positive biodiversity-productivity relationship...

Seediscussionsstatsandauthorprofilesforthispublicationathttpswwwresearchgatenetpublication309127324

Positivebiodiversity-productivityrelationshippredominantinglobalforests

ArticleinSciencemiddotOctober2016

DOI101126scienceaaf8957

CITATIONS

0

READS

1695

84authorsincluding

Someoftheauthorsofthispublicationarealsoworkingontheserelatedprojects

EcologicalfoodprintofforestryViewproject

InformationoncurrentresearchprojectsseewebsiteViewproject

Gert-JanNabuurs

WageningenUniversityampResearch

262PUBLICATIONS5877CITATIONS

SEEPROFILE

HanYHChen

LakeheadUniversityThunderBayCampus


SEEPROFILE

XdLei

ChineseAcademyofForestry


SEEPROFILE

BernhardSchmid

UniversityofZurich


SEEPROFILE

Allin-textreferencesunderlinedinbluearelinkedtopublicationsonResearchGate

lettingyouaccessandreadthemimmediately

AvailablefromHuicuiLu

Retrievedon07November2016

RESEARCH ARTICLE SUMMARY

FOREST ECOLOGY

Positive biodiversity-productivityrelationship predominantin global forestsJingjing Liang Thomas W Crowther Nicolas Picard Susan Wiser Mo ZhouGiorgio Alberti Ernst-Detlef Schulze A David McGuire Fabio Bozzato Hans PretzschSergio de-Miguel Alain Paquette Bruno Heacuterault Michael Scherer-LorenzenChristopher B Barrett Henry B Glick Geerten M Hengeveld Gert-Jan NabuursSebastian Pfautsch Helder Viana Alexander C Vibrans Christian Ammer Peter SchallDavid Verbyla Nadja Tchebakova Markus Fischer James V Watson Han Y H ChenXiangdong Lei Mart-Jan Schelhaas Huicui Lu Damiano Gianelle Elena I ParfenovaChristian Salas Eungul Lee Boknam Lee Hyun Seok Kim Helge BruelheideDavid A Coomes Daniel Piotto Terry Sunderland Bernhard SchmidSylvie Gourlet-Fleury Bonaventure Sonkeacute Rebecca Tavani Jun Zhu Susanne BrandlJordi Vayreda Fumiaki Kitahara Eric B Searle Victor J Neldner Michael R NgugiChristopher Baraloto Lorenzo Frizzera Radomir Bałazy Jacek OleksynTomasz Zawiła-Niedźwiecki Olivier Bouriaud Filippo Bussotti Leena FineacuterBogdan Jaroszewicz Tommaso Jucker Fernando Valladares Andrzej M JagodzinskiPablo L Peri Christelle Gonmadje William Marthy Timothy OrsquoBrienEmanuel H Martin Andrew R Marshall Francesco Rovero Robert BitarihoPascal A Niklaus Patricia Alvarez-Loayza Nurdin Chamuya Renato ValenciaFreacutedeacuteric Mortier Verginia Wortel Nestor L Engone-Obiang Leandro V FerreiraDavid E Odeke Rodolfo M Vasquez Simon L Lewis Peter B Reich

INTRODUCTION Thebiodiversity-productivityrelationship (BPR the effect of biodiversity onecosystem productivity) is foundational to ourunderstanding of the global extinction crisisand its impacts on the functioning of naturalecosystems The BPR has been a prominentresearch topicwithin ecology in recent decadesbut it is only recently that we have begun todevelop a global perspective

RATIONALE Forests are the most importantglobal repositories of terrestrial biodiversitybut deforestation forest degradation climatechange and other factors are threatening

approximately one half of tree species world-wide Although there have been substantialefforts to strengthen the preservation andsustainable use of forest biodiversity through-out the globe the consequences of this di-versity loss pose amajor uncertainty for ongoinginternational forest management and conser-vation efforts The forest BPR represents acritical missing link for accurate valuation ofglobal biodiversity and successful integrationof biological conservation and socioeconomicdevelopment Until now there have been limitedtree-based diversity experiments and the forestBPR has only been explored within regional-

scale observational studies Thus the strengthand spatial variability of this relationship re-mains unexplored at a global scale

RESULTS We explored the effect of treespecies richness on tree volume productivity atthe global scale using repeated forest invento-

ries from 777126 perma-nent sample plots in 44countries containingmorethan 30million trees from8737 species spanningmostof the global terrestrial bi-omes Our findings reveal a

consistent positive concave-down effect of bio-diversity on forest productivity across the worldshowing that a continued biodiversity losswouldresult in an accelerating decline in forestproductivity worldwideThe BPR shows considerable geospatial var-

iation across theworld The same percentage ofbiodiversity loss would lead to a greater relative(that is percentage) productivity decline in theboreal forests of North America NortheasternEurope Central Siberia East Asia and scatteredregions of South-central Africa and South-centralAsia In the Amazon West and SoutheasternAfrica Southern China Myanmar Nepal andthe Malay Archipelago however the same per-centage of biodiversity losswould lead to greaterabsolute productivity decline

CONCLUSION Our findings highlight thenegative effect of biodiversity loss on forestproductivity and the potential benefits fromthe transition of monocultures to mixed-speciesstands in forestry practices The BPR we dis-cover across forest ecosystems worldwidecorresponds well with recent theoretical ad-vances as well as with experimental and ob-servational studies on forest and nonforestecosystems On the basis of this relationshipthe ongoing species loss in forest ecosystemsworldwide could substantially reduce forest pro-ductivity and thereby forest carbon absorptionrate to compromise the global forest carbonsink We further estimate that the economicvalue of biodiversity in maintaining commer-cial forest productivity alone is $166 billion to$490 billion per year Although representingonly a small percentage of the total value ofbiodiversity this value is two to six times asmuch as it would cost to effectively implementconservation globally These results highlightthe necessity to reassess biodiversity valuationand the potential benefits of integrating andpromoting biological conservation in forestresource management and forestry practicesworldwide

RESEARCH

196 14 OCTOBER 2016 bull VOL 354 ISSUE 6309 sciencemagorg SCIENCE

The list of author affiliations is available in the full article onlineCorresponding author Email albecalianggmailcomCite this article as J Liang et al Science 354 aaf8957(2016) DOI 101126scienceaaf8957

Global effect of tree species diversity on forest productivity Ground-sourced data from 777126global forest biodiversity permanent sample plots (dark blue dots left)which cover a substantial portionof the global forest extent (white) reveal a consistent positive and concave-down biodiversity-productivity relationship across forests worldwide (red line with pink bands representing 95 con-fidence interval right)

ON OUR WEBSITE

Read the full articleat httpdxdoiorg101126scienceaaf8957

on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

RESEARCH ARTICLE

FOREST ECOLOGY

Positive biodiversity-productivityrelationship predominantin global forestsJingjing Liang1 Thomas W Crowther23dagger Nicolas Picard4 Susan Wiser5 Mo Zhou1

Giorgio Alberti6 Ernst-Detlef Schulze7 A David McGuire8 Fabio Bozzato9

Hans Pretzsch10 Sergio de-Miguel1112 Alain Paquette13 Bruno Heacuterault14

Michael Scherer-Lorenzen15 Christopher B Barrett16 Henry B Glick3

Geerten M Hengeveld1718 Gert-Jan Nabuurs1719 Sebastian Pfautsch20

Helder Viana2122 Alexander C Vibrans23 Christian Ammer24 Peter Schall24

David Verbyla25 Nadja Tchebakova26 Markus Fischer2728 James V Watson1

Han Y H Chen29 Xiangdong Lei30 Mart-Jan Schelhaas17 Huicui Lu19

Damiano Gianelle3132 Elena I Parfenova26 Christian Salas33 Eungul Lee34

Boknam Lee35 Hyun Seok Kim35363738 Helge Bruelheide3940 David A Coomes41

Daniel Piotto42 Terry Sunderland4344 Bernhard Schmid45 Sylvie Gourlet-Fleury46

Bonaventure Sonkeacute47 Rebecca Tavani48 Jun Zhu4950 Susanne Brandl1051

Jordi Vayreda5253 Fumiaki Kitahara54 Eric B Searle29 Victor J Neldner55

Michael R Ngugi55 Christopher Baraloto5657 Lorenzo Frizzera31 Radomir Bałazy58

Jacek Oleksyn5960 Tomasz Zawiła-Niedźwiecki6162 Olivier Bouriaud6364

Filippo Bussotti65 Leena Fineacuter66 Bogdan Jaroszewicz67 Tommaso Jucker41

Fernando Valladares6869 Andrzej M Jagodzinski5970 Pablo L Peri717273

Christelle Gonmadje7475 William Marthy76 Timothy OrsquoBrien76 Emanuel H Martin77

Andrew R Marshall7879 Francesco Rovero80 Robert Bitariho81 Pascal A Niklaus45

Patricia Alvarez-Loayza82 Nurdin Chamuya83 Renato Valencia84 Freacutedeacuteric Mortier46

Verginia Wortel85 Nestor L Engone-Obiang86 Leandro V Ferreira87

David E Odeke88 Rodolfo M Vasquez89 Simon L Lewis9091 Peter B Reich2060

The biodiversity-productivity relationship (BPR) is foundational to our understanding of theglobal extinction crisis and its impacts on ecosystem functioning Understanding BPR is criticalfor the accurate valuation and effective conservation of biodiversity Using ground-sourced datafrom 777126 permanent plots spanning 44 countries and most terrestrial biomes we reveala globally consistent positive concave-down BPR showing that continued biodiversity losswould result in an accelerating decline in forest productivity worldwideThe value of biodiversityin maintaining commercial forest productivity alonemdashUS$166 billion to 490 billion per yearaccording to our estimationmdashis more than twice what it would cost to implement effectiveglobal conservationThis highlights the need for a worldwide reassessment of biodiversityvalues forest management strategies and conservation priorities

The biodiversity-productivity relationship(BPR) has been a major ecological researchfocus over recent decades The need tounderstand this relationship is becomingincreasingly urgent in light of the global

extinction crisis because species loss affects thefunctioning and services of natural ecosystems(1 2) In response to an emerging body of evidencethat suggests that the functioning of natural eco-systemsmaybe substantially impairedby reductionsin species richness (3ndash10) global environment-al authorities including the IntergovernmentalPlatform on Biodiversity and Ecosystem Services(IPBES) and United Nations Environment Pro-gramme (UNEP) have made substantial effortsto strengthen the preservation and sustainable useof biodiversity (2 11) Successful international

collaboration however requires a systematic asses-sment of the value of biodiversity (11) Quantifi-cation of the global BPR is thus urgently neededto facilitate the accurate valuation of biodiversity(12) the forecast of future changes in ecosystemservices worldwide (11) and the integration ofbiological conservation into international socio-economic development strategies (13)The evidence of a positive BPR stems primarily

from studies of herbaceous plant communities(14) In contrast the forest BPR has only beenexplored at the regional scale [(3 4 7 15) andreferences therein] or within a limited numberof tree-based experiments [(16 17) and referencestherein] and it remains unclear whether theserelationships hold across forest types Forestsare the most important global repositories of

terrestrial biodiversity (18) but deforestationclimate change and other factors are threat-ening a considerable proportion (up to 50) oftree species worldwide (19ndash21) The consequencesof this diversity loss pose a critical uncertainty forongoing international forest management andconservation efforts Conversely forest manage-ment that convertsmonocultures tomixed-speciesstands has often seen a substantial positive effecton productivity with other benefits (22ndash24) Al-though forest plantations are predicted tomeet 50to 75of the demand for lumber by 2050 (25 26)nearly all are still planted as monocultures high-lighting the potential of forest management instrengthening the conservation and sustainableuse of biodiversity worldwideHere we compiled in situ remeasurement data

most of which were taken at two consecutiveinventories from the same localities from 777126permanent sample plots [hereafter global forestbiodiversity (GFB) plots] across 44 countries andterritories and 13 ecoregions to explore the forestBPR at a global scale (Fig 1) GFBplots encompassforests of various origins (from naturally re-generated to planted) and successional stages(from stand initiation to old-growth) A total ofmore than 30 million trees across 8737 specieswere tallied and measured on two or more con-secutive inventories from theGFB plots Samplingintensity was greater in developed countrieswhere nationwide forest inventories have beenfully or partially funded by governments Inmostother countries national forest inventories werelacking andmost ground-sourced data were col-lected by individuals and organizations (table S1)On the basis of ground-sourced GFB data we

quantified BPR at the global scale using a data-driven ensemble learning approach (Materialsand methods Geospatial random forest) Ourquantification of BPR involved characterizingthe shape and strength of the dependency func-tion through the elasticity of substitution (q)which represents the degree to which speciescan substitute for each other in contributing toforest productivity q measures the marginalproductivitymdashthe change in productivity resultingfrom one unit decline of species richnessmdashandreflects the strength of the effect of tree diversityon forest productivity after accounting for cli-matic soil and plot-specific covariates A higherq corresponds to a greater decline in productivitydue to one unit loss in biodiversity The niche-efficiency (N-E) model (3) and several precedingstudies (27ndash30) provide a framework for inter-preting the elasticity of substitution and approx-imating BPR with a power function model

P = a middot f(X) middot Sq (1)

where P and S signify primary site productivityand tree species richness (observed on a 900-m2

area basis on average) (Materials and methods)respectively f(X) is a function of a vector of con-trol variables X (selected from stand basal areaand 14 climatic soil and topographic covariates)and a is a constant This model is capable ofrepresenting a variety of potential patterns of

RESEARCH

SCIENCE sciencemagorg 14 OCTOBER 2016 bull VOL 354 ISSUE 6309 aaf8957-1

on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

BPR 0 lt q lt 1 represents a positive and concavedown pattern (a degressively increasing curve)which is consistent with the N-E model and pre-ceding studies (3 27ndash30) whereas other q valuescan represent alternative BPR patterns includingdecreasing (q lt 0) linear (q = 1) convex (q gt 1) orno effect (q = 0) (Fig 2) (14 31) The model (Eq 1)was estimated by using the geospatial randomforest technique based onGFBdata and covariatesacquired from ground-measured and remote-sensing data (Materials and methods)We found that a positive BPR predominated

in forests worldwide Out of 10000 randomly se-lected subsamples (each consisting of 500 GFBplots) 9987 had a positive concave-down rela-tionship with relative species richness (0 lt q lt 1)whereas only 013 show negative trends andnone was equal to zero or greater than or equalto 1 (Fig 2) Overall the global forest productiv-ity increased with a declining rate from 27 to118 m3 haminus1 yearminus1 as relative tree species richnessincreased from the minimum to the maximum

value which corresponds to a q value of 026(Fig 3A)At the global scale we mapped the magnitude

of BPR (as expressed by q) using geospatialrandom forest and universal kriging By plottingvalues of q onto a global map we revealed con-siderable geospatial variation across the world(Fig 3B) The highest q (029 to 030) occurredin the boreal forests of North America North-eastern Europe Central Siberia and East Asiaand the sporadic tropical and subtropical forestsof South-central Africa South-central Asia andthe Malay Archipelago In these areas of thehighest elasticity of substitution (32) the samepercentage of biodiversity loss would lead to agreater percentage of reduction in forest produc-tivity (Fig 4A) In terms of absolute productivitythe same percentage of biodiversity loss wouldlead to the greatest productivity decline in theAmazon West Africarsquos Gulf of Guinea South-eastern Africa including Madagascar SouthernChina Myanmar Nepal and the Malay Archi-

pelago (Fig 4B) Because of a relatively narrowrange of the elasticity of substitution (32) esti-mated from the global-level analysis (02 to 03)the regions of the greatest productivity declineunder the same percentage of biodiversity losslargely matched the regions of the greatest pro-ductivity (fig S1) Globally a 10 decrease in treespecies richness (from 100 to 90)would cause a2 to 3 decline in productivity and with a de-crease in tree species richness to one (Materialsand methods Economic analysis) this decline inforest productivity would be 26 to 66 even ifother things such as the total number of treesand forest stocking remained the same (fig S4)

Discussion

Our global analysis provides strong and consistentevidence that productivity of forests would de-crease at an accelerating rate with the loss ofbiodiversity The positive concave-down patternwe discovered across forest ecosystemsworldwidecorrespondswell with recent theoretical advances

aaf8957-2 14 OCTOBER 2016 bull VOL 354 ISSUE 6309 sciencemagorg SCIENCE

1School of Natural Resources West Virginia University Morgantown WV 26505 USA 2Netherlands Institute of Ecology Droevendaalsesteeg 10 6708 PB Wageningen Netherlands 3Yale Schoolof Forestry and Environmental Studies Yale University 195 Prospect Street New Haven CT 06511 USA 4Forestry Department Food and Agriculture Organization of the United Nations RomeItaly 5Landcare Research Lincoln 7640 New Zealand 6Department of Agri-Food Animal and Environmental Sciences University of Udine via delle Scienze 206 Udine 33100 Italy 7Max-PlanckInstitut fuumlr Biogeochemie Hans-Knoell-Strasse 10 07745 Jena Germany 8US Geological Survey Alaska Cooperative Fish and Wildlife Research Unit University of Alaska Fairbanks FairbanksAK 99775 USA 9Architecture and Environment Department Italcementi Group 24100 Bergamo Italy 10Institute of Forest Growth and Yield Science School of Life Sciences WeihenstephanTechnical University of Munich (TUM) Hans-Carl-von-Carlowitz-Platz 2 85354 Freising Germany 11Departament de Produccioacute Vegetal i Ciegravencia Forestal Universitat de Lleida-Agrotecnio Center(UdL-Agrotecnio) Avinguda Rovira Roure 191 E-25198 Lleida Spain 12Centre Tecnologravegic Forestal de Catalunya (CTFC) Carretera De St Llorenccedil de Morunys km 2 E-25280 Solsona Spain13Centre deacutetude de la forecirct (CEF) Universiteacute du Queacutebec agrave Montreacuteal Montreacuteal QC H3C 3P8 Canada 14Centre de Coopeacuteration Internationale en la Recherche Agronomique pour leDeacuteveloppement (CIRAD) UMR Joint Research Unit Ecology of Guianan Forests (EcoFoG) AgroParisTech CNRS INRA Universiteacute des Antilles Universiteacute de la Guyane Kourou French Guiana15University of Freiburg Faculty of Biology Geobotany D-79104 Freiburg Germany 16Charles H Dyson School of Applied Economics and Management Cornell University Ithaca NY 14853 USA17Wageningen University and Research (Alterra) Team Vegetation Forest and Landscape Ecologyndash6700 AA Netherlands 18Forest and Nature Conservation Policy Group Wageningen Universityand Research 6700 AA Wageningen Netherlands 19Forest Ecology and Forest Management Group Wageningen University 6700 AA Wageningen UR Netherlands 20Hawkesbury Institute forthe Environment Western Sydney University Richmond NSW 2753 Australia 21Center for Studies in Education Technologies and Health (CIampDETS) Research CentreDepartamento de Ecologiae Agricultura Sustentaacutevel (DEAS)ndashEscola Superior Agraacuteria de Viseu (ESAV) Polytechnic Institute of Viseu Portugal 22Centre for the Research and Technology of Agro-Environmental andBiological Sciences (CITAB) University of Traacutes-os-Montes and Alto Douro (UTAD) Quinta de Prados 5000-801 Vila Real Portugal 23Departamento de Engenharia Florestal UniversidadeRegional de Blumenau Rua Satildeo Paulo 3250 89030-000 Blumenau-Santa Catarina Brazil 24Department of Silviculture and Forest Ecology of the Temperate Zones Georg-August UniversityGoumlttingen Buumlsgenweg 1 D-37077 Goumlttingen Germany 25School of Natural Resources and Extension University of Alaska Fairbanks Fairbanks AK 99709 USA 26V N Sukachev Institute ofForests Siberian Branch Russian Academy of Sciences Academgorodok 5028 660036 Krasnoyarsk Russia 27Institute of Plant Sciences Botanical Garden and Oeschger Centre for ClimateChange Research University of Bern 3013 Bern Switzerland 28Senckenberg Gesellschaft fuumlr Naturforschung Biodiversity and Climate Research Centre (BIK-F) 60325 Frankfurt Germany29Faculty of Natural Resources Management Lakehead University Thunder Bay ON P7B 5E1 Canada 30Research Institute of Forest Resource Information Techniques Chinese Academy ofForestry Beijing 100091 China 31Sustainable Agro-Ecosystems and Bioresources Department Research and Innovation Centre - Fondazione Edmund Mach Via E Mach 1 38010ndashS MicheleallrsquoAdige (TN) Italy 32Foxlab Joint CNRndashFondazione Edmund Mach Initiative Via E Mach 1 38010 - SMichele allrsquoAdige Adige (TN) Italy 33Departamento de Ciencias Forestales Universidad deLa Frontera Temuco Chile 34Department of Geology and Geography West Virginia University Morgantown WV 26506 USA 35Research Institute of Agriculture and Life Sciences Seoul NationalUniversity Seoul Republic of Korea 36Department of Forest Sciences Seoul National University Seoul 151-921 Republic of Korea 37Interdisciplinary Program in Agricultural and ForestMeteorology Seoul National University Seoul 151-744 Republic of Korea 38National Center for AgroMeteorology Seoul National University Seoul 151-744 Republic of Korea 39Institute ofBiologyGeobotany and Botanical Garden Martin Luther University Halle-Wittenberg Am Kirchtor 1 06108 Halle (Saale) Germany 40German Centre for Integrative Biodiversity Research (iDiv)Halle-Jena-Leipzig Deutscher Platz 5e 04103 Leipzig Germany 41Forest Ecology and Conservation Department of Plant Sciences University of Cambridge Cambridge CB2 3EA UK42Universidade Federal do Sul da Bahia Ferradas Itabuna 45613-204 Brazil 43Sustainable Landscapes and Food Systems Centre for International Forestry Research Bogor Indonesia 44Schoolof Marine and Environmental Studies James Cook University Australia 45Institute of Evolutionary Biology and Environmental Studies University of Zurich CH-8057 Zurich Switzerland 46UPRFampS Montpellier 34398 France 47Plant Systematic and Ecology Laboratory Department of Biology Higher Teachers Training College University of Yaounde I Post Office Box 047 YaoundeCameroon 48Forestry Department Food and Agriculture Organization of the United Nations Rome 00153 Italy 49Department of Statistics University of WisconsinndashMadison Madison WI 53706USA 50Department of Entomology University of WisconsinndashMadison Madison WI 53706 USA 51Bavarian State Institute of Forestry Hans-Carl-von-Carlowitz-Platz 1 Freising 85354 Germany52Center for Ecological Research and Forestry Applications (CREAF) Cerdanyola del Vallegraves 08193 Spain 53Universitat Autogravenoma Barcelona Cerdanyola del Valleacutes 08193 Spain 54ShikokuResearch Center Forestry and Forest Products Research Institute Kochi 780-8077 Japan 55Ecological Sciences Unit at Queensland Herbarium Department of Science Information Technologyand Innovation Queensland Government Toowong Qld 4066 Australia 56International Center for Tropical Botany Department of Biological Sciences Florida International University Miami FL33199 USA 57INRA UMR EcoFoG Kourou French Guiana 58Forest Research Institute Sekocin Stary Braci Lesnej 3 Street 05-090 Raszyn Poland 59Institute of Dendrology Polish Academy ofSciences Parkowa 5 PL-62-035 Kornik Poland 60Department of Forest Resources University of Minnesota St Paul MN 55108 USA 61Warsaw University of Life Sciences (SGGW) Faculty ofForestry ul Nowoursynowska 159 02-776 Warszawa Poland 62Polish State Forests ul Grojecka 127 02-124 Warszawa Poland 63Forestry Faculty University Stefan Cel Mare of Suceava 13Strada Universitații720229 Suceava Romania 64Institutul Național de Cercetare-Dezvoltare icircn Silvicultură 128 Bd Eroilor 077190 Voluntari Romania 65Department of Agri-Food Production andEnvironmental Science University of Florence P le Cascine 28 51044 Florence Italy 66Natural Resources Institute Finland 80101 Joensuu Finland 67Białowieża Geobotanical Station Faculty ofBiology University of Warsaw Sportowa 19 17-230 Białowieża Poland 68Museo Nacional de Ciencias Naturales Consejo Superior de Investigaciones Cientiacuteficas Serrano 115 dpdo E-28006Madrid Spain 69Universidad Rey Juan Carlos Mostoles Madrid Spain 70Poznan University of Life Sciences Department of Game Management and Forest Protection Wojska Polskiego 71c PL-60-625 Poznan Poland 71Consejo Nacional de Investigaciones Cientiacuteficas y Tecnicas (CONICET) Rivadavia 1917 (1033) Ciudad de Buenos Aires Buenos Aires Argentina 72Instituto Nacional deTecnologiacutea Agropecuaria (INTA) Estacioacuten Experimental Agropecuaria (EEA) Santa Cruz Mahatma Ghandi 1322 (9400) Riacuteo Gallegos Santa Cruz Argentina 73Universidad Nacional de la PatagoniaAustral (UNPA) Lisandro de la Torre 1070 (9400) Riacuteo Gallegos Santa Cruz Argentina 74Department of Plant Ecology Faculty of Sciences University of Yaounde I Post Office Box 812Yaounde Cameroon 75National Herbarium Post Office Box 1601 Yaoundeacute Cameroon 76Wildlife Conservation Society Bronx NY 10460 USA 77College of African Wildlife ManagementDepartment of Wildlife Management Post Office Box 3031 Moshi Tanzania 78Environment Department University of York Heslington York YO10 5NG UK 79Flamingo Land Malton NorthYorkshire YO10 6UX 80Tropical Biodiversity Section MUSE-Museo delle Scienze Trento Italy 81Institute of Tropical Forest Conservation Kabale Uganda 82Center for Tropical ConservationDurham NC 27705 USA 83Ministry of Natural Resources and Tourism Forestry and Beekeeping Division Dar es Salaam Tanzania 84Escuela de Ciencias Bioloacutegicas Pontificia UniversidadCatoacutelica del Ecuador Apartado 1701-2184 Quito Ecuador 85Forest Management Department Centre for Agricultural Research in Suriname (CELOS) Paramaribo Suriname 86Institut deRecherche en Ecologie Tropicale Institut de Recherche en Ecologie Tropicale (IRET)Centre National de la Recherche Scientifique et Technologique (CENAREST) B P 13354 Libreville Gabon87Museu Paraense Emilio Goeldi Coordenacao de Botanica Belem PA Brasil 88National Forest Authority Kampala Uganda 89Prolongacioacuten Bolognesi Mz-E-6 Oxapampa Pasco Peru90Department of Geography University College London UK 91School of Geography University of Leeds UKCorresponding author Email albecalianggmailcom daggerPresent address Netherlands Institute of Ecology Droevendaalsesteeg 10 6708 PB Wageningen Netherlands

RESEARCH | RESEARCH ARTICLE

on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

in BPR (3 28ndash30) as well as with experimental(27) and observational (14) studies on forest andnonforest ecosystems The elasticity of substitu-tion (32) estimated in this study (ranged between02 and 03) largely overlaps the range of valuesof the same exponent term (01 to 05) fromprevious theoretical and experimental studies[(10) and references therein] Furthermore ourfindings are consistent with the global estimatesof the biodiversity-dependent ecosystem servicedebt under distinct assumptions (10) and withrecent reports of the diminishing marginal ben-efits of adding a species as species richness in-creases based on long-term forest experimentsdating back to 1870 [(15 33) and references therein]Our analysis relied on stands ranging from

unmanaged to extensively managed forestsmdashmanaged forests with low operating and invest-

ment costs per unit area Conditions of naturalforests would not be comparablewith intensivelymanaged forests because timber production inthe latter systems often focuses on a single orlimited number of highly productive tree speciesIntensively managed forests where saturated re-sources canweaken the effects of niche efficiency(3) are shown in some studies (34 35) to havehigher productivity than that of natural diverseforests of the same climate and site conditions(fig S3) In contrast other studies (6 22ndash24) com-pared diverse stands with monocultures at thesame level of management intensity and foundthat the positive effects of species diversity ontree productivity and other ecosystem servicesare applicable to intensively managed forestsAs such there is still an unresolved debate on theBPR of intensively managed forests Nevertheless

because intensively managed forests only accountfor a minor (lt7) portion of global forests (18)our estimated BPR would be minimally affectedby such manipulations and thus should reflectthe inherent processes governing the vastmajorityof global forest ecosystemsWe focused on the effect of biodiversity on

ecosystem productivity Recent studies on the op-posite causal direction [productivity-biodiversityrelationship (14 36 37)] suggest that theremay bea potential two-way causality between biodiversityand productivity It is admittedly difficult to usecorrelative data to detect and attribute causal ef-fects Fortunately substantial progress has beenmade to tease the BPR causal relationship fromother potentially confounding environmentalvariables (14 38 39) and this study made con-siderable efforts to account for these otherwise


Fig 1 GFB ground-sourced data were collected from in situ remeasure-mentof 777126permanent sampleplots consistingofmore than30milliontrees across 8737 species GFB plots extend across 13 ecoregions [verticalaxis delineated by theWorldWildlife Fundwhere extensive forests occur withinall the ecoregions (72)] and 44 countries and territories Ecoregions are namedfor theirdominant vegetation types but all contain some forestedareasGFBplotscovera substantial portionof theglobal forest extent (white) includingsomeof themostdistinct forest conditions (a) thenorthernmost (73degNCentral SiberiaRussia)

(b) southernmost (52degS Patagonia Argentina) (c) coldest (ndash17degC annual meantemperature Oimyakon Russia) (d) warmest (28degC annual mean temperaturePalau United States) and (e) most diverse (405 tree species on the 1-ha plotBahia Brazil) Plots in war-torn regions [such as (f)] were assigned fuzzed co-ordinates to protect the identity of the plots and collaboratorsThe box plots showthe mean and interquartile range of tree species richness and primary site pro-ductivity (bothonacommon logarithmicscale)derived fromground-measured tree-and plot-level recordsThe complete list of species is presented in table S2


on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

potentially confounding environmental covariatesin assessing likely causal effects of biodiversityon productivityBecause taxonomic diversity indirectly incor-

porates functional phylogenetic and genomicdiversity our results that focus on tree speciesrichness are likely applicable to these other ele-ments of biodiversity all of which have beenfound to influence plant productivity (1) Ourstraightforward analysis makes clear the taxo-nomic contribution to forest ecosystem produc-tivity and functioning and the importance ofpreserving species diversity to biological conser-vation and forest managementOur findings highlight the necessity to reassess

biodiversity valuation and reevaluate forestman-agement strategies and conservation priorities inforests worldwide In terms of global carbon cycleand climate change the value of biodiversity maybe considerable On the basis of our global-scaleanalyses (Fig 4) the ongoing species loss in forestecosystemsworldwide (1 21) could substantiallyreduce forest productivity and thereby forest car-bon absorption rate which would in turn com-promise the global forest carbon sink (40) Wefurther estimate that the economic value of bio-diversity in maintaining commercial forest pro-

ductivity is $166 billion to $490 billion per year(166 times 1011 to 490 times 1011 yearminus1 in 2015 US$) (Ma-terials and methods Economics analysis) By it-self this estimate does not account for other valuesof forest biodiversity (including potential valuesfor climate regulation habitat water flow regula-tion and genetic resources) and represents only asmall percentage of the total value of biodiversity(41 42) However this value is already betweentwo to six times the total estimated cost that wouldbe necessary if we were to effectively conserve allterrestrial ecosystems at a global scale [$761 billionper year (43)] The high benefit-to-cost ratio under-lines the importance of conserving biodiversityfor forestry and forest resource managementAmid the struggle to combat biodiversity loss

the relationship between biological conservationand poverty is gaining increasing global atten-tion (13 44) especially with respect to rural areaswhere livelihoods depend most directly on eco-systemproducts Given the substantial geographicoverlaps between severe multifaceted povertyand key areas of global biodiversity (45) theloss of species in these areas has the potentialto exacerbate local poverty by diminishing forestproductivity and related ecosystem services (44)For example in tropical and subtropical regions

many areas of high elasticity of substitution (32)overlapped with biodiversity hotspots (46) in-cluding EasternHimalaya andNepal Mountainsof Southwest China EasternAfromontaneMadreanpine-oakwoodlands Tropical Andes andCerradoFor these areas only a few species of commercialvalue are targeted by logging As such the risk oflosing species through deforestation would farexceed the risk through harvesting (47) De-forestation and other anthropogenic drivers ofbiodiversity loss in these biodiversity hotspotsare likely to have considerable impacts on theproductivity of forest ecosystems with the po-tential to exacerbate local poverty Furthermorethe greater uncertainty in our results for thedeveloping countries (Fig 5) reflects the well-documented geographic bias in forest samplingincluding repeated measurements and reiteratestheneed for strongcommitments toward improvingsampling in the poorest regions of the worldOur findings reflect the combined strength

of large-scale integration and synthesis of eco-logical data andmodernmachine learningmethodsto increase our understanding of the global forestsystem Such approaches are essential for gen-erating global insights into the consequences ofbiodiversity loss and the potential benefits of


Fig 2 Theoretical positive and concave-down biodiversityndashproductivityrelationship supported byempirical evidence drawn from theGFBdata (Left)The diagram demonstrates that under the theoretical positive and concave-down(monotonically and degressively increasing) BPR (3 27 28) loss in tree speciesrichness may reduce forest productivity (73) (Middle) Functional curves rep-resent different BPR under different values of elasticity of substitution (q) q values

between 0 and 1 correspond to the positive and concave-down BPR (blue curve)(Right) The three-dimensional scatter plot shows q values we estimated from ob-served productivity (P) species richness (S) and other covariates Out of 5000000estimatesofq (mean=026SD=009)4993500fellbetween0and1(blue)whereasonly 6500were negative (red) and nonewas equal to zero orgreater than or equal to1 the positive and concave-down BPRwas supported by 999 of our estimates


on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

integrating and promoting biological conserva-tion in forest resource management and forestrypracticesmdasha common goal already shared by in-tergovernmental organizations such as the Mon-treacuteal andHelsinki ProcessWorking Groups Thesefindings should facilitate efforts to accuratelyforecast future changes in ecosystem servicesworldwide which is a primary goal of IPBES(11) and provide baseline information necessaryto establish international conservation objectivesincluding the United Nations Convention on Bio-logical Diversity Aichi targets the United NationsFrameworkConventiononClimateChangeREDD+goal and theUnitedNationsConvention toCombatDesertification land degradation neutrality goalThe success of these goals relies on the under-

standing of the intrinsic link between biodiversityand forest productivity

Materials and methodsData collection and standardization

Our current study used ground-sourced forestmeasurement data from 45 forest inventoriescollected from 44 countries and territories (Fig1 and table S1) Themeasurementswere collectedin the field from predesignated sample area unitsie Global Forest Biodiversity permanent sampleplots (hereafter GFB plots) For the calculation ofprimary site productivity GFB plots can be cat-egorized into two tiers Plots designated as ldquoTier1rdquo have been measured at two or more points intime with aminimum time interval betweenmea-

surements of two years ormore (globalmean timeinterval is 9 years see Table 1) ldquoTier 2rdquo plots wereonlymeasured once and primary site productivitycan be estimated from known stand age or den-drochronological records Overall our study wasbased on 777126 GFB plots of which 597179 (77)were Tier 1 and 179798 (23) were Tier 2 GFBplots primarily measured natural forests rangingfrom unmanaged to extensively managed forestsie managed forests with low operating and in-vestment costs per unit area Intensively man-aged forests with harvests exceeding 50 percentof the stocking volume were excluded from thisstudyGFBplots represent forests of variousorigins(fromnaturally regenerated toplanted) and succes-sional stages (from stand initiation to old-growth)


Fig 3 The estimated global effect of biodiversity on forest productivitywas positive and concave-down and revealed considerable geospatial var-iation across forest ecosystems worldwide (A) Global effect of biodiversity onforest productivity (red linewith pink bands representing95confidence interval)correspondstoaglobalaverageelasticityofsubstitution(q)valueof026withclimaticsoil andother plot covariatesbeingaccounted forandkept constant at samplemean

Relativespecies richness (Š) is in thehorizontal axis andproductivity (Pm3haminus1 yearminus1)is in theverticalaxis(histogramsof thetwovariables on topand right in the logarithmscale) (B) q represents the strength of the effect of tree diversity on forest produc-tivitySpatiallyexplicit valuesofqwereestimatedbyusinguniversal kriging (Materialsandmethods) across the current global forest extent (effect sizes of the estimatesare shown in Fig 5) whereas blank terrestrial areas were nonforested


on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Table 1 Definition unit and summary statistics of key variables

Variable Definition Unit Mean Standard

deviation

Source Nominal

resolution

Response variables

P Primary forest

productivity measured in

periodic annual increment in stem

volume (PAI)

m3 haminus1 yearminus1 757 1452 Author-generated

from ground-measured

data

Plot attributes

S Tree species

richness the number of live tree

species observed on

the plot

unitless 579 864 ground-measured

A Plot size

area of the

sample plot

ha 004 012 ground-measured

Y Elapsed time

between two

consecutive

inventories

year 863 1162 ground-measured

G Basal area

total cross-sectional

area of live trees

measured at 13

to 14 m above ground

m2 haminus1 1900 1894 Author-generated


data

E Plot elevation m 46930 56592 GSRTM (74)

I1 Indicator of plot tier

I1 = 1 if a plot was

Tier-2

I1 = 0 if otherwise

unitless 023 042 Author-generated


data

I2 Indicator of plot size

I2 = 1 when 001 le ps lt 005

I2 = 2 when 005 le ps lt015

I2 = 3 when 015 le ps lt 050

I2 = 4 when 050 le ps lt 100

where ps was plot

size (hectares)



data

Climatic covariates

T1 Annual mean temperature 01degC 1084 5592 WorldClim v1 (75) 1 km2

T2 Isothermality unitless

index100

3543 705 WorldClim v1 1 km2

T3 Temperature seasonality Std(0001degC) 778600 209239 WorldClim v1 1 km2

C1 Annual precipitation mm 102000 38835 WorldClim v1 1 km2

C2 Precipitation seasonality

(coefficient of variation)

unitless 2754 1638 WorldClim v1 1 km2

C3 Precipitation of warmest

quarter

mm 28200 12088 WorldClim v1 1 km2

PET Global Potential Evapotranspiration mm yearminus1 106343 27180 CGIAR-CSI (76) 1 km2

IAA Indexed Annual Aridity unitless index10minus4 991509 451299 CGIAR-CSI 1 km2

Soil covariates

O1 Bulk density g cmminus3 070 057 WISE30sec v1 (77) 1 km2

O2 pH measured in water unitless 372 280 WISE30sec v1 1 km2

O3 Electrical conductivity dS mminus1 044 076 WISE30sec v1 1 km2

O4 CN ratio unitless 964 778 WISE30sec v1 1 km2

O5 Total nitrogen g kgminus1 271 462 WISE30sec v1 1 km2

Geographic coordinates and classification

x Longitude in WGS84 datum degree

y Latitude in WGS84 datum degree

Ecoregion Ecoregion defined by World Wildlife Fund (78)


on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

For each GFB plot we derived three key at-tributes from measurements of individual treesmdashtree species richness (S) stand basal area (G) andprimary site productivity (P) Because for each ofall the GFB plot samples S and P were derivedfrom the measurements of the same trees thesampling issues commonly associated with bio-diversity estimation (48) had little influence onthe SndashP relationship (ie BPR) in this studySpecies richness S represents the number of

different tree species alive at the time of inventorywithin the perimeter of a GFB plot with an aver-age size of approximately 900 m2 Ninety-fivepercent of all plots fall between 100 and 1100 m2

in size To minimize the species-area effect (49)we studied theBPRhere using a geospatial randomforest model in which observations from nearbyGFB plots would be more influential than plotsthat are farther apart (see sectGeospatial randomforest) Because nearby plots aremost likely fromthe same forest inventory data set and there wasno or little variation of plot area within each dataset the BPR derived from this model largely re-flected patterns under the same plot area basis

To investigate the potential effects of plot size onour results we plotted the estimated elasticity ofsubstitution (q) against plot size and found thatthe scatter plot was normally distributed with nodiscernible pattern (fig S2) In addition the factthat the plot size indicator I2 had the secondlowest (08) importance score (50) among allthe covariates (Fig 6) further supports that theinfluence of plot size variation in this studywas negligibleAcross all theGFBplots therewere 8737 species

in 1862 genera and 231 families and S valuesranged from 1 to 405 per plot We verified allthe species names against 60 taxonomic data-bases includingNCBI GRINTaxonomy for PlantsTropicosndashMissouri Botanical Garden and theInternational Plant Names Index using thelsquotaxizersquo package in R (51) Out of 8737 speciesrecorded in the GFB database 7425 had verifiedtaxonomic information with a matching score(51) of 0988 or higher whereas 1312 species namespartially matched existing taxonomic databaseswith a matching score between 050 and 075indicating that these species may have not been

documented in the 60 taxonomic databases Tofacilitate inter-biome comparison we furtherdeveloped relative species richness (Š) a con-tinuous percentage score converted from speciesrichness (S) and the maximal species richness ofa set of sample plots (S) using

S frac14 S

Seth2THORN

Stand basal area (G in m2 haminus1) represents thetotal cross-sectional area of live trees per unitsample area G was calculated from individualtree diameter-at-breast-height (dbh in cm)

G frac14 0000079 sdotX

i

dbh2i sdotki eth3THORN

where ki denotes the conversion factor (haminus1) ofthe ith tree viz the number of trees per harepresented by that individual G is a key bioticfactor of forest productivity as it representsstand densitymdashoften used as a surrogate for re-source acquisition (through leaf area) and stand


Fig 4 Estimated percentage and absolute decline in forest productivity under 10 and 99 decline in current tree species richness (values in par-entheses correspond to 99) holding all the other terms constant (A) Percent decline in productivity was calculated according to the general BPRmodel (Eq1) andestimatedworldwidespatiallyexplicit valuesof theelasticityof substitution (Fig 3B) (B)Absolutedecline inproductivitywasderived fromtheestimatedelasticityofsubstitution (Fig 3B) and estimates of global forest productivity (fig S1)The first 10 reduction in tree species richness would lead to a 0001 to 0597 m3 haminus1 yearminus1

decline in periodic annual increment which accounts for 2 to 3 of current forest productivityThe raster data are displayed in 50-km resolution with a 3 SD stretch


on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

competition (52) Accounting for basal area as acovariate mitigated the artifact of different min-imumdbh across inventories and the artifact ofdifferent plot sizesPrimary site productivity (P in m3 haminus1 yrminus1)

was measured as tree volume productivity interms of periodic annual increment (PAI) cal-culated from the sum of individual tree stemvolume (V in m3)

P frac14

X

i2

Vi2 sdotkiminusX

i1

Vi1 sdotki thornM

Yeth4THORN

where Vi1 and Vi2 (in m3) represent total stemvolume of the ith tree at the time of the firstinventory and the second inventory respectivelyM denotes total removal of trees (including mor-tality harvest and thinning) in stem volume(in m3 haminus1) Y represents the time interval (inyears) between two consecutive inventories Paccounted for mortality ingrowth (ie recruit-ment between two inventories) and volumegrowth Stem volume valueswere predominantlycalculated using region- and species-specific al-lometric equations based on dbh and other tree-and plot-level attributes (Table 1) For the regionslacking an allometric equation we approximated

stem volume at the stand level from basal areatotal tree height and stand form factors (53) Incase of missing tree height values from theground measurement we acquired alternativemeasures from a global 1-km forest canopy heightdatabase (54) For Tier 2 plots that lacked re-measurement Pwasmeasured inmean annualincrement (MAI) based on total stand volumeand stand age (52) or tree radial growth mea-sured from increment cores Since the traditionalMAI metric does not account for mortality wecalculated P by adding to MAI the annual mor-tality based on regional-specific forest turnoverrates (55) The small and insignificant correla-tion coefficient between P and the indicator ofplot tier (I1) together with the negligible variableimportance of I1 (18 Fig 6) indicate that PAIandMAIwere generally consistent such thatMAIcould be a good proxy of PAI in our study Al-though MAI and PAI have considerable uncer-tainty in any given stand it is difficult to seehow systematic bias across diversity gradientscould occur on a scale sufficient to influence theresults shown hereP although only representing a fraction of total

forest net primary production has been an im-portant and widely used measure of forest pro-

ductivity because it reflects the dominantaboveground biomass component and the long-lived biomass pool in most forest ecosystems(56) Additionally although other measures ofproductivity (eg net ecosystem exchange pro-cessed to derive gross and net primary produc-tion direct measures of aboveground net primaryproduction including all components and remotelysensed estimates of LAI and greenness coupledwith models) all have their advantages and dis-advantages none would be feasible at a similarscale and resolution as in this studyTo account for abiotic factors that may in-

fluence primary site productivity we compiled14 geospatial covariates based on biological rel-evance and spatial resolution (Fig 6) These co-variates derived fromsatellite-based remote sensingandground-based survey data can be grouped intothree categories climatic soil and topographic(Table 1) We preprocessed all geospatial covar-iates using ArcMap 103 (57) and R 2153 (58) Allcovariates were extracted to point locations ofGFB plots with a nominal resolution of 1 km2

Geospatial random forest

We developed geospatial random forestmdasha data-drivenensemble learningapproachmdashto characterize


Fig 5 Standard error and generalized R2 of the spatially explicit estimates of elasticity of substitution (q) across the current global forest extentin relation to Š (A) Standard error increased as a location was farther from those sampled (B) The generalized R2 values were derived with a geostatisticalnonlinear mixed-effects model for GFB sample locations and thus (B) only covers a subset of the current global forest extent


on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

the biodiversityndashproductivity relationship (BPR)and to map BPR in terms of elasticity of sub-stitution (31) on all sample sites across the worldThis approach was developed to overcome twomajor challenges that arose from the size andcomplexity of GFB data without assuming anyunderlying BPR patterns or data distributionFirst we need to account for broad-scale dif-ferences in vegetation types but global classi-fication andmapping of homogeneous vegetationtypes is lacking (59) and secondly correlationsand trends that naturally occur through space(60) can be significant and influential in forestecosystems (61) Geostatistical models (62) havebeen developed to address the spatial auto-correlation but the size of the GFB data set farexceeds the computational constraints of mostgeostatistical softwareGeospatial random forest integrated conven-

tional random forest (50) and a geostatisticalnonlinear mixed-effects model (63) to estimateBPR across the world based on GFB plot dataand their spatial dependence The underlyingmodel had the following form

logPijethuTHORN frac14 qi sdotlogSijethuTHORN thorn ai sdotXijethuTHORN

thorn eijethuTHORNuisinDsubreal2 eth5THORNwhere logPij(u) and logŠij (u) represent naturallogarithm of productivity and relative speciesrichness (calculated from actual species rich-ness and the maximal species richness of thetraining set) of plot i in the jth training set atpoint locations u respectively The model wasderived from the nichendashefficiency model and qcorresponds to the elasticity of substitution(31) aimiddotXij(u)=ai0 + ai1middotxij1+hellip+ ainmiddotxijn repre-sents n covariates and their coefficients (Fig 6and Table 1)To account for potential spatial autocorrelation

which can bias tests of significance due to theviolation of independence assumption and isespecially problematic in large-scale forest eco-system studies (60 61) we incorporated aspherical variogram model (62) into the residualterm eij(u) The underlying geostatistical assump-tion was that across the world BPR is a second-order stationary processmdasha common geographicalphenomenon in which neighboring points aremore similar to each other than they are to pointsthat are more distant (64) In our study we foundstrong evidence for this gradient (Fig 7) indi-cating that observations from nearby GFB plotswould be more influential than plots that arefarther away The positive spherical semivariancecurves estimated from a large number of boot-strapping iterations indicated that spatial de-pendence increasedasplots becamecloser togetherThe aforementioned geostatistical nonlinear

mixed-effects model was integrated into randomforest analysis (50) bymeans ofmodel selectionandestimation In themodel selection process randomforest was employed to assess the contribution ofeach of the candidate variables to the dependentvariable logPij(u) in terms of the amount of in-crease in prediction error as one variable is per-muted while all the others are kept constant We

used the randomForest package (65) in R to ob-tain importance measures for all the covariatesto guide our selection of the final variables in thegeostatistical nonlinearmixed-effectsmodelXij(u)We selected stand basal area (G) temperatureseasonality (T3) annual precipitation (C1) pre-cipitation of the warmest quarter (C3) potentialevapotranspiration (PET) indexed annual aridity(IAA) and plot elevation (E) as control variablessince their importancemeasureswere greater thanthe 9 percent threshold (Fig 6) preset to ensure

that the final variables accounted for over 60 per-cent of the total variable importance measuresFor geospatial random forest analysis of BPR

we first selected control variables based on thevariable importance measures derived from ran-dom forests (50) We then evaluated the values ofelasticity of substitution (32) which are expectedto be real numbers greater than 0 and less than 1against the alternatives ie negative BPR (H01q lt 0) no effect (H02 q = 0) linear (H03 q = 1)and convex positive BPR (H04 q gt 1) We


Fig 6 Correlation matrix and importance values of potential variables for the geospatial randomforest analysis (A)There were a total of 15 candidate variables from three categories namely plot at-tributes climatic variables and soil factors (a detailed description is provided in Table 1) Correlationcoefficients between these variables were represented by sizes and colors of circles and ldquotimesrdquo marks co-efficients not significant at a = 005 level (B) Variable importance () values were determined by thegeospatial random forest (Materials and methods) Variables with importance values exceeding the 9threshold line (blue)were selected as control variables in the final geospatial random forestmodels Elasticityofsubstitution (coefficient) productivity (dependent variable) and species richness (key explanatory variable)were not ranked in the variable importance chart because they were not potential covariates

Fig 7 Semivariance and esti-mated spherical variogrammodels (blue curves) obtainedfrom geospatial random forestin relation to ŠGray circlessemivariance blue curves esti-mated spherical variogrammodelsThere was a general trendthat semivariance increased withdistance spatial dependence of qweakened as the distance betweenany two GFB plots increasedThefinal sphericalmodels had nugget =08 range = 50 degrees and sill =13To avoid identical distances allplot coordinates were jittered byadding normally distributedrandom noises


on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

examined all the coefficients by their statisticalsignificance and effect sizes using Akaike in-formation criterion (AIC) Bayesian informationcriterion (BIC) and the generalized coefficient ofdetermination (66)

Global analysis

For the global-scale analysis we calibrated thenonlinear mixed-effects model parameters (q andarsquos) using training sets of 500 plots randomlyselected (with replacement) from the GFB globaldataset according to the bootstrap aggregating(bagging) algorithm We calibrated a total of10000 models based on the bagging samplesusing our own bootstrapping program and thenonlinear package nlme (63) of R to calculatethe means and standard errors of final modelestimates (Table 2) This approach overcamecomputational limits by partitioning the GFBsample into smaller subsamples to enable thenonlinear estimation The size of training setswas selected based on the convergence and effectsize of the geospatial random forest models Inpilot simulations with increasing sizes of trainingsets (Fig 8) the value of elasticity of substitution(32) fluctuated at the start until the convergencepoint at 500 plots GeneralizedR2 values declinedas the size of training sets increased from 0 to350 plots and stabilized at around 035 as train-ing set size increased further Accordingly weselected 500 as the size of the training sets forthe final geospatial random forest analysis Basedon the estimated parameters of the globalmodel(Table 2) we analyzed the effect of relative species

richness on global forest productivity with a sen-sitivity analysis by keeping all the other variablesconstant at their samplemeans for each ecoregion

Mapping BPR across globalforest ecosystems

For mapping purposes we first estimated thecurrent extent of global forests in several stepsWe aggregated the ldquotreecover2000rdquo and ldquolossrdquodata (67) from 30mpixels to 30 arc-second pixels(~1 km) by calculating the respective means Theresult was ~1 km pixels showing the percentageforest cover for the year 2000 and the percentageof this forest cover lost between 2000 and 2013respectively The aggregated forest cover loss wasmultiplied by the aggregated forest cover to pro-duce a single raster value for each ~1 km pixelrepresenting a percentage forest lost between2000 and 2013 This multiplication was neces-sary since the initial loss values were relative toinitial forest cover Similarly we estimated thepercentage forest cover gain by aggregating theforest ldquogainrdquo data (67) from 30 m to 30 arc-seconds while taking a mean Then this gainlayer was multiplied by 1 minus the aggregatedforest cover from the first step to produce asingle value for each ~1 km pixel that signifiespercentage forest gain from 2000ndash2013 Thismultiplication ensured that the gain could onlyoccur in areas that were not already forestedFinally the percentage forest cover for 2013 wascomputed by taking the aggregated data fromthe first step (year 2000) and subtracting thecomputed loss and adding the computed gain

We mapped productivity P and elasticity ofsubstitution (32) across the estimated currentextent of global forests here defined as areaswith 50 percent or more forest cover BecauseGFB ground plots represent approximately 40percent of the forested areas we used universalkriging (62) to estimate P and q for the areaswith no GFB sample coverage The universalkriging models consisted of covariates speci-fied in Fig 6B and a spherical variogrammodelwith parameters (ie nugget range and sill)specified in Fig 7 We obtained the best linearunbiased estimators of P and q and their stan-dard error in relation to Š across the currentglobal forest extent with the gstat package of R(68) By combining q estimated from geospatialrandom forest anduniversal krigingwe producedthe spatially continuous maps of the elasticity ofsubstitution (Fig 3B) and forest productivity(fig S1) at a global scale The effect sizes of thebest linear unbiased estimator of q (in terms ofstandard error and generalized R2) are shownin Fig 5 We further estimated percentage andabsolute decline in worldwide forest produc-tivity under two scenarios of loss in tree speciesrichnessmdash low (10 loss) and high (99 loss)These levels represent the productivity decline(in both percentage and absolute terms) if localspecies richness across the global forest extentwould decrease to 90 and 1 percent of the cur-rent values respectively The percentage declinewas calculated based on the general BPR model(Eq 1) and estimated worldwide spatially explicitvalues of the elasticity of substitution (Fig 3B)


Table 2 Parameters of the global geospatial random forest model in 10000 iterations of 500 randomly selected (with replacement) GFB plotsMean and SE of all the parameters were estimated by using bootstrapping Effect sizes were represented by the Akaike information criterion (AIC) Bayesian

information criterion (BIC) and generalized R2 (G-R2) Const constant

Coefficients

Loglik AIC BIC G-R2 const q G T3 C1 C3 PET IAA E

Mean ndash76141 154671 159708 0354 3816 02625243 0014607 ndash0000106 0001604 0001739 ndash0002566 ndash0000134 ndash0000809

SE 054 110 113 0001 0011 00009512 0000039 0000001 0000008 0000008 0000009 0000001 0000002

Iteration

1 ndash75689 153778 158835 0259 4299 0067965 0014971 ndash0000100 0002335 0001528 ndash0003019 ndash0000185 ndash0000639






7 ndash77026 156453 161510 0421 4691 0353071 0009602 ndash0000127 0002390 ndash0001151 ndash0003337 ndash0000172 ndash0000441








9997 ndash77508 157417 162474 0313 1589 0193865 0012525 ndash0000056 ndash0000589 0000550 ndash0000066 ndash0000155 ndash0000839





on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

The absolute declinewas the product of theworld-wide estimates of primary forest productivity(fig S1) and the standardized percentage declineat the two levels of biodiversity loss (Fig 4A)

Economic analysis

Estimates of the economic value-added fromforests employ a range of methods One promi-nent recent global valuation of ecosystem services(69) valued global forest production [in terms oflsquoraw materialsrsquo (including timber fiber biomassfuels and fuelwood and charcoal] provided byforests (table S1) (69) in 2011 at US$ 649 billion(649 times 1011 in constant 2007 dollars) Using analternative method the UN FAO (25 26) esti-mates gross value-added in the formal forestrysector a measure of the contribution of forestrywood industry and pulp and paper industry tothe worldrsquos economy at US$606 billion (606 times1011 in constant 2011 dollars) Because these tworeasonably comparable values are directly im-pacted by and proportional to forest productivitywe used them as bounds on our coarse estimateof the global economic value of commercial forestproductivity converted to constant 2015 US$based on the US consumer price indices (70 71)As indicated by our global-scale analyses (Fig 4A)a 10 percent decrease of tree species richnessdistributed evenly across the world (from 100to 90) would cause a 21 to 31 percent declinein productivity which would equate to US$13ndash23 billion per year (constant 2015 US$) For theassessment of the value of biodiversity in main-

taining forest productivity a drop in speciesrichness from the current level to one specieswould lead to 26ndash66 reduction in commercialforest productivity in the biomes that contributesubstantially to global commercial forestry (figS4) equivalent to 166ndash490 billion US$ per year(166 times 1011 to 490 times 1011 constant 2015 US$ cal-culated by multiplying the foregoing economicvalue-added from FAO and the other study by 26and 66 respectively) Therefore we estimatedthat the economic value of biodiversity in main-taining commercial forest productivity worldwidewould be 166 billion to 490 billion US$ per yearWe held the total number of trees global forest

area and stocking and other factors constant toestimate the value of productivity loss solely dueto a decline in tree species richness As such theseestimates did not include the value of land con-verted from forest and losses due to associatedfauna and flora decline or forest habitat reductionThis estimate only reflects the value of biodiver-sity in maintaining commercial forest productiv-ity that contributes directly to forestry woodindustry and pulp and paper industry and doesnot account for other values of biodiversity in-cluding potential values for climate regulationhabitat water flow regulation genetic resourcesetc The total global value of biodiversity could ex-ceed this estimate by orders ofmagnitudes (41 42)

REFERENCES AND NOTES

1 S Naeem J E Duffy E Zavaleta The functions of biologicaldiversity in an age of extinction Science 336 1401ndash1406(2012) doi 101126science1215855 pmid 22700920

2 Millennium Ecosystem Assessment ldquoEcosystems and HumanWell-being Biodiversity Synthesisrdquo (World Resources Institute2005)

3 J Liang M Zhou P C Tobin A D McGuire P B ReichBiodiversity influences plant productivity through niche-efficiency Proc Natl Acad Sci USA 112 5738ndash5743 (2015)doi 101073pnas1409853112 pmid 25901325

4 M Scherer-Lorenzen in Forests and Global ChangeD Burslem D Coomes W Simonson Eds (Cambridge UnivPress 2014) pp 195ndash238

5 B J Cardinale et al Biodiversity loss and its impact onhumanity Nature 486 59ndash67 (2012) doi 101038nature11148 pmid 22678280

6 Y Zhang H Y H Chen P B Reich Forest productivityincreases with evenness species richness and trait variationA global meta-analysis J Ecol 100 742ndash749 (2012)doi 101111j1365-2745201101944x

7 A Paquette C Messier The effect of biodiversity on treeproductivity From temperate to boreal forests Glob EcolBiogeogr 20 170ndash180 (2011) doi 101111j1466-8238201000592x

8 P Ruiz‐Benito et al Diversity increases carbon storage andtree productivity in Spanish forests Glob Ecol Biogeogr 23311ndash322 (2014) doi 101111geb12126

9 F van der Plas et al Biotic homogenization can decreaselandscape-scale forest multifunctionality Proc Natl Acad SciUSA 113 3557ndash3562 (2016) doi 101073pnas1517903113pmid 26979952

10 F Isbell D Tilman S Polasky M Loreau The biodiversity-dependent ecosystem service debt Ecol Lett 18 119ndash134(2015) doi 101111ele12393 pmid 25430966

11 S Diacuteaz et al The IPBES Conceptual FrameworkmdashConnectingnature and people Curr Op Environ Sust 14 1ndash16 (2015)doi 101016jcosust201411002

12 United Nations vol COP 10 Decision X2 (Nagoya Japan2010)

13 W M Adams et al Biodiversity conservation and theeradication of poverty Science 306 1146ndash1149 (2004)doi 101126science1097920 pmid 15539593

14 J B Grace et al Integrative modelling reveals mechanismslinking productivity and plant species richness Nature 529390ndash393 (2016) doi 101038nature16524 pmid 26760203

15 D I Forrester H Pretzsch Tamm Review On the strength ofevidence when comparing ecosystem functions of mixtureswith monocultures For Ecol Manage 356 41ndash53 (2015)doi 101016jforeco201508016

16 C M Tobner et al Functional identity is the main driver ofdiversity effects in young tree communities Ecol Lett 19638ndash647 (2016) doi 101111ele12600 pmid 27072428

17 K Verheyen et al Contributions of a global network of treediversity experiments to sustainable forest plantations Ambio45 29ndash41 (2016) doi 101007s13280-015-0685-1pmid 26264716

18 FAO ldquoGlobal Forest Resources Assessment 2015mdashHow are theworldrsquos forests changingrdquo (Food and Agriculture Organizationof the United Nations 2015)

19 H Ter Steege et al Estimating the global conservation statusof more than 15000 Amazonian tree species ScienceAdvances 1 e1500936 (2015) doi 101126sciadv1500936pmid 26702442

20 R Fleming N Brown J Jenik P Kahumbu J Plesnik in UNEPYear Book 2011 United Nations Environment Program Ed(UNEP Nairobi Kenya 2011) pp 46ndash59

21 International Union for Conservation of Nature (IUCN) IUCNRed List Categories and Criteria Version 31 Version 20111(Gland Switzerland and Cambridge ed 2 2012) vol iv p 32

22 H Pretzsch G Schuumltze Transgressive overyielding in mixedcompared with pure stands of Norway spruce and Europeanbeech in Central Europe Evidence on stand level andexplanation on individual tree level Eur J For Res 128183ndash204 (2009) doi 101007s10342-008-0215-9

23 A Bravo-Oviedo et al European Mixed Forests Definition andresearch perspectives For Syst 23 518ndash533 (2014)

24 K B Hulvey et al Benefits of tree mixes in carbon plantingsNature Clim Change 3 869ndash874 (2013) doi 101038nclimate1862

25 FAO ldquoContribution of the forestry sector to nationaleconomies 1990ndash2011rdquo (Food and Agriculture Organization ofthe United Nations 2014)

26 Value-added has been adjusted for inflation and is expressed inUSD at 2011 prices and exchange rates

27 B J Cardinale et al The functional role of producer diversityin ecosystems Am J Bot 98 572ndash592 (2011) doi 103732ajb1000364 pmid 21613148

28 P B Reich et al Impacts of biodiversity loss escalate throughtime as redundancy fades Science 336 589ndash592 (2012)doi 101126science1217909 pmid 22556253

29 M Loreau A Hector Partitioning selection andcomplementarity in biodiversity experiments Nature 41272ndash76 (2001) doi 10103835083573 pmid 11452308

30 D Tilman C L Lehman K T Thomson Plant diversity andecosystem productivity Theoretical considerations Proc NatlAcad Sci USA 94 1857ndash1861 (1997) doi 101073pnas9451857 pmid 11038606

31 H Y H Chen K Klinka Aboveground productivity of westernhemlock and western redcedar mixed-species stands insouthern coastal British Columbia For Ecol Manage 18455ndash64 (2003) doi 101016S0378-1127(03)00148-8

32 The elasticity of substitution (q) which represents the degreeto which species can substitute for each other in contributingto stand productivity reflects the strength of the effect ofbiodiversity on ecosystem productivity after accounting forclimatic soil and other environmental and local covariates

33 H Pretzsch et al Growth and yield of mixed versus purestands of Scots pine (Pinus sylvestris L) and European beech(Fagus sylvatica L) analysed along a productivity gradientthrough Europe Eur J For Res 134 927ndash947 (2015)doi 101007s10342-015-0900-4

34 E D Schulze et al Opinion paper Forest management andbiodiversityWeb Ecol 14 3ndash10 (2014) doi 105194we-14-3-2014

35 R Waring et al Why is the productivity of Douglas-fir higher inNew Zealand than in its native range in the Pacific NorthwestUSA For Ecol Manage 255 4040ndash4046 (2008)doi 101016jforeco200803049

36 J Liang J V Watson M Zhou X Lei Effects of productivityon biodiversity in forest ecosystems across the United Statesand China Conserv Biol 30 308ndash317 (2016) doi 101111cobi12636 pmid 26954431

37 L H Fraser et al Worldwide evidence of a unimodalrelationship between productivity and plant species richnessScience 349 302ndash305 (2015) doi 101126scienceaab3916pmid 26185249

38 M Loreau Biodiversity and ecosystem functioning Amechanistic model Proc Natl Acad Sci USA 955632ndash5636 (1998) doi 101073pnas95105632pmid 9576935


Fig 8 Effect of the size of training sets usedin the geospatial random forest on estimatedelasticity of substitution (q) and generalized R2

in relation toŠMean (solid line) andSEband (greenarea) were estimated with 100 randomly selected(with replacement) training sets for each of the 20size values (between 50 and 1000 GFB plots withan increment of 50)


on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

39 F Isbell et al Nutrient enrichment biodiversity loss andconsequent declines in ecosystem productivity Proc NatlAcad Sci USA 110 11911ndash11916 (2013) doi 101073pnas1310880110 pmid 23818582

40 C Le Queacutereacute et al Global carbon budget 2015 Earth Syst SciData 7 349ndash396 (2015) doi 105194essd-7-349-2015

41 A Hector R Bagchi Biodiversity and ecosystemmultifunctionality Nature 448 188ndash190 (2007) doi 101038nature05947 pmid 17625564

42 L Gamfeldt et al Higher levels of multiple ecosystem servicesare found in forests with more tree species Nat Commun 41340 (2013) doi 101038ncomms2328 pmid 23299890

43 D P McCarthy et al Financial costs of meeting globalbiodiversity conservation targets Current spending and unmetneeds Science 338 946ndash949 (2012) doi 101126science1229803 pmid 23065904

44 C B Barrett A J Travis P Dasgupta On biodiversityconservation and poverty traps Proc Natl Acad Sci USA108 13907ndash13912 (2011) doi 101073pnas1011521108pmid 21873176

45 B Fisher T Christopher Poverty and biodiversity Measuringthe overlap of human poverty and the biodiversity hotspotsEcol Econ 62 93ndash101 (2007) doi 101016jecolecon200605020

46 N Myers R A Mittermeier C G MittermeierG A B da Fonseca J Kent Biodiversity hotspots forconservation priorities Nature 403 853ndash858 (2000)doi 10103835002501 pmid 10706275

47 S Gourlet-Fleury J-M Guehl O Laroussinie Ecology andManagement of a Neotropical Rainforest Lessons Drawn fromParacou A Long-Term Experimental Research Site in FrenchGuiana (Elsevier 2004)

48 J C Tipper Rarefaction and rarefictionmdashThe use and abuse ofa method in paleoecology Paleobiology 5 423ndash434 (1979)doi 101017S0094837300016924

49 M L Rosenzweig Species Diversity in Space and Time(Cambridge Univ Press 1995)

50 L Breiman Random forests Mach Learn 45 5ndash32 (2001)doi 101023A1010933404324

51 S A Chamberlain E Szoumlcs taxize Taxonomic search andretrieval in R F1000Res 2 191 (2013)pmid 24555091

52 B Husch T W Beers J A Kershaw Jr Forest Mensuration(John Wiley amp Sons ed 4 2003)

53 M G R Cannell Woody biomass of forest stands For EcolManage 8 299ndash312 (1984) doi 1010160378-1127(84)90062-8

54 M Simard N Pinto J B Fisher A Baccini Mapping forestcanopy height globally with spaceborne lidar J Geophys Res116 G04021 (2011) doi 1010292011JG001708

55 N L Stephenson P J Mantgem Forest turnover rates followglobal and regional patterns of productivity Ecol Lett 8524ndash531 (2005) doi 101111j1461-0248200500746xpmid 21352456

56 D A Clark et al Measuring net primary production in forestsConcepts and field methods Ecol Appl 11 356ndash370 (2001)doi 1018901051-0761(2001)011[0356MNPPIF]20CO2

57 ESRI ldquoRelease 103 of Desktop ESRI ArcGISrdquo (EnvironmentalSystems Research Institute 2014)

58 R Core Team ldquoR A language and environment for statisticalcomputingrdquo (R Foundation for Statistical Computing Vienna 2013)

59 A Di Gregorio L J Jansen ldquoLand Cover Classification System(LCCS) classification concepts and user manualrdquo (FAODepartment of Natural Resources and Environment 2000)

60 P Legendre Spatial autocorrelation Trouble or new paradigmEcology 74 1659ndash1673 (1993) doi 1023071939924

61 J Liang Mapping large-scale forest dynamics A geospatialapproach Landscape Ecol 27 1091ndash1108 (2012) doi 101007s10980-012-9767-7

62 N A C Cressie Statistics for Spatial Data (J Wiley 1993)63 J Pinheiro D Bates S DebRoy D Sarkar R Development

Core Team nlme Linear and Nonlinear Mixed Effects Models(2011) vol R package version 31-101

64 P Legendre N L Oden R R Sokal A Vaudor J KimApproximate analysis of variance of spatially autocorrelatedregional data J Classif 7 53ndash75 (1990) doi 101007BF01889703

65 A Liaw M Wiener Classification and regression byrandomForest R News 2 18ndash22 (2002)

66 L Magee R2 measures based on Wald and likelihood ratio jointsignificance tests Am Stat 44 250ndash253 (1990)

67 M C Hansen et al High-resolution global maps of 21st-century forest cover change Science 342 850ndash853 (2013)doi 101126science1244693 pmid 24233722

68 E J Pebesma Multivariable geostatistics in S The gstatpackage Comput Geosci 30 683ndash691 (2004) doi 101016jcageo200403012

69 R Costanza et al Changes in the global value of ecosystemservices Glob Environ Change 26 152ndash158 (2014)doi 101016jgloenvcha201404002

70 BLS in BLS Handbook of Methods (US Department of LaborWashington DC 2015)

71 We used the online CPI calculator httpdatablsgovcgi-bincpicalcpl

72 T W Crowther et al Mapping tree density at a global scaleNature 525 201ndash205 (2015) doi 101038nature14967pmid 26331545

73 The tree drawings in Fig 2 were based on actual species fromthe GFB plots The scientific names of these species are(clockwise from the top) Abies nebrodensis Handroanthusalbus Araucaria angustifolia Magnolia sinica Cupressussempervirens Salix babylonica Liriodendron tulipiferaAdansonia grandidieri Torreya taxifolia and Quercus mongolicaFive of these 10 species (A nebrodensis A angustifoliaM sinica A grandidieri and T taxifolia) are listed asendagered or critically endangered species in the IUCN RedList Hand drawings were made by R K Watson

74 Elevation consists mostly of ground-measured data and themissing values were replaced with the highest-resolutiontopographic data generated from NASArsquos Shuttle RadarTopography Mission (SRTM)

75 R J Hijmans S E Cameron J L Parra P G Jones A JarvisVery high resolution interpolated climate surfaces for globalland areas Int J Climatol 25 1965ndash1978 (2005)doi 101002joc1276

76 A Trabucco R J Zomer in CGIAR Consortium for SpatialInformation (CGIAR 2009)

77 N Batjes ldquoWorld soil property estimates for broad-scalemodelling (WISE30sec)rdquo (ISRIC-World Soil Information 2015)

78 D M Olson E Dinerstein The Global 200 Priority ecoregionsfor global conservation Ann Mo Bot Gard 89 199ndash224(2002)

ACKNOWLEDGMENTS

We are grateful to all the people and agencies that helped incollection compilation and coordination of the field data includingbut not limited to T Malone J Crowe M Sutton J LovettP Munishi M Rautiainen staff members from the Seoul NationalUniversity Forest and all persons who made the two SpanishForest Inventories possible especially the main coordinatorsR Villaescusa (IFN2) and J A Villanueva (IFN3) This work wassupported in part by West Virginia University under the UnitedStates Department of Agriculture (USDA) McIntire-Stennis FundsWVA00104 and WVA00105 US National Science Foundation(NSF) Long-Term Ecological Research Program at Cedar Creek(DEB-1234162) the University of Minnesota Department of ForestResources and Institute on the Environment the Architectureand Environment Department of Italcementi Group Bergamo(Italy) a Marie Skłodowska Curie fellowship Polish NationalScience Center grant 201102ANZ900108 the FrenchLrsquoAgence Nationale de la Recherche (ANR) (Centre drsquoEacutetude de laBiodiversiteacute Amazonienne ANR-10-LABX-0025) the GeneralDirectory of State Forest National Holding DB General Directorateof State Forests Warsaw Poland (Research Projects 107and OR2717311) the 12th Five-Year Science and TechnologySupport Project (grant 2012BAD22B02) of China the USGeological Survey and the Bonanza Creek Long Term EcologicalResearch Program funded by NSF and the US Forest Service(any use of trade firm or product names is for descriptivepurposes only and does not imply endorsement by the USgovernment) National Research Foundation of Korea (grantNRF-2015R1C1A1A02037721) Korea Forest Service (grantsS111215L020110 S211315L020120 and S111415L080120) andPromising-Pioneering Researcher Program through Seoul NationalUniversity (SNU) in 2015 Core funding for Crown ResearchInstitutes from the New Zealand Ministry of Business Innovationand Employmentrsquos Science and Innovation Group the DeutscheForschungsgemeinschaft (DFG) Priority Program 1374 BiodiversityExploratories Chilean research grants Fondo Nacional deDesarrollo Cientiacutefico y Tecnoloacutegico (FONDECYT) 1151495 and11110270 Natural Sciences and Engineering Research Council ofCanada (grant RGPIN-2014-04181) Brazilian Research grantsCNPq 3120752013 and FAPESC 2013TR441 supporting SantaCatarina State Forest Inventory (IFFSC) the General Directorate ofState Forests Warsaw Poland the Bavarian State Ministry forNutrition Agriculture and Forestry project W07 the BavarianState Forest Enterprise (Bayerische Staatsforsten AoumlR) German

Science Foundation for project PR 29212-1 the European Unionfor funding the COST Action FP1206 EuMIXFOR FEDERCOMPETEPOCI under Project POCI-01-0145-FEDER-006958and FCTndashPortuguese Foundation for Science and Technologyunder the project UIDAGR040332013 Swiss National ScienceFoundation grant 310030B_147092 the EU H2020 PEGASUSproject (no 633814) EU H2020 Simwood project (no 613762)and the European Unionrsquos Horizon 2020 research and innovationprogram within the framework of the MultiFUNGtionality MarieSkłodowska-Curie Individual Fellowship (IF-EF) under grantagreement 655815 The expeditions in Cameroon to collect thedata were partly funded by a grant from the Royal Society and theNatural Environment Research Council (UK) to Simon L LewisPontifica Universidad Catoacutelica del Ecuador offered working facilitiesand reduced station fees to implement the census protocol inYasuni National Park We thank the following agencies andorganization for providing the data USDA Forest Service School ofNatural Resources and Agricultural Sciences University of AlaskaFairbanks the Ministegravere des Forecircts de la Faune et des Parcs duQueacutebec (Canada) the Alberta Department of Agriculture andForestry the Saskatchewan Ministry of the Environment andManitoba Conservation and Water Stewardship (Canada) theNational Vegetation Survey Databank (New Zealand) Italian andFriuli Venezia Giulia Forest Services (Italy) Bavarian State ForestEnterprise (Bayerische Staatsforsten AoumlR) and the ThuumlnenInstitute of Forest Ecosystems (Germany) Queensland Herbarium(Australia) Forestry Commission of New South Wales (Australia)Instituto de Conservaccedilatildeo da Natureza e das Florestas (Portugal)MBaiumlki data were made possible and provided by the ARF Project(Appui la Recherche Forestiegravere) and its partners AFD (AgenceFranccedilaise de Deacuteveloppement) CIRAD (Centre de CoopeacuterationInternationale en Recherche Agronomique pour le Deacuteveloppement)ICRA (Institut Centrafricain de Recherche Agronomique)MEDDEFCP (Ministegravere de lEnvironnement du DeacuteveloppementDurable des Eaux Forecircts Chasse et Pecircche) SCACMAE (Servicede Coopeacuteration et drsquoActions Culturelles Ministegravere des AffairesEtrangegraveres) SCAD (Socieacuteteacute Centrafricaine de Deacuteroulage) andthe University of Bangui All TEAM data were provided by theTropical Ecology Assessment and Monitoring (TEAM) Networkmdashacollaboration between Conservation International the SmithsonianInstitute and the Wildlife Conservation Societymdashand partiallyfunded by these institutions the Gordon and Betty MooreFoundation the Valuing the Arc Project (Leverhulme Trust) andother donors The Exploratory plots of FunDivEUROPE receivedfunding from the European Union Seventh Framework Programme(FP72007-2013) under grant agreement 265171 The ChineseComparative Study Plots (CSPs) were established in theframework of BEF-China funded by the German ResearchFoundation (DFG FOR891) The Gabon data set was provided bythe Institut de Recherche en Ecologie Tropicale (IRET)CentreNational de la Recherche Scientifique et Technologique(CENAREST) Dutch inventory data collection was done with thehelp of Probos Silve Bureau van Nierop and Wim Daamenfinanced by the Dutch Ministry of Economic Affairs Data collectionin Middle Eastern countries was supported by the Spanish Agencyfor International Development Cooperation [Agencia Espantildeola deCooperacioacuten Internacional para el Desarrollo (AECID)] andFundacioacuten Biodiversidad in cooperation with the governments ofSyria and Lebanon We are grateful to the Polish State ForestHolding for the data collected in the project ldquoEstablishment of aforest information system covering the area of the Sudetes andthe West Beskids with respect to the forest condition monitoringand assessmentrdquo financed by the General Directory of State ForestNational Holding We thank two reviewers who providedconstructive and helpful comments to help us further improvethis paper The data used in this manuscript are summarized inthe supplementary materials (tables S1 and S2) All data neededto replicate these results are available at httpsfigsharecomand wwwgfbinitiativeorg New Zealand data (doi107931V13W29)are available from SW under a materials agreementwith the National Vegetation Survey Databank managed byLandcare Research New Zealand Access to Poland data needsadditional permission from Polish State Forest National Holdingas provided to TZ-N

SUPPLEMENTARY MATERIALS

wwwsciencemagorgcontent3546309aaf8957supplDC1Figs S1 to S4Tables S1 to S2References (79ndash134)

6 May 2016 accepted 22 August 2016101126scienceaaf8957



on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

RESEARCH ARTICLE SUMMARY

FOREST ECOLOGY

Positive biodiversity-productivityrelationship predominantin global forestsJingjing Liang Thomas W Crowther Nicolas Picard Susan Wiser Mo ZhouGiorgio Alberti Ernst-Detlef Schulze A David McGuire Fabio Bozzato Hans PretzschSergio de-Miguel Alain Paquette Bruno Heacuterault Michael Scherer-LorenzenChristopher B Barrett Henry B Glick Geerten M Hengeveld Gert-Jan NabuursSebastian Pfautsch Helder Viana Alexander C Vibrans Christian Ammer Peter SchallDavid Verbyla Nadja Tchebakova Markus Fischer James V Watson Han Y H ChenXiangdong Lei Mart-Jan Schelhaas Huicui Lu Damiano Gianelle Elena I ParfenovaChristian Salas Eungul Lee Boknam Lee Hyun Seok Kim Helge BruelheideDavid A Coomes Daniel Piotto Terry Sunderland Bernhard SchmidSylvie Gourlet-Fleury Bonaventure Sonkeacute Rebecca Tavani Jun Zhu Susanne BrandlJordi Vayreda Fumiaki Kitahara Eric B Searle Victor J Neldner Michael R NgugiChristopher Baraloto Lorenzo Frizzera Radomir Bałazy Jacek OleksynTomasz Zawiła-Niedźwiecki Olivier Bouriaud Filippo Bussotti Leena FineacuterBogdan Jaroszewicz Tommaso Jucker Fernando Valladares Andrzej M JagodzinskiPablo L Peri Christelle Gonmadje William Marthy Timothy OrsquoBrienEmanuel H Martin Andrew R Marshall Francesco Rovero Robert BitarihoPascal A Niklaus Patricia Alvarez-Loayza Nurdin Chamuya Renato ValenciaFreacutedeacuteric Mortier Verginia Wortel Nestor L Engone-Obiang Leandro V FerreiraDavid E Odeke Rodolfo M Vasquez Simon L Lewis Peter B Reich

INTRODUCTION Thebiodiversity-productivityrelationship (BPR the effect of biodiversity onecosystem productivity) is foundational to ourunderstanding of the global extinction crisisand its impacts on the functioning of naturalecosystems The BPR has been a prominentresearch topicwithin ecology in recent decadesbut it is only recently that we have begun todevelop a global perspective

RATIONALE Forests are the most importantglobal repositories of terrestrial biodiversitybut deforestation forest degradation climatechange and other factors are threatening

approximately one half of tree species world-wide Although there have been substantialefforts to strengthen the preservation andsustainable use of forest biodiversity through-out the globe the consequences of this di-versity loss pose amajor uncertainty for ongoinginternational forest management and conser-vation efforts The forest BPR represents acritical missing link for accurate valuation ofglobal biodiversity and successful integrationof biological conservation and socioeconomicdevelopment Until now there have been limitedtree-based diversity experiments and the forestBPR has only been explored within regional-

scale observational studies Thus the strengthand spatial variability of this relationship re-mains unexplored at a global scale

RESULTS We explored the effect of treespecies richness on tree volume productivity atthe global scale using repeated forest invento-

ries from 777126 perma-nent sample plots in 44countries containingmorethan 30million trees from8737 species spanningmostof the global terrestrial bi-omes Our findings reveal a

consistent positive concave-down effect of bio-diversity on forest productivity across the worldshowing that a continued biodiversity losswouldresult in an accelerating decline in forestproductivity worldwideThe BPR shows considerable geospatial var-

iation across theworld The same percentage ofbiodiversity loss would lead to a greater relative(that is percentage) productivity decline in theboreal forests of North America NortheasternEurope Central Siberia East Asia and scatteredregions of South-central Africa and South-centralAsia In the Amazon West and SoutheasternAfrica Southern China Myanmar Nepal andthe Malay Archipelago however the same per-centage of biodiversity losswould lead to greaterabsolute productivity decline

CONCLUSION Our findings highlight thenegative effect of biodiversity loss on forestproductivity and the potential benefits fromthe transition of monocultures to mixed-speciesstands in forestry practices The BPR we dis-cover across forest ecosystems worldwidecorresponds well with recent theoretical ad-vances as well as with experimental and ob-servational studies on forest and nonforestecosystems On the basis of this relationshipthe ongoing species loss in forest ecosystemsworldwide could substantially reduce forest pro-ductivity and thereby forest carbon absorptionrate to compromise the global forest carbonsink We further estimate that the economicvalue of biodiversity in maintaining commer-cial forest productivity alone is $166 billion to$490 billion per year Although representingonly a small percentage of the total value ofbiodiversity this value is two to six times asmuch as it would cost to effectively implementconservation globally These results highlightthe necessity to reassess biodiversity valuationand the potential benefits of integrating andpromoting biological conservation in forestresource management and forestry practicesworldwide

RESEARCH

196 14 OCTOBER 2016 bull VOL 354 ISSUE 6309 sciencemagorg SCIENCE

The list of author affiliations is available in the full article onlineCorresponding author Email albecalianggmailcomCite this article as J Liang et al Science 354 aaf8957(2016) DOI 101126scienceaaf8957

Global effect of tree species diversity on forest productivity Ground-sourced data from 777126global forest biodiversity permanent sample plots (dark blue dots left)which cover a substantial portionof the global forest extent (white) reveal a consistent positive and concave-down biodiversity-productivity relationship across forests worldwide (red line with pink bands representing 95 con-fidence interval right)

ON OUR WEBSITE

Read the full articleat httpdxdoiorg101126scienceaaf8957

on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

RESEARCH ARTICLE

FOREST ECOLOGY


































RESEARCH


on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from






Discussion





on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from










on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from












on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from










on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from




deviation

Source Nominal

resolution

Response variables

P Primary forest



volume (PAI)



data

Plot attributes

S Tree species


species observed on

the plot


A Plot size

area of the

sample plot


Y Elapsed time

between two

consecutive

inventories


G Basal area


area of live trees

measured at 13




data




Tier-2

I1 = 0 if otherwise



data


I2 = 1 when 001 le ps lt 005


I2 = 3 when 015 le ps lt 050

I2 = 4 when 050 le ps lt 100

where ps was plot

size (hectares)



data

Climatic covariates



index100








quarter




Soil covariates











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from







S frac14 S

Seth2THORN



i







on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from



P frac14

X

i2

Vi2 sdotkiminusX

i1

Vi1 sdotki thornM

Yeth4THORN











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from














on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Global analysis









Coefficients



SE 054 110 113 0001 0011 00009512 0000039 0000001 0000008 0000008 0000009 0000001 0000002

Iteration




















on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Economic analysis















































on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from









































ACKNOWLEDGMENTS








on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

RESEARCH ARTICLE

FOREST ECOLOGY


































RESEARCH


on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from






Discussion





on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from










on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from












on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from










on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from




deviation

Source Nominal

resolution

Response variables

P Primary forest



volume (PAI)



data

Plot attributes

S Tree species


species observed on

the plot


A Plot size

area of the

sample plot


Y Elapsed time

between two

consecutive

inventories


G Basal area


area of live trees

measured at 13




data




Tier-2

I1 = 0 if otherwise



data


I2 = 1 when 001 le ps lt 005


I2 = 3 when 015 le ps lt 050

I2 = 4 when 050 le ps lt 100

where ps was plot

size (hectares)



data

Climatic covariates



index100








quarter




Soil covariates











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from







S frac14 S

Seth2THORN



i







on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from



P frac14

X

i2

Vi2 sdotkiminusX

i1

Vi1 sdotki thornM

Yeth4THORN











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from














on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Global analysis









Coefficients



SE 054 110 113 0001 0011 00009512 0000039 0000001 0000008 0000008 0000009 0000001 0000002

Iteration




















on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Economic analysis















































on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from









































ACKNOWLEDGMENTS








on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from






Discussion





on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from










on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from












on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from










on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from




deviation

Source Nominal

resolution

Response variables

P Primary forest



volume (PAI)



data

Plot attributes

S Tree species


species observed on

the plot


A Plot size

area of the

sample plot


Y Elapsed time

between two

consecutive

inventories


G Basal area


area of live trees

measured at 13




data




Tier-2

I1 = 0 if otherwise



data


I2 = 1 when 001 le ps lt 005


I2 = 3 when 015 le ps lt 050

I2 = 4 when 050 le ps lt 100

where ps was plot

size (hectares)



data

Climatic covariates



index100








quarter




Soil covariates











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from







S frac14 S

Seth2THORN



i







on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from



P frac14

X

i2

Vi2 sdotkiminusX

i1

Vi1 sdotki thornM

Yeth4THORN











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from














on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Global analysis









Coefficients



SE 054 110 113 0001 0011 00009512 0000039 0000001 0000008 0000008 0000009 0000001 0000002

Iteration




















on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Economic analysis















































on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from









































ACKNOWLEDGMENTS








on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from










on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from












on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from










on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from




deviation

Source Nominal

resolution

Response variables

P Primary forest



volume (PAI)



data

Plot attributes

S Tree species


species observed on

the plot


A Plot size

area of the

sample plot


Y Elapsed time

between two

consecutive

inventories


G Basal area


area of live trees

measured at 13




data




Tier-2

I1 = 0 if otherwise



data


I2 = 1 when 001 le ps lt 005


I2 = 3 when 015 le ps lt 050

I2 = 4 when 050 le ps lt 100

where ps was plot

size (hectares)



data

Climatic covariates



index100








quarter




Soil covariates











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from







S frac14 S

Seth2THORN



i







on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from



P frac14

X

i2

Vi2 sdotkiminusX

i1

Vi1 sdotki thornM

Yeth4THORN











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from














on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Global analysis









Coefficients



SE 054 110 113 0001 0011 00009512 0000039 0000001 0000008 0000008 0000009 0000001 0000002

Iteration




















on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Economic analysis















































on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from









































ACKNOWLEDGMENTS








on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from












on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from










on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from




deviation

Source Nominal

resolution

Response variables

P Primary forest



volume (PAI)



data

Plot attributes

S Tree species


species observed on

the plot


A Plot size

area of the

sample plot


Y Elapsed time

between two

consecutive

inventories


G Basal area


area of live trees

measured at 13




data




Tier-2

I1 = 0 if otherwise



data


I2 = 1 when 001 le ps lt 005


I2 = 3 when 015 le ps lt 050

I2 = 4 when 050 le ps lt 100

where ps was plot

size (hectares)



data

Climatic covariates



index100








quarter




Soil covariates











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from







S frac14 S

Seth2THORN



i







on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from



P frac14

X

i2

Vi2 sdotkiminusX

i1

Vi1 sdotki thornM

Yeth4THORN











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from














on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Global analysis









Coefficients



SE 054 110 113 0001 0011 00009512 0000039 0000001 0000008 0000008 0000009 0000001 0000002

Iteration




















on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Economic analysis















































on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from









































ACKNOWLEDGMENTS








on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from










on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from




deviation

Source Nominal

resolution

Response variables

P Primary forest



volume (PAI)



data

Plot attributes

S Tree species


species observed on

the plot


A Plot size

area of the

sample plot


Y Elapsed time

between two

consecutive

inventories


G Basal area


area of live trees

measured at 13




data




Tier-2

I1 = 0 if otherwise



data


I2 = 1 when 001 le ps lt 005


I2 = 3 when 015 le ps lt 050

I2 = 4 when 050 le ps lt 100

where ps was plot

size (hectares)



data

Climatic covariates



index100








quarter




Soil covariates











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from







S frac14 S

Seth2THORN



i







on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from



P frac14

X

i2

Vi2 sdotkiminusX

i1

Vi1 sdotki thornM

Yeth4THORN











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from














on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Global analysis









Coefficients



SE 054 110 113 0001 0011 00009512 0000039 0000001 0000008 0000008 0000009 0000001 0000002

Iteration




















on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Economic analysis















































on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from









































ACKNOWLEDGMENTS








on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from




deviation

Source Nominal

resolution

Response variables

P Primary forest



volume (PAI)



data

Plot attributes

S Tree species


species observed on

the plot


A Plot size

area of the

sample plot


Y Elapsed time

between two

consecutive

inventories


G Basal area


area of live trees

measured at 13




data




Tier-2

I1 = 0 if otherwise



data


I2 = 1 when 001 le ps lt 005


I2 = 3 when 015 le ps lt 050

I2 = 4 when 050 le ps lt 100

where ps was plot

size (hectares)



data

Climatic covariates



index100








quarter




Soil covariates











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from







S frac14 S

Seth2THORN



i







on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from



P frac14

X

i2

Vi2 sdotkiminusX

i1

Vi1 sdotki thornM

Yeth4THORN











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from














on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Global analysis









Coefficients



SE 054 110 113 0001 0011 00009512 0000039 0000001 0000008 0000008 0000009 0000001 0000002

Iteration




















on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Economic analysis















































on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from









































ACKNOWLEDGMENTS








on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from







S frac14 S

Seth2THORN



i







on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from



P frac14

X

i2

Vi2 sdotkiminusX

i1

Vi1 sdotki thornM

Yeth4THORN











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from














on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Global analysis









Coefficients



SE 054 110 113 0001 0011 00009512 0000039 0000001 0000008 0000008 0000009 0000001 0000002

Iteration




















on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Economic analysis















































on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from









































ACKNOWLEDGMENTS








on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from



P frac14

X

i2

Vi2 sdotkiminusX

i1

Vi1 sdotki thornM

Yeth4THORN











on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from














on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Global analysis









Coefficients



SE 054 110 113 0001 0011 00009512 0000039 0000001 0000008 0000008 0000009 0000001 0000002

Iteration




















on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Economic analysis















































on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from









































ACKNOWLEDGMENTS








on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from














on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Global analysis









Coefficients



SE 054 110 113 0001 0011 00009512 0000039 0000001 0000008 0000008 0000009 0000001 0000002

Iteration




















on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Economic analysis















































on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from









































ACKNOWLEDGMENTS








on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Global analysis









Coefficients



SE 054 110 113 0001 0011 00009512 0000039 0000001 0000008 0000008 0000009 0000001 0000002

Iteration




















on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Economic analysis















































on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from









































ACKNOWLEDGMENTS








on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from


Economic analysis















































on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from









































ACKNOWLEDGMENTS








on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from









































ACKNOWLEDGMENTS








on

Oct

ober

14

201

6ht

tp

scie

nce

scie

ncem

ago

rg

Dow

nloa

ded

from

positive biodiversity-productivity relationship...

Documents