landscape trajectory of natural boreal forest loss as an...

Contributed Paper

Landscape trajectory of natural boreal forest loss asan impediment to green infrastructure

Johan Svensson ,1 ∗ Jon Andersson,2 Per Sandström,3 Grzegorz Mikusiński,4,5

and Bengt Gunnar Jonsson1,6

1Department of Wildlife, Fish and Environmental Studies, Swedish University of Agricultural Sciences, 901 83 Umeå, Sweden2Sweco Environment AB, Umestan Företagspark Hus 12, Box 110, 901 03 Umeå, Sweden3Department of Forest Resource Management, Swedish University of Agricultural Sciences, 901 83 Umeå, Sweden4Grimsö Wildlife Research Station, Department of Ecology, Swedish University of Agricultural Sciences, 730 91 Riddarhyttan, Sweden5School for Forest Management, Swedish University of Agricultural Sciences, Box 43, 739 21 Skinnskatteberg, Sweden6Department of Natural Sciences, Mid Sweden University, 851 70 Sundsvall, Sweden

Abstract: Loss of natural forests by forest clearcutting has been identified as a critical conservation chal-lenge worldwide. This study addressed forest fragmentation and loss in the context of the establishmentof a functional green infrastructure as a spatiotemporally connected landscape-scale network of habitatsenhancing biodiversity, favorable conservation status, and ecosystem services. Through retrospective analysisof satellite images, we assessed a 50- to 60-year spatiotemporal clearcutting impact trajectory on naturaland near-natural boreal forests across a sizable and representative region from the Gulf of Bothnia tothe Scandinavian Mountain Range in northern Fennoscandia. This period broadly covers the whole forestclearcutting period; thus, our approach and results can be applied to comprehensive impact assessmentof industrial forest management. The entire study region covers close to 46,000 km2 of forest-dominatedlandscape in a late phase of transition from a natural or near-natural to a land-use modified state. We founda substantial loss of intact forest, in particular of large, contiguous areas, a spatial polarization of remainingforest on regional scale where the inland has been more severely affected than the mountain and coastalzones, and a pronounced impact on interior forest core areas. Salient results were a decrease in area of thelargest intact forest patch from 225,853 to 68,714 ha in the mountain zone and from 257,715 to 38,668 hain the foothills zone, a decrease from 75% to 38% intact forest in the inland zones, a decrease in largest patchcore area (assessed by considering 100-m patch edge disturbance) from 6114 to 351 ha in the coastal zone,and a geographic imbalance in protected forest with an evident predominance in the mountain zone. Theseresults demonstrate profound disturbance of configuration of the natural forest landscape and disruptedconnectivity, which challenges the establishment of functional green infrastructure. Our approach supportsthe identification of forests for expanded protection and conservation-oriented forest landscape restoration.

Keywords: change detection, clearcutting, continuity forest, continuous cover forest, forest core areas, forestfragmentation, landscape configuration, satellite image, Sweden

Trayectoria Natural de la Pérdida de Bosque Boreal Natural como Impedimento para la Infraestructura Verde

Resumen: La pérdida de bosques naturales por causa de la tala uniforme de árboles en los mismos ha sidoidentificada como un reto muy importante para la conservación global. Este estudio abordó la fragmentacióny pérdida de bosques en el contexto del establecimiento de una infraestructura verde funcional como una redde hábitats a escala de paisaje conectados espacio-temporalmente que mejoren la biodiversidad, los estadosfavorables de conservación y los servicios ambientales. Por medio de un análisis retrospectivo de imágenessatelitales evaluamos una trayectoria de impacto espacio-temporal de 50 a 60 años de tala uniforme sobrebosques boreales naturales y casi naturales en una región considerable y representativa desde el Golfo de

∗email [email protected] impact statement: Profound transformation of natural forests into human-made landscapes hampers Aichi targets and practical greeninfrastructure implementation.Paper submitted December 13, 2017; revised manuscript accepted May 25, 2018.

152Conservation Biology, Volume 33, No. 1, 152–163C© 2018 Society for Conservation BiologyDOI: 10.1111/cobi.13148

http://orcid.org/0000-0002-0427-5699

Svensson et al. 153

Botnia hasta la Cordillera Escandinava en el norte de Fenoescandia. Este rango cubre todo el periodo detala uniforme en el bosque en términos generales; por lo tanto, nuestra metodoloǵıa y resultados puedenaplicarse a la evaluación completa del impacto del manejo industrial de bosques. Toda la región de estudiocubŕıa hasta 46,000 km2 de paisaje dominado por bosque en una etapa tardı́a de la transición entre el estadonatural o casi natural y el estado de uso de suelo modificado. Encontramos una pérdida sustancial de bosqueintacto, particularmente para áreas grandes y contiguas, una polarización espacial del bosque restante a unaescala regional en la que tierra adentro hay mayores afectaciones que en las zonas montañosas y costeras, yun impacto pronunciado sobre las áreas nucleares de los bosques interiores. Los resultados salientes fueronuna disminución en el área del fragmento más grande de bosque intacto de 225, 853 a 68, 714 ha enla zona montañosa y de 257, 715 a 38, 668 ha en la zona de pie de monte, una disminución del 75%al 38% de bosque intacto en las zonas tierra adentro, una disminución en el área nuclear del fragmentomás grande (valorada al considerar 100-m de perturbación al borde del fragmento) de 6, 114 a 351 haen la zona costera, y un desbalance geográfico en los bosques protegidos con una evidente mayoŕıa en lazona montañosa. Estos resultados demuestran una perturbación profunda de la configuración del paisajede bosque natural y una conectividad interrumpida, lo que presenta un reto para el establecimiento de unainfraestructura verde funcional. Nuestro enfoque sustenta la identificación de bosques para su protecciónexpandida y la restauración del paisaje de bosque orientada hacia la conservación.

Palabras Clave: áreas nucleares de bosque, cobertura continua de bosque, configuración de paisaje, con-tinuidad del bosque, detección de cambios, fragmentación de bosque, imagen satelital, Suecia, tala uniforme deárboles

��: ��,��,�� (Fennoscandia)�� (Gulf of Bothnia)�� (Scandinavian Mountain Range)�� 50-60��,�� 46,000 ��, ��,��,��,��,�� : �� 225,853 �� 68,714 ��, �� 257,715�� 38,668��,�� 75%�� 38%,�� (�� 6,114�� 351 �� 100 ��), �� (��) ��, ��, ��, ��: ��; ��: ��

��:��,��,��,��,��,��,��,��,�

Introduction

Globally, the ongoing loss of natural boreal forests hasbeen assessed as the second largest after the forest lossin the tropics, in both absolute and proportional terms(Hansen et al. 2013). In Fennoscandia, this loss is pre-dominantly caused by forest clearcutting (Kouki et al.2001), whereas in North America and parts of Eurasiaalso beyond-baseline levels of forest fires are a large for-est loss source (de Groot et al. 2013). Although intactforests still persist over vast areas in many boreal regions(Potapov et al. 2008; Gauthier et al. 2015), the frontiersof natural forest landscapes are being modified and arecontracting at high rates (Potapov et al. 2017). In theFennoscandian boreal biome, where forestry has had ma-jor and widespread impacts (Kouki et al. 2001), clearcut-ting continues in the remaining fragments of natural andnear-natural forests (Forest Europe 2015), notwithstand-ing policies that advocate increasing protection rates,

landscape-context approaches, and awareness that favor-able conservation status for many target forest habitatsand species is not secured (e.g., Van Teeffelen et al. 2012;Sverdrup-Thygeson et al. 2014; Orlikowska et al. 2016).The continued expansion of the human footprint (Tuckeret al. 2018) and loss of intact forest landscapes impedesconservation of biodiversity and ecosystem services (e.g.,Watson et al. 2018). Reaching environmental policy goalsin the Fennoscandian forest landscape, such as the AichiBiodiversity Targets (CBD 2010), thus demands rigorousefforts.

The broad-scale and long-term impacts of forestry haveraised much concern about the ecological integrity of theremaining natural forest fragments (Jönsson et al. 2009;Kuuluvainen 2009; Moen et al. 2014). Remnant forestswith temporal and spatial continuity of key habitat andecosystem attributes function as biodiversity hotspots(e.g., Paillet et al. 2010) and thus play a critical role inforest ecosystem conservation (Hanski 1999; Ranius &

Conservation BiologyVolume 33, No. 1, 2019

154 Boreal Forest and Green Infrastructure

Kindvall 2006; Nordén et al. 2014). Forest continuity im-plies attributes of old-growth habitats existing for severaltree generations within a defined patch and an uninter-rupted presence of such patches in a landscape matrix(e.g., Nordén et al. 2014). Forest continuity is associatedwith interior core areas, which are less influenced byproximity to peripheral and external disturbance factorsand thus may provide a refuge for natural old-growthstructures and processes (Riitters et al. 2016; Pfeifer et al.2017). The conservation significance of core areas withcontinuity attributes is very high on both habitat and land-scape scales, which is reflected in nature conservationpolicy and planning (e.g., Angelstam et al. 2011; Aksenovet al. 2014; Müller et al. 2018).

In addressing forest conservation and sustainability inlandscapes dominated by managed forests, argumentsand knowledge have accumulated on the need to in-crease forest protection and to expand forest restora-tion, green tree retention, multifunctional forestry andother conservation-oriented management approaches(e.g., Gustafsson et al. 2012; Lindenmayer et al. 2012).Both remaining core areas and the surrounding matrixneed to be considered in landscape sustainability plan-ning and practice to secure persistence and resilienceof ecosystem functions, biodiversity, and ecosystem ser-vices (Mikusiński et al. 2007; Swift & Hannon 2010;Van Teeffelen et al. 2012). In a forest landscape char-acterized by management, forests in a modified state arealso needed as pathways for species movement and ex-pansion of habitats (Bengtsson et al. 2003) to supportthe metapopulation capacity of the landscape (Hanski &Ovaskainen 2000).

The concept of green infrastructure (GI) has expandedfrom promoting ecosystem values and human well-beingin urban environments (Tzoulas et al. 2007) to amainstream EU environmental policy (EC 2013). The EUmember states are presently implementing, or preparingimplementation of, GI (e.g., Snäll et al. 2016). Green in-frastructure is a strategic and operation planning networkof spatiotemporally connected natural and seminaturalhabitats that support and improve ecological connec-tivity, favorable conservation status, ecosystem services,and ecosystem multifunctionality at multiple scales evenas climate and land use change (e.g., Benedict & MacMa-hon 2002; Mehtälä & Vuorisalo 2007; Liquete et al. 2015).For managed landscapes, biodiversity conservation andsustainability of ecosystems and their services require ap-proaches that address and mitigate habitat fragmentation(Johnstone et al. 2016). Forest areas suitable for protec-tion and restoration need to be identified (Halme et al.2013; Rybicki & Hanski 2013; Müller et al. 2018). Hence,accurate mapping of the most important forest habitatsand patches provides important input for GI implemen-tation. For mapped gross data based on remote-sensinginformation, which is currently widely applied forlandscape impact and change detection (e.g., Sverdrup-

Thygeson et al. 2016; Tyukavina et al. 2016; Potapovet al. 2017), it is of particular value to consider and assessecologically relevant parameters for defining the mostimportant GI components. Because ecological connec-tivity may be used to assess the ecological performanceof forest habitats (Lindenmayer et al. 2006), mappingconnectivity of continuity forests thus provides neededinput.

We addressed the challenges of establishing afunctional GI across a large geographic region extendingfrom the Gulf of Bothnia to the Scandinavian MountainRange in northern Sweden. The region exemplifies asignificant and representative part of the Fennoscandianboreal biome, which has been greatly affected by forestmanagement. Only 4% of the productive and 7% of allforest land in Sweden is formally protected at presentwith the majority located in a narrow zone in themountainous area (Swedish University of AgriculturalSciences 2017). This is very far from the 17% called forin the Aichi target 11. Through retrospective analyses ofsatellite images, we sequentially detected clearcuts overthe last 50–60 years and mapped the forest-landscapechange trajectory of lost and remaining forest. Thisperiod broadly covers the industrial forest clearcuttingera in the study region (Lundmark et al. 2013). Weregard the initiation of widespread clearcutting in themiddle of the 20th century, with its large harvestingareas, soil scarification, and artificial regeneration (Eckeet al. 2013), as the onset of the transition to a managedforest landscape. Our study rationale was that remainingforest fragments, with or without traces of earliermanagement, represent components of a functional GIto which protection, restoration, and other conservation-oriented management should be directed. Because ouranalyses were based exclusively on remote-sensing data,we considered forest patches identified as not beingclearcut as proxy continuity forest (pCF). In a similarapproach in pan-tropical forests, Tyukavina et al. (2016)applied the term hinterland forest to patches identifiedthrough remote sensing as not recently disturbed. Thespatiotemporal resolution we applied allowed for op-erational planning approaches that complement earliermapping of intact boreal landscape at pan-national andlarger scales, such as those by Potapov et al. (2008, 2011).

Our main research questions concerned the spa-tiotemporal changes in landscape-level configuration offorests as a consequence of long-term forest clearcutting:How are remaining pCF distributed across the gradientfrom coast to mountain? How have the amount anddistribution of pCF changed over time? How does thedistribution of pCF relate to protected forest and tototal forest land area over time? And, can spatiotemporalforest core area be assessed by considering the extentof edge disturbance depths? We considered our resultsrelative to boreal forest loss and fragmentation and toprospects for establishing a functional GI in a landscape


Svensson et al. 155

that has been and most likely will continue to bedominated by forest management.

Methods

Study Region

The 45,755-km2 study region represents a forest-landscape transition extending from the coastal mid-boreal to the northern boreal and the birch-dominated(Betula pubescens ssp. czerepanovii) subalpine zones(Gustafsson & Ahlén 1996). The predominant treespecies are Scots pine (Pinus sylvestris) and Norwayspruce (Picea abies), 46% and 22% of standing volumerespectively, on productive forest in the VästerbottenCounty (Swedish University of Agricultural Sciences2017), which covers most of the study region. The el-evational gradient extends from sea level to about 900 masl at the alpine tree line, and with associated macrocli-matic and forest site productivity changes as elevationincreases.

The study region is characterized by a managedforest landscape (Fig. 1), where clearcutting has beena dominant management method since the mid-1900s(Ecke et al. 2013; Lundmark et al. 2013). The coast tomountain gradient represents a historical progressionof the south-north and east-west movement of modernforestry. More evident forest exploitation has occurredsince the mid-1800s, including the first wave of large-diameter saw-timber harvesting followed by a period ofselective logging and some clearcutting.

Data Sources

Across a large landscape with multiple land forms, land-cover types, and landowner categories, remote sensingwith ancillary data presents an opportunity to compileholistic information (e.g., Kennedy et al. 2014). The Land-sat program was launched in 1972 as the first programtailored for global cover (Wulder et al. 2012). Satelliteimage-based change detection has since been used suc-cessfully to map, for example, land-use change (Muukko-nen et al. 2012), deforestation (Potapov et al. 2017), andminimally disturbed forests (Tyukavina et al. 2016).

We used 24 Landsat images from 1973 to 2014 to iden-tify and define remaining forest that was not clearcut(Supporting Information). The red spectral band wasused because this wavelength is suitable for distinguish-ing changes in forest cover (Potapov et al. 2008, 2011).We aimed to create a spatiotemporally continuous dataset along a gradient from the coast to the mountains, soscenes with minimal amount of clouds were pooled intoa patchwork of 7 satellite scene batches that togetherdetermined the extent of the study region and definedthe zones. For correcting remaining minor cloud overlay,we used supplementary images from the same year. How-

ever, because recent clearcuts and older clearcuts withor without young regenerating forests can be detectedand the site productivity and thus tree growth capacityis rather poor in the study region, clearcuts prior to theacquisition year of the earliest available images could bedetected. This allowed us to interpret 1 (coastal area) to2 (mountain area) decades farther back and to generatea 50- to 60-year forest landscape change sequence.

Data on formal protection were downloaded from Sky-ddad natur (Swedish Environmental Protection Agency2017). Data on land cover were downloaded from theGSD-Road Map (National Land Survey 2017), which is acontinuously updated database.

Change Detection, Spatiotemporal Stratification, and Analyses

We applied spectral change detection through maxi-mum likelihood classification for each image pair to de-tect clearcuts (Supporting Information). To make batchpairs compatible for change detection, we stretched andhistogram-matched each image. As training samples forthe supervised classification, we randomly selected 6training polygons per 1000 km2 with forest areas thathad been clearcut and forest areas that had not beenclearcut, respectively. For each time step, a new set of6 random training polygons was selected covering eitherclearcut or not-clearcut forests. For detecting not-clearcutforest areas in the earliest images, we used new clearcutdetected in the next later images. Through this proce-dure, we sequentially detected and withdrew clearcutsand a net pCF data set of remaining intact forests wasgenerated. To avoid including very small pCF fragmentsin the analyses, all pCF polygons


Figure 1. Landsat 8 satelliteimage (2013) of an area centralin the western inland zone of thestudy region showing thedistribution of proxy continuityforests (green), legally protectedareas (black line), clearcut forest(light green to yellow; brightercolor indicating more recentclearcut), water bodies (darkblue), and open mires,agriculture land, and othernonforested surfaces (pink). Therelatively large protected area inthe lower right of center is the2369-ha Björnlandet NationalPark (63°58’N, 18°1’E).

Figure 2. (a) Northern Sweden with vegetationbiomes according to Gustafsson and Ahlén (1996)(blue, alpine; green, northern boreal; beige, middleboreal; yellow, southern boreal; black line, studyregion). (b) Compiled satellite scene batches 1–7(batches 1–5 build up the mountain [1], foothill [2],western inland [3], eastern inland [4], and coastal [5]zones; 6 and 7 are substantially smaller than theothers). (c) Distribution in 2013 of remaining proxycontinuity forest (green), clearcuts (yellow), andnonforested surfaces (gray and white, alpine area,mires, water bodies, agriculture, or urban lands)projected on top of a gray-scale map of thesurrounding land cover.

patch-area distribution, all pCF patches that through thissize reduction became

Svensson et al. 157

Figure 3. (a) Percentage of remaining proxycontinuity forest (pCF) relative to the total forest landin 20% classes calculated for 1-km2 raster squares(n = 41,734) for 5 time steps following satellite imageyears and (b) examples (15 × 15 km) of pCF,clearcuts, and other surfaces (white, i.e. alpine, mires,water bodies, agriculture, or urban lands) for themountain, foothill, western inland, eastern inland,and coastal zones.

zones. The share of protection increased gradually forall zones but remained low, particularly in the easterninland and coastal zones. The mountain zone, with 14.3%protection, contributed more than 50% of the protectedarea in the study region.

The spatial characteristics and thus ecological func-tionality of the forest landscape has been altered; re-maining functional pCF core areas relative to differentedge depth assessments were greatly reduced in sizeand fragmented compared with the pCF patches (Fig. 5).The largest patch, mean patch size, and proportion ofcore area decreased considerably over time for all zones(Table 1 & Supporting Information). In the 1973 timestep, the largest patch in all zones was from around225,000 to just below 300,000 ha. The largest patch(225,853 ha) in the mountain zone in 1973 encompassed40% of all forest land in the zone (Supporting Informa-

Figure 4. Change over time in (a) the percentage ofproxy continuity forest (pCF) and (b) percentage offormally protected forest land (including nationalparks, nature reserves, biotope protection areas, andnature protection agreements) for the mountain,foothill, western inland, eastern inland, and coastalzones.

tion), whereas the largest patch in 2014 encompassedonly 1%, based on 100-m edge depth assessment. Witha 100-m edge depth, the largest patch sizes decreasedfrom 23,468 to 6102 ha (mountain) and from 6114 to351 ha (coastal). For the mountain and foothill zones, thedecrease in forest patch area between the 2 earliest timesteps was considerable, whereas for the other zones thedecrease in size of the largest patches was more gradual.A salient result was the decrease in the largest patch inthe foothills zone from 257,715 ha in 1973 to 38,668 hain 1983.

The mean patch size decreased continually over timefor all zones (Table 1). However, for some zones andtime steps, the mean area increased when patch area wasdetermined by edge depth assessment. This is understoodas an effect of fragmentation of large patches into severalsmaller patches of relatively large individual size. For ex-ample, for the foothill zone, we identified 3668 patches of


Tabl

e1.

The

larg

esta

ndm

ean

patc

hsi

zes

ofpr

oxy

cont

inui

tyfo

rest

(pCF

)an

dth

epe

rcen

tage

offu

nctio

nalc

ore

area

rela

tive

toto

talf

ores

tlan

dar

eafo

rth

epC

Fpa

tche

san

dfo

rth

epC

Fpa

tche

s�

1ha

with

redu

ced

patc

har

ea25

-,50

-,an

d10

0-m

dist

ance

from

patc

hed

gefo

r5

time

step

sfo

llow

ing

sate

llite

imag

eye

ars

for

mou

ntai

n,fo

othi

ll,w

este

rnin

land

,eas

tern

inla

nd,a

ndco

asta

lzo

nes.

pC

Fpa

tch

Core

are

a2

5m

from

edge

Core

are

a5

0m

from

edge

Core

are

a1

00

mfr

om

edge

pa

tch

size

(ha

)pa

tch

size

(ha

)pa

tch

size

(ha

)pa

tch

size

(ha

)

Zon

eY

ear

larg

est

mea

n%

core

are

ala

rges

tm

ean

%co

rea

rea

larg

est

mea

n%

core

are

ala

rges

tm

ean

%co

rea

rea

Mo

un

tain

1973

225,

853

139

9013

2,76

315

672

60,4

5212

959

23,4

6892

4019

8614

4,03

889

7959

,989

9460

38,0

9080

4794

5157

3020

0178

,356

5569

43,4

1853

4917

,138

4636

6484

3521

2005

75,2

8850

6742

,223

4847

16,1

4142

3564

8433

1920

1468

,714

3962

39,6

1538

4213

,829

3530

6102

3017

Foo

thill

1973

257,

715

8779

48,3

3111

263

41,9

4595

5115

,838

8035

1986

38,6

8847

6413

,221

5547

5548

4736

3141

4022

1990

30,6

6441

6177

7846

4344

3340

3231

4135

1920

0218

,989

2751

5021

2733

4074

2322

2693

2011

2013

7036

1944

4529

1725

3695

1616

2586

157

Wes

tern

inla

nd

1973

288,

553

7571

93,9

2595

5659

,297

8545

13,8

3869

3019

8022

3,68

656

6421

,250

4840

25,3

4665

3871

5553

2519

9034

,917

3856

8033

4440

8033

4430

2474

3718

2002

14,3

6723

4526

5924

2814

8222

1973

819

1020

1349

0216

3815

0616

2192

015

1371

413

6Ea

ster

nin

lan

d19

7629

2,74

810

975

188,

821

125

6043

,686

9949

10,0

8570

3219

8024

6,63

379

6760

,263

9652

21,1

6772

4179

0351

2619

9016

3,17

456

6015

,269

6345

9126

4834

4244

3520

2002

28,8

1235

5137

3931

3425

2924

2378

618

1120

1362

4622

4313

2318

2586

515

1545

711

6C

oas

tal

1976

245,

623

159

8418

0,96

217

867

97,7

5913

154

6114

6835

1990

200,

586

101

7512

2,61

598

5726

,595

6943

2775

3625

1999

181,

846

7769

62,1

0266

5012

,458

4436

1115

2519

2005

115,

038

5763

13,3

9641

4346

8228

2944

017

1420

1358

,789

3857

6806

2635

1420

1923

351

1310


Svensson et al. 159

(Supporting Information). For a core area assessed byedge depth of 100 m, there was between 40% (mountain)and 30% (western inland) total remaining core area rela-tive to total forest area in the 1973 step, compared withbetween 17% (mountain) and 6% (western and easterninland) in 2013. For the 50- and 25-m edge assessments,remaining core area was slightly higher but constantlydecreased over time. From 1973 to 2013, the relative de-crease in pCF proportion of total forest land was from 54%(western inland) to 69% (mountain) (original pCF patch[Table 1]); in mean patch size was from 20% (easterninland) to 30% (mountain); and in largest patch size wasfrom 2% to 3% (western and eastern inland, and foothills)to 28% (mountain). The decreasing trend line for meanpatch size approached the decreasing trend line of largestpatch (Supporting Information). Information on relativedecrease in assessment of 100-m edge is given in Support-ing Information. For mean patch size and largest patchsize, the most profound changes came sequentially laterin time from the foothills (1983) to the western inland(1993) and the eastern inland (2003) zones.

Thus, in addition to a regional polarization with thecentral inland areas more heavily affected than the moun-tain and coastal areas, we found that fragmentation, patchsize, and core-area reduction were extensive and con-tinuous. Our results also indicated an eastward move-ment of the most profound impact over time, that thelargest patches were mostly affected, and a homogeniza-tion of pCF patch-area distribution across the entire studyregion.

Discussion

The boreal forest biome has a relatively high proportionof intact forests and low degree of human footprint rel-ative to other major forest biomes (Gauthier et al. 2015;Watson et al. 2018). In Fennoscandia, the boreal forestlandscape escaped major and widespread forest loss for along time and has been perceived as Europe’s last wilder-ness area (Kuuluvainen et al. 2017). The continuing im-pact of forest clearcutting and other land use, however,has generated substantial attention on degradation, de-cline, and fragmentation of forest landscapes and habitatswith presumed or actual high nature conservation values(e.g., Moen et al. 2014; Potapov et al. 2017; Watson et al.2018). As concluded by many (e.g., Aune et al. 2005;Lindenmayer et al. 2006; Potapov et al. 2017), the ecolog-ical qualities and spatial connectivity of remaining intactforests need to be mapped and assessed for conservationactions such as forest protection and green infrastruc-ture (GI) planning and implementation. Further, in-depthspatial analyses aimed at identifying the most efficientways to improve the GI functionality are necessary (e.g.,Mönkkönen et al. 2014). To minimize the adverse effectsof fragmentation and increase the conservation benefits, a

general recommendation is that habitat fragments shouldbe protected in clusters rather than scattered randomly(Rybicki & Hanski 2013). The spatiotemporal forest-landscape change and forest-fragmentation trajectory wehave documented, across a representable and sizable re-gion of the boreal biome, reveal recent and pronouncedimpacts. Loss of intact forest continues, which challengesGI implementation and conservation attainment, even inthe part of the European boreal biome where significantlevels of forests of high conservation value still exist.

Our results demonstrate a substantial and rapidloss of natural and near-natural forests over the last50–60 years of intensive forest management. The remain-ing pCF patches are largely fragmented, the pCF-patchareas and functional core areas have decreased substan-tially, and the natural landscape configuration has beendisrupted. Salient examples are the reduction in the areaof the largest patches during the study period and thedramatic effects when considering forest core area byassessing depths of edge disturbance. We expect similarpatterns in other regions of the northern European bo-real biome. The national average harvest rate in Swedenexceeds that of the study region (0.85% [Swedish Uni-versity of Agricultural Sciences 2017]) and is similar tothe rate in Finland (0.7–0.8% [Luke 2017]). Our resultsfurther show a polarization on a regional scale with aparticularly low proportion of remaining pCF patches inthe inland relative to the coastal and in particular thefoothill and mountain zones, a general homogenizationof patch-size distribution, and a gradual loss of the largestremaining patches across the study region. The generic,combined effects of these changes may result in that theexpected loss of biodiversity and ecosystem services as-sociated with intact natural and near-natural forest in theinland could be particularly severe. Following the pre-dictions from the species–area relationship (Rybicki &Hanski 2013) and immediate as well as future extinctiondebts (Hanski & Ovaskainen 2002), increased fragmenta-tion and smaller area of remaining intact forest patchesmay cause extirpations and overall decline of speciesdiversity, particularly for species that are disturbancesensitive and dependent on old-growth and continuityforests. The observed loss of pCF core area, representinga key entity with interior-ecosystem habitat qualities lessinfluenced by edge disturbance, needs direct attentionin strategic and operational GI implementation. In thiscontext, our results are based on automatic pCF identi-fication in remote-sensing-generated data; thus, field in-ventories are needed to validate the actual ecological andbiodiversity premises before GI planning is conducted.

A higher proportion of pCF relative to total forestland in the mountain zone was expected due to thelater arrival of modern forestry to this area relative tothe coastal and inland areas and due to the advance ofnature conservation since the 1970s and the protectionemphasis on northwestern Sweden. However, our results



Figure 5. (a) Distribution in 2013 ofproxy continuity forests (pCFs) functionalcore area (shaded). The functional corearea was estimated for pCF patches�1 ha by assessing edge disturbanceeffects by reducing pCF patch area with25-, 50-, and 100-m distance from patchedge. (b) Examples (15 × 15 km, sameexamples as in Fig. 2) of distribution offunctional pCF core area in mountain,foothill, western inland, eastern inland,and coastal zones.

also indicates that coastal areas had a higher proportion ofpCF patches relative to the inland. Because clearcuttingwas initiated earlier in the coastal area, mature reforestedareas may have been detected as pCF. In fact, a recentpilot survey on a similar type of data in the coastal area tothe south of our study region indicated that about 40% ofdetected pCF areas may be managed stands (Ahlcronaet al. 2017). Hence, extended retrospective temporalsequence and complementary methods are required toidentify remaining pCF in the coastal area. Given the nor-mal rotation period of forest harvesting of 80–100 years inthe coastal area, it needs to be evaluated whether matureregenerated forests can support conservation functionsand thus should be included in a functional GI. However,the actual presence of pCF in the coastal zone and alongthe more accessible, low-elevation and more fertile forestsites along the river valleys inland from the coast, shouldbe assumed to be lower than reported in this study, inparticular in terms of their ecological-continuity core-areaqualities. Furthermore, the abovementioned pilot surveyindicates that open and semiopen forests are a source oferror. Hence, we excluded such forests in our study toimprove data consistency. However, because open andsemiopen forests on both productive and nonproductiveforest land can harbor significant continuity values, dataand methods should be improved to also allow assess-ment of such lands.

Remaining pCF patches were detected within, and,more importantly, also outside already protected areas.Thus, our results provide information on forest areathat based on intrinsic conservation value or spatiallocation can complement protected areas and contributeto building a functional GI. Managed forest can alsocontribute in a GI context. Red-listed forest bryophytesand lichens may survive and possibly colonize harvested

forest areas if adequate conservation measures areapplied (Perhans et al. 2014) and if dispersal sourcesexist within close proximity (Hanski 1999, 2011). Inmanaged and fragmented forest landscapes, remainingminimally disturbed continuity forest patches supportspecies and ecological processes that require more stableold-growth conditions (e.g., Paillet et al. 2010; Dondinaet al. 2017). Such patches need to occur on a significantportion of the landscape (e.g., Gustafsson et al. 2012). Inline with the Convention on Biological Diversity (2010),threshold levels of 10–30% protected area have beensuggested (Hanski 2011).

We have highlighted the importance of forest conti-nuity core areas as key components in a functional GI.Edge disturbance sensitivity varies by species and spatialcharacteristics of the patches (Murcia 1995; Aune et al.2005; Siitonen et al. 2005). Superimposing effects of for-est loss, fragmentation, and loss of core area create ag-gravating circumstances for conservation values (Riitterset al. 2016; Pfeifer et al. 2017). Even though about halfof the forest land was not subject to clearcutting duringthe period we studied, the net effect on the remainingfunctional pCF core areas was pronounced. We stress thatacknowledging edge disturbance in conservation plan-ning and design of a functional GI, regardless of assessededge disturbance depth, should be a standard procedurein particular in management-dominated forest landscapeswhere buffering the most valuable forest areas is needed.Forecasting the trends of fragmentation and forest land-scape alteration demonstrated in this study into the futurejeopardizes achievement of Aichi Biodiversity targets, inparticular target 7, on sustainable management, biodi-versity, and conservation; target 11, on setting aside aminimum of 17% of terrestrial areas; and target 15, onrestoring degraded ecosystems (CBD 2010).


Svensson et al. 161

The current distribution of formally protected forestsis biased toward the remote mountain area, where forestsites with subalpine characteristics, including mountainbirch-dominated forests, are generally less productive.Protected forest area in the more accessible and produc-tive boreal eastern and coastal parts of the study regionincreased slowly over time but is still at very low levelsrelative to the Aichi 17% target. Furthermore, the frag-mentation and loss of intact forest has been particularlydramatic in these parts of the region. With the south-eastto north-west river valley terrain profile of the landscapeof northern Sweden, which determines the orientationof the main natural forest habitat connectivity patternsand species migration routes, the disrupted functionalGI links and networks from the coast to the mountainimply critical strategic and planning challenges. Conse-quently, we argue that conservation emphasis, includ-ing conservation-oriented forest restoration, also needsto be placed on the inland and coastal regions to se-cure connected GI components across the east-to-westgradient.

We provided evidence for extensive, rapid, and recentloss of natural or near-natural forest patches, fragmenta-tion, and pronounced forest landscape change across asizable region of the boreal forest biome. As a result offorest clearcutting, the landscape is in a late transitionalstage to a land-use-modified state. With ongoing climatechange, which is expected to affect the ecology andresilience of boreal forest (Kuuluvainen et al. 2017), thistransition needs attention to not jeopardize ecosystemadaptation capacity. To support strategic and operationalplanning for functional GI in forest landscapes and to ful-fill the quantitative and qualitative goals of the EU habitatand species directives and the CBD Aichi targets, thereis an urgent need for identification and directed actiontoward those valuable habitats that still exist. In the con-text of national and international conservation policies,it should be noted that pronounced fragmentation andloss of intact forest, in particular of the largest intactpatches in the foothills and inland region, are ongoingat great rates. Despite increasing overall rate of forestprotection, the proportion of protected forests remainsat very low levels relative to the global target of 17% anddisplays a marked geographical imbalance; the vast ma-jority of protected area is in the remote and relatively un-productive mountain zone. Complementary protectionis critical, as is conservation-oriented management andrestoration in the surrounding managed landscape. Theremaining intact forest patches represent optional targetentities for such directed actions. Our findings can informimplementation of GI as a conceptual approach to ad-dress connectivity of forest habitats and landscape-scalestrategic conservation planning aiming to strengthenand complement current networks of protectedforests.

Acknowledgments

This study has been completed with funding providedby the Swedish Environmental Protection Agency, Stock-holm, grant NV-03501-15. Initial steps in preparing thedata were performed with a grant (251-2011-1702) fromthe Swedish Research Council Formas. Both projectsare hosted by the Swedish University of AgriculturalSciences. We acknowledge assistance by Metria AB,Stockholm.

Supporting Information

Mapping, change detection, and data-management pro-cedures (Appendix S1), supporting study data (AppendixS2), and supporting data analyses (Appendices S3–S6) areavailable online. The authors are solely responsible forthe content and functionality of these materials. Queries(other than absence of the material) should be directedto the corresponding author.

Literature cited

Ahlcrona E, Giljam C, Keskitalo C, Klein J, Naumov V. 2017. Precis-erad kartering av kontinuitetsskog i Västernorrlands län. Metria påuppdrag av Naturvårdsverket (in Swedish).

Aksenov D, Kuhmonen A, Mikkola J, Sobolev N, editors. 2014. Thecharacteristics and representativeness of the protected area net-work in the Barents region. Vol. 29, pp. 189. Reports of the FinnishEnvironment Institute.

Angelstam P, Andersson K, Axelsson R, Elbakidze M, Jonsson, B-G,Roberge J-M. 2011. Protecting forest areas for biodiversity in Sweden1991–2010: the policy implementation process and outcomes onthe ground. Silva Fennica 45:1111–1133.

Aune K, Jonsson B-G, Moen J. 2005. Isolation and edge effects amongwoodland key habitats in Sweden: Is forest policy promoting frag-mentation? Biological Conservation 124:89–95.

Benedict MA, MacMahon ET. 2002. Green infrastructure: smart con-servation for the 21st century. Renewable Resources Journal 20:12–17.

Bengtsson J, Angelstam P, Elmqvist T, Emanuelsson U, Folke C, IhseM, Moberg F, Nyström M. 2003. Reserves, resilience and dynamiclandscapes. Ambio 32:389–396.

CBD (Convention on Biological Diversity). 2010. Strategic plan for bio-diversity 2011–2020 and the Aichi targets. Convention on BiologicalDiversity, Montreal. Available from https://www.cbd.int/sp/targets/(accessed March 2017).

de Groot WJ, Cantin AS, Flannigan MD, Soja AJ, Gowman LM, NewberyA. 2013. A comparison of Canadian and Russian boreal forest fireregimes. Forest Ecology and Management 294:23–34.

Dondina O, Orioli V, D’Occhio P, Luppi M, Bani L. 2017. How doesforest species specialization affect the application of the island bio-geography theory in fragmented landscapes? Journal of Biogeogra-phy 44:1041–1052.

EC (European Commission). 2013. Building a green infrastructurefor Europe. Publications Office of the European Union. 24 pp.https://doi.org/10.2779/54125.

Ecke F, Magnusson M, Hörnfeldt B. 2013. Spatiotemporal changes inthe landscape structure of forests in northern Sweden. ScandinavianJournal of Forest Research 28:651–667.


https://www.cbd.int/sp/targets/https://doi.org/10.2779/54125


Forest Europe. 2015. State of Europe’s forests 2015. Ministerial Confer-ence on the Protection of Forests in Europe, Madrid, Spain.

Gauthier S, Bernier P, Kuuluvainen T, Shvidenko AZ, SchepaschenkoDG. 2015. Boreal forest health and global change. Science 349:819–822.

Gustafsson L, Ahlén I, editors. 1996. Geography of plants and animals.National atlas of Sweden. SNA Publishing, Stockholm.

Gustafsson L, et al. 2012. Retention forestry to maintain multifunctionalforests: a world perspective. BioScience 62:633–645.

Halme P, et al. 2013. Challenges of ecological restoration: lessons fromforests in northern Europe. Biological Conservation 167:248–256.

Hansen MC, et al. 2013. High-resolution global maps of 21st-centuryforest cover change. Science 342:850–853.

Hanski I. 1999. Habitat connectivity, habitat continuity, and metapop-ulations in dynamic landscapes. Oikos 87:209–219.

Hanski I. 2011. Habitat loss, the dynamics of biodiversity, and a per-spective on conservation. Ambio 40:248–255.

Hanski I, Ovaskainen O. 2000. The metapopulation capacity of a frag-mented landscape. Nature 404:755–758.

Hanski I, Ovaskainen O. 2002. Extinction debt at extinction threshold.Conservation Biology 16:666–673.

Johnstone JF, et al. 2016. Changing disturbance regimes, ecologicalmemory, and forest resilience. Frontiers in Ecology and the Envi-ronment 14:369–378.

Jönsson MT, Fraver S, Jonsson B-G. 2009. Forest history and the devel-opment of old-growth characteristics in fragmented boreal forests.Journal of Vegetation Science 20:91–106.

Kennedy RE, et al. 2014. Bringing an ecological view of change toLandsat-based remote sensing. Frontiers in Ecology and the Envi-ronment 12:339–346.

Kouki J, Löfman S, Martikainen P, Rouvinen S, Uotila A. 2001. Forest frag-mentation in Fennoscandia: linking habitat requirements of wood-associated threatened species to landscape and habitat changes.Scandinavian Journal of Forest Research 16:27–37.

Kuuluvainen T. 2009. Forest management and biodiversity conserva-tion based on natural ecosystem dynamics in Northern Europe: thecomplexity challenge. Ambio 38:309–315.

Kuuluvainen T, Hofgaard A, Aakala T, Jonsson B-G. 2017. NorthFennoscandian mountain forests: history, composition, disturbancedynamics and the unpredictable future. Forest Ecology and Manage-ment 385:140–149.

Lindenmayer DB, et al. 2012. A major shift to the retention approachfor forestry can help resolve some global forest sustainability issues.Conservation Letters 5: 421–431.

Lindenmayer DB, Franklin JF, Fischer J. 2006. General managementprinciples and a checklist of strategies to guide forest biodiversityconservation. Biological Conservation 131:433–445.

Liquete C, Kleeschulte S, Dige G, Maes J, Grizetti B, Olah B, ZulianG. 2015. Mapping green infrastructure based on ecosystem servicesand ecological networks: a Pan-European case study. EnvironmentalScience & Policy 54:268–280.

Luke. 2017. Statistics database of the Natural Resources InstituteFinland. Available from http://stat.luke.fi/en (accessed March 2018).

Lundmark A, Josefsson T, Östlund L. 2013. The history of clear-cutting innorthern Sweden—driving forces and myths in boreal silviculture.Forest Ecology and Management 307:112–122.

Mehtälä J, Vuorisalo T. 2007. Conservation policy and the EU Habi-tats Directive: favourable conservation status as a measure ofconservation success. Environmental Policy and Governance 17:363–375.

Mikusiński G, Pressey RL, Edenius L, Kujala H, Moilanen A, NiemeläJ, Ranius T. 2007. Conservation planning in forest landscapes ofFennoscandia and an approach to the challenge of Countdown2010. Conservation Biology 21:1445–1454.

Moen J, et al. 2014. Eye on the Taiga: removing global policy imped-iments to safeguard the boreal forest. Conservation Letters 7:408–418.

Mönkkönen M, Juutinen A, Mazziotta A, Miettinen K, Podkopaev D,Reunanen P, Salminen H, Tikkanen O-P. 2014. Spatially dynamicforest management to sustain biodiversity and economic returns.Journal of Environmental Management 134:80–89.

Müller J, Noss RF, Thorn S, Bässler C, Leverkus AB, Linden-mayer DB. 2018. Increasing disturbance demands new policiesto conserve intact forest. Conservation Letters: https://doi.org/10.1111/conl.1244.

Murcia C. 1995. Edge effects in fragmented forests: implications forconservation. Trends in Ecology & Evolution 10:58–62.

Muukkonen P, Angervuori A, Virtanen T, Merila J. 2012. Loss and frag-mentation of Siberian jay (Perisoreus infaustus) habitats. BorealEnvironment Research 17:59–71.

National Land Survey. 2017. GSD-road data. Available from https://www.lantmateriet.se/sv/Kartor-och-geografisk-information/ (acce-ssed August 2017).

Nordén B, Dahlberg A, Brandrud TE, Fritz O, Ejrnaes R, OvaskainenO. 2014. Effects of ecological continuity on species richnessand composition in forests and woodlands: a review. Ecoscience21:34–45.

Orlikowska E, Roberge J-M, Blicharska M, Mikusiński G. 2016.Gaps in ecological research on the world’s largest internationallycoordinated network of protected areas: a review of Natura 2000.Biological Conservation 200:216–227.

Paillet Y, et al. 2010. Biodiversity differences between managed andunmanaged forests: meta-analysis of species richness in Europe.Conservation Biology 24:101–112.

Perhans K, Haight RG, Gustafsson L. 2014. The value of information inconservation planning: selecting retention trees for lichen conser-vation. Forest Ecology and Management 318:175–182.

Pfeifer M, et al. 2017. Creation of forest edges has a global impact onforest vertebrates. Nature 551:187–191.

Potapov P, et al. 2008. Mapping the world’s intact forest landscapes byremote sensing. Ecology and Society 13:51.

Potapov P, et al. 2017. The last frontiers of wilderness: tracking lossof intact forest landscapes from 2000 to 2013. Science Advances3:1-13.

Potapov P, Turubanova S, Hansen MC. 2011. Regional-scale boreal for-est cover and change mapping using Landsat data composites forEuropean Russia. Remote Sensing of Environment 115:548–561.

Ranius T, Kindvall O. 2006. Extinction risk of wood-living model speciesin forest landscapes as related to forest history and conservationstrategy. Landscape Ecology 21:687–698.

Riitters K, Wickham J, Costanza JK, Vogt P. 2016. A global evaluationof forest interior area dynamics using tree cover data from 2000 to2012. Landscape Ecology 31:137–148.

Ruete A, Snäll T, Jonsson M. 2016. Dynamic anthropogenic edge effectson the distribution and diversity of fungi in fragmented old-growthforests. Ecological Applications 26:1475–1485.

Rybicki J, Hanski I. 2013. Species area relationships and extinctioncaused by habitat loss and fragmentation. Ecology Letters 16(Supplement 1):27–38.

Siitonen P, Lehtinen A, Siitonen M. 2005. Effects of forest edges on thedistribution, abundance, and regional persistence of wood-rottingfungi. Conservation Biology 19:250–260.

Snäll T, Lehtomäki J, Arponen A, Elith J, Moilanen A. 2016. Green in-frastructure design based on spatial conservation prioritization andmodeling of biodiversity features and ecosystem services. Environ-mental Management 57:251–256.

Sverdrup-Thygeson A, Örka HO, Gobakken T, Naesset E. 2016. Canairborne laser scanning assist in mapping and monitoring naturalforests? Forest Ecology and Management 369:116–125.

Sverdrup-Thygeson A, Sörgard G, Rusch GM, Barton DN. 2014. Spa-tial overlap between environmental policy instruments and ar-eas of high conservation value in forest. PLOS ONE 9 (e115001)https://doi.org/10.1371/journal.pone.0115001.


http://stat.luke.fi/enhttps://doi.org/10.1111/conl.1244https://doi.org/10.1111/conl.1244https://www.lantmateriet.se/sv/Kartor-och-geografisk-information/https://www.lantmateriet.se/sv/Kartor-och-geografisk-information/https://doi.org/10.1371/journal.pone.0115001

Svensson et al. 163

Swedish Environmental Protection Agency. 2017. Available fromhttp://skyddadnatur.naturvardsverket.se/ (accessed August 2017).

Swedish University of Agricultural Sciences. 2017. Forest statistics 2017.Official Statistics of Sweden. Swedish University of AgriculturalSciences, Umeå.

Swift TJ, Hannon SJ. 2010. Critical thresholds associated with habitatloss: a review of the concepts, evidence and applications. BiologicalReviews 85:35–53.

Tucker MA, et al. 2018. Moving in the Anthropocene: global reductionsin terrestrial mammalian movement. Science 359:466–469.

Tyukavina A, Hansen MC, Potapov PV, Krylov AM, Goetz SJ. 2016. Pan-tropical hinterland forests: mapping minimally disturbed forests.Global Ecology and Biogeography 25:151–163.

Tzoulas K, Korpela K, Venn S, Yli-Pelkonen V, Kazmierczak A, NiemelaJ, James P. 2007. Promoting ecosystem and human health in urbanareas using green infrastructure: a literature review. Landscape andUrban Planning 81:167–178.

Van Teeffelen AJA, Vos CC, Opdam P. 2012. Species in a dynamicworld: consequences of habitat network dynamics on conservationplanning. Biological Conservation 153:239–253.

Watson JEM, et al. 2018. The exceptional value of intact forest ecosys-tems. Nature Ecology and Evolution 2:599–610.

Wulder MA, Masek JG, Cohen WB, Loveland TR, Woodcock CE. 2012.Opening the archive: how free data has enabled the science andmonitoring promise of Landsat. Remote Sensing of Environment122:2–10.

http://skyddadnatur.naturvardsverket.se/

landscape trajectory of natural boreal forest loss as an...

Documents