Modelling the Distribution of the Stag Beetle (Lucanus Cervus) in Belgium
Total Page:16
File Type:pdf, Size:1020Kb
1 Thomaes, A., Kervyn, T. & Maes, D 2008. Applying species 2 distribution modelling for the conservation of the threatened 3 saproxylic Stag Beetle (Lucanus cervus). Biological conservation, 4 141: 1400-1410 5 Authors a b c 6 Arno THOMAES , Thierry KERVYN & Dirk MAES 7 Affiliations a 8 Research Institute for Nature and Forest (INBO), Gaverstraat 4, B-9500 9 Geraardsbergen, Belgium, email: [email protected] b 10 Directorate General for Nature Resources and Environment (DGRNE), Avenue Prince 11 de Liège 15, B-5100 Jambes, Belgium, [email protected] c 12 Research Institute for Nature and Forest (INBO), Kliniekstraat 25, B-1070 Brussels, 13 Belgium, [email protected] 14 *Full address for correspondence 15 Dirk MAES, Research Institute for Nature and Forest (INBO), Kliniekstraat 25, B-1070 16 Brussels, Belgium, Tel. +32 2 558 18 37; fax +32 2 558 18 05; [email protected] 17 18 Abstract 19 Despite its size and attractiveness, many Lucanus cervus sites remain undetected in 20 NW Europe because of its short flight period and its nocturnal activity. Therefore, 21 present-day designated conservation areas for L. cervus are probably insufficient for a 22 sustainable conservation of the species. We applied eight species distribution 23 modelling techniques (Artificial Neural Networks, Classification Tree Analysis, 24 Generalised Additive Models, Generalised Boosting Models, Generalised Linear 25 Models, Mixture Discriminant Analysis, Multiple Adaptive Regression Splines and 26 Random Forests) to predict the distribution of L. cervus in Belgium using ten randomly 27 generated calibration and evaluation sets. We used AUC, sensitivity (% correctly 28 predicted presences in the evaluation set) and specificity (% correctly predicted 29 absences in the evaluation set) and Kappa statistics to compare model performances. 30 To avoid the incorporation of only marginally suitable woodland sites into the Natura 31 2000 network, we, conservatively, considered the species as being present only in grid 32 cells where all ten randomly generated model sets predicted the species as such. 33 Model performance was, on average, good allowing to predict the potential distribution 34 of L. cervus accurately. According to the predicted distribution using the more robust 35 prevalence threshold, only 5 731 ha (11% of the potentially suitable area) is protected 36 under the Natura 2000 scheme in Belgium. Subsequently, we categorised the 37 potentially suitable woodlands into three conservation priority categories based on their 38 surface area and the already designated Natura 2000 area. Including the most suitable 39 L. cervus woodlands previously not included in the Natura 2000 sites within such 40 network would require protecting an area of 15 260 ha. Finally, we discuss the 41 implications of using species distribution modelling for nature policy decisions in 42 designating conservation networks. 43 44 Keywords: Belgium, saproxylic beetle, Natura 2000, predictive modelling, AUC, Kappa 45 statistics, model comparison, protected area effectiveness 46 47 1. Introduction 48 The accelerating decline and extinction of many species has made species 49 conservation in particular and nature conservation in general a globally important issue 50 (Thomas et al., 2004). This resulted in the ratification of the Convention on Biological 51 Diversity (CBD) in which world leaders agreed to halt biodiversity loss by 2010 1 52 (Secretariat of the Convention on Biological Diversity, 2006). Although they make up 53 75-80% of all known species, invertebrate focal species are rarely used for hotspot 54 analyses or site designation (McGeoch, 1998; Samways, 2005). However, coincidence 55 of invertebrate hotspots with those from the more commonly used taxonomic groups 56 (vertebrates and/or plants) could be rather low (Prendergast et al., 1993; Maes et al., 57 2005). 58 In Europe, two directives oblige member states to protect sites where focal species 59 occur: the Habitats Directive (92/43/EEC) and Birds Directive (79/409/EEC). 60 Furthermore, member states have to adequately map the distribution and to monitor 61 the abundance of known populations of species that are in the annexes of both 62 directives. Together, sites protected under the European Bird Directive and Habitat 63 Directive form the so called Natura 2000 network (see Wätzold and Schwerdtner, 64 2005). To designate sites on the basis of focal species, information on their distribution 65 is essential (Kareiva and Levin, 2003; Hortal et al., 2007). Even in well studied regions, 66 species can remain undetected because of their inconspicuous behaviour (e.g., 67 nocturnal), small size (e.g., many invertebrates), low abundance, taxonomical problems 68 (difficult to classify) or spatial biases in distribution data (Dennis and Thomas, 2000). 69 Furthermore, if distribution data are present, sites are usually only designated if the 70 species is actually known to be present. This approach almost always results in a too 71 limited area of protected sites, because sites with a high potential for the occurrence of 72 the species but without its documented permanent presence, are rarely designated 73 (Decleer, 2007). 74 The threatened saproxylic Stag Beetle Lucanus cervus (LINNAEUS 1758) is one of the 75 invertebrate species that is listed in Annex II of the European Habitat Directive. 76 Saproxylic invertebrates in general and beetles in particular have been identified as 77 one of the most threatened invertebrate communities in Europe (Speight, 1989; Berg et 78 al., 1994) and many saproxylic species have been used as indicators for the quality of 79 woodlands (e.g., Fowles et al., 1999; Ranius, 2002). Despite its size and 80 attractiveness, distribution data of L. cervus are not available on a scale that allows an 81 adequate designation of protected areas in many NW European countries. This is due 82 to the fact that L. cervus occurs in low numbers and is only active during a short period 83 of warm nights in June and July (Smith, 2003; Smit, 2004). 84 When distribution data are scarce, modelling techniques are increasingly used to fill in 85 gaps in distribution maps or to target conservation efforts towards sites with a 86 potentially high conservation value (e.g., Guisan and Zimmermann, 2000; Luoto et al., 87 2002; Wilson et al., 2005; Heikkinen et al., 2007). Modelling techniques use distribution 88 data to link species to a set of biotic (e.g., land cover, species interactions) and/or 89 abiotic variables (e.g., soil type, climate data) and permit to predict presence 90 probabilities for un-surveyed sites (Guisan and Zimmermann, 2000). Furthermore, 91 predictive modelling is advocated to be an increasingly powerful tool in conservation 92 biology because it allows to incorporate un-surveyed areas into nature policy decision 93 making (Pullin et al., 2004; Rushton et al., 2004). A prerequisite, however, is that a 94 minimum number of data are available to build and evaluate the species distribution 95 models (Pearson et al., 2007). Despite their attractiveness and recommended use in 96 conservation biology, models always need to be critically evaluated and validated if 97 they are to be applied to designate sites (Hortal et al., 2007). The use of different 98 modelling techniques and different methods to test model efficiency is, therefore, highly 99 recommended (Thuiller, 2003). They can, at most, help to prioritise sites and to help 100 policy makers in their decision to designate the most cost-effective sites for the 101 protection of species or habitats (Chefaoui et al., 2005). 102 Here, we use the threatened saproxylic Stag Beetle Lucanus cervus in Belgium as an 103 example of how modelling techniques can be used to help designate ecological 104 networks. First, we describe the characteristics of the known distribution pattern of L. 105 cervus in Belgium. Secondly, we apply different modelling techniques to predict the 106 potential distribution of this species. Subsequently, we assign conservation priority 107 values to woodland sites that have a high probability of harbouring the species and 108 suggest their incorporation in the Natura 2000 network. Finally, we discuss the use of 2 109 species distribution modelling to detect un-surveyed but potentially suitable sites for the 110 focal species and how such sites can be incorporated in ecological networks and 111 conservation policy making. 112 113 2. Material & Methods 114 2.1. Study area 115 Belgium is a strongly industrialised NW European country with high human population 116 density (335 inhabitants/km², Van Goethem, 2001) and, consequently, intense 117 pressure on nature (OECD, 1998). The general landscape and topography differ 118 considerably between the two administrative regions of Belgium: Flanders and 119 Wallonia. Flanders, the northern part, is a lowland zone (mean elevation = 38 m) and 120 only has a limited total area of nature reserves (1.6% of the territory, Van Goethem, 121 2001) and forest (8% - CEC, 1994). The highest amount of woodland in Flanders is 122 found in the north-eastern part of the country and in the surroundings of Brussels (e.g., 123 Sonian Forest, Haller Forest, Meerdaal Forest). Wallonia, the southern part and 124 comparatively an upland region (mean elevation = 310 m) has a similar total area of 125 nature reserves (ca. 1% of the territory, Van Goethem, 2001). However, it has a 126 considerably higher amount of woodlands, mainly coniferous (31%, CEC, 1994). 127 128 2.2. Study species and data 129 The saproxylic Stag Beetle Lucanus cervus (LINNAEUS 1758) is one of the largest 130 beetle species in Europe (Luce, 1996). In the past, the species was thought to be 131 confined to large woodlands (Tochtermann, 1992), but more recent studies in NW 132 Europe have shown that L. cervus can also occur in open and more urban habitats 133 such as gardens, parks, open forests, hollow ways, orchards and afforested slopes in 134 the vicinity of large woodlands (Rink and Sinsch, 2006).