the interactive effects of livestock exclusion and mammalian pest control on the restoration of...

This article was downloaded by: [Adams State University]On: 03 November 2014, At: 10:01Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office:Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

New Zealand Journal of ZoologyPublication details, including instructions for authors and subscriptioninformation:http://www.tandfonline.com/loi/tnzz20

The interactive effects of livestock exclusionand mammalian pest control on the restorationof invertebrate communities in small forestremnantsRaphael K. Didham a , Gary M. Barker c , Jessica A. Costall d , Lisa H. Denmeadb , Christopher G. Floyd c & Corinne H. Watts ca School of Biological Sciences , University of Canterbury , Private Bag 4800,Christchurch, 8140, New Zealand E-mail:b School of Biological Sciences , University of Canterbury , Private Bag 4800,Christchurch, 8140, New Zealandc Landcare Research , Private Bag 3127, Hamilton, 3240, New Zealandd Ecology Group, Institute of Natural Resources , Massey University , PrivateBag 11222, Palmerston North, 4442, New ZealandPublished online: 19 Feb 2010.

To cite this article: Raphael K. Didham , Gary M. Barker , Jessica A. Costall , Lisa H. Denmead , Christopher G.Floyd & Corinne H. Watts (2009) The interactive effects of livestock exclusion and mammalian pest control onthe restoration of invertebrate communities in small forest remnants, New Zealand Journal of Zoology, 36:2,135-163, DOI: 10.1080/03014220909510148

To link to this article: http://dx.doi.org/10.1080/03014220909510148

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”)contained in the publications on our platform. However, Taylor & Francis, our agents, and ourlicensors make no representations or warranties whatsoever as to the accuracy, completeness, orsuitability for any purpose of the Content. Any opinions and views expressed in this publication arethe opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis.The accuracy of the Content should not be relied upon and should be independently verified withprimary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoevercaused arising directly or indirectly in connection with, in relation to or arising out of the use of theContent.

This article may be used for research, teaching, and private study purposes. Any substantialor systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or

http://www.tandfonline.com/loi/tnzz20

http://www.tandfonline.com/action/showCitFormats?doi=10.1080/03014220909510148

http://dx.doi.org/10.1080/03014220909510148

distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use canbe found at http://www.tandfonline.com/page/terms-and-conditions

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014

http://www.tandfonline.com/page/terms-and-conditions

New Zealand Journal of Zoology, 2009, Vol. 36: 135-1630301-4223/09/3602-0135 © The Royal Society of New Zealand 2009

135

The interactive effects of livestock exclusion and mammalian pestcontrol on the restoration of invertebrate communities in smallforest remnants

RAPHAEL K. DIDHAM1,*

GARY M. BARKER2

JESSICA A. COSTALL3

LISA H. DENMEAD1

CHRISTOPHER G. FLOYD2

CORINNEH. WATTS2

1School of Biological SciencesUniversity of CanterburyPrivate Bag 4800Christchurch 8140, New Zealand

2Landcare ResearchPrivate Bag 3127Hamilton 3240, New Zealand

3Ecology GroupInstitute of Natural ResourcesMassey UniversityPrivate Bag 11222Palmerston North 4442, New Zealand

*Author for correspondence:[email protected]

Abstract In many agricultural landscapes, signifi-cant biodiversity gains can be made by improvingthe ecological condition of degraded remnants ofsemi-natural habitat. Recent emphasis has been onthe level of management intervention required toinitiate vegetation recovery in small forest remnants,but no comparable emphasis has been placed onbenefits for invertebrate communities. In the Wai-kato region, New Zealand, we tested the effects oflivestock exclusion, mammalian pest control, andtheir interaction, on leaf-litter invertebrate commu-nities in 30 forest remnants, using a space-for-timesubstitution approach. Atotal of 87 376 invertebrates

Z09003; Online publication date 23 April 2009Received 30 January 2009; accepted 28 March 2009

were extracted from 964 leaf-litter samples. Inver-tebrate density was an order of magnitude lowerin remnants than in nearby large forest reserves.For key taxa, such as Diplopoda, Isopoda, Coleop-tera and Mollusca, 10- to 100-fold lower densitieswere recorded in remnants with no pest control,particularly where livestock were not excluded. Bycontrast, other taxa such as Thysanoptera and For-micidae (Hymenoptera) had up to 100-fold greaterdensities in remnants with recent stock exclusionand pest control. These changes led to a significantlivestock exclusion x pest control interaction effecton the degree of invertebrate community dissimi-larity between forest remnants and forest reserves.Using structural equation modelling, we found thattreatment effects were largely mediated by a cascad-ing series of indirect causal paths involving alteredsoil chemistry, vegetation composition, and littermass relative to large forest reserves, although thelivestock exclusion × pest control interaction wasinadvertently confounded with differing slopes andareas of remnants in different treatments. Livestockexclusion and mammalian pest control have sig-nificant, but contrasting, effects on invertebrates inthe first 10-20 years following livestock exclusionfrom forest remnants, with mammalian pest controlhaving limited benefit for the leaf-litter invertebratefauna without livestock exclusion.

Keywords Berlese funnel; conservation manage-ment; edge effects; habitat fragmentation; habitatrestoration; leaf-litter invertebrates; patch area; pestcontrol; species invasion; structural equation model-ling

INTRODUCTION

New Zealand justifiably celebrates its place amongthe world's top 25 "biodiversity hotspots" (regionsharbouring more than 1% of the world's endemicplants; Myers et al. 2000). Yet, by and large, wechoose to ignore the fact that these are also hotspotsof extinction threat, defined equally by Myers' second

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014

136 New Zealand Journal of Zoology, 2009, Vol. 36

criterion of severe (>70 %) habitat loss (myers etal. 2000). Nearly half of the administrative districtsof New Zealand have less than 10-15% remainingnative habitat cover, and most of that is heavilyfragmented (ewers et al. 2006). This is particularlytrue of the once extensive lowland forests of NewZealand (Thompson et al. 1983; mfe 2000). Thenon-random pattern of severe native habitat loss infertile lowland regions (ewers et al. 2006), combinedwith a protected areas network built on the acquisi-tion of high-elevation land of low commercial value(Norton 1999; park 2000), has greatly exacerbatedconservation threats in "the other 70%" of the coun-try used for agricultural production. across wideswathes of our lowland agricultural landscapes theextent of indigenous biodiversity loss is unparalleledanywhere in the world (Norton & miller 2000).correspondingly, the conservation value of even thesmallest remnants of semi-natural habitat in theselandscapes is extremely high.

Recognition that lowland forests are poorly repre-sented in the conservation estate has prompted callsto shift the balance of protection and managementtowards small, under-represented forest remnants inlandscapes that have suffered the highest amountsof deforestation (leathwick et al. 2003a). overthe past decade, however, it has become clear thatconservation management on private land bringswith it a unique set of challenges for conservationmanagers (Norton & miller 2000; Newburn et al.2005). For instance, the rich ecological debatesover reserve selection and design criteria in the1980s and 1990s (e.g., Simberloff & abele 1982;Saunders et al. 1991) are moot when there is littlealternative but to accept the remaining habitat in thelandscape, regardless of spatial location or configu-ration (Schwartz 1999; Tscharntke et al. 2002b).

Recent evidence suggests that the relative conser-vation value of small remnants on private land can besubstantial (abensperg-Traun & Smith 1999; oliveret al. 2006; arroyo-Rodríguez et al. 2009). evensmall areas of remnant forest can sustain diverseassemblages of native plants (Whaley et al. 1997)and animals (harris & Burns 2000), compared tothe surrounding agricultural landscape (harris &Burns 2000; Derraik et al. 2005; ewers et al. 2007).Furthermore, c. 20% of New Zealand's threatenedvascular plants are confined to private land, and foran additional 60% of species private land constitutesa significant proportion of their habitat (Norton2000). Nevertheless, it must be acknowledged thatthe small size of many of the remaining lowlandforest remnants limits the number of species they

contain (ogle 1987; Saunders et al. 1991), and thetotal population sizes of remaining species (connoret al. 2000), particularly where species are sensi-tive to edge effects (ewers & Didham 2007; ewerset al. 2007). To complicate matters, the effects ofthese fragmentation processes may only becomefully evident in small remnants after considerablelag-times (the so-called "extinction debt", Tilmanet al. 1994), with an estimated half-life of speciesdecline of 25-100 years for birds (Brooks et al. 1999;Ferraz et al. 2003) and 50-100 years for prairie-dwelling plants (leach & Givnish 1996). Somelong-term studies of vegetation dynamics in smallisolated forest remnants in New Zealand paint a grimpicture with regard to the maintenance of nativeplant diversity. For example, Whaley et al. (1997)recorded the local extinction of one-third of the 122native species from a single 5.2 ha forest remnant(claudelands Bush) in the Waikato region between1956 and 1980. a s a result, plant species whichoccupied habitats that have undergone the greatestreductions in area are more likely to have becomerare or been extirpated than species in habitats thatwere already sparsely distributed (Duncan & young2000). clearly, the legacy of historic habitat loss stillvery much influences current populations (Ewers &Didham 2006b; ewers et al. 2006; ewers & Didham2007), notwithstanding recent statements that "clas-sical problems of ecosystem loss and fragmentationhave largely been countered in some regions byreservation of 30% of total land area" (craig et al.2000, p. 61).

historical habitat loss can have further indirecteffects on native species in forest remnants by alter-ing the relative magnitude of effects of a wide rangeof other threatening processes. For example, theimpact of intermittent livestock browsing and soildisturbance (Jane 1983; Buxton et al. 2001; miller2006), anthropogenic fertiliser inputs (Stevenson2004), altered hydrological regimes (Burns et al.2000), and the impact of exotic mammals (atkinson2001; Wardle et al. 2001) and plant pests (Timmins& Williams 1991) are all likely to be exacerbatedin small remnants with a high proportion of edgehabitat (ewers & Didham 2006b). Recent workhas provided compelling evidence for synergisticinteractions between multiple drivers of global en-vironmental change, including land-use intensifi-cation and species invasion (Didham et al. 2005,2007; Tylianakis et al. 2008). For example, Wiseret al. (1998) found that the invasive exotic herbHieracium lepidulum (Stenstroem) omang (aster-aceae) had the highest probability of occurrence

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014

Didham et al.—Invertebrate communities in forest remnants 137

near forest edges, and Timmins & Williams (1991)noted the invasion of a wide range of exotic plantsinto forest remnants with greater levels of anthro-pogenic disturbance. Invasion of the exotic Trad-escantia fluminensis Vell. (commelinaceae) intoforest remnants caused a significant reduction inspecies richness, abundance and survival rates ofnative forest seedlings (Standish et al. 2001), and asignificant change in the diversity and compositionof terrestrial invertebrate communities (Toft et al.2001; yeates & Williams 2001; Standish 2004), withsubstantial flow-on effects for litter decompositionrates and nutrient availability (Standish et al. 2004).clearly, in the face of severe threats from multiple,interacting drivers of ecological change, preserva-tion alone is not likely to be sufficient to stem theloss of biodiversity from forest remnants withoutmanagement intervention (leathwick et al. 2003a;chazdon 2008).

land managers now recognise these issues morewidely in New Zealand, and recent initiatives ex-plicitly address the need for and importance of na-ture conservation on private land (Norton 2000).Small privately-owned remnants are a key focusof territorial local authorities in the management ofbiodiversity, and in the development of guidelinesfor regional biodiversity protection. open space cov-enants through the New Zealand Queen elizabeth IINational Trust (QeII Trust 1984) allow landownersto protect parts of their property in perpetuity, andthe government has provided funds for the pur-chase or protection of native remnants on privateland through the Nature heritage Fund and NgaWhenua Rahui Trust (Norton & miller 2000). Theseinitiatives have been matched by increases in com-munity involvement in conservation and restorationprojects (mfe 2000; Ritchie 2000; p c e 2001; mfe2002). The remnants managed by >63 000 rural landowners (Statistics NZ 2008) provide unparalleledopportunities to address local-scale restoration oflowland forests, while at the same time achievinghigh-level national goals for conservation. mostimportant of all, there is a steadily growing desireby landowners to maintain and improve the condi-tion of biodiversity on their land (Davis & cocklin2001; mfe 2002). For example, as of November2008, QeII National Trust had over 3500 registeredcovenants, with active management of threats to over100000 ha of privately-owned remnants (anon.2008). These initiatives recognise that managementactions such as fencing will often be required toprevent grazing and trampling by livestock (Burnset al. 2000), and exotic weeds and pests may need

to be controlled (porteus 1993) in order to reducethreats to native biodiversity within remnants. Thesefigures are encouraging and reflect the enthusiasmNew Zealanders have for the protection of theirnatural heritage. however, the degree to which thisenthusiasm has translated into significant conserva-tion gains in small remnants has not been tested.

In this context, the goal of our study was to testwhether livestock exclusion and mammalian pestcontrol, the two management actions most com-monly applied in the conservation management offorest remnants on private land (QeII Trust 1984;Porteus 1993), are sufficient to promote the recoveryof terrestrial invertebrate abundance and compo-sition towards the condition observed in the fewrelatively large forest reserves remaining in thelandscape. although there have been no previousattempts to test the potential conservation benefits ofmanagement intervention for the restoration of na-tive invertebrate communities in severely degradedforest remnants, recent studies testing the effectsof livestock exclusion on plant community dynam-ics and physical soil properties in forest remnants(Burns et al. 2000; Smale et al. 2005; Dodd & power2007; Dodd et al. 2008; Smale et al. 2008) suggestthat strong responses might also be expected amongsoil- and litter-dwelling invertebrates.

We test for direct effects of time since livestockexclusion and mammalian pest control on litterinvertebrate density and community structure, high-lighting in particular the antagonistic interactioneffects that can occur between management treat-ments. We also attempt to tease apart the indirectmechanisms by which management treatments mightaffect invertebrates via altered soil structure and geo-chemistry due to trampling, altered plant structureand composition due to browsing, and altered litterstructure and biomass resulting from more com-plex feedbacks between soil, vegetation and litterprocesses (milchunas & lauenroth 1993; Wardleet al. 2001). Here, we present the first phase in ourinvestigation, in which we take a higher taxonomicapproach to the detection of the key invertebratetaxa responding to habitat fragmentation and sub-sequent conservation management intervention. Indoing so, we adopt the approach of Biaggini et al.(2007) and others (Kremen et al. 1993; pik et al.1999; andersen et al. 2002; Nakamura et al. 2007)in identifying general trends across the entire inver-tebrate fauna, laying the foundation for species-levelanalyses testing management effects on detritivorousinvertebrates and their role in litter decompositionprocesses (Barker, Watts & Didham, unpubl.).

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


Fig. 1 map of the study area inthe Waikato region, North Island,New Zealand, showing the spatiallocations of the 30 forest remnantsand three forest reserves (Te miroScenic Reserve, maungatautarimountain Scenic Reserve andKarakariki Scenic Reserve) inwhich leaf-litter invertebrates weresampled.

METHODS

Study areaThe study was conducted in the Waikato region ofthe central North Island of New Zealand (Fig. 1).In the central portions of the region (the Waikato,Waipa, matamata-piako and South Waikato admin-istrative districts) surrounding the city of hamilton,only 9% of the original 800 000 ha forest area hassurvived human settlement (ewers et al. 2006).much of the remaining forest consists of small forestremnants dominated by tawa (Beilschmiedia tawa(a.cunn.) Benth. et hook.f. ex Kirk; lauraceae) onmoderate hill country from 100-400 m a.s.l. (landEnvironments of New Zealand, LENZ, classifica-tion D2, F1, F6, F7; leathwick et al. 2003b). mostof the 5000 or so separate remnants in the lowlandsand rolling hill country of the Waikato region are onprivate land, and 96% of these are less than 25 ha inarea, with little or no conservation protection.

In addition to the significant component of tawain the canopy and subcanopy at all sites sampledin this study, other important tree species in oneor more remnants included Knightia excelsa R.Br.(proteaceae), Laurelia novae-zealandiae a.cunn.(atherospermataceae), Litsea calicaris (a.cunn.)Benth. et hook.f. ex Kirk (lauraceae), Dysoxy-lum spectabile (G.Forst.) hook.f. (meliaceae), andMelicytis ramiflorus J.R.Forst. et G.Forst. (Viol-aceae), as well as several other canopy species found

predominantly at the forest reserve sites, includingElaeocarpus dentatus (J.R.Forst. et G.Forst.) Vahl(elaeocarpaceae), Metrosideros robusta a.cunn.(myrtaceae), Weinmannia racemosa l.f. (cuno-niaceae) and the conifers Prumnopitys taxifolia(D.Don) de laub., Prumnopitysferruginea (D.Don)laubenf. (prumnopityaceae), Dacrydium cupressi-num lamb., Dacrycarpus dacrydioides (a.Rich.)de laub., and Podocarpus totara/hallii G.Benn. exD.Don (podocarpaceae) (Smale et al. 2008; Floyd,Burns, Smale & arnold unpubl.).

Sampling designA Geographic Information System (GIS) analysiswas used to identify all forest remnants in l e N Zclasses D2, F1, F6 and F7 in the Waikato, Waipa,matamata-piako and South Waikato administrativedistricts. Subsequently, extensive phone interviewsand site visits with landowners were conducted forover 100 forest remnants in the c. 0.5-50 ha sizerange, in order to select a final subset of 47 remnantsthat had been subject to a factorial combination ofdiffering types and intensities of management ac-tion: (1) four levels of livestock exclusion (fencing),crossed with (2) two levels of control of introducedmammalian omnivores (pest control). In the live-stock exclusion treatment, remnants were selectedbased on four nominal classes of time since livestockexclusion: F0, fenced 0-1 years ago (i.e., no fence orlivestock excluded by fencing within the past year);

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


F6, fenced 2-9 years ago; F12, fenced 10-15 yearsago; and F42, fenced 22-67 years ago. a remnantwas assigned to a livestock exclusion treatment classonly if it was known that a high quality fence hadbeen present, and had evidently excluded livestockgrazing for the entire treatment period. Within eachnominal time since livestock exclusion class, rem-nants were subsequently selected on the basis ofmammalian pest control treatments: p0, no or little(sporadic, and/or low-intensity) pest control; andPI intensive, sustained pest control (i.e., involvingcontinual use of poison or trapping stations for aminimum of the previous 2 years, but usually >10years, with at least annual repeats using at least 1trap/bait station per hectare, with full coverage of theremnant). pest control was predominantly targetedat brushtail possums (Trichosurus vulpecula (Kerr);phalangeridae), but may also have impacted otheromnivorous species such as ship rats (Rattus rattus(linnaeus); muridae) which are abundant in forestremnants (Innes, King, Bridgman, Fitzgerald &arnold, unpubl.). Due to the limited availability ofremnants that had been subject to some combinationsof pre-existing management treatments, the numberof remnant replicates varied from 3 to 8 within theeight livestock exclusion × pest control treatmentcombinations. Because of their clumped spatialdistribution in the landscape, remnants tended to beclustered in three major areas (Te miro, Whatawhata,and maungatautari-Te Waotu; Fig. 1), and therewas potentially confounding spatial autocorrelationof remnant location across treatments (particularlydue to widespread, landscape-scale possum cullingfor tuberculosis control in the maungatautari-TeWaotu area), and this was explicitly factored intomultivariate analyses.

The effects of livestock exclusion and pest controltreatments on plant community structure (Floyd,Burns, Smale & arnold, unpubl.) and leaf-litterdecomposition rates (Barker, Watts and Didham,unpubl.) were tested across all 47 remnants, but itwas not logistically feasible to sample invertebratesat all sites. Therefore, invertebrates were sampledfrom leaf litter in 30 forest remnants selected fromthe full set of 47 remnants, based on accessibility,while ensuring the inclusion of three or four repli-cates in each treatment combination.

a s a reference point for changes in invertebratecommunity composition in forest remnants, inverte-brate communities were also sampled in three forestreserves, representing some of the largest availablereference sites classified in the same LENZ cat-egories as the forest remnants throughout the study

area: Te miro Scenic Reserve (402.8 ha) in the Temiro area, Karakariki Scenic Reserve (5500 ha) inthe Whatawhata area, and maungatautari mountainScenic Reserve (3363 ha) in the maungatautari-TeWaotu area (Fig. 1). The reserves had larger areasand lower intensities of recent anthropogenic distur-bance than remnants.

In each of the 30 remnants and three forest re-serves, 30 sampling points were selected along amatrix-to-forest interior gradient using a fully-ran-domised block design. The forest edge (0 m) wasdefined by the position of the trunks of the outermosttrees that formed an unbroken canopy. Negative dis-tances from edge were assigned to sites outside theforest edge, while positive distances were assignedto sites within the forest. a l l edges were selected onthe northern side of the remnant, and were adjacentto open pasture matrix habitat. Sampling points wereselected at five random distances from the forestedge, within each of six fixed distance intervals cen-tred at 0, 5,10,20,40 m and "core" distances fromedge; i.e., within distance intervals of-2.5 to 2.4 m,2.5-7.4 m, 7.5-14.9 m, 15-29.9 m, 30-59.9 m and60 m-core. each sampling point was randomly off-set by ±10 m perpendicular to the edge-to-interiorgradient (i.e., parallel to the forest edge) in order togive a good degree of spatial separation betweensamples. The distance to the remnant core variedwith patch area, from 15 to 200 m from the edge.a s sampling effort was equal at the patch level (n= 30 sampling points), small remnants in which thecore was less than 60 m from the edge had a greaternumber of replicates allocated (proportionally) intothe near-edge distance intervals. Sampling intensitywas purposely higher close to the forest edge thanin the forest interior, because the rate of change ininvertebrate abundance and composition is known tobe greatest near the forest edge (Didham et al. 1998;ewers et al. 2007).

Sampling leaf-litter invertebrates

one leaf-litter sample was collected in a 33 cm di-ameter circular frame (0.086 m2) at the 30 selectedsampling distances in each of the 30 remnants andthree forest reserves, giving a total of 990 leaf-littersamples collected from 4 December 2007 to 19February 2008. a t each remnant all 30 leaf-littersamples were collected on the same day so thatputative edge gradients in invertebrate density orcomposition were reliably captured, without intro-ducing potential bias from sampling different partsof the edge gradient on different days. Within logis-tical constraints of researcher movement throughout

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


the study area, daily sampling was randomly allo- exclusion by pest control treatment combinationcated between different treatment classes to prevent ranged from 3 to 8 weeks). a l l leaf litter and friablebias arising from seasonal variation in activity pat- humus was scraped rapidly from the frame andterns of invertebrates (variation in the seasonal placed in a large bag-sieve to minimise invertebratespread of sampling dates within a given livestock escape. The material was immediately sieved over

Table 1 Treatment variables, potential confounding variables and sample-level, plot-level and patch-level envi-ronmental correlates used in the ordination analyses and structural equation model. NZmG = New Zealand metricGrid mapping system; SSh-mDS = semi-strong hybrid multidimensional scaling;† = variable removed in ordinationanalysis due to multicollinearity.

code explanation Units

Treatment variables and associated interaction effectsl vexc l Number of years since livestock were excluded from the remnant (0 = not excluded) yearspstctr presence (p1) or absence (p0) of pest control binaryl x p livestock exclusion × pest control interaction (cross-product of deviance scores)edge Distance from forest edge into forest interior ml x e livestock exclusion by distance from edge interaction (cross-product of deviance scores)pxe pest control by distance from edge interaction (cross-product of deviance scores)a x e patch area by distance from edge interaction (cross-product of deviance scores)l x a livestock exclusion by patch area interaction (cross-product of deviance scores)p x a pest control by patch area interaction (cross-product of deviance scores)

Potential confounding variablesDate Sampling datex Modified spatial coordinate representing longitude (eastings in NZmG units divided by 1000) (see text)y Modified spatial coordinate representing latitude (northings in NZmG units divided by 1000) (see text)xy linear component of spatial trend surfacex2 Quadratic longitudinal component of spatial trend surfacey2 Quadratic latitudinal component of spatial trend surfacex2y Quadratic latitudinal and longitudinal component of spatial trend surfacexy2 Quadratic latitudinal and longitudinal component of spatial trend surfacex3 cubic longitudinal component of spatial trend surfacey3 cubic latitudinal component of spatial trend surface

Environmental correlatesSample-level variables

Smplit Dry-weight litter mass of the sample from which invertebrates were extracted gPlot-level variables

pltSdl Number of seedlings in a 0.5 m radius plotpltSpl Number of saplings in a 2.5 m radius plotpltasp aspect at each seedling plot (absolute deviations from North) °pltSlp Slope at each seedling plot °plt l i t average dry-weight litter mass in the vicinity of each seedling plot g m–2

Patch-level variablesl ga rea log1 0 patch area haSI Shape index (see text for formula)5kcore proportion of indigenous forest cover within a 5 km radius of the "core" vegetation plotelev elevation above mean sea level mmDScli axis 1 scores of an SSh-mDS ordination on nine climate variables (see text)mDSsoi axis 1 scores of an SSh-mDS ordination on 10 edaphic variables (see text)Basal Total tree basal area in an 11.3 m radius plot at the remnant "core" m2 ha–1

VegRic Total plant species richness in an 11.3 m radius plot at the remnant "core"opcan proportion of open canopy above an 11.3 m radius plot at the remnant "core"mDSveg axis 1 scores of an SSh-mDS ordination on the basal areas of 38 plant species (see text)mDShis Dissimilarity of axis 1 and 2 scores of current versus historical predicted vegetation

composition in SSh-mDS ordination

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


a 10 mm mesh by vigorously shaking the bag-sievefor approximately 5 min. The fine, sieved littercontaining invertebrates was then transported to thelaboratory in individual cotton bags. after the siev-ing process, the coarse "top fraction" of remaininglitter was placed in a plastic bag and later dried andweighed to obtain a combined estimate of samplelitter mass.

Invertebrates were extracted from the sievedleaf litter over a 72-h period using Berlese funnels(BioQuip® collapsible bag design #2832, RanchoDominguez, california) (Wheeler & mchugh 1987).We operated 60 Berlese funnels continuously for3 months. Subsequently, invertebrates were sortedand identified to invertebrate Phylum, Class andorder, and counted. Two samples were excludeddue to labelling error, and 24 samples were notsorted due to time constraints, giving a total of 964samples sorted (with four sites having 20,21,25 and28 samples sorted out of 30).

Measurement of explanatory variablesIn addition to nine treatment variables imposed bythe study design (livestock exclusion, pest control,distance from forest edge and their interactions witheach other and with patch area), we measured 10 po-tential confounding variables and 17 sample-, plot-,and patch-level environmental variables to identifypotential determinants of invertebrate response tomanagement action (Table 1).

Treatment variables

To test for potential interaction effects, three vari-ables were calculated representing livestock exclu-sion × pest control, livestock exclusion × distancefrom edge, and pest control × distance from edgeinteractions. In addition, three interaction variableswere calculated between the treatment design vari-ables and patch area (livestock exclusion × patcharea, pest control × patch area, and distance fromedge × patch area). To avoid problems of linear de-pendency and collinearity in multivariate analyses,the interaction terms were calculated by multiplyingdeviation scores (i.e., each of the values minus theirrespective grand mean), rather than the raw scorecross-products (Kline & Dunn 2000).

Potential confounding variables

although every effort was made to randomise the or-der of sampling of treatment replicates across dates,the potential confounding effect of sampling date oninvertebrate community composition was explicitlytaken into account in multivariate analyses.

Further, a cubic trend surface approach was usedto remove potentially confounding spatial autocor-relation from the data (Borcard et al. 1992; Davies etal. 2003). This approach utilises nine variables rep-resenting linear, quadratic, and cubic combinationsof New Zealand metric Grid (NZmG) northing andeasting values (x, y, xy, x2, y2, x2y, xy2, x3, y3) of theinterior sampling site at each remnant or reserve.The NZmG coordinates were expressed relative tothe minimum northing and easting values within thedataset, and divided by 1000 prior to analysis. Theuse of fitted trend surfaces is a conservative approachto dealing with potentially autocorrelated data (seeewers et al. 2007).

Sample-level correlatesFor each of the 964 leaf-litter samples, the coarse topfraction of litter removed during the sieving proc-ess in the field, and the fine sieved fraction of litterreturned to the laboratory for invertebrate extraction,were both oven dried at 80°c for 24 h and the result-ing dry mass values combined to give a single meas-ure of sample litter mass (range: 13.2-1402.5 g).

Plot-level correlates

a t (up to) six distance intervals (0, 5, 10, 20, 40 mand "core") into the 30 remnants and three forest re-serves (184 plots in total), the number of woody plantseedlings (15 cm < height < 1.4 m) was recorded ina 0.5 m radius area (0.79 m2) (range: 0-40) and thenumber of woody plant saplings (>0.5 m in heightand <3 cm DBh) was recorded in a 2.5 m radiusarea (19.6 m2) (range: 0-55). In addition, the aspect(recorded as absolute deviations from North) (range:0-180°) and the slope (range: 0-45°) were recordedat each plot, and standing mass of leaf litter on theforest floor was estimated by collecting two leaf-littersamples with a 33 cm diameter circular frame, anddrying them (80°c, 24 h) to obtain a single averagevalue of dry mass (range: 64.7-6219.4 g m–2).

Patch-level correlates

For each of the 30 remnants and three forest reserves,patch area (range: 0.9-27.3 ha for remnants, and402.8-5500.0 ha for forest reserves) was determinedfrom a GIS analysis of the land cover Database(lcDB2) (Terralink 2004), and patch shape index(SI) was calculated using the formula SI = perimeter•*• 200(π.Area)0.5 (patton 1975), where patch perim-eter (m) was determined from lcDB2. The SI pro-vides a measure of deviations from circularity, witha perfectly circular remnant having an SI value of1.0, and remnants with increasingly more complex

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


shapes having greater SI values (range: 1.1-23.0).A 5 km radius circular buffer around the core ofeach remnant and reserve was created in the GIS,and the proportion of the landscape in indigenousforest cover was calculated within this buffer area(range: 0.03-0.52).

a t the "core" plots, a range of topographic, cli-matic, edaphic and vegetation variables were alsomeasured, or derived from spatial databases of en-vironmental attributes maintained as GIS grids forgeographic locations in NZmG coordinates. First,elevation (range: 99-356 m a.s.l.) was derived fromthe New Zealand 25 m resolution digital elevationmodel. Next, climate parameters for each remnantand reserve were derived from 100 m resolution GISsurfaces interpolated from meteorological stationclimate data (30-year period 1950-80) (leathwick& Stephens 1998): mean annual temperature (range:7.0-13.9°c); mean temperature of the coldest month(July) (range: 0.4-4.6°c); mean temperature in driestmonth (march) (range: 10.7-17.1°c); mean annualsolar radiation (range: 74.8-165.6 mJ m–2 day–1);mean solar radiation during winter (June) (range:4.5-75.0 mJ m–2 day–1); vapour pressure deficit inthe windiest month (october) (range: 27.0-34.0kpa); mean annual rainfall (range: 1339-1613 mm);absolute variation in annual rainfall (range: 148—153 mm); and the ratio of mean annual rainfall topotential evapotranspiration (range: 2.7-3.0). Themeteorological stations were predominantly on flatterrain, whereas actual solar radiation on hilly terrainis strongly influenced by topography (Antonic 1998),so the estimates of solar radiation and temperaturefor each location were corrected for slope and aspectusing an empirical method developed for the NorthIsland, New Zealand (mcaneney & Noble 1976).Similarly, since increasing slope reduces soil mois-ture status (Radcliffe & lefever 1981), the estimatesof rainfall received at sites were adjusted by thecosine of the slope angle. a composite measure ofoverall variation in climate among remnants wasobtained from the axis 1 scores of a semi-stronghybrid multidimensional scaling ordination (Belbin1991), using the Gower (1971) dissimilarity metric,implemented in the paTN software package (Bel-bin 1995). The climate ordination had a good fit tothe data (stress = 0.076), and of the nine measuredvariables the gradient in ordination scores was moststrongly correlated with mean annual temperature(r = 0.95), variation in solar radiation in winter (r= 0.94), mean temperature in the driest month (r =0.95), mean temperature of the coldest month (r =0.94) and mean annual solar radiation (r = 0.94).

edaphic parameters were measured in a singlebulked soil sample taken from the "core" in eachremnant and forest reserve with a 25 mm diameterhoffer soil-corer (0-10 cm depth), using standardanalytical procedures (Blakemore et al. 1987; seewww.landcareresearch.co.nz/services/laboratories/eclab/eclabmethods_soils): ph (range: 3.8-6.3); to-tal nitrogen (range: 0.2-1.8%); total carbon (range:3.6-29.8%); available phosphorus (olsen, range:2.7-139.7 mg/kg); exchangeable calcium (range:2.9-24.2 cmol/kg); exchangeable potassium (range:0.4-2.7 cmol/kg); exchangeable magnesium (range:1.2-7.2 cmol/kg); exchangeable sodium (range:0.1-0.8 cmol/kg); total exchangeable cation capac-ity (range: 16.2-54.9 cmol/kg); and base saturation(range: 13.7-80.6%). a composite measure of over-all variation in edaphic conditions among remnantswas obtained from the axis 1 scores of a SSh-mDSordination, using the Gower dissimilarity metric,implemented in paTN. The edaphic ordination hada good fit to the data (stress = 0.085), and of the10 measured variables the gradient in ordinationscores was most strongly correlated with varia-tion in total carbon (r = -0.92), total nitrogen (r =-0.86), base saturation (r = 0.71), p h (r = 0.71) andexchangeable cation capacity (r = -0.66).

Vegetation structure and composition were meas-ured in one 11.3 m radius plot (400 m2) at the "core"distance inside each remnant and forest reserve.combined measures of total tree basal area (range:42.2-151.9 m2 ha–1), total plant species richness(range: 20-55 species per plot), and proportion opencanopy (range: 0.05-0.17) were recorded in eachplot, as well as basal areas of each of 38 woodyplant species (>3 cm DBh). a composite measure ofoverall variation in tree species composition amongremnants was obtained from the axis 1 scores ofa SSh-mDS ordination on the basal areas of the38 tree species, using the Bray & curtis (1957)dissimilarity metric, implemented in paTN. Thevegetation ordination had a moderate fit to the data(stress = 0.164), and of the 38 measured variablesthe gradient in ordination scores was most stronglycorrelated with variation in B. tawa cover (r = 0.60),P. totara cover (r = 0.52), Pseudopanax crassifolius(Sol. ex a.cunn.) c.Koch cover (r = 0.51), Macro-piper excelsum (G.Forst.) cover (r = 0.44) and D.dacrydioides cover (r = 0.42).

a s a measure of the long-term history of canopydisturbance at each site, current tree species com-position observed in the "core" plot was comparedwith the predicted tree species composition at thatlocation, estimated from 100 m resolution spatial

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


predictions of the potential natural vegetation of eachsite derived from generalised additive regressionmodels (leathwick 2001; J. overton & c. Dischingerunpubl.; www.derivedbd.landcareresearch.co.nz).For the same 38 tree species as observed in currentvegetation plots, we compared variation in predictedbasal area (m2) of tree species among remnants us-ing a SSh-mDS ordination, with the Bray-curtisdissimilarity metric, implemented in paTN. Fromthe biplot of SSh-mDS axes 1 and 2 scores, weestimated the historical change in vegetation com-position in each remnant as the euclidean distancebetween observed and predicted vegetation composi-tion in ordination space.

ANALYSES

Variation in invertebrate density withinand among forest remnantsThe number of individual invertebrates per quadratwas converted to no. m–2 prior to analysis. We testedtwo a priori expectations for variation in inverte-brate density within and among forest remnants.First, we expected that the impact of livestock en-croachment and mammalian omnivores should begreatest at forest edges, and in small remnants, dueto spatial subsidisation in the surrounding landscapematrix, resulting in positive slopes of edge effects oninvertebrate abundance in the absence of livestockexclusion or pest control (F0p0); i.e., edges mightbe degraded relative to remnant interiors. Therefore,we expected that the greatest immediate responseto management action would be observed at edgelocations, resulting in a decreasing slope of edge ef-fect on invertebrate abundance with increasing timesince livestock exclusion and commencement ofintensive pest control (although disturbance-adaptedtaxa might show the opposite response compared toforest-dependent taxa). consequently, slopes of edgeeffects were calculated using simple least squaresregression on log10 (x + 1) transformed no. m–2, withdistance from edge on a linear scale, and the slopeswere compared across treatment classes.

We also expected that management actions shouldimprove "ecological condition" within remnant inte-riors over time. Differences in invertebrate densityin the interiors of forest remnants subject to differ-ent livestock exclusion and pest control treatmentswere best compared using the fitted edge responsefunctions to calculate standardised, model-predictedabundances at a nominally-selected distance of 40 m

from the forest edge (which was the maximum coredistance of many of the remnants). predicted valueswere derived from the best-fit edge response func-tion (null or linear in all cases) (ewers & Didham2006a), in order to overcome the differing sizes offorest remnants and the differing edge responses ofdifferent taxa at different sites. Negative predicteddensity values were assigned a zero value.

Variation in invertebrate communitycomposition among remnantsVariation in the relative abundance of invertebrateclasses and orders between the 30 remnants andthree forest reserves was quantified using a con-strained canonical correspondence analysis ( c c a )ordination on log10 (x + 1) transformed sample-levelabundance data. First, an unconstrained Detrendedcorrespondence analysis (Dca) ordination wascarried out to confirm that the gradient lengths inthe species-abundance data were appropriate to theunimodal (chi-square) distance metric underlyingthe c c a ( D c a axis 1: 3.858). We next assessedcollinearity among the 36 treatment, environmentaland potential confounding variables using a correla-tion matrix, and found that there were seven pairsof variables that were highly intercorrelated (r >0.9). one of each pair of intercorrelated variableswas removed (the spatial variables x2, y2, x2y, xy2,x3, y3, and the shape index, SI), leaving a total of 29variables in the analysis. Sampling date and spatialautocorrelation among sites (x, y, xy) had significantconfounding effects on invertebrate composition andwere included as covariables in a subsequent partialCCA (pcca ) analysis. We used a forward selectionprocedure to rank the measured variables in order oftheir importance in explaining invertebrate composi-tion (ter Braak & Verdonschot 1995). This process isanalogous to forward stepwise regression, and sig-nificance was tested at each step using a Monte Carlopermutation test with 999 random permutations. Theoverall significance of the final pCCA ordination wasalso tested with a monte carlo permutation test, us-ing 999 permutations. a l l ordination analyses wereconducted using c a N o c o version 4.02 (ter Braak& Smilauer 1997).

although the p c c a ordination provides a usefulcomparison of invertebrate composition betweenforest remnants and the large-area reference sites, itdoes not allow a direct test of the effects of manage-ment actions on the remnants themselves, becausethe inclusion of the reference sites in the ordinationalters the relative weighting of treatment and envi-ronmental correlations. Therefore, treatment effects

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


were more effectively discriminated in subsequentanalyses utilising patch-level dissimilarity valuescalculated from the ordination. From the p c c abiplot of axes 1 and 2 sample scores, the degree ofdissimilarity in invertebrate community compositionwas calculated between each of the 30 remnantsand their geographically nearest continuous forestreference site. To do this, we calculated the meanand 95% confidence limits of the euclidean distancebetween the weighted average (Wa) sample scoresfor each leaf-litter sample in a given remnant and theaverage (centroid) of the Wa sample scores for thegeographically nearest large forest area that acted asa local-area reference point.

Discriminating the direct and indirect effectsof management actionsWe used structural equation modelling (Sem) to dis-criminate the relative direct, indirect and total effectsof livestock exclusion, pest control, and a livestockexclusion × pest control interaction on invertebratecommunity dissimilarity, using amos version 16.0(arbuckle 2007). From the p c c a ordination analy-sis it was clear that there was not only a significantinteraction effect between the livestock exclusionand pest control treatments, but the treatment effectsalso covaried with the effects of patch area and slope,and were intercorrelated with multiple environmen-tal variables. In situations such as this, structuralequation modelling offers a superior alternative fordiscriminating the underlying mechanisms of effect,as it allows variables in each model to be specifiedas both a potential predictor variable and a potentialresponse variable, enabling the causal structure of acomposite path model to be tested (Kline 2005).

In our Sem, we recognised three causal pathwaysof treatment effects: (1) indirect effects mediatedby environmental variables, (2) indirect effects viashared correlations with patch slope and area, and(3) direct effects (of unknown mechanism). First,we were primarily interested in determining therelative importance of the indirect factors mediatingthe influence of livestock exclusion, mammalian pestcontrol and their interaction on invertebrate commu-nity dissimilarity. Because the number of potentialindirect mediating variables that we measured wasmuch larger than the number of degrees of freedomavailable to test multiple paths in the model, westarted from the premise of selecting only one vari-able to represent each of the soil, vegetation and litterstructural variation in the system. We compared theraw correlations between environmental variablesand invertebrate dissimilarity as an initial indicator of

the variables most likely to mediate treatment effectsin the Sem model, testing correlations of both abso-lute values of the environmental variables, as wellas relative values calculated as dissimilarity to thenearest continuous forest reference site (as was donefor invertebrate dissimilarity). The three variablesselected to have the greatest explanatory power werea soil variable represented by the absolute values ofaxis 1 scores from the soils SSh-mDS ordination,a vegetation variable represented by relative dis-similarity values calculated between axis 1 and 2scores of forest remnants and their nearest continuousforest reference site in the plant species SSh-mDSordination, and a litter structure variable representedby absolute values of average sample litter mass.

Second, in addition to indirect mediated path-ways in the Sem, the total effects of the treatmentvariables may also be influenced by indirect effectsvia joint correlations with underlying spatial ortopographic characteristics of the remnants that mayalso influence invertebrate composition. Moderatelystrong correlations were observed between the treat-ment variables and patch slope and patch area, sothese variables were entered into the model with allpossible combinations of their effects on invertebratedissimilarity via soil chemistry, vegetation dissimi-larity and litter mass.

Third, there could also be residual direct effectsof livestock exclusion and pest control that representvariance in unmeasured proximate mechanisms ofeffect, such as the provision of dung resources bylivestock (abensperg-Traun et al. 1996; hanski etal. 2008), or the direct predation of invertebrates byintroduced mammals (Daniel 1973; cowan & moeed1987; Dugdale 1996; Fitzgerald & Gibb 2001).

The full Sem model was tested using a maxi-mum likelihood (ml) approach (Kline 2005). of thenine variables in the full Sem, time since livestockexclusion, patch area, patch slope and litter masswere log10-transformed prior to analysis to meetthe assumptions of m l and multivariate normality.The SEM determines standardised path coefficientsamong variables, which are equivalent to standard-ised regression coefficients, and these are used toquantify the direct effects of an independent vari-able on a dependent variable, while controlling forthe effects of other independent variables (mitchell2001). To find the most parsimonious SEM modelwith the minimum adequate suite of paths neces-sary to explain variation in invertebrate composi-tion among forest remnants, we compared multiplehierarchical models using a stepwise specificationsearch in amos 16.0.1 (arbuckle 2007), where all

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


paths directly and indirectly affecting compositionwere fixed as optional. To select the best fittingmodel from the stepwise specification search re-sults, we used the minimum discrepancy function( MIN), adjusted for sample size, MIN /d.f. (Grace2006). lower MIN/d.f. values indicate good modelfit, where ( ^ /d.f. 2 is an acceptable minimumvalue (Bollen 1989). aprobability value of P 0.05indicates that the null hypothesis cannot be rejectedand the predicted correlations and covariances ofthe model equal their observed counterparts (Kline2005). Finally, we used bootstrapping with 1000random samples generated from the observed covari-ance matrix to estimate 90th percentile confidenceintervals and significance values for the standardiseddirect, indirect and total effects (Kline 2005) in thefinal, most parsimonious model.

RESULTS

Variation in invertebrate density withinand among forest remnantsA total of 87 376 invertebrates in 31 higher taxonom-ic groups (not including acari or collembola) wereextracted from the 964 leaf-litter samples (range:1-2518 individuals per sample) (Table 2). The faunawas numerically dominated by the phylum annelida(6011), the myriapod class Diplopoda (15 429), thecrustacean orders amphipoda (7351) and Isopoda(2739), and by the insect orders coleoptera (11 880adults, 3837 larvae), Thysanoptera (11 671), Diptera(352 adults, 10 140 larvae), hymenoptera (2783Formicidae, 1290 other families) and hemiptera(3842). The average densities of most taxa variedwidely across sites, with some taxa varying in den-sity by up to two orders of magnitude between the30 forest remnants and the three forest reserves, andby one to two orders of magnitude between remnantsin the different livestock exclusion and pest controltreatment classes (Table 2).

For total invertebrate density, there was no strongevidence of significant edge effects in either theforest reserves (except a weakly significant positiveedge effect at maungatautari mountain SR), or inthe 30 forest remnants (except for weakly significantnegative edge effects in four remnants) (Fig. 2a).consequently, there was no evidence that livestockexclusion or pest control treatments had any influenceon the slope of edge effects in total invertebrate den-sity (Fig. 2a). however, total invertebrate density inthe "remnant interior" (i.e., density predicted by the

best-fit null or linear edge-model at a standardiseddistance of 40 m from the edge) was dramaticallylower in many of the forest remnants compared withthe forest reserves (Fig. 2B), and there was a clearindication that this was related to the absence ofpest control in these remnants (squares indicate nopest control, circles indicate pest control in Fig. 2B).eight out of the 10 remnants with the lowest totalinvertebrate densities (100-500 m–2) were in theno pest control treatment, whereas nine out of the10 remnants with the highest invertebrate densities(1000-2500 m–2) were in the pest control treatment(Fig. 2B). By contrast, there was no clear indicationof an effect of time since livestock exclusion on totalinvertebrate density (Fig. 2).

a s observed for total invertebrate density, therewere predominantly non-significant or relativelyweak edge effects on the densities of most indi-vidual taxa in the majority of forest remnants andforest reserves (representative examples shown inFig. 3A—D). However, when significant edge ef-fects were detected, these were almost invariablyobserved in the smallest forest remnants (indicatingan interaction between patch area and distance fromforest edge), and these also tended to be remnantsin the p1 pest control treatment class. positive edgeeffects indicated more degraded edge habitat for taxasuch as coleoptera (Fig. 3a), Isopoda (Fig. 3B),mollusca (Fig. 3c) and Diplopoda (not shown), andan increase in suitable habitat near edges for taxasuch as Thysanoptera (Fig. 3D) and Formicidae (notshown), in the smallest forest remnants. only formollusca (Fig. 3c) was there also some indicationthat time since livestock exclusion had a detectableeffect on the slopes of edge effects for invertebratetaxa, with weak positive slopes for recently-fencedremnants (F0 and F6) and weak negative slopes forlong-fenced remnants (F12 and F42).

In contrast to the weak evidence for edge effectson invertebrate density, there was strong evidencefor substantial changes in the densities of inverte-brate taxa in the interiors of most forest remnants,compared to densities at the same distance from theedge of forest reserves (Fig. 3e-h) . For example,coleoptera (Fig. 3e) and mollusca (Fig. 3G) showeda 10-fold decline in density in forest remnants, andIsopoda (Fig. 3F), hemiptera (not shown) andDiplopoda (not shown) had up to a 100-fold declinein density compared to the interior of the forest re-serves. meanwhile, Formicidae (not shown) had upto a 10-fold increase in density in forest remnants,and Thysanoptera (Fig. 3h) had up to a 100-foldincrease in density compared to the forest reserves.

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014

Table 2 Mean (±1 SD) densities (no. nr2) of terrestrial invertebrates (excluding Acari and Collembola) extracted from 33 cm diameter leaf-litter samples in theWaikato Region, New Zealand, using Berlese funnels. Data are collated for multiple forest remnants sampled in each of eight livestock exclusion by mammalian pestcontrol treatment classes (PO, no pest control; P1, pest control) (FO - F42, time since livestock exclusion, indicating the median number of years since livestock wereexcluded from remnants in each class). TMRes = Te Miro Scenic Reserve; MauRes = Maungatautari Mountain Scenic Reserve; KaraRes = Karakariki Scenic Reserve."Other taxa" comprises Protura (n = 39), Symphyla (24), Isoptera (16), Diplura (14), Neuroptera adults (2) and larvae (11), Archaeognatha (10), Orthoptera (9),Blattodea (7), Dermaptera (7), Hymenoptera larvae (3), Platyhelminthes (3), Trichoptera (3), Pauropoda (2), Siphonaptera (2), Cladocera (1) and unknown (4).

Pest control

Livestock exclusionNo. forest remnants

No. samples

Phylum ArthropodaInsectaColeoptera Adults

Larvae

Diptera Adults

Larvae

Hemiptera

Hymenoptera:Formicidae

Other

Lepidoptera Adults

Larvae

Psocoptera

Thysanoptera

MyriapodaChilopoda

Diplopoda

FO4

120

94.6(149.6)

31.2(39.8)

1.9(5.7)75.7

(131.8)19.4

(32.7)20.5

(39.6)13.8

(20.4)0.1

(1.1)23.4

(29.2)18.0

(36.0)43.6

(72.3)

0.8(3.3)80.6

(119.7)

F64

120

128.2(154.8)

30.0(53.4)

4.2(8.1)

103.3(123.5)

32.2(44.2)25.7

(79.7)14.2

(19.2)0.5

(2.3)15.1

(24.6)6.3

(18.7)85.5

(128.3)

2.5(7.6)

168.7(245.5)

P0

F12388

129.9(246.1)

29.5(37.6)

5.4(14.9)190.0

(274.7)29.1

(49.6)24.6

(47.9)9.6

(13.6)0.4

(2.1)15

(19.7)5.6

(14.6)45.2

(54.7)

1.9(4.7)

170.3(216.0)

F42390

72.7(155.2)

24.0(30.2)

1.7(7.1)67.8

(76.4)30.8

(48.5)5.3

(15.5)10.8

(25.5)

29.0(29.2)

4.8(10.4)131.3

(237.2)

0.3(1.7)60.3

(203.2)

F04

120

62.2(76.8)72.4

(114.1)2.3

(6.5)75.0

(133.9)61.2

(82.5)140.9

(323.8)26.6

(34.6)0.3

(1.8)62.3

(80.4)46.5

(121.8)261.3

(295.2)

0.6(2.6)65.7

(143.5)

F64

115

118.4(128.1)

45.2(96.0)

6.7(13.9)132.8

(173.9)71.2

(121.8)28.3

(59.5)21.4

(27.7)0.2

(1.5)43.2

(71.9)17.3

(38.8)193.3

(214.9)

2.2(7.2)

204.9(287.8)

PI

F124

110

196.5(253.4)

58.2(134.2)

2.7(6.7)

102.7(96.4)34.5

(47.5)15.7

(68.9)10.9

(18.5)0.3

(1.9)13.3

(15.2)7.0

(15.6)165.4

(459.6)

2.0(6.3)

210.5(350.6)

F424

111

194.3(202.3)

73.4(87.1)

6.7(16.7)184.9

(227.2)53.8

(92.7)15.8

(32.8)14.2

(20.2)0.1

(1.1)46.2

(44.3)7.2

(16.1)291.3

(339.8)

4.1(9.9)

544.1(747.1)

Reference sites

TMRes1

30

454.4(428.1)

60.8(38.0)

3.9(7.1)

215.5(222.4)144.2

(120.2)3.1

(7.5)16

(18.5)0.8

(3.0)35.1

(29.8)4.3

(8.4)4.7

(9.0)

1.2(3.6)

212.4(198.2)

MauRes1

30

326.6(292.8)

62.0(79.2)

7.8(13.5)227.6

(231.1)95.9

(132.8)17.9

(40.9)30.4

(23.7)1.2

(4.7)23

(25.8)3.9

(11.2)27.3

(43.5)

10.9(14.4)243.6

(453.9)

KaraRes1

30

215.9(176.1)

21.0(29.8)10.1

(27.8)162.5

(125.9)30.0

(27.2)2.7

(5.0)2.7

(5.0)2.3

(8.9)6.6

(11.4)1.6

(5.1)3.1

(6.1)

2.3(5.7)46.8

Totalcount

33964

11880

3837

352

10 140

3842

2783

1290

29

2525

1135

11671

173

15 429(42.4)

(continued)

2;

1

o

N

8o*no

<£too©in>, V

ol. 36

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014

Table 2 {continued)

Pest control

Livestock exclusionNo. forest remnants

No. samples

CrustaceaAmphipoda

Isopoda

ArachnidaAraneae

Opiliones

Pseudoscorpiones

Other PhylaAnnelida

Mollusca

Nematoda

"Other taxa"

All taxa combined

FO4

120

45.1(86.9)

7.6(12.9)

13.8(20.0)

6.8(12.1)

16.8(43.1)12.9

(24.7)26.0

(141.9)

0.4(2.1)

552.9(418.7)

F64

120

113.1(142.7)

40.5(109.3)

20.9(23.4)

9.1(14.7)

50.0(151.2)

9.7(17.8)17.6

(56.6)

1.5(4.4)

878.8(639.1)

P0

F12388

186.4(282.0)

34.4(78.7)

20.1(25.4)

27.9(54.9)

28.0(55.0)10.6

(16.9)1.6

(15.0)

0.5(3.0)

966.0(1028)

F42390

8.8(23.3)

6.8(19.0)

18.1(23.3)

8.3(10.8)

1.6(7.9)16.1

(23.9)0.5

(3.0)

2.5(6.9)

501.4(562.6)

F04

120

55.2(131.9)

2.8(11.7)

36.7(48.0)

0.1(1.1)7.3

(19.5)

31.4(117.4)

8.1(16.9)

1.1(6.9)

0.6(2.6)

1020.5(731.3)

F64

115

61.0(152.7)

18.2(37.4)

75.2(132.5)

7.5(15.4)

45.1(99.4)18.8

(27.9)

2.1(10.6)

1113.1(861.4)

PI

F124

110

153.1(241.3)

39.5(90.9)

20.5(46.4)

9.9(19.9)

330.1(2486.9)

17.3(26.6)

2.4(21.6)

0.7(2.9)

1393.5(2998.8)

F424

111

147.6(303.5)

37.1(64.7)

31.3(36.1)

4.0(41.1)30.0

(36.7)

111.5(310.5)

17.7(24.2)19.2

(93.4)

2.6(11.9)

1837.3(1430.4)

Reference sites

TMRes1

30

18.7(23.5)125.5

(146.3)

23.4(21.9)

0.4(2.1)48.3

(51.4)

2.3(5.7)20.3

(35.8)1.2

(3.6)

2.7(5.9)

1399.1(936.2)

MauRes1

30

75.6(67.7)242.0

(207.0)

30.8(24.7)

2.3(4.8)76.4

(74.3)

60.8(69.4)104.1

(181.7)24.6

(43.4)

16.4(24.3)

1710.9(1191.3)

KaraRes1

30

2.3(4.8)23.0

(26.7)

17.9(19.4)

0.4(2.1)36.6

(35.5)

3.9(15.1)25.3

(23.4)1.6

(5.1)

2.7(7.9)

621.6(365.3)

Totalcount

33964

7351

2739

2438

47

1378

6011

1419

750

157

87 376

Didham

e

1oQ

Is"

%sj1at

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


A 0.03

0.02

CDCD 0.01TO

•a

CD

2 0.00oQ

CO

-0.01 -

-0.02

All invertebrate taxa

B

Ed

ro 1000

p

ICO

100

TeMiroReserve

MaungatautariReserve

KarakarikiReserve

1.0 10 100

Patch area (ha)

1,000 10,000

Fig.2 Variationintotalleaf-litterinvertebrate density (excludingacari and collembola) in forestremnants of differing sizes andmanagementtreatments: (A) slopes(±1 Se) of edge effects calculatedusing simple least squares regres-sion on log10 (x + 1) transformedinvertebrate no. m–2, with distancefrom edge on a linear scale; and(B) mean (±1 Se) invertebratedensity in the "remnant interior"(i.e., density predicted by thebest-fit null or linear edge-modelat a standardised distance of 40 mfromtheedge). open symbols=un-fencedorrecently-fencedremnants(livestock exclusion categories F0and F6); solid symbols=remnantsfenced for a moderately-long tolong period of time (F12 andF42).Squares = no pest control; circles= pest control; triangles = forestreserves. Note: a slope of edgeeffect of 0.01 on a log10 scale ofinvertebrate density indicates anorder of magnitude change in den-sity for every 100 m distance alongthe edge-to-interior gradient, anda slope of 0.02 indicates an orderof magnitude change in densityevery 50 m.

These extremely low fragment-interior densities ofcoleoptera, hemiptera and Diplopoda, in particular,were observed predominantly in remnants in the p0pest control treatment class, and the extremely highdensities of Thysanoptera were observed predomi-nantly in remnants in the p1 pest control treatmentclass (Fig. 3e-h). again, only mollusca showed asubstantial time since livestock exclusion effect ondensities in the remnant interior, with the lowestdensities observed for recently-fenced remnants(F0 and F6) (Fig. 3G). These contrasting patternsof variation in the abundance of individual taxa areindicative of a significant shift in both the overall

density (Fig. 2) as well as the composition of inver-tebrate communities (Fig. 3) in forest remnants.

Variation in invertebrate communitycomposition among remnants

In an initial c c a ordination, all four variablesconsidered to be potential confounding factors inthe analysis had significant effects on invertebratecommunity composition, with substantial varianceexplained by spatial autocorrelation in latitude (λ =0.009, F= 10.05, P = 0.001), longitude (λ = 0.020,F = 21.31, P = 0.001) and the linear component oflatitude × longitude (λ = 0.010, F= 10.40,P=0.001)

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014

Thysanoptera IoQ

3

r

10 100 1,000

Patch area (ha)

1.0 10 100 1,000 10,000

Patch area (ha)

10 100 1,000 10,000

Patch area (ha)

1.0 10 100 1,000 10,000

Patch area (ha)

Fig. 3 Variation in the densities of selected leaf-litter invertebrate taxa in forest remnants of differing sizes and management treatments: (A-D) slopes (±1 SE) ofedge effects for Coleoptera, Isopoda, Mollusca and Thysanoptera, respectively; and (E-H) mean (±1 SE) density in the "remnant interior" for Coleoptera, Isopoda,Mollusca and Thysanoptera, respectively (see text and Fig. 2 for explanation of how slopes and interior values were calculated). Symbols as in Fig. 2.

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


of sampling locations, and to a lesser extent by sam-pling date (λ = 0.010, F = 10.97, P = 0.001). Thesevariables were entered as covariables into a partialc c a , in which they collectively explained 5.9% ofthe variance in invertebrate composition.

After partialling out the covariables, the first fourp c c a axes explained 14.1% of the remaining vari-ance in invertebrate composition among leaf-littersamples (λ1 = 0.072, λ2= 0.022, λ3= 0.018 and λ4=0.009 for axes 1-4, respectively; Table 3), with theforward selection procedure identifying patch area,vegetation richness, time since livestock exclusionand a livestock exclusion × pest control interactionas the major predictors of invertebrate composi-tion, explaining more than half of the total effect(Table 3). Site ordering along p c c a axes 1 and 2(Fig. 4) shows the strong dissimilarity in invertebratecomposition between forest remnants and the forest

reserves, with the relative densities of Isopoda (r =-0.45), pseudoscorpiones (r = -0.40), coleoptera (r= -0.37), amphipoda (r = -0.28) and Diplopoda (r= -0.25) strongly negatively correlated with axis 1,and the relative densities of psocoptera (r = 0.29),lepidoptera (r = 0.25), Thysanoptera (r = 0.22) andFormicidae (r = 0.22) strongly positively correlatedwith axis 1 (all P < 0.001). only Formicidae densitywas strongly positively correlated with axis 2 (r =0.50, P< 0.001).

The strong patch area gradient was also closelyaligned with the effects of time since livestock ex-clusion and the livestock exclusion × pest controlinteraction on invertebrate composition (Table 3).Unfenced or recently-fenced remnants (F0 andF6; open symbols in Fig. 4) tended to have higherdissimilarity to the forest reserves than did the long-fenced remnants (F12 and F42; closed symbols in

Table 3 (a) Results of a forward selection procedure to determine which of the 25 treatment and environmentalvariables in a partial Canonical Correspondence Analysis (pCCA) explained significant variation in invertebrate com-munity composition among the 964 leaf-litter samples (see Fig. 4). Effects were calculated after first partialling out theconfounding effects of sampling date and spatial autocorrelation among sampling locations. (b) Interset correlationsbetween variation in the environmental variables and variation in invertebrate community composition on each of thethree canonical pCCA axes. Variables are ordered from most to least significant in the forward selection procedure (allwith a Bonferroni corrected P-value of P < 0.002). λa represents the additional variance explained by environmentalvariables as they are sequentially added into the model. Correlations in bold are significant at P < 0.001. codes forenvironmental variables as in Table 1.

Variable

lgareaVegRiclvexcll x ppltSplopcanBasalmDShismDSsoipxamDSvegpstctrelev5kcorel x aSmplitDistl x epltSdlmDScli

) :Total inertia:

(a) Forward selection

K0.0420.0180.0100.0090.0090.0060.0050.0090.0070.0050.0050.0050.0050.0040.0030.0030.0020.0030.0030.003

0.1560.911

F

49.56021.30012.18010.83011.1207.2305.74011.7508.7106.1206.1006.5806.0505.9104.5604.1302.7104.5403.6503.350

P

0.0010.0010.0010.0010.0010.0010.0010.0010.0010.0010.0010.0010.0010.0010.0010.0010.0010.0010.0010.002

eigenvalue:

Species-environment correlation:

axis 1

-0.5350.122

-0.436-0.471-0.206

0.021-0.323

0.0400.148

-0.5420.378

-0.051-0.265-0.439-0.395-0.210-0.176-0.024-0.048-0.152

0.072

0.734

(b) Interset correlations

axis 2

0.1880.3170.066

-0.003-0.026

0.114-0.108

0.101-0.026

0.115-0.061

0.188-0.105

0.1510.1850.0180.0810.103

-0.0230.077

0.022

0.540

axis 3

0.0680.220

-0.2270.1070.076

-0.1100.2170.1230.2790.0620.125

-0.2090.036

-0.1670.107

-0.0030.090

-0.050-0.040-0.026

0.018

0.535

axis 4

0.052-0.089

0.0230.0460.131

-0.041-0.008

0.032-0.125

0.0340.014

-0.0130.1920.0140.0970.0370.0410.0740.1440.127

0.009

0.433

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


CMCMO©

II

CM

1.0

0.5

0.0

-0.5

Maungatautari Karakarik

-0.5

Fig. 4 partial canonical cor-respondence analysis ( p c c a )ordination biplot showing vari-ation in invertebrate communitycomposition among 964 leaf-littersamples collected from 30 forestremnants and three forest reserves(Te miro Scenic Reserve, maun-gatautari mountain Scenic Reserveand Karakariki Scenic Reserve) inthe Waikato region, North Island,New Zealand. Symbols as inFig. 2.For clarity, variation in invertebratecomposition among the individualleaf-litter samples within each rem-nant or forest reserve is representedas 95% confidence limits aroundthe mean (centroid) of the axis 1and 2 scores for that site. Weightedaverage sample scores were derivedfrom the pCC A analysis after firstpartialling out the confounding ef-fects of sampling date and spatialautocorrelation (x, y, and xy) amongsampling locations. Lambda (λ) is ameasure of the variance explainedby site ordering along the ordinationaxes (see Table 3).

Fig. 4), but this was only evident for forest remnantswith pest control (p1; circles in Fig. 4) and not inremnants without pest control (p0; squares in Fig. 4).plotting the livestock exclusion × pest control inter-action effect on invertebrate dissimilarity betweenthe 30 forest remnants and their (geographically)nearest forest reserve (Fig. 5a), it is clear that thereason for the treatment intercorrelation with patcharea along p c c a axis 1 is because the pest controltreatment was inadvertently confounded with patcharea in the study design (Fig. 5B; see also Fig. 6).The highest invertebrate dissimilarity values wereobserved in remnants in the youngest time sincelivestock exclusion class with good pest control(Fig. 5a), but these were also the smallest of all theforest remnants sampled (Fig. 5B), making it dif-ficult to discriminate the relative treatment effectsfrom the patch area effect.

Finally, interset correlations between environ-mental variables and site ordering along axis 1 werealso high for various components of vegetationchange (particularly tree basal area and plant speciescomposition) and sample litter mass (Table 3), sug-gesting further complex interdependencies amongtreatment and environmental variables.

JL

0.0 0.5 1.0

pCCAaxis1 (λ = 0.072)

Discriminating the direct and indirect effectsof management actions

To discriminate the direct and indirect causal re-lationships among these variables, a stepwisespecification search was used to select the mostparsimonious structural equation model (Sem) ex-plaining variation in invertebrate community dis-similarity (Fig. 6). The reduced model (Fig. 6) hadan acceptable minimum discrepancy function, C^f/d.f. = 0.543, and no significant difference betweenthe predicted covariance structure of the model andthe observed covariance structure in the data (P =0.933). In this final reduced model, three correlationsamong variables and 16 causal paths were found tomake significant contributions to overall model fit(Fig. 6), although these paths varied in their strengthand the statistical significance of their partial regres-sion or covariance coefficients.

The major proximate mechanisms affecting in-vertebrate community dissimilarity were a negativeeffect of litter mass on invertebrate dissimilarity (i.e.,lower litter mass in the remnant led to a greater dis-similarity of invertebrate composition between theremnant and the nearest forest reserve), a positiveeffect of vegetation dissimilarity on invertebratedissimilarity (i.e., the more dissimilar the vegetationcomposition was to the nearest forest reserve, themore dissimilar the invertebrate composition was),as well as a cascading series of causal paths betweenaltered soil chemistry (higher p h and base saturation,and lower total carbon, nitrogen and exchangeablecation capacity), increased vegetation dissimilarity,and lower litter mass (Fig. 6), with substantial total

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


2.0 n

1.5

I

O

1.0

0.5

0.0 4

No pestI control

0 10 20 30 40 50 60 70 80 90 100

Time since livestock exclusion (years)

Pestcontrol

1.0 10

Patch area (ha)

100 Forestreserves

Fig. 5 Dissimilarity of invertebrate community composition between each of the 30 forest remnants and their (geo-graphically) nearest forest reserve, plotted with respect to (A) time since livestock exclusion, and (B) patch area. meandissimilarity (±95% confidence limits) was calculated from the pCCA analysis in Fig. 4, using the euclidean distancebetween the weighted average (Wa) sample scores for each leaf-litter sample in a given remnant and the average(centroid) of the Wa sample scores for the nearest forest reserve that acted as a local-area reference point. mean dis-similarity values (±95% confidence limits) are also shown for all possible pairwise comparisons among (not within)forest reserve sites in (b). Symbols as in Fig. 2. overlapping data points offset for clarity.

Livestockexclusion

Past

-= i. Alt

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


effect sizes for vegetation dissimilarity and litter pest control were typically smaller) was mediated bymass in particular (Table 4). a strong positive influence (somewhat surprisingly)

Livestock exclusion and pest control influenced on vegetation dissimilarity, which affected inverte-these proximate relationships via weak to moderate brate dissimilarity both directly and indirectly viaeffects of time since livestock exclusion and the its effect on litter mass (Fig. 6).livestock exclusion × pest control interaction on Finally, time since livestock exclusion, pestsoil chemistry, and via strong significant effects control, and the livestock exclusion × pest controlof pest control on vegetation dissimilarity (Fig. 6, interaction all had direct effects on invertebrateTable 4). In addition, the pest control treatment had dissimilarity that could not be explained by the in-further indirect effects on invertebrate dissimilarity direct mediating pathways incorporated in the modelvia joint correlations with underlying remnant at- (although only the direct effect of the livestocktributes of both slope and area (Fig. 6). The patch exclusion × pest control interaction was stronglyslope effect (i.e., remnants with pest control were significant). These direct effects represented thetypically on shallower slopes) was mediated by a dominant components of the total effect sizes of thestrong influence on soil chemistry parameters, and treatment variables (Table 4).by a significant negative effect on litter mass (Fig. 6). overall, the Sem model indicates that the longermeanwhile, the patch area effect (i.e. remnants with the period of time that livestock have been excluded

Table 4 Standardised path coefficients from the structural equation model in Fig. 6, showing thedirect effects, indirect effects (product of the component direct effects within each indirect pathway,summed for all indirect pathways between the independent and dependent variables) and total effects(sum of direct and indirect effects) of factors influencing the dissimilarity of invertebrate communitycomposition between the 30 forest remnants and their nearest forest reserve (Fig. 4). Standardisedpath coefficients express the number of standard deviations of change in the dependent variable forevery one standard deviation of change in the independent variable (negative values indicate thatthe causal variable resulted in a greater similarity of the invertebrate community between remnantsand the nearest continuous forest reference site). Dashes indicate paths that were removed in thefinal, most parsimonious structural equation model. Bootstrapped confidence intervals were used toestimate significance levels for each of the total effects below. †, P 0.10; *, P 0.05.

causal variable

Time since livestock exclusionpest controllivestock exclusion × pest controlpatch slopepatch areaSoil chemistryVegetation dissimilaritylitter mass

Direct effects

-0.290†0.211

-0.335†-——

0.156-0.250

Indirect effects

-0.0300.1210.0170.177†0.1000.1030.160-

Total effects

-0.320*0.332*

-0.318†0.177†0.1000.1030.315†

-0.250

Fig. 6 Final reduced structural equation model testing the direct and indirect effects of the livestock exclusion andmammalian pest control treatments on the dissimilarity of invertebrate community composition between each of the30 forest remnants and their (geographically) nearest forest reserve. Single-headed arrows represent causal paths frompredictor to response variables, and the number on each path is the value of the unstandardised partial regression coef-ficient, indicating whether the relationship is positive or negative. The statistical significance of individual regressioncoefficients is indicated by the colour of the line (black, P 0.05; dark grey 0.05 > P 0.10; light grey, P > 0.10). Thethickness of the line indicates the magnitude of the standardised path coefficients, which relates to the effect sizes pre-sented in Table 4. Double-headed arrows indicate covariance between variables, with solid lines indicating significantcovariance (P 0.05) and dashed lines indicating non-significant covariance (P > 0.05) which nevertheless improvesthe overall model fit. Correlation coefficients, r, are shown for covariance paths. For the four endogenous variablesin the model, squared multiple correlations (R2) are given to represent the variance explained by all the associatedpathways linking that variable. Note that vegetation dissimilarity is not the absolute vegetation composition of the site,but is instead the dissimilarity of vegetation composition between each remnant and the nearest forest reserve.

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


from a remnant the greater the reduction in inverte-brate community dissimilarity between the remnantand the forest reserves (lvexcl, total effect=-0.320;Table 4). meanwhile, the direct and indirect effectsof pest control on vegetation dissimilarity, com-bined with the inadvertent confounding of the pestcontrol treatment with patch area and slope, led to asignificant increase in invertebrate community dis-similarity in remnants with pest control comparedto remnants without pest control (pstctr, total effect= 0.332; Table 4), but this effect only appears to besignificant for unfenced or recently fenced remnants(lxp, total effect = -0.318; Table 4) (see Fig. 5a).

DISCUSSION

The impact of habitat fragmentation on leaf-litterinvertebrate communities in managed versusunmanaged remnantsIn the heavily fragmented lowland forests of theWaikato region, New Zealand, the density and com-position of leaf-litter invertebrate communities var-ied significantly between managed and unmanagedforest remnants, suggesting that management inter-vention can lead to substantial conservation gains ineven the most degraded forest remnants. We identi-fied three broad trends in the response of leaf-litterinvertebrate communities to habitat fragmentationand subsequent restoration management: (1) theleaf-litter invertebrate fauna is severely degradedin unmanaged forest remnants; (2) managementintervention generally increases the density of leaf-litter invertebrates and reduces the dissimilarityof community composition between remnants andnearby forest reserves, but different managementactions have contrasting effects on the recovery ofleaf-litter invertebrate communities; and (3) even themost intensive management intervention in forestremnants does not result in complete convergenceof community composition with that found in forestreserves, over the time period studied here.

First, unmanaged forest remnants (i.e., unfencedremnants without significant management interven-tion) exhibited the lowest invertebrate densities anda substantial shift in overall community compositioncompared to forest reserves, even when composi-tion was measured as total counts of invertebrateswithin higher taxonomic units. The higher taxonapproach taken here thus provides strong evidenceof widespread restructuring of litter-dwelling inver-tebrate communities in small remnants, involving in

particular key detritivore taxa such as mollusca, Di-plopoda, Isopoda and coleoptera. There have beensurprisingly few comparable studies of the degreeto which habitat fragmentation affects overall in-vertebrate community composition in New Zealand(ewers et al. 2002; Norton 2002; ewers 2004), butelsewhere similar levels of effect have been widelyreported (reviewed in Didham 1997 and ewers &Didham 2006b). Given that invertebrates comprisethe largest component of biomass and biodiversityin terrestrial systems (Wilson 1987), these effectsare likely to have a significant impact on ecosys-tem functioning in forest remnants (e.g., Didhamet al. 1996; larsen et al. 2005; Snyder & hendrix2008).

of course, we recognise that the use of highertaxon data will inevitably leave some questions un-resolved until the corresponding species-level databecome available. For instance, here we interpret theobserved trends in higher taxon abundance as beingindicative of a general decline in many, if not most,dominant species within each of these orders, butthis could easily mask contrasting species-specificvariation in the densities of rarer species (ewers &Didham 2006b). For example, at the hope RiverForest Fragmentation project in the South Island,New Zealand, trends in the abundance of higher taxain forest remnants (ewers & Didham 2006a) wereconsistent with declining abundances of the major-ity of individual species (e.g., ewers & Didham2008), and yet a substantial number of rarer speciesshowed contrasting responses (ewers et al. 2007).Furthermore, contrasting species-specific responsesto habitat fragmentation have been recorded for arange of other invertebrate species in New Zealand(ogle 1987; Burns et al. 2000; harris & Burns 2000;Bach & Kelly 2004) and elsewhere (see reviews inFoggo et al. 2001; Tscharntke et al. 2002a; ewers& Didham 2006b).

Second, although management intervention wasgenerally beneficial for invertebrate communitiesinhabiting remnants, different management actionshad contrasting effects on the densities of differentinvertebrate taxa. In some treatments there was asignificant reduction in community dissimilaritybetween remnants and their nearest forest reserve(convergence, or recovery, in community structure),whereas in other treatments there was a significantincrease in invertebrate community dissimilarity inremnants (divergence in community structure). Forinstance, time since livestock exclusion had a con-sistently positive effect on the densities of most taxa,particularly Isopoda, pseudoscorpiones, coleoptera,

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


amphipoda, Diplopoda and mollusca, but the mag-nitude of the effect was strongly dependent on thelevel of mammalian pest control applied. In un-fenced or recently-fenced remnants, pest controlactually had an adverse effect on invertebrate com-munities, with intensive pest control associated withan increase in invertebrate dissimilarity comparedto the nearest large forest area. In particular, inten-sive pest control in recently-fenced forest remnantspromoted unusually high densities of Thysanopteraand Formicidae (hymenoptera) that were atypicalof the densities normally found in forest reserves.

There was no consistent evidence that these con-trasting management effects on invertebrate com-munities were mediated by altered edge responsefunctions of different taxa. We had expected theimpact of livestock encroachment and feral mam-malian omnivores to be greater at forest edges thanin the forest interior (e.g., Bach & Kelly 2004),and therefore predicted that the greatest immediateresponse to management action would be observedat edge locations. In general, though, we found thatthe densities of most invertebrate taxa were notsignificantly related to distance from forest edgein most remnants, despite recent evidence of thelarge magnitude and extent of edge effects for manyinvertebrate taxa (ewers & Didham 2006a,b, 2008;but see Kotze & Samways 2001; Dangerfield et al.2003). In the relatively few cases where significantedge effects were detected, they did tend to be inthe direction expected, with a decrease in the slopesof edge effects for forest-dependent invertebrates,and an increase in the slopes of edge effects fordisturbance-adapted invertebrates, with increasingtime since livestock exclusion in the intensive pestcontrol treatment. however, a more parsimoniousexplanation for the observed edge responses is thatthere was an inadvertent bias in the selection of for-est remnants across management treatment classes,with recently-fenced remnants in the intensive pestcontrol treatment tending to have the smallest patchareas of all the remnants sampled. The greater slopesof edge effects in the smallest remnants (typically<1.5 ha) are consistent with recent evidence thatnon-linear interactions between patch area and edgeeffects (ewers et al. 2007; Fletcher et al. 2007)exacerbate the impact of habitat fragmentation oninvertebrate community decline in very small rem-nants.

The confounding effect of patch area on inver-tebrate community structure was pronounced, eventhough remnants were selected to have only a narrowrange of patch areas (from c. 1-30 ha) reflecting the

real range of remnant sizes that are typically consid-ered for conservation management on private land(QeII Trust 1984; porteus 1993; anon. 2008). a s agenerality, larger remnants (greater than c. 4-5 ha)tended to have densities of most invertebrate taxathat were more similar to forest reserves than thosein smaller remnants, regardless of management treat-ment. The smallest remnants (less than c. 1.5 ha)were most dissimilar in community composition andappeared to exhibit divergent trajectories of commu-nity change in response to pest control treatment.

Third, even in forest remnants with the greatestlevel of management intervention (i.e., a long periodof livestock exclusion and a history of intensivemammalian pest control), there was still a signifi-cant residual difference in invertebrate communitycomposition compared to that observed in forestreserves. This most likely represents the inevitablechange in composition associated with a dramaticreduction in habitat area (i.e., this is "as good as itgets" for a forest remnant), but it might also be that67 years of livestock exclusion is still not enoughtime for complete community recovery. Deforesta-tion and land-use intensification in the surroundinglandscape may have forced remnant communitiesover some significant biotic or abiotic thresholds,suggesting that there might still be some componentof "recoverable" variation in community composi-tion in remnants that requires more extreme manage-ment intervention (such as faunal translocations orimprovement of habitat quality) (hobbs & cramer2008).

Discriminating the interactive effects of livestockexclusion and mammalian pest control oninvertebrate community recovery in degradedforest remnantsOur most striking finding was the degree to whichthe recovery in invertebrate abundance and commu-nity composition towards the condition observed innearby forest reserves was influenced by a complex,antagonistic interaction between livestock exclu-sion and mammalian pest control. Using structuralequation modelling to discriminate the direct andindirect drivers of management effects, we foundthat livestock exclusion and pest control affectedlitter invertebrate communities via a cascading seriesof changes in soil geochemistry, leading to alteredvegetation composition and significant variation inleaf-litter mass among remnants, with the strongestrelative effects mediated by the direct and indirecteffects of vegetation dissimilarity on invertebratedissimilarity. Interestingly, both the main effect of

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


time since livestock exclusion and the main effectof pest control, had an almost identical magnitude oftotal (direct and indirect) effects, but with oppositesigns. The livestock exclusion × pest control interac-tion was itself of equivalent magnitude of total effectto the main effects of the two separate treatmentvariables, with a net effect in the same direction asthe time since livestock exclusion treatment. Thissuggests that although the two management actionshad contrasting effects in the short term, it appearsthat long-term livestock exclusion results in thegreatest recovery of invertebrate composition (low-est dissimilarity), irrespective of whether intensivepest control is applied or not.

In terms of the livestock exclusion treatment, it issurprising that there have been only anecdotal stud-ies investigating the effects of livestock tramplingand browsing on invertebrate communities in NewZealand forests (Burns et al. 2000; but see Scrim-geour & Kendall 2003; Doledec et al. 2006; Schonet al. 2008 for studies of New Zealand grasslandecosystems, and abensperg-Traun et al. 1996; Bro-mham et al. 1999; Woinarski et al. 2002 for a rangeof similar studies in forest ecosystems elsewhere).Furthermore, we are aware of no studies that thathave explicitly tested the effects of livestock exclu-sion on invertebrate communities in native forestremnants in New Zealand. even for plant communi-ties, there have been remarkably few studies of theeffects of livestock grazing in New Zealand forests(Burns et al. 2000; Buxton et al. 2001; Timmins2002; miller 2006; Dodd et al. 2008; Smale et al.2008) despite the severe damage caused by livestockin the forest understorey. o f these, we know ofonly two previous studies that have measured thelong-term recovery of plant communities followinglivestock exclusion from forest remnants (Smale etal. 2005; Dodd & power 2007; but see also pettit etal. 1995, Spooner et al. 2002, close et al. 2008 andDodd et al. 2008 for short-term temporal studies oflivestock exclusion from forest remnants in NewZealand and australia). In both kahikatea-dominatedforests in the Waikato Basin (Smale et al. 2005) andtawa-dominated forests in the Rotorua Basin (Dodd& power 2007), time since livestock exclusion hada significant positive effect on litter cover (up to10-15 years) and native vegetation recovery (up to30-35 years), and a significant negative effect onpopulations of exotic plants in Waikato remnants.Smale et al. (2008) concluded that the outlook forplant communities was bleak in the absence of man-agement intervention, and they argued that livestockexclusion is the single most important measure that

managers can take to improve the long-term viabilityof plant communities in forest remnants.

For invertebrates, our principal knowledge of thepotential effects of browsing mammals on nativeforests comes from the landmark study by Wardleet al. (2001) comparing the community- and ecosys-tem-level effects of introduced browsing mammals(predominantly feral deer, goats and pigs) inside andoutside fenced exclosures throughout New Zealand.Wardle et al. (2001) found very strong increases inthe abundances of all dominant leaf-litter inver-tebrate taxa following browser exclusion. Unlikeour study, though, the response of most leaf-litterinvertebrates to browser exclusion in Wardle et al.(2001) was not significantly related to the effect ofbrowsers on vegetation density or composition, orto the age of fenced exclosure. Instead, variation ininvertebrate density was better explained by com-binations of variables reflecting soil geochemistryand litter structure (Wardle et al. 2001). In particular,invertebrate community dissimilarity inside versusoutside browser exclosures was most strongly cor-related with measures of litter quality and heteroge-neity (Wardle et al. 2001).

Wardle et al. (2001) concluded that browsingmammals clearly induced changes in the structureand community composition of dead leaves in thelitter layer that must ultimately have resulted fromchanges in the composition of aboveground vegeta-tion, yet they could not detect a significant cascadingseries of causal links between soil, vegetation andlitter structural effects on litter invertebrates. In thepresent study, we did observe a weak direct effect ofplant community dissimilarity on invertebrate com-munity dissimilarity across remnants, and a strongassociation between vegetation dissimilarity andlitter mass, that together represent the major indirectpathways mediating mammalian browser effects oninvertebrate dissimilarity in forest remnants. Thesechanges in invertebrate abundance and compositionmay result from the negative effect of browsers onplant properties allowing increased light incidenceat the ground layer (Suominen et al. 1999) and fromadverse microclimatic changes in the ground layer(Kielland & Bryant 1998). however, it is morelikely that these effects operate primarily throughalteration of litter structure and resource availability.Unfortunately, our measure of litter mass is only aweak surrogate for more complex changes in litterquality, litter structure and litter heterogeneity. Withno direct effect of soil geochemistry on invertebratedissimilarity, and in the absence of appropriate mea-sures of litter resource quality and heterogeneity,

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


we follow Wardle et al. (2001) in considering thatadverse effects of browsing mammals on leaf-litterinvertebrates might have more to do with the physi-cal trampling and scuffing effects of browsers, thanthe indirect effects via vegetation change. Wardle etal. (2001) consider that the intensity of scuffing andtreading (and resultant disturbance, compaction, andreduced substrate porosity) caused by hoof pressurefrom deer and goats can be considerable (see alsoDuncan & holdaway 1989), and the intensity ofthese effects by domestic livestock is likely to beeven greater because of their larger mass and greaterdensity. Similarly, Dodd & power (2007) attributethe effects of livestock exclusion on plant commu-nity recovery in tawa-dominated forest remnantsto the reduction in trampling and soil compaction,and to the reduction in browsing effects on removalof leaves and subsequent litter inputs and soil geo-chemistry.

In terms of the mammalian pest control treatment,the Sem results indicate an increase in invertebratedissimilarity in remnants with pest control comparedto remnants without pest control, mediated largelyby the indirect effects of pest control on vegetationdissimilarity. The exact mechanism underlying thisadverse effect of pest control is uncertain (Floyd,Burns, Smale & arnold, unpubl.), but as it is ob-served only in remnants with continued livestockbrowsing, or very recent livestock exclusion, it ispossible that it is associated with differing succes-sional trajectories resulting from the differentialsusceptibility of plants to livestock versus omnivore(predominantly possum) browsing. For example,across the larger set of 47 tawa-dominated rem-nants sampled as a component of the present study,Floyd et al. (unpubl.) found that the p1 pest controltreatment was associated with an increase in exoticplant species richness (Floyd, Burns, Smale & ar-nold, unpubl.). It may also be that a primary focuson reducing possum numbers in the pest controloperations unintentionally caused an increase inother omnivorous mammals, such as mice and rats(e.g., Tompkins & Veltman 2006; Sweetapple &Nugent 2007). certainly, in some of the same for-est remnants that we sampled, Innes et al. (unpubl.)found very high densities of ship rats (Innes, King,Bridgman, Fitzgerald & arnold, unpubl.). onceagain, the exact mechanism by which the changes inplant community dissimilarity might have resultedin the observed changes in invertebrate communitydissimilarity is unknown. The most striking changein invertebrate composition in small remnants withP1 pest control management was a massive increase

in the densities of Thysanoptera (thrips) and For-micidae (ants) in the leaf litter. Whether this resultsfrom direct changes in plant resource availability, oris only indirectly associated with other soil, vegeta-tion or litter properties remains to be tested.

lastly, the Sem was also able to tease out the de-gree to which management effects were confoundedwith inadvertent differences in the area and slope ofremnants in different treatments. Both patch areaand patch slope had significant indirect effects oninvertebrate dissimilarity via their influence on soilgeochemistry, vegetation dissimilarity and littermass, but the magnitude of these indirect effects wassubstantially smaller than either of the managementtreatments or their interaction. of course, there isonly a limited capacity for statistical analyses toovercome the inherent bias in remnant sizes amongtreatments (Grace 2006), and these conclusionswill inevitably remain somewhat speculative in thisstudy.

Trajectories of invertebrate community recoveryin heavily fragmented landscapescareful management consideration needs to be givento the strikingly divergent short term trends in bothplant and invertebrate composition exhibited inunfenced or recently-fenced forest remnants withintensive pest control compared to remnants withno pest control. For plants, short-term trajectories ofweed invasion and the succession of a non-randomsubset of native species persisting in forest rem-nants could drive these dynamics (Smale et al. 2005;Floyd, Burns, Smale & arnold, unpubl.). For inver-tebrates, initial indications are that the unusuallyhigh densities of Thysanoptera and Formicidae inthese treatments might represent high levels of inva-sion of disturbance-adapted taxa when intensive pestcontrol is applied without first having 10-20 yearsof recovery from livestock exclusion, and this war-rants further species-level investigation. The ironywould be if mammalian pest control promoted anincrease in invertebrate pest abundance, whether asan indirect result of pest control effects on vegetationstructure, or due to the confounding effects of patcharea on invertebrate community structure.

Finally, it is interesting to speculate whether theconfounding of intensive pest control treatmentswith recent livestock exclusion from remnants ofsmaller average area, on shallower average slopes,was due solely to an inadvertent bias in remnantselection from a moderately small subsample ofavailable forest remnants in this study, or whetherit actually stems from a genuine trend in the degree

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


to which management actions covary with remnantattributes in real landscapes. Both livestock exclu-sion and pest control appear to be applied intention-ally to non-random subsets of remnants in order tomaximise conservation gain, and perhaps the moreintensive and costly the management action, thegreater the bias in the selection of the remnantsdeemed to be of greatest value. Whether the criteriaused by land managers to select appropriate rem-nants for management action actually matches thedegree of improvement in ecological condition ofthe fragment, or even whether this correlates withany wider perceived benefits for conservation, isan open question. Certainly, our data suggest thatintensive pest control had unexpected, adverse short-term effects on the trajectory of invertebrate com-munity recovery, both independently of patch areavia changes in plant community structure (Floyd,Burns, Smale & Arnold, unpubl.), and also becausemost of the unfenced and recently-fenced remnantswith intensive pest control happened to be of smallaverage area (in our dataset).

CONCLUSION

Not only is livestock exclusion from native for-est remnants an important management tool forthe recovery of vegetation structure and composi-tion, it also has substantial positive effects on therecovery of invertebrate community compositionon the forest floor. However, the relative effects oflivestock exclusion are strongly dependent on thelevel of mammalian pest control. There appears tobe a significant adverse effect of mammalian pestcontrol on invertebrate community composition inunfenced remnants, and for the first 10-20 yearsfollowing livestock exclusion, although a significantcomponent of this livestock exclusion x pest controlinteraction is confounded with differing patch slopesand patch areas among treatments. Partitioning of theapparent pest control effect on invertebrate compo-sition using structural equation modelling suggeststhat it operates primarily through an alteration ofvegetation biomass or vegetation successional tra-jectories, leading to unusually high densities oftaxa such as Thysanoptera and Formicidae, withoutsubstantive recovery of populations of taxa suchas Isopoda, Diplopoda, Coleoptera and Molluscathat are typical of relatively undisturbed forests.In the longer term, recovery trajectories followinglivestock exclusion converge on similar levels of

"effectiveness" at about 30 years after livestockexclusion, whether intensive pest control is appliedor not. Nevertheless, given that there are also likelyto be substantial species-specific benefits of pestcontrol for the recovery of individual invertebrate(as well as plant and vertebrate) taxa that are notcaptured by our higher-taxon approach, we cannotdiscount the possibility that a combination of thesemanagement actions may be most effective in thelong term. With the available evidence, though,priority should be given to livestock exclusion whenconservation management is targeted at leaf-litterinvertebrate communities.

ACKNOWLEDGMENTS

This study was funded by the Foundation for Research,Science andTechnologyinNewZealand(UOWX0609). Wethank the Department of Conservation and 33 landholdersfor allowing us access to land under their management(Alec and Yvonne Adams, Brett and Karen Bennett, Pauland Jo Bodle, Ian and Trisha Brennan, Andrew Colson,Bruce and Beverley Dean, David Findlay, Peter Gilmour,Kevin and Linda Goodman, Wallace and Sally Hall, JohnHewitt, Ag Research, Merv Hunt, Morris King, John andGaye Lamb, Gary Large, Swap Quarries, Bill and GayleMcMillan, Hamish and Selena McMullin, John and KarenMeredith, Dane and Katherine Miller, Brett and MichelleMiller, Trustees of Barnett Reserve, Michael and MargaretOliver, Steve Pemberton, Duncan and Mary Scott, Jackand Lois Scott, Gordon and Celia Stephenson, John andDeborah Stretton, John andHeather Taylor, Kevin Wiltshireand Waotu Lands Trust). Toni Johnston, Jane Johnston, andScott Bartlam provided invaluable assistance in the field,and Scott Bartlam, Lucy Bridgman, Neil Fitzgerald andDanny Thornburrow provided logistical and technicalsupport. James Crowe, Ruth Hudson, Melissa Jacobsen,Shaun Nielsen, Andre Siebers and Geraldine Teng helpedwith invertebrate sorting. We thank Tanya Blakely andJason Tylianakis for advice on the SEM, and Bruce Burns,John Innes, Carolyn King, Mark Smale, and the late GregArnold for support and critical feedback. Two anonymousreviewers provided valuable feedback on an earlier draftof the manuscript.

REFERENCES

Abensperg-Traun M, Smith GT 1999. How small is toosmall for small animals? Four terrestrial arthropodspecies in different-sized remnant woodlands inagricultural Western Australia. Biodiversity andConservation 8(5): 709-726.

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


Abensperg-Traun M, Smith GT, Arnold GW, Steven DE1996. The effects of habitat fragmentation andlive stock-grazing on animal communities in rem-nants of gimlet Eucalyptus salubris woodland inthe Western Australian wheatbelt. I. Arthropods.Journal of Applied Ecology 33(6): 1281-1301.

Andersen AN, Hoffmann BD, Muller WJ, Griffiths AD2002. Using ants as bioindicators in land manage-ment: simplifying assessment of ant communityresponses. Journal of Applied Ecology 39(1):8-17.

Anonymous 2008. QEII Trust: helping you protect the spe-cial character of your land. Open Space: Magazineof the Queen Elizabeth II National Trust 74: 31.

Antonic O 1998. Modelling daily topographic solar radia-tion without site-specific hourly radiation data.Ecological Modelling 113(1-3): 31-40.

Arbuckle JL 2007. Amos 16.0 user's guide. Chicago, USA,Amos Development Corporation.

Arroyo-Rodriguez V, Pineda E, Escobar F, Benitez-Malvido J in press. Value of small patches in theconservation of plant-species diversity in highlyfragmented rainforest. Conservation Biology, doi:101111/j.1523-1739.2008.01120.x

Atkinson IAE 2001. Introduced mammals and models forrestoration. Biological Conservation 99: 81-96.

Bach CE, Kelly D 2004. Effects of forest edges onherbivory in a New Zealand mistletoe, Alepisflavida. New Zealand Journal of Ecology 28(2):195-205.

Belbin L 1991. Semi-strong hybrid scaling, a new ordi-nation algorithm. Journal of Vegetation Science2(4): 491-496.

Belbin L 1995. PATN analysis package. Canberra,Australia, Division of Sustainable Ecosystems,CSIRO.

Biaggini M, Consorti R, Dapporto L, Dellacasa M, Pag-getti E, Corti C 2007. The taxonomic level orderas a possible tool for rapid assessment of arthropoddiversity in agricultural landscapes. AgricultureEcosystems & Environment 122(2): 183-191.

Blakemore LC, Searle PL, Daly BK 1987. Methods forchemical analysis of soils Scientific Report 80.103 p.

Bollen KA 1989. Structural equations with latent variables.New York, USA, Wiley.

Borcard D, Legendre P, Drapeau P 1992. Partialling outthe spatial component of ecological variation.Ecology 73(3): 1045-1055.

Bray JR, Curtis JT 1957. An ordination of the upland forestcommunities of southern Wisconsin. EcologicalMonographs 27(4): 326-349.

Bromham L, Cardillo M, Bennett AF, Elgar MA 1999.Effects of stock grazing on the ground invertebratefauna of woodland remnants. Australian Journalof Ecology 24(3): 199-207.

Brooks TM, Pimm SL, Oyugi JO 1999. Time lag be-tween deforestation and bird extinction in tropi-cal forest fragments. Conservation Biology 13:1140-1150.

Burns BR, Barker G, Harris RJ, Innes J 2000. Conifersand cows: forest survival in a New Zealand dairylandscape. In: Craig JL, Mitchell N, Saunders DAed. Nature Conservation 5: Nature conservation inproduction environments: managing the matrix.Australia, Surrey Beatty & Sons. Pp. 81-89.

Buxton RP, Timmins SM, Burrows LE, Wardle P 2001.Monitoring the impact of animals on Departmentof Conservation grazing leases in South Westland:1989-99. Science for Conservation 179: 1-64.

Chazdon RL 2008. Beyond deforestation: restoring forestsand ecosystem services on degraded lands. Science320(5882): 1458-1460.

Close DC, Davidson NJ, Watson T 2008. Health ofremnant woodlands in fragments under distinctgrazing regimes. Biological Conservation 141(9):2395-2402.

Connor EF, Courtney AC, Yoder JM 2000. Individuals-area relationships: the relationship betweenanimal population density and area. Ecology 81:734-748.

Cowan PE, Moeed A 1987. Invertebrates in the diet ofbrushtail possums, Trichosurus vulpecula, inlowland podocarp broadleaf forest, OrongorongoValley, Wellington, New Zealand. New ZealandJournal of Zoology 14(2): 163-177.

Craig JL, Anderson S, Clout M, Creese B, Mitchell N,Ogden J, Roberts M, Ussher G 2000. Conservationissues in New Zealand. Annual Review of Ecologyand Systematics 31: 61-78.

Dangerfield JM, Pik AJ, Britton D, Holmes A, GillingsM, Oliver I, Briscoe D, Beattie AJ 2003. Patternsof invertebrate biodiversity across a natural edge.Austral Ecology 28: 227-236.

Daniel MJ 1973. Seasonal diet of the ship rat (Rattus r.rattus) in lowland forest in New Zealand. Pro-ceedings of the New Zealand Ecological Society20: 21-30.

Davies RG, Hernandez LM, Eggleton P, Didham RK,Fagan LL, Winchester NN 2003. Environmentaland spatial influences upon species composition ofa termite assemblage across neotropical forest is-lands. Journal of Tropical Ecology 19: 509-524.

Davis P, Cocklin C 2001. Protecting habitats on privateland: perspectives from Northland, New Zealand.Science for Conservation 181: 1-69.

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014

1-60 New Zealand Journal of Zoology, 2009, Vol. 36

Derraik JGB, Rufaut CG, Closs GP, Dickinson KJM 2005.Ground invertebrate fauna associated with nativeshrubs and exotic pasture in a modified rural land-scape, Otago, New Zealand. New Zealand Journalof Ecology 29(1): 129-135.

Didham RK 1997. An overview of invertebrate responsesto forest fragmentation. In: Watt A, Stork NE,Hunter M ed. Forests and insects. London, Chap-man and Hall. Pp. 303-320.

Didham RK, Ghazoul J, Stork NE, Davis AJ 1996. Insectsin fragmented forests: a functional approach.Trends in Ecology & Evolution 11(6): 255-260.

Didham RK, Hammond PM, Lawton JH, Eggleton P,Stork NE 1998. Beetle species responses to tropi-cal forest fragmentation. Ecological Monographs68(3): 295-323.

Didham RK, Tylianakis JM, Hutchison MA, Ewers RM,Gemmell NJ 2005. Are invasive species the driv-ers of ecological change? Trends in Ecology &Evolution 20(9): 470-474.

Didham RK, Tylianakis JM, Gemmell NJ, Rand TA, EwersRM 2007. Interactive effects of habitat modifica-tion and species invasion on native species decline.Trends in Ecology & Evolution 22(9): 489-496.

Dodd MB, Power IL 2007. Recovery of tawa-dominatedforest fragments in the Rotorua Basin, New Zea-land, after cessation of livestock grazing. Ecologi-cal Management and Restoration 8: 208-216.

Dodd MB, Quinn JM, Thorrold BS, Parminter TG, Wed-derburn ME 2008. Improving the economic andenvironmental performance of a New Zealand hillcountry farm catchment: 3. Short-term outcomesof land-use change. New Zealand Journal of Ag-ricultural Research 51(2): 155-169.

Doledec S, Phillips N, Scarsbrook M, Riley RH, TownsendCR 2006. Comparison of structural and functionalapproaches to determining landuse effects ongrassland stream invertebrate communities. Jour-nal of the North American Benthological Society25(1): 44-60.

Dugdale JS 1996. Natural history and identification oflitter-feeding Lepidoptera larvae (Insecta) inbeech forests, Orongorongo Valley, New Zealand,with especial reference to the diet of mice (Musmusculus). Journal of the Royal Society of NewZealand 26(2): 251-274.

Duncan K, Holdaway R 1989. Footprint pressures andlocomotion of moas and ungulates and their effectson the New Zealand indigenous biota throughtrampling. New Zealand Journal of Ecology 12:97-101.

Duncan RP, Young JR 2000. Determinants of plantextinction and rarity 145 years after Europeansettlement of Auckland, New Zealand. Ecology81(11): 3048-3061.

Ewers RM 2004. The extent of forest fragmentation inNew Zealand and its effects on biodiversity. Un-published PhD thesis, University of Canterbury,Christchurch, New Zealand.

Ewers RM, Didham RK 2006a. Continuous response func-tions for quantifying the strength of edge effects.Journal of Applied Ecology 43(3): 527-536.

Ewers RM, Didham RK 2006b. Confounding factors in thedetection of species responses to habitat fragmen-tation. Biological Reviews 81(1): 117-142.

Ewers RM, Didham RK 2007. The effect of fragmentshape and species' sensitivity to habitat edges onanimal population size. Conservation Biology21(4): 926-936.

Ewers RM, Didham RK 2008. Pervasive impact of large-scale edge effects on a beetle community. Proceed-ings of the National Academy of Sciences USA105:5426-5429.

Ewers RM, Didham RK, Ranson LH 2002. The HopeRiver forest fragmentation project. The Weta24(1): 25-34.

Ewers RM, Thorpe S, Didham RK 2007. Synergisticinteractions between edge and area effects in aheavily fragmented landscape. Ecology 88(1):96-106.

Ewers RM, Kliskey AD, Walker S, Rutledge D, HardingJS, Didham RK 2006. Past and future trajectoriesof forest loss in New Zealand. Biological Conser-vation 133(3): 312-325.

Ferraz G, Russell GJ, Stouffer PC, Bierregaard RO, PimmSL, LovejoyTE 2003. Rates of species loss fromAmazonian forest fragments. Proceedings ofthe National Academy of Sciences of the USA100(24): 14069-14073.

Fitzgerald BM, Gibb JA 2001. Introduced mammals ina New Zealand forest: long-term research in theOrongorongo Valley. Biological Conservation99: 97-108.

Fletcher RJ, Ries L, Battin J, Chalfoun AD 2007. The roleof habitat area and edge in fragmented landscapes:definitively distinct or inevitably intertwined? Ca-nadian Journal of Zoology 85(10): 1017-1030.

Foggo A, Ozanne CMP, Speight MR, Hambler C 2001.Edge effects and tropical forest canopy inverte-brates. Plant Ecology 153: 347-359.

Gower JC 1971. General coefficient of similarity and someof its properties. Biometrics 27(4): 857-871.

Grace JB 2006. Structural equation modeling and naturalsystems. Cambridge, UK, Cambridge UniversityPress.

Hanski I, Wirta H, Nyman T, Rahagalala P 2008. Resourceshifts in Malagasy dung beetles: contrastingprocesses revealed by dissimilar spatial geneticpatterns. Ecology Letters 11(11): 1208-1215.

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


Harris RJ, Burns BR 2000. Beetle communities of ka-hikatea forest patches in a pasture dominatedlandscape. New Zealand Journal of Ecology 24:57-67.

Hobbs RJ, Cramer VA 2008. Restoration ecology:interventionist approaches for restoring andmaintaining ecosystem function in the face ofrapid environmental change. Annual Review ofEnvironment and Resources 33: 39-61.

Jane G 1983. The impact of introduced herbivores on low-land forests in the North Island. In: Thompson K,H odder APW, Edmonds AS, Irving Red. Lowlandforests in New Zealand. Hamilton, New Zealand,University of Waikato. Pp. 135-152.

Kielland K, Bryant JP 1998. Moose herbivory in taiga: ef-fects on biogeochemistry and vegetation dynamicsin primary succession. Oikos 82(2): 377-383.

Kline RB 2005. Principles and practice of structural equa-tion modeling. New York, USA, Guilford Press.

Kline TJB, Dunn B 2000. Analysis of interaction termsin structural equation models: a non-technicaldemonstration using the deviation score approach.Canadian Journal of Behavioural Science 32(2):127-132.

Kotze DJ, Samways MJ 2001. No general edge effectsfor invertebrates at Afromontane forest/grasslandecotones. Biodiversity and Conservation 10(3):443-466.

Kremen C, Colwell RK, Erwin TL, Murphy DD, NossRF, Sanjayan MA 1993. Terrestrial arthropodassemblages: their use in conservation planning.Conservation Biology 7(4): 796-808.

Larsen TH, Williams NM, Kremen C 2005. Extinctionorder and altered community structure rapidlydisrupt ecosystem functioning. Ecology Letters8: 538-547.

Leach MK, Givnish TJ 1996. Ecological determinantsof species loss in remnant prairies. Science 273:1555-1558.

Leathwick JR 2001. New Zealand's potential forest patternas predicted from current species-environmentrelationships. New Zealand Journal of Botany39(3): 447-464.

Leathwick JR, Stephens RTT 1998. Climate surfacesfor New Zealand contract report LC9798/126.1-19 p.

Leathwick JR, Overton JM, McLeod M 2003a. An envi-ronmental domain classification of New Zealandand its use as a tool for biodiversity management.Conservation Biology 17(6): 1612-1623.

Leathwick JR, Wilson G, Rutledge D, Wardle P, MorganF, Johnston K, McLeod M, Kirkpatrick R 2003b.Land environments of New Zealand. Auckland,New Zealand, David Bateman Ltd.

McAneney KJ, Noble PF 1976. Estimating solar radia-tion on sloping surfaces. New Zealand Journal ofExperimental Agriculture 4: 195-202.

MfE 2000. Final report of the Ministerial AdvisoryCommittee on biodiversity and private land.Wellington, New Zealand, Ministry for the En-vironment.

MfE 2002. Protecting our native plants and animals:communities in action. Wellington, New Zealand,Ministry for the Environment.

Milchunas DG, Lauenroth WK 1993. Quantitative effectsof grazing on vegetation and soils over a globalrange of environments. Ecological Monographs63(4): 327-366.

Miller C 2006. Does cattle grazing facilitate or hinderthe recruitment of podocarps in floodplain forestpatches, south Westland, New Zeal and? EcologicalManagement and Restoration 7: 232-233.

Mitchell RJ 2001. Path analysis: pollination. In: SchinerSM, Gurevitch J ed. Design and analysis ofecological experiments. Oxford, UK, OxfordUniversity Press. Pp. 211-231.

Myers N, Mittermeier RA, Mittermeier CG, Fonseca GAB,Kent J 2000. Biodiversity hotspots for conserva-tion priorities. Nature 403: 853-858.

Nakamura A, Catterall CP, House APN, Kitching RL,Burwell CJ 2007. The use of ants and other soil andlitter arthropods as bio-indicators of the impactsof rainforest clearing and subsequent land use.Journal of Insect Conservation 11(2): 177-186.

Newburn D, Reed S, Berck P, Merenlender A 2005.Economics and land-use change in prioritizingprivate land conservation. Conservation Biology19(5): 1411-1420.

Norton DA 1999. Forest reserves. In: Hunter MLJ ed.Maintaining biodiversity in forest ecosystems.Cambridge, UK, Cambridge University Press.Pp. 525-555.

Norton DA 2000. Conservation biology and private land:shifting the focus. Conservation Biology 14:1221-1223.

Norton DA 2002. Edge effects in a lowland temperateNew Zealand rainforest. DOC Science InternalSeries 27: 1-33.

Norton DA, Miller CJ 2000. Some issues and options forthe conservation of native biodiversity in ruralNew Zealand. Ecological Management and Res-toration 1: 26-34.

Ogle CC 1987. The incidence and conservation of animaland plant species in remnants of native vegetationwithin New Zealand. In: Saunders DA, Arnold GW,Burbidge AA, Hopkins AJM ed. Nature conserva-tion: the role of remnants of native vegetation.Australia, Surrey Beatty & Sons Pty Limited inassociation with CSIRO and CALM. Pp. 79-87.

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


Oliver I, Pearce S, Greenslade PJM, Britton DR 2006.Contribution of paddock trees to the conservationof terrestrial invertebrate biodiversity within grazednative pastures. Austral Ecology 31(1): 1-12.

Park GN 2000. New Zealand as ecosystems: the ecosystemconcept as a tool for environmental managementand conservation. Wellington, New Zealand, De-partment of Conservation.

Patton DR 1975. A diversity index for quantifying habitat"edge". Wildlife Society Bulletin 3(4): 171-173.

PCE 2001. Weaving resilience into our working lands:future roles for native plants on private land.Wellington, New Zealand, Parliamentary Com-misioner for the Environment.

Pettit NE, FroendRH, LaddPG 1995. Grazing in remnantwoodland vegetation—changes in species compo-sition and life form groups. Journal of VegetationScience 6(1): 121-130.

Pik AJ, Oliver I, Beattie AJ 1999. Taxonomic sufficiencyin ecological studies of terrestrial invertebrates.Australian Journal of Ecology 24(5): 555-562.

Porteus T 1993. Native forest restoration: a practical guidefor landowners. Wellington, New Zealand, QueenElizabeth II National Trust.

QEII Trust 1984. Open space covenants. Wellington, NewZealand, Queen Elizabeth II National Trust.

Radcliffe JE, Lefever KR 1981. Aspect influences onpasture microclimate at Coopers Creek, NorthCanterbury. New Zealand Journal of AgriculturalResearch 24(1): 55-66.

Ritchie H 2000. Reconciling conservation and productionvalues: a case study of the role of Landcare groupsin the Waikato region, New Zealand. In: Craig JL,Mitchell N, Saunders DA ed. Nature Conservation5: Nature conservation in production environ-ments: managing the matrix. Australia, SurreyBeatty& Sons. Pp. 524-533.

Saunders DA, Hobbs RJ, Margules CR 1991. Biologicalconsequences of ecosystem fragmentation: a re-view. Conservation Biology 5(1): 18-32.

Schon NL, Mackay AD, Minor MA, Yeates GW, HedleyMJ 2008. Soil fauna in grazed New Zealand hillcountry pastures at two management intensities.Applied Soil Ecology 40(2): 218-228.

Schwartz MW 1999. Choosing the appropriate scale ofreserves for conservation. Annual Review of Ecol-ogy and Systematics 30: 83-108.

Scrimgeour GJ, Kendall S 2003. Effects of livestockgrazing on benthic invertebrates from a nativegrassland ecosystem. Freshwater Biology 48(2):347-362.

Simberloff DS, Abele LG 1982. Refuge designandislandbiogeographic theory: effects of fragmentation.The American Naturalist 120(1): 41-50.

Smale MC, Ross CW, Arnold GC 2005. Vegetation reco-very in rural kahikatea (Dacrycarpus dacrydio-ides) forest fragments in the Waikato region, NewZealand, following retirement from grazing. NewZealand Journal of Ecology 29(2): 261-269.

Smale MC, Dodd MB, Burns BR, Power IL 2008. Long-term impacts of grazing on indigenous forestremnants on North Island hill country, New Zea-land. New Zealand Journal of Ecology 32(1):57-66.

Snyder BA, Hendrix PF 2008. Current and potentialroles of soil macroinvertebrates (earthworms,millipedes, and isopods) in ecological restoration.Restoration Ecology 16(4): 629-636.

Spooner P, Lunt I, Robinson W 2002. Is fencingenough? The short-term effects of stock exclu-sion in remnant grassy woodlands in southernNSW. Ecological Management and Restoration3(2): 117-126.

Standish RJ 2004. Impact of an invasive clonal herb onepigaeic invertebrates in forest remnants in NewZealand. Biological Conservation 116: 49-58.

Standish RJ, Robertson AW, Williams PA 2001. The impactof an invasive weed Tradescantia fluminensis onnative forest regeneration. Journal of AppliedEcology 38: 1253-1263.

Standish RJ, Williams PA, Robertson AW, Scott NA,Hedderley DI 2004. Invasion by a perennial herbincreases decomposition rate and alters nutrientavailability in warm temperate lowland forestremnants. Biological Invasions 6: 71-81.

Statistics New Zealand 2008. 2007 Agricultural productioncensus. Available online at: http://www.stats.govt.nz/economy/primary-production/2007+Agriculture+Production+Census.html (date accessed 6April 2009).

Stevenson BA 2004. Changes in phosphorus availabilityand nutrient status of indigenous forest fragmentsin pastoral New Zealand hill country. Plant andSoil 262(1-2): 317-325.

Suominen O, Danell K, Bergstrom R 1999. Moose,trees, and ground-living invertebrates: indirectinteractions in Swedish pine forests. Oikos 84(2):215-226.

Sweetapple PJ, Nugent G 2007. Ship rat demographyand diet following possum control in a mixedpodocarp-hardwood forest. New Zealand JournalofEcology31: 186-201.

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014


ter Braak CJF, Verdonschot PFM 1995. Canonical Cor-respondence Analysis and related multivariatemethods in aquatic ecology. Aquatic Sciences57(3): 255-289.

ter Braak CJF, Smilauer P 1997. Canoco for Windows.Version 4.02. Wageningen, The Netherlands,Centre for Biometry.

Terralink 2004. New Zealand land cover database(LCDB2). Wellington, New Zealand, TerralinkInternational Limited. 14 p.

Thompson K, Hodder APW, Edmonds AS, Irving R1983.Lowland forests in New Zealand. Hamilton, NewZealand, University of Waikato.

Tilman D, May RM, Lehman CL, Nowak MA 1994. Ha-bitat destruction and the extinction debt. Nature371: 65-66.

Timmins SM 2002. Impact of cattle on conservation landlicensed for grazing in South Westland, NewZealand. New Zealand Journal of Ecology 26(2):107-120.

Timmins SM, Williams PA 1991. Weed numbers in NewZealand forest and scrub reserves. New ZealandJournal of Ecology 15(2): 153-162.

Toft RJ, Harris RJ, Williams PA 2001. Impacts of the weedTradescantia fluminensis on insect communitiesin fragmented forests in New Zealand. BiologicalConservation 102: 31-46.

Tompkins DM, Veltman CJ 2006. Unexpected conse-quences of vertebrate pest control: predictionsfrom a four-species community model. EcologicalApplications 16(3): 1050-1061.

Tscharntke T, Steffan-Dewenter I, Kruess A, Thies C2002a. Characteristics of insect populations onhabitat fragments: a mini review. Ecological Re-search 17(2): 229-239.

Tscharntke T, Steffan-Dewenter I, Kruess A, Thies C2002b. Contribution of small habitat fragments toconservation of insect communities of grassland-cropland landscapes. Ecological Applications 12:354-363.

Tylianakis JM, Didham RK, Bascompte J, Wardle DA2008. Global change and species interactionsin terrestrial ecosystems. Ecology Letters 11:1351-1363.

Wardle DA, Barker GM, Yeates GW, Bonner KI, GhaniA 2001. Introduced browsing mammals in NewZealand natural forests: aboveground and below-ground consequences. Ecological Monographs71: 587-614.

Whaley PT, Clarkson BD, Smale MC 1997. ClaudelandsBush: ecology of an urban kahikatea (Dacrycarpusdacrydioides) forest remnant in Hamilton, NewZealand. Tane 36: 131-155.

Wheeler QD, McHugh JV 1987. A portable and conver-tible "Moczarski/Tullgren" extractor for fungusand litter Coleoptera. The Coleopterists Bulletin41: 9-12.

Wilson EO 1987. The little things that run the world (theimportance and conservation of invertebrates).Conservation Biology 1(4): 344-346.

Wiser SK, Allen RB, Clinton PW, Platt KH 1998. Commu-nity structure and forest invasion by an exotic herbover 23 years. Ecology 79(6): 2071-2081.

Woinarski JCZ, Andersen AN, Churchill TB, Ash AJ 2002.Response of ant and terrestrial spider assemblagesto pastoral and military land use, and to landscapeposition, in a tropical savanna woodland in nort-hern Australia. Austral Ecology 27(3): 324-333.

Yeates GW, Williams PA2001. Influence of three invasiveweeds and site factors on soil microfauna in NewZealand. Pedobiologia 45: 367-383.

Dow

nloa

ded

by [

Ada

ms

Stat

e U

nive

rsity

] at

10:

01 0

3 N

ovem

ber

2014

the interactive effects of livestock exclusion and mammalian pest control on the restoration of...

Documents