Variable

Table 3 (continued )

7.28 (6.04±9.0) 5.33 (4.3±7.0) 22% Empetrm nigrum, Ledum palustre, Hylocomium splendens

Fine sandy till or sandy till

3

21.2 (5.0±65.0) No

Dryish Plain

2

61±150 2

Class 10

Fine sandy till or sandy-till Partly drained 5.48 (3.22±9.59) 3.82 (2.0±6.1) 195 Melampyrum pratense, Solidago virgaurea, Carex globularis, Dryopteris carthusiana, Sphagnum riparium

2

Mesic Plain, sunny Hillside 20.2 (4.1±36.2) Yes

3

51±210 2

Class 11

Fine sandy till or sandy till Drained 8.98 (4.67±11.97) 5.56 (2.9±8.5) 19% Ebilobium angustifolium, Trientalis europaea

2

Mesic,dryish Plain sunny Hillside 8.2 (4.0±14.0) No

0

21±30 1

Class 12

11.62 (7.27±15.91) 7.90 (3.7±11.0) 25% Empetrum nigrum, Vaccinium vitis-idaea, Deschampsia flexuosa, Dicranum Polysetum

Fine sandy till or sandy till

2

Dryish Plain sunny Hillside 8.3 (3.0±11.6) No

1

11±30 1

Class 13

13.88 (11.42±16.34) 9.70 (7.7±11.6) 32% Equistetum sylvativum, Dryopteris carthusiana, Sphagnum girgensohnii

Fine sandy till or sandy till

3

21.2 (1.6±36.0) Yes

Mesic Not measured

0

80±140 1

Class 14

130 S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137 Table 4 Result of cross-validation on the fourth Class 1 2 3 4 5 6 7 8 9

TWINSPAN

classification level. Total count of errors 26.09%

1 2 3 4 5 7 (100%) 3 (60%) 2(100%) 3 8 (100%) 1 (33.33%) 1 (25%)

10 11

6

7

8

9

1 8 (94.74%)

2 (50%)

4 (80%)

0

0

100

12

5.26

2 9 (100%)

1 (25%) 1 (20%) 1 (25%)

1 (12.5%)

1 (12.5%) 40

11

2 (66.67%)

12 13

10

13

14

2

3 (75%)

14 Error% 0

131

50

20

0

75

1 0 (100%)

0

7 (87.5%)

12.5

1 (5.26%)

7 (87.5%) 12.5

1 (50%) 50

Table 5 Mean, minimum and maximum values for alpha diversity indices in the various classes Class

1 2 3 4 5 6 7 8 9 10 11 12 13 14

Shannon's index H

Simpson's reciproal

Q-statistics

Hill's diversity index N1

Mean

Min.

Max.

Mean

Min.

Max.

Mean

Min.

Max.

Mean

Min.

Max.

1.83 1.99 2.81 1.58 2.19 1.83 2.17 1.98 1.50 1.98 1.64 2.16 2.42 2.62

1.55 1.55 2.76 1.00 1.82 1.04 1.96 1.83 0.97 1.80 1.17 1.54 1.98 2.44

2.12 2.40 2.85 2.08 2.47 2.39 2.31 2.40 2.19 2.20 2.26 2.48 2.77 2.79

4.30 5.10 10.65 3.43 6.07 4.12 5.85 4.86 3.58 5.33 3.82 5.56 7.90 9.70

3.7 2.4 9.8 1.6 3.5 1.7 4.2 3.2 2.0 4.3 2.0 2.9 3.7 7.7

5.0 8.5 11.5 5.7 8.4 7.9 7.2 7.5 6.3 7.0 6.1 8.5 11.0 11.6

2.96 3.90 7.88 2.52 5.02 4.76 4.03 3.87 2.23 3.46 3.24 5.70 5.39 5.95

1.76 1.84 7.57 1.23 2.96 1.95 3.25 2.52 0.69 1.98 1.73 3.46 3.55 5.54

4.34 5.28 8.18 5.58 7.41 6.70 5.29 4.75 5.81 5.34 5.77 11.50 7.55 6.36

6.32 7.80 16.51 5.07 9.29 6.73 8.81 7.39 4.69 7.28 5.48 8.98 11.62 13.88

4.72 4.70 15.76 2.72 6.17 2.83 7.12 6.21 2.64 6.04 3.22 4.67 7.27 11.42

8.35 11.05 17.26 8.02 11.86 10.92 10.04 10.99 8.93 9.00 9.59 11.97 15.91 16.34

diversity indices (Table 5). Since the maximum value of Simpson's reciprocal is the number of species growing on the plot, the mean values were regarded as representing proportions of the values for the number of species (Table 3). The highest proportions recorded in the data were for classes 3, 13 and 14 (27%, 25% and 32%, respectively).

The lowest values for N1, Simpson's reciprocal, Shannon's H0 and Q-statistics were obtained for class 9, which represents a mature forest on a mesic site with mosses dominating the ground vegetation. The values for the class were 4.69 (2.64±8.93), 3.58 (2.0± 6.3), 1.5 (0.97±2.19) and 2.23 (0.69±5.81), respectively. The values of the alpha diversity indices for the

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137

132

Table 6 Mean, minimum and maximum values of the evenness indices in the various classes Class y

Pielou's eveness index J0 Mean

Min.

Max.

Mean

Min.

Max.

Mean

Min.

Max.

1 2 3 4 5 6 7 8 9 10 11 12 13 14

0.60 0.60 0.76 0.56 0.65 0.56 0.67 0.61 0.54 0.63 0.56 0.65 0.70 0.77

0.57 0.48 0.73 0.38 0.60 0.37 0.62 0.54 0.39 0.57 0.41 0.52 0.61 0.72

0.66 0.73 0.79 0.75 0.72 0.74 0.70 0.72 0.67 0.68 0.82 0.81 0.78 0.81

0.64 0.57 0.63 0.58 0.59 0.52 0.61 0.60 0.69 0.69 0.62 0.56 0.63 0.67

0.52 0.38 0.60 0.36 0.49 0.38 0.52 0.41 0.48 0.55 0.38 0.43 0.42 0.64

0.79 0.74 0.65 0.83 0.68 0.70 0.70 0.68 0.83 0.75 0.78 0.73 0.78 0.69

0.27 0.22 0.24 0.21 0.21 0.15 0.24 0.23 0.33 0.33 0.26 0.19 0.27 0.30

0.14 0.06 0.21 0.05 0.12 0.05 0.14 0.07 0.11 0.17 0.06 0.08 0.08 0.26

0.46 0.40 0.27 0.51 0.31 0.34 0.34 0.31 0.51 0.40 0.44 0.37 0.45 0.33

Alatalo's eveness index F

other older classes were similar regardless of the site type. Slightly different classes emerged as possessing the highest values for the evenness measures, Alatalo's F, Molinari's G, and Pielou's index J0 , the three best in terms of Alatalo's F being classes 10, 14, and 1, those for Molinari's G classes 10, 14, and 13 and those for Pielou's J0 classes 14, 3, and 13 (Table 6). In summary, the evenness results suggested that the highest diversity was achieved in class 14, a mature forest on a mesic site. Class 10, also representing mature forest but on a dryish site with a sparse canopy cover (11± 20%), also had quite a high alpha diversity (Table 5). Class 1, in turn, was a young forest on a dry site with a crown cover of 1±10%, and had alpha diversity indices that were among the lowest in the data (Table 5). When the species richness and dominance were considered in each class separately, high species richness was usually related to low dominance and vice versa (Fig. 5). The reason for this is that the less the two most abundant species dominate the growing space, the greater the opportunity there is for other species to exist and the more even the vegetation can be. The indices complied exactly with the changes in species richness in this data set (Fig. 6). The most readily distinguishable classes according to Duncan's test were 3, 6, 9, 12, 13 and 14. Four of these represent young forests with quite a high alpha diversity, but class 9 represents an old forest on a

Molinari's eveness index G

fertile site which has the lowest alpha diversity in the data. The low diversity was due to the dominance of the mosses Pleurozium schreberii and Hylocomium splendens in the ground layer. Class 14 was old forest on a fertile site with a high alpha diversity. In this class none of the species was superior to the others. The only class that showed no difference compared to the others was class 4, which represented an old stand on dry forest site with low alpha diversity. The alpha diversity indices that separated these classes were Hill's N1, Q-statistics, Shannon's H0 and Simpson's reciprocal. The calculated measures of beta diversity showed only a slight increase from dry forests towards more fertile forest sites and from younger stands towards older ones. This indicated differences between these biotopes, although no clear trend could be identi®ed when the different measures were considered plot by plot, suggesting that the beta diversity does not change markedly (increase or decrease) along the two main gradients of GNMDS except at their end points (Fig. 7). This may be due to the nature of the data, in that forest sites of intermediate fertility prevailed and no really old forests occurred in the area concerned, the age structure being dominated by stands under 80 years of age. Thus, as a whole, the data can be considered to be very even. Low beta diversity was also obvious when considering the overall species turnover along the GNMDS

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137

133

Fig. 5. Variations in dominances of the two most frequent species and species richness between classes. Classes are arranged in descending order of species richness.

gradients. The values for the measures of Whittaker, Cody, and Wilson and Shmida were 6.2, 135.5 and 5.92, respectively, for the ®rst axis (site fertility) and 6.12, 147.0, and 6.42 for the second (stage in the succession). The DCA ordination-based method for calculating beta diversity gave the same result as the beta diversity indices; that beta diversity in the data was very low. All the values were below 4 S.D. units, which usually indicates total turnover of species (Table 7). The Table 7 Estimates of beta diversity (compositional turnover) on gradients corresponding to axes 1 and 2 in the ordination of the plots by detrended correspondence analysis (DCA) yAxis

Gradient

Total

Tree layer

Field layer

Ground layer

1 2

Fertility Succession

2.760 2.432

2.530 2.411

2.649 2.262

2.185 2.344

highest beta diversity for both axes was found for the total number of species. The ground layer and ®eld layer had the lowest beta diversity for the ®rst and second axes: 2.185 and 2.262, respectively (Table 7). 4. Discussion The aim of this work was to develop a method for assessing forest biodiversity. The results showed that stand structure and diversity of the ground vegetation are correlated, and that this connection can be described by forest classes, for which diversity measures can be calculated. The classi®cation was needed because such classes are useful in practical forestry for assessing the biodiversity of stands. Each stand can be classi®ed into a certain class once its variables have been measured, and information on biodiversity can be obtained from the resulting tables.

134

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137

Fig. 6. Effect of species richness on Hill's ratio N1, Simpson's reciprocal Shannon's H0 and Q-statistics in classes arranged in descending order of species richness.

The calculation of several diversity measures and comparison of the results has been considered as desirable method of studying biodiversity (Baev and Penev, 1995). In this work, a total of ten measures of diversity were calculated, all of which are regarded as being among the most useful ones (Peet, 1974, Baev and Penev, 1995). In addition, beta diversity was studied using the DCA ordination of plots. The results did not depend on the particular measure of alpha or beta diversity (Tables 5 and 6, Fig. 6), as the same classes had the highest and lowest values of the diversity indices, as could also be concluded from the correlation analyses between the indices, in which no discrepancies were found. Beta diversity was evaluated using various existing measures and by DCA ordination. These methods again supported one another, the results indicating low beta diversity in all cases (Fig. 7, Table 7). This is partly due to the fact that the environmental factors were quite similar throughout the area concerned, and partly because the forests were managed. Zobel et al.

(1993) found that managed communities have low variance in diversity between their biotopes. The fact that all the species recorded were generalists reduced the diversity even more. On the other hand, the low values for the beta diversity indices may result from the indices used, which make no actual reference to the gradients but are more a measure of total heterogeneity within the plots (Oksanen and Tonteri, 1995). On the other hand, the DCA ordination-based method offers a sound basis for determining beta diversity (ékland, 1990a). Of the various alpha diversity indices, Hill's N1, Qstatistics, Shannon's H0 and Simpson's reciprocal discriminated between the classes when assessed with Duncan's test and also conformed to the variations in species richness (Fig. 6). One could conclude from this that species richness alone would be enough to distinguish between the classes, but the mean values for species richness did not in fact differ signi®cantly. The indices obviously had higher variability in their mean values, which may be due to the fact that they

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137

135

Fig. 7. Variations in beta diversity. Classes are arranged in order of site fertility.

were in¯uenced to varying extents by the evenness in the species abundances. Thus, the indices evidently do take into account the two important factors in biodiversity, species richness and evenness, and differences between the classes can be found. One of the most interesting issues here was whether different vegetation classes differ in terms of stand characteristics and diversity. This was studied using discriminant analysis and Duncan's test. Also Scheffe"s test for multiple comparison of means has been used for this purpose (PitkaÈnen, 1997), but since Scheffe's test is based on F-distribution, Duncan's test was more effective in this case when the variables (the different indices) were normally distributed. Discriminant analysis has been used successfully by Lindholm and Tuominen (1989) to classify old natural forests on the basis of structural stand characteristics, and by Kuusipalo (1985) to study site classi®cation in southern Finland. Lindholm and Tuominen (1989), using discriminant analysis to assess how well the TWINSPAN classes worked as a method of classify-

ing old forests, found that 79.6% of the classi®cation was correct and concluded that TWINSPAN was a suf®cient method for the purpose. The present comparable classi®cation proved to be 73.92% correct. Kuusipalo (1985), in turn, used discriminant analysis to study the variation in vegetation from site to site and its connection with the variation in tree characteristics, and regarded discriminant analysis as an effective method for testing his fertility classes. Thus, in this study also it was possible to use the results of discriminant analysis as a criterion for the usefulness of the forest classi®cation. The 14 classes out of the original 21 that were left after discriminant analysis (Figs. 2±4) were based on differences in stand variables perceived in this analysis, so they differed in both their ground vegetation and their stand characteristics. Duncan's test showed that not all the classes differed from each other in diversity, however, the best distinguishable being the classes with the lowest and highest diversity indices. Since these classes can be distinguished by stand

136

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137

variables alone, they could also be identi®ed in the ®eld. In addition to stand variables, differential species and their evenness should be taken into account. On the other hand, most of the classes were of similar diversity. Thus, if only diversity is considered, these classes should be combined. This would result in highly heterogeneous stand variables within them. Although they do not differ in terms of diversity, they can be identi®ed in the ®eld and information can be obtained about the diversity of the ground vegetation. Due to this the classes were not combined. Certain forest management measures, prescribed burning, drainage and arti®cial regeneration, were signi®cant in the analysis. The ®rst vegetation following prescribed burning is typical of mesic sites, but later, as the canopy closes in, it becomes typical of dry sites (Lindholm and Vasander, 1987). Fire in¯uences the vegetation for almost thirty years. The stands involved here that had been managed by ®re were young, and thus, it is natural that prescribed burning distinguished them from the others. Drainage likewise has an impact on the vegetation (Rey Benayas, 1995). Regeneration through planting accelerates canopy closure and together with intensive treatment designed to favour planted seedlings, promotes the growth of species that are less sensitive to disturbances in the vegetation. These changes, which could be seen in the vegetation-based classi®cation, contribute to an explanation of why drainage and arti®cial regeneration are signi®cant in discriminant analysis. It should be remembered in this context that although a method may exist for measuring biodiversity in forests, it is useless unless the goals of forest management are known. In order to promote biodiversity in general, forests should be managed the way that supports it in unmanaged areas and maintains it as high as possible within managed areas. The term `unmanaged' refers here to forests excluded from timber production or other silvicultural operations and the term `managed' to forests manipulated in some way for the purposes of timber production and where ®res are controlled. The goals of forestry depend on the forest owner and his preferences, however, and this aspect must be taken into account. It is shown here that it is possible to manage forests for biodiversity if this is one goal accepted by the forest owner. Factors were identi®ed in the stand structure,

such as tree species composition, number of canopy layers and mean diameter of trees, that affect the diversity of the vegetation as a whole and which can be adjusted by means of management. The present ®ndings support the notion that it is possible to classify forests in terms of biodiversity based on stand characteristics. A most interesting ®eld of research in future will be the assessment of diversity in unmanaged forests and comparison of managed and unmanaged forests. Data entailing greater variability in terms of environmental factors would also offer an opportunity to study both alpha and beta diversity in more detailed, as together they can be said to describe quite well the total diversity in a forest area consisting of numerous stands. Acknowledgements The author wishes to thank MetsaÈmiesten SaÈaÈtioÈ, the Joensuu Research Station of the Finnish Forest Research Institute and the Faculty of Forestry, University of Joensuu. Special thanks are due to Professors Seppo KellomaÈki and Pekka NiemelaÈ, JuhaPekka Hotanen, Lic. Phil., and Dr. Rune H. ékland for their valuable advice and comments provided during the calculation and writing phases. The contribution of Mr. Malcolm Hicks in editing the English of the manuscript is also appreciated. References Afifi, A.A., Clark, V., 1990. Computer-aided multivariate analysis, 2nd ed. Chapman and Hall, London, 505 pp. Alatalo, R.V., 1981. Problems in the measurement of evenness in ecology. Oikos 37, 199±204. Attiwill, P.M., 1994a. The disturbance of forest ecosystems: the ecological basis for conservative management. For. Ecol. Manage. 63, 247±300. Attiwill, P.M., 1994b. Ecological disturbance and the conservative management of eucalyptus forests in Australia. For. Ecol. Manage. 63, 301±346. Baev, P.V., Penev, L.D., 1995. BIODIV. Program for calculating biological diversity parameters, similarity, niche overlap, and cluster analysis. Version 5.1. Pensoft, 57 pp. Butterfield, R.P., 1995. Promoting biodiversity: advances in evaluating native species for reforestation. For. Ecol. Manage. 75, 111±121. Halpern, C.B., Spies, T.A., 1995. Plant species diversity in natural and managed forests of the Pacific Northwest. Ecol. Appl. 5(4), 913±934.

S. PitkaÈnen / Forest Ecology and Management 112 (1998) 121±137 Hill, M.O., 1973. Diversity and evenness: a unifying notation and its consequences. Ecology 54(2), 427±432. Hill, M.O., 1979. TWINSPAN ± a Fortran program for arranging multivariate data in an ordered two-way table by classification of the individuals and attributes. Cornell University, Ithaca, New York, 48 pp. Hotanen, J.-P., Nousiainen, H., 1990. MetsaÈ ± ja suokasvillisuuden numeerisen ryhmittelyn ja kasvupaikkatyyppien rinnastettavuus. Summary: The parity between numerical units and site types of forest and mire vegetation. Folia Forestalia 763, 1±54 (in Finnish with English summary). Huston, M.A., 1994. Biological diversity. The coexistence of species on changing landscapes. Cambridge University Press, Cambridge, 681 pp.. Kaila, L., Martikainen, P., Punttila, P., Yakolev, E., 1994. Saproxylic beetles (Coleoptera) on dead birch trunks decayed by different polypore species. Ann. Zool. Fennici 31(1), 97± 107. Kempton, R.A., Taylor, L.R., 1976. Models and statistics for species diversity. Nature 262, 818±820. Kuusipalo, J., 1985. An ecological study of upland forest site classification in southern Finland. Seloste: Ekologinen tutkimus EtelaÈ-Suomen kangasmetsien kasvupaikkaluokituksesta. Acta Forestalia Fennica 192, 1±77. Larsen, J.B., 1995. Ecological stability of forests and sustainable silviculture. For. Ecol. Manage. 73, 85±96. Lindholm, T., Vasander, H., 1987. Vegetation and stand development of mesic forest after prescribed burning. Seloste: Kasvillisuuden ja puuston kehitys tuoreella kankaalla kulotuksen jaÈlkeen. Silva Fennica 21(3), 259±278. Lindholm and Tuominen, S., 1989. Vanhojen luonnonmetsien rakennetyypit eraÈillaÈ etelaÈboreaalisilla luonnonsuojelualueilla. Summary: The structure classes of southern boreal natural forests in some Finnish nature protection areas. In: PoikajaÈrvi, H., Sepponen, P., Varmola, M. (Eds.), Tutkimus luonnonsuojelualueilla. Research activities on the nature conservation areas. Folia Forestalia 736, 46±52 (in Finnish, with English summary). Minchin, P., 1987. An evaluation of the relative robustness of techniques for ecological ordination. Vegetatio 69, 89± 107.

137

Minchin, P., 1991. DECODA ± Database for ecological community data. Preliminary documentation notes. Version 2.05. Australian National University. Canberra, 14 pp.. Molinari, J., 1989. A calibrated index for the measurement of evenness. Oikos 56, 319±326. Nieppola, J., 1992. Long-term vegetation changes in stands of Pinus sylvestris in southern Finland. J. Veg. Sci. 3, 475±484. ékland, R.H., 1990a. A phytoecological study of the mire Northern Kisselbergmosen, SE Norway. III. Diversity and habitat niche relationships Nordic. J. Bot. 10, 191±220. ékland, R.H., 1990b. Vegetation ecology: theory,methods and applications with reference to Fennoscandia Sommerfeltia supplement 1, 1±216. Oksanen, J., Tonteri, T., 1995. Rate of compositional turnover along gradients and total gradient length. J. Veg. Sci. 6, 815± 824. Peet, R.K., 1974. The measurement of species diversity. Ann. Rev. Ecol. Systematics 5, 285±307. PitkaÈnen, S., 1997. Correlation between stand structure and ground vegetation: an analytical approach. Plant Ecol. 131, 109±126. Rescia, A.J., Schmitz, M.F., Martin de Agar, P., de Pablo, C.L., Atauri, J.A., Pineda, F.D., 1994. Influence of landscape complexity and land management on woody plant diversity in northern Spain. J. Veg. Sci. 5, 505±516. Rey Benayas, J.M., 1995. Patterns of diversity in the strata of boreal montane forest in British Columbia. J. Veg. Sci. 6, 95± 98. Tonteri, T., 1994. Species richness of boreal understorey forest vegetation in relation to site type and successional stage. Ann. Zoologici Fennici 31(1), 53±60. ToÂthmeÂreÂsz, B., 1995. Comparison of different methods for diversity ordering. J. Veg. Sci. 6, 283±290. Whittaker, R.H., 1972. Evolution and measurement of species diversity. Taxon 21(2/3), 213±251. Wilson, M.V., Mohler, C.L., 1983. Measuring compositional change along the gradients. Vegetatio 54, 129±141. Wilson, M.V., Shmida, A., 1984. Measuring beta diversity with presence-absence data. J. Ecol. 72, 1055±1064. Zobel, K., Zobel, M., Peet, R.K., 1993. Change in pattern diversity during secondary succession in Estonian forests. J. Veg. Sci. 4, 489±498.