Agriculture, Ecosystems and Environment 122 (2007) 366–376 www.elsevier.com/locate/agee
Determinants of local species richness of diurnal Lepidoptera in boreal agricultural landscapes Mikko Kuussaari a,*, Janne Helio¨la¨ a, Miska Luoto b, Juha Po¨yry a a
Finnish Environment Institute, Research Programme for Biodiversity, P.O. Box 140, FIN-00251 Helsinki, Finland b Thule Institute, P.O. Box 7300, FIN-90014 University of Oulu, Finland Received 12 October 2006; received in revised form 6 February 2007; accepted 9 February 2007 Available online 9 April 2007
Abstract The determinants of local species richness of diurnal Lepidoptera were studied in 68 agricultural landscapes in five geographic regions by sampling field margins, road verges, forest edges and patches of semi-natural grassland. Butterflies and day-active moths were counted and habitat quality was measured in 1191 separate 50 m transects. Generalized linear mixed models (GLMM) were built to explain patterns of species richness by first adjusting for the effects of geographic location and weather conditions during the butterfly counts. Habitat type and quality largely determined local species richness, whereas the explanatory power of the local habitat quantity and management variables was limited. Species richness was highest in patches of semi-natural grassland, second highest in forest edges and lowest in road and field margins surrounded by cultivated fields. Species richness increased with abundance of nectar flowers and margin width, and decreased with increasing soil moisture and within-patch openness. Landscape openness showed a unimodal response with a maximum species richness at intermediate values. Largely the same variables affected species richness of moths and butterflies, but moths appeared to be more sensitive to negative effects of windiness and intensive management and less dependent on abundant flowers and increasing margin width than butterflies. The results highlight the significance of even small patches of semi-natural grasslands and open, sunny forest edges for species richness in typical modern farmland with relatively intensive agriculture. The effectiveness of agri-environment schemes in promoting biodiversity could be increased by directing more efforts to maintenance and restoration of these habitats. # 2007 Elsevier B.V. All rights reserved. Keywords: Agri-environment scheme; Biodiversity; Butterflies; Day-active moths; Habitat quality; Habitat type
1. Introduction In northern Europe the focus of biodiversity conservation efforts has traditionally been on forests (Esseen et al., 1992), and relatively little attention has been paid to agricultural landscapes. Recently also concern for agricultural environments has grown, because several studies have reported severe loss of habitats maintaining biodiversity in agricultural landscapes (Eriksson et al., 2002; Lindborg and Eriksson, 2004). In Finland the area of semi-natural grasslands has decreased to less than 1% of the area that existed only 100 years ago (Luoto et al., 2003). Not * Corresponding author. Tel.: +358 20490 2248; fax: +358 20490 2290. E-mail address:
[email protected] (M. Kuussaari). 0167-8809/$ – see front matter # 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.agee.2007.02.008
surprisingly, the loss of open uncultivated habitats has caused the decline and endangerment of several groups of organisms in northern Europe (Ga¨rdenfors, 2000; Rassi et al., 2001). During the last 10 years national agri-environment schemes partly funded by the European Union (EU) have offered economic support to farmers for employing measures aiming to promote farmland biodiversity (Kleijn and Sutherland, 2003; Berendse et al., 2004). The success of these measures has been very variable, which underlines the importance of monitoring their effectiveness (Kleijn and Sutherland, 2003). There is also a need for more information on the relative importance of different farmland habitats for biodiversity, and better understanding of the factors affecting variation in species richness in agricultural
M. Kuussaari et al. / Agriculture, Ecosystems and Environment 122 (2007) 366–376
landscapes. Such knowledge will facilitate comparison of alternative scenarios in developing agri-environment schemes towards better and more cost-efficient promotion of farmland biodiversity in the future. For example in the Finnish agri-environment scheme obligatory measures to promote biodiversity have been vaguely defined and there is a need to develop the scheme towards more specific measures in the future (Puurunen, 2004). Boreal agricultural environment is typically a mosaic of cultivated areas and forests (Virkkala et al., 2004). It differs from temperate agricultural landscapes (e.g. Ouin and Burel, 2002) in its extensive cover of forests and lack of hedgerows. This was reflected in the present study design, which focused on the four most important habitats for pollinator insects commonly available in northern European agricultural landscapes: semi-natural grasslands, forest edges, road verges and field margins. In this paper results of an extensive, quantitative survey of biodiversity in Finnish farmland (covering plant, insect, bird and habitat diversity) are presented with the ultimate aim of providing an appropriate and cost-effective approach to monitoring the impacts of the Finnish agri-environment support scheme (Kuussaari et al., 2004). Results are reported of the survey for butterflies and day-active moths by focusing on the question: how do habitat type, local habitat quantity, habitat quality and management affect species richness at the local scale of a habitat patch? Generalized linear mixed models (GLMM) (Venables and Ripley, 2002) were used to explain patterns of variation in species richness separately for butterflies and moths, and the variation partitioning method (Borcard et al., 1992) to evaluate the relative roles of the variable groups. Factors affecting species richness at the landscape level of 0.25 km2 study squares have been addressed in a paper by Kivinen et al. (2006) based on the same sampling scheme.
2. Material and methods In order to sample representatively typical modern Finnish farmland, a total of 68 agricultural landscapes in five geographic regions in Finland were selected using stratified random sampling (Fig. 1). In order to facilitate butterfly counts in two relatively adjacent squares during the same day, pairs of two randomly selected 1 km2 squares within 10–20 km distance from each other were selected within all except the smallest geographic region. The study regions included the three most important and rather intensively cultivated agricultural areas of Finland in the south, south-west and west, but also less intensively ˚ land islands. cultivated areas in eastern Finland and in the A For a detailed description of the structure of the studied landscapes and their dominant agricultural production types see Luoto (2000) and Kivinen et al. (2006). Each study area was 1 km2 in size with a minimum of 11.6% of its area consisting of arable land (mean = 46.1%, max = 90.8%).
367
Cover of forests varied from 1.7 to 81.0% between the study areas (mean = 41.4%). Altitudinal variation between the study areas was low, ranging from five to 205 m above sea level. In each 1 km2 study area, butterflies and day-active moths of the families Zygaenidae, Sphingidae, Lasiocampidae, Lymantriidae, Drepanidae, Geometridae, Arctiidae and Noctuidae (nomenclature according to Kullberg et al., 2002) were studied in 20 separate 50 m transects which were located in different habitat patches and in at least 50 m distance from each other (Fig. 1). In patches of semi-natural grassland the study transects were placed in central areas avoiding patch edges. In order to distinguish day-active moths from other moth species, a classification based on available literature and a large database of moth records from transect counts conducted in Finland was used (the classification is available from the authors). The transects were placed in open and semi-open uncultivated habitats, most of which could be classified into the following four types: field margins (n = 511), road verges (n = 168), forest edges (n = 322) and patches of semi-natural grassland (n = 190). The focus of this paper is solely on the 1191 (out of all the sampled 1355) transects which fit these four focal habitat types (see Kuussaari et al., 2004). Data on diurnal Lepidoptera were collected by applying the widely used transect-method, in which butterflies are counted along a permanent route within a 5 m 5 m square ahead of the counter (Pollard and Yates, 1993). Transect counts were conducted in each 50 m transect in all study areas seven times during the summer at ca two week intervals from mid May until the end of August. This period covers the flight season of most butterfly and day-active moth species in Finland. In mainland Finland the data (n = 1010 transects) were gathered in the year 2001 and in ˚ land islands (n = 181 transects) in 2002. Transect the A counts were carried out in weather conditions when butterflies were actively in flight. Counts were not made when the temperature was below 13 8C in sunny and below 17 8C in overcast weather, or wind speed exceeded five on the Beaufort scale. Sixteen environmental variables were used to explain the observed variation in species richness of butterflies and moths (Table 1). Environmental variables were divided into five groups: (1) adjusting, (2) habitat type, (3) quantity, (4) quality and (5) management variables. Adjusting variables included geographic location and average weather conditions during the transect counts. Geographic location was measured as the longitudinal and latitudinal coordinates of the midpoints of the 1 km2 study squares. Weather variables included the averages of temperature (8C), proportion of direct sunshine (% of the length of a transect) and wind speed (on the Beaufort scale) during the seven counts of each 50 m transect. The proportion of sunshine and wind speed were estimated separately for each transect during each count. For each count the temperature was calculated as the mean of two
368
M. Kuussaari et al. / Agriculture, Ecosystems and Environment 122 (2007) 366–376
Fig. 1. Sampling design at the geographic and landscape scale. (a) A total of 68 agricultural landscapes (black dots) were sampled within five geographic ˚ land islands, D = south-western Finland and E = southern Finland. Grey shading indicates regions in Finland: A = western Finland, B = eastern Finland, C = A the distribution of cultivated fields. (b) Within each 1 km2 study area a total of 20 separate 50 m long transects (black vectors) were studied in open, uncultivated habitats. See Kuussaari et al. (2004) for more details of the sampling design.
measurements, made at the beginning and the end of each count of a study square comprising 20 transects. Habitat type was measured as a categorical variable which included four types: semi-natural grassland, forest
edge, field margin and road margin. Semi-natural grasslands were patches of uncultivated and typically non-fertilized open habitat which differed from the other habitat types by their non-linear shape. The three linear habitat types differed
Table 1 Univariate effects of the 16 environmental variables on species richness of butterflies and day-active moths Environmental variable
Species richness of butterflies
Species richness of day-active moths
Estimate
% of total deviance explained
Estimate
1.14 0.11 1.10 0.41 5.58 12.33
1.122** 0.000* 0.002 0.058 0.006** 0.450*** 0, 0.17, 0.52,
Adjusting variables Latitude Latitude2 Longitude Temperature during transect counts Sunshine during transect counts Windiness during transect counts Habitat type
0.007 0.138*** 0.003* 0.247*** 0, 0.06, 0.31,
Habitat quantity Area of semi-natural grassland Margin width
0.000* 0.034***
2.72 2.92
Habitat quality Soil moisture Vegetation height Vegetation height2 Flower abundance Flower abundance 2 Within-patch openness Within-patch openness2 Landscape openness Landscape openness 2 Slope angle
0.224*** 0.006 0.000** 0.178*** 0.010** 0.513*** 0.078*** 0.035*** 0.062*** 0.072***
3.58
0.106**
3.25
Habitat management Grazing intensity Grazing intensity2 Mowing *
**
0.023***
0.099** ***
0.26***
n
% of total deviance explained 1191 0.15 0.00 0.02 0.61 9.14 20.79
1191 1191 1191 1191 1191
0.000 0.024
1.03 1.05
190 1001
0.132** 0.004***
0.55 1.75
1191 1191
0.090***
4.47
1191
0.53***
0.85 9.62 4.80 12.26 1.14
0.72
0.221*** 0.057** 0.050*** 0.074** 0.107*** 0.638*** 0.163* 0.085
1191 9.52 1191 16.20 1.25
1191 190
12.83 0.31
1001
P < 0.05; P < 0.01; P < 0.001. Results for the quadratic functions (2) are also shown in cases where the addition of the quadratic term after the linear term significantly increased the deviance explained by the model. The estimates for habitat type refer to the coefficients of forest edge, semi-natural grassland, field margin and road margin, respectively.
M. Kuussaari et al. / Agriculture, Ecosystems and Environment 122 (2007) 366–376
from each other by their surrounding habitat. Forest edges were linear margin habitats between a forest and a cultivated field. The edge between a forest and an uncultivated open strip was typically rather sharp with a closed forest starting immediately behind the open uncultivated strip. The surrounding forests varied in size and type. They were typically dominated either by Norway spruce (Picea abies) or Scotch pine (Pinus sylvestris) with various amount of deciduous trees such as birches (Betula pendula and B. pubescens), aspen (Populus tremula) and willows (Salix spp.). In order to standardize light conditions suitable for butterfly transect counts only forest edges facing south or west were included in the sampling. Road margins were linear, open habitats surrounded by a road on one side and a cultivated field on the other. Only road margins surrounded by cultivated fields on both sides of the road were included in the analyses. These roads were typically small gravel roads with a low traffic density. Field margins were open, linear habitats completely surrounded by cultivated fields. For a description of typical plant communities on Finnish agricultural field margins see Tarmi et al. (2002). Local habitat quantity was measured as the area of each studied patch of semi-natural grassland (m2) and the width of each studied linear margin habitat (m). Size of grassland patches varied between 0.01 and 11.70 ha (mean = 0.8, median = 0.2 ha). Linear margin habitats were typically narrow uncultivated and non-fertilized strips, the width varying between 0.5 and 10.0 m (mean 2.1 m) in forest edges, 0.5–7.0 m (mean 2.0 m) in road margins and 0.5– 9.0 m (mean 2.1 m) in field margins. Habitat quality was measured by six variables. Soil moisture conditions were classified into three types, (1) dry, (2) mesic and (3) moist, which correspond to the three main vegetation types in Finnish semi-natural grasslands. Vegetation height was estimated as the average height of the dominant vegetation of the study transect in late July, with an accuracy of 10 cm. Nectar flower abundance was estimated twice, in late June and late July. On both occasions we estimated the total abundance of flowering nectar plants using a scale from 0 (no nectar flowers) to 4 (nectar flowers abundant). The sum of the two measurements was used as an index describing the total abundance of nectar flowers. Two measures of openness were estimated for the studied habitat patch of each transect. Within-patch openness was estimated on the scale 0–4, from (0) most of the area covered by bushes and trees to (4) open habitat without any bushes or trees. Landscape openness was estimated by considering the habitats immediately surrounding the study habitat on the scale 0–3, from (0) habitat completely surrounded by closed habitats, usually forest, to (3) habitat completely surrounded by open habitats, usually cultivated field. Topographical variation was measured as a four-class variable indicating the average slope angle within the studied habitat, ranging from (0) flat terrain to (3) steep slope. Habitat management was measured differently in seminatural grassland and in linear habitats because the
369
prevailing management type differed between these two habitats. Semi-natural grasslands, if managed at all, were usually managed by cattle grazing (25% of the patches of semi-natural grassland). Grazing intensity was measured during the last butterfly count in late August using the scale: 0 = no grazing, 1 = light, 2 = moderate and 3 = heavy grazing. The corresponding average vegetation heights in the differently grazed semi-natural grasslands in late July were 70 cm (no grazing), 39 cm (light grazing), 22 cm (moderate grazing) and 9 cm (heavy grazing). The prevailing management in linear habitats was mowing, which was measured as a binary variable: 0 = no mowing and 1 = mown at least once during the summer (June–August). Mowing was more common in road verges (57% mown) than in field margins (21%) and forest edges (22%). Fertilizers and pesticides were generally not used in the studied field and road margins. Nevertheless some drift of fertilizers and pesticides occasionally happens from conventionally cultivated fields to surrounding field margins. The highest pair-wise Spearman rank correlation between the 16 environmental variables was relatively low, rs = 0.43 between latitude and the average temperature during transect counts. In the analyses habitat type was treated as a categorical variable and the other 15 environmental variables as continuous variables. 2.1. GLMM analyses We built generalized linear mixed models with penalized quasi-likelihood (GLMM PQL) (Venables and Ripley, 2002) to explain observed variation in species richness of butterflies and day-active moths. This statistical modelling approach has the advantage of allowing the incorporation of random terms that control for spatial non-independence in the data, arising from grouped observations (Pinheiro and Bates, 2000; Venables and Ripley, 2002). Our sampling design included potentially spatially correlated data on two scales: in the five geographic study regions (each of which included separate study squares) and in the 68 study squares (each of which included separate study transects). To overcome the problems of lacking spatial independence of data points (Legendre, 1993), we fitted the geographic region and the square kilometer study area as two categorical random terms before adding any other variables into the GLMMs. This procedure also takes into account the ˚ land potential effect of the different sampling year in the A islands (2002) compared with the other four geographic regions (2001). We built three separate GLMMs for both butterflies and day-active moths. The first model was built to test the effect of habitat type in the whole data set (n = 1191), and the two other models to examine the effects of habitat quantity, quality and management separately in the subsets of seminatural grasslands (n = 190) and linear margins (n = 1001). Because the straightforward application of automatic stepwise procedures in regression model construction has
370
M. Kuussaari et al. / Agriculture, Ecosystems and Environment 122 (2007) 366–376
been criticised (e.g. James and McCulloch, 1990; Mac Nally, 2000), we applied a mixture of an ecologically focused and a traditional stepwise procedure in the model building. The significant adjusting geographic and weather variables were added into the model first before considering the environmental variables of primary interest. The variable groups were entered into the GLMMs in the following order: (1) random effects, (2) adjusting variables, (3) habitat type, (4) habitat quantity, (5) habitat quality and (6) habitat management. The entering order of the first four variable groups was based on our study design. For the last two variable groups we chose to test the effects of habitat quality before the effects of habitat management because we considered that the measures of habitat quality are likely to have more direct effects on our study organisms than habitat management per se. The pure effects of habitat management were considered separately in the variation partition analyses (see below). At each step of testing the effects of one variable group the statistically significant environmental variables were entered into the model in the order of their explanatory power, i.e. using the forward selection procedure. Because of the large sample sizes and six different models built, we used P < 0.01 instead of the conventional P < 0.05 as the significance limit for the inclusion of variables into the models. For each variable both linear and quadratic effects were tested. At each step, all variables were retested for possible deletion. Because overdispersion was observed we used quasi-Poisson error distribution and an F ratio test in all the GLMMs (Venables and Ripley, 2002). Paired comparisons of species richness among the habitat types were conducted by building separate univariate GLMMs with the six possible combinations of the four habitat types. All statistical calculations were performed using the statistical package R version 2.0.1 (R Development Core Team, 2004). GLMM are not shown for the number of individuals, because abundance of both butterflies and day-active moths was strongly correlated with species richness (Spearman rank correlation coefficients: butterflies 0.83, day-active moths 0.87). 2.2. Variation partitioning The variation partitioning (VP) approach was used to decompose the variation in the species richness of butterflies and moths among the three groups of predictors: local habitat quantity (A), habitat quality (Q), and management (M) variables. VP was run separately for the species richness of butterflies and moths in the four habitats, semi-natural grasslands, field margins, road margins and forest edges. Variation in the species richness was partitioned using a series of partial GLMMs after the statistically significant adjustive variables were included into the models. As a first step, within each of the three groups of predictors forward selection of predictor variables was performed to include variables that contributed significantly (P < 0.01) to the
explained deviance (Borcard et al., 1992; Heikkinen et al., 2004). The goodness-of-fit for each added variable was measured by the deviance statistics and the change in deviance was tested by an F-ratio test. Variation partitioning with three explanatory matrices has been described in detail by Heikkinen et al. (2004). Here, it leads to eight fractions: pure effect of habitat quantity (A), pure effect of habitat quality (Q), pure effect of management (M); combined variation due to the joint effects of habitat quantity and habitat quality (AQ), habitat quantity and management (AM), habitat quality and management (QM), the three groups of explanatory variables (AQM), and finally unexplained variation.
3. Results A total of 19,250 individuals of 60 species of butterflies and 10,858 individuals of 118 species of moths were recorded. Sixty-seven (56.8%) of the moth species and 10,705 individuals (98.6%) were considered day-active. The average species richness of butterflies in the 50 m study transects varied from a minimum of 3.7 in the west coast ˚ land islands lowlands to a maximum of 7.4 species in the A and in southern Finland. In moths the average species richness varied rather similarly: from 2.7 in western Finland to 4.6 in southern Finland. The four most abundant butterfly species (Aphantopus hyperantus, Pieris napi, Nymphalis urticae and Thymelicus lineola) comprised 65.0% of all recorded butterfly individuals. In day-active moths the four most abundant species (Scotopteryx chenopodiata, Rheumaptera hastata, Chiasmia clathrata and Xanthorhoe montanata) accounted for 65.9% of all observations. The majority of the recorded species were common species of boreal agri-environment, but also seven nationally threatened species were observed (five butterfly and two moth species). With the exception of ˚ land islands (with 85 recorded Maniola jurtina in the A individuals), threatened species were extremely scarce, comprising 0.5 and 0.02% of the total number of observed butterflies and moths, respectively. Table 1 summarises the univariate effects of all the 16 explanatory environmental variables on butterfly and moth species richness when the effect of each environmental variable was modelled separately by adding the variable to a GLMM already including the two categorical random terms. Three of the four variables with the highest explanatory power were the same for butterflies and moths: habitat type, landscape openness and average windiness during the transect counts. These variables are interrelated, because the amount of forest surrounding the focal habitat affects the value of each variable, e.g. habitats along a forest edge were less windy than habitats surrounded by more open landscape. In addition to these three variables, the univariate positive effect of the abundance of nectar flowers was especially
M. Kuussaari et al. / Agriculture, Ecosystems and Environment 122 (2007) 366–376 Table 2 Results for GLMMs examining the effects of adjusting variables and habitat type on species richness of (a) butterflies and (b) moths Variable
Estimate
(a) Butterflies Intercept 12.214 Windiness 0.114 Latitude 0.019 Sunshine 0.005 Temperature 0.122 Habitat type 0, 0.04, 0.28, 0.23 n = 1191, 15.6% of deviance explained (b) Day-active moths Intercept 9.524 Windiness 0.160 Latitude 0.011 Habitat type 0, 0.14, 0.47, 0.47 n = 1191, 21.5% of deviance explained
F
P
Table 3 Results for GLMMs examining the effects of environmental variables on species richness of (a) butterflies and (b) moths in patches of semi-natural grassland Variable
50.51,1183 184.41,1183 19.61,1183 16.01,1183 41.13,1183
152.01,1185 12.21,1185 76.43,1185
