FILIPE MACHADO FRANÇA ECOLOGICAL IMPACTS OF SELECTIVE LOGGING IN THE AMAZON: LESSONS FROM DUNG BEETLES LAVRAS – MG 2015 FILIPE MACHADO FRANÇA ECOLOGICAL IMPACTS OF SELECTIVE LOGGING IN THE AMAZON: LESSONS FROM DUNG BEETLES Thesis submitted for the degree of Doctor of Philosophy as a Dual PhD with Applied Ecology Postgraduate Program, Federal University of Lavras, Brazil and Lancaster Environment Centre, Lancaster University, United Kingdom Supervisors Dr. Júlio Louzada Dr. Jos Barlow LAVRAS - MG 2015 Ficha catalográfica elaborada pelo Sistema de Geração de Ficha Catalográfica da Biblioteca Universitária da UFLA, com dados informados pelo(a) próprio(a) autor(a). França, Filipe Machado. Ecological impacts of selective logging in the Amazon: lessons from dung beetles/ Filipe Machado França. – Lavras : UFLA, 2016. 190 p. : il. Tese (Doutorado)–Universidade Federal de Lavras, 2015. Orientadores: Júlio Neil Cassa Louzada e Jos Barlow. Bibliografia. 1. Tropical forest. 2. Selective logging. 3. Amazon rainforest. 4. Dung beetles. 5. Ecosystem processes. I. Universidade Federal de Lavras. II. Título. O conteúdo desta obra é de responsabilidade do(a) autor(a) e de seu orientador(a) FILIPE MACHADO FRANÇA ECOLOGICAL IMPACTS OF SELECTIVE LOGGING IN THE AMAZON: LESSONS FROM DUNG BEETLES Thesis submitted for the degree of Doctor of Philosophy as a Dual PhD with Applied Ecology Postgraduate Program, Federal University of Lavras, Brazil and Lancaster Environment Centre, Lancaster University, United Kingdom APPROVED on 2nd of December, 2015 Dr. Carla Ribas Universidade Federal de Lavras, Brazil Dr. David P. Edwards The University of Sheffield, UK Dr. Luiz Fernando S. Magnago Universidade Federal de Lavras, Brazil Dr. Rosa Menéndez Lancaster University, UK Dr. Júlio Louzada Dr. Jos Barlow Supervisors LAVRAS - MG 2015 DECLARATION I hereby declare that this work has been originally produced by myself for this thesis and it has not been submitted for the award of a higher degree to any other institution. Inputs from co-authors are acknowledged throughout. Filipe França, Lavras, December 2015. Suggested citation: França, F. M. (2015). Ecological impacts of selective logging in the Amazon: lessons from dung beetles. PhD Thesis. Postgraduate Program in Applied Ecology, Lavras, Brazil. Dua PhD with Lancaster Environment Centre, Lancaster University, Lancaster, United Kingdom. “Earth provides enough to satisfy every man's needs, but not every man's greed.” ‒ Mahatma Gandhi I dedicate this thesis to my mom Maria de Lourdes for her unconditional love and constant support. Dedico esse tese à minha mãe, Maria de Lourdes, por todo amor incondicional e constante suporte. DEDICATION AKNOWLEDGEMENTS / AGRADECIMENTOS First of all, I thank God for life and for letting me finish this PhD. “There is no knowledge without unity.” (Irish proverb) This way, I am very grateful to my supervisors Julio Louzada and Jos Barlow, who kept me on track to build this thesis. I thank Julio for believing in me and for his confidence on my potential; for allowing me to achieve the dream of working in the Amazon forest. His supervision and friendship in the last years guided me to be a better person and professional. His passion for dung beetles contaminated me, and now I am very proud to be a ‘rola-bosteiro’. Muito obrigado Júlio! Definitely not least, I am also extremely thankful to Jos for accepting to supervise me; for always giving me his time to read my work and for providing incredibly insightful feedbacks that amazingly improved this thesis. Aside from academic contributions, I really appreciated his exceptional support and interest on my wellbeing, which motivated me a lot to keep trying my best to write this thesis, even when I thought that I could not handle the pressure. I also thanks to Vanesca Korasaki for advincing me in the first time I went to the field work; Juliana Silveira for logistical support during all field works; and Ronara Ferreira, Toby Gardner Amanda Arcanjo and Fernando Vaz-de-Mello, who helped me during different stages of planning this thesis. Additionally, I ackonwledge to Dr. Carla Ribas, Dr. David Edwards, Dr. Luiz Fernando Magnago and Dr. Rosa Menéndez for accepting to be part of a dream that hopefully will become true. An enormously thanks to my family, who have been a great support throughout my entire life. Palavras não podem expressar a gratidão que tenho pelos ‘meus velhos’ Fausto e Doxa, sem o seu amor e orações eu jamais teria chegado até aqui. Em especial a minha mãe, sem seu suporte nos últimos meses (e anos) eu não teria chegado até aqui! A minha vovó Déa, por plantar em mim o amor pela natureza. Meus irmãos, Fausto e Fabrício, por suas orações e suas respectivas esposas, Dessa e Renata, pelo carinho e amizade. Especialmente à Giovana, sobrinha mais linda, pelas mensagens que enchem meus olhos de alegria. Aos irmãos da 1a IEQ de Varginha minha gratidão é eterna, em especial D. Sandra e D. Jandira (e famílias) e meus queridos irmãos Ariana, André & Samira. This PhD would not have been possible without the support of the many lovely people around me. In Brazil, I would like to thank UFLA and Applied Ecology postgraduate program for being substantial in my professional development. I thank the professors for the positive criticism and contributions during my career development. Particularly so, to Dr. Carla Ribas, an exemplar person and scientist, who has given me countless advices thorough these years. Thereby, I thank to Prof. Pedro Castro and Prof. Antônio Carlos Fraga, for the great friendship and technical support in the dung beetle fat extractions. Also, an enormous thanks to all colleagues and friends from UFLA, especially the NIECO and ‘Rola-Bosteiros’ friends for their continuous understanding and support during many critical moments. To name just a few: Ananza & Wallace, André, Tonho, Amanda, Bárbara Lemes, Cla, Crises2, Carol, Cavalo & Vanessa, Stenio & Jéssica, Ernesto, Fer Tanure, Grazi, Gui, Java (eco), Java & Thalita, Ju Tuller, Julius & Marina, Lisi & Toru, Lívia, Luizinha, Poli, Rafa(s), Raquel, Rodrigo, Ruanny & Ivo, Renan & Lilian and Yoyo. Above all, I have no words to thanks my best friend and brother Fábio Frazão for being such a positive influence in my life. His friendship, teaching and advices were fundamental to help me to try to be a better person. Lembre-se: ‘o mundo é pequeno e o tempo voa’. I acknowledge to UFLA and Lancaster University for the opportunity of being part of such innovative partnership and for all support during the dual PhD scheme. Fieldwork was founded by CAPES and CNPq-PELD site 23 and Jari Florestal gave us logistical support. Essa tese nunca seria possível sem a colaboração de Edivar D. Correia (Rei da Floresta), Jucelino Alves (Irmão) e da querida Maria Orlandina. Muito obrigado por fazerem parte da minha história, sem vocês eu não seria mestre, muito menos doutor. Gratitude also I have to the Amazon, for being more than a ‘study site’; for allowing me to explore its beauty and dung beetles; for making me sure I chose the right profession. In Lancaster, I am truly grateful for all the great people I’ve met: LEC’s Tropical Research group, thank you very much for being so welcoming and for accepting me in this unique group. I have learned lots from you, and all this knowledge and love made me improve myself. Namely, Andrés (de Moraes e Lima), Anne, Ananza & Wallace, Ciça, Érika & Darren, Gina & Luke (fofos), Jesus, Julia, Mari Piva, Natalie Swan, Patricia, Ruanny, Rach, Sâmia and Sam. Remarkably this ‘Farfer’ is very grateful to ‘Maritaca’ (Hannah Griffiths) and ‘Dannyboy’ (Daniel Tregidgo) for giving me the love and strength I needed many times throughout this PhD. Thank you for making me feel loved just as the crazy person I am, obrigado for showing me a new way to see the world and for laughing, crying and dancing crazily during the rain in the jungle. I miss, love and keep you both in my heart. Indeed, I am also extremely thankful to my housemates and ‘siblings’: Rachel Marshal, Alistair Campbell (Alzão) and Ciça Leal. Muito obrigado de verdade! Por tudo! Eu amo vocês! There are also too many ‘Lancastrians’ who were absolutely essential to made my life happier, but those who taught me countless ‘words of the day’, or even have loved, hugged, laughed, drunk, misbehaved, and cried with me deserve extra gratitude: Ben, Daichi, Daniel, Fran, Harry, James, Joe, Jonny, Liv, Lucie (mercy), Medwards, Molly, Seb, Tekle, Vicky and Molly dog (from Block 3). As well as Ali, Alzão, Amy, Antonio (grazie), Arthur, Arlete & Caio, Caley, Cath & Daniel Baxendale, Eduardo (gracias), Emily (danke), Gareth & Nick, Isabel (obrigado), Iain, John, Kelly Manson, Kirsty, Lucas Gent, Mark, Mia, Nath, Pedro, Phill, Rosa (gracias), Rowan and Tom Walker (from LEC and CEH). I am dreadfully grateful to all of you, thanks for showing me that ‘home is where my heart is’. I also give extra gratitude to Victor and Lisiane (for all the love and uncountable help trhough these years); Rachael Carrie (for landing the ‘lady bug’, which provided me great experiences); Anne Toomey (the best craziest and greatest dance partner I ever had). Therefore, to Teotônio Soares, Tom Walker, Fábio Frazão, Hannah Griffiths, Ricardo Solar, André Tavares, Lisiane Zanella and Ali Birket for their gorgeous smiles when helping with doubts in stats, R & figures; and mainly to those who took part of their time (even during holydays) to read the many drafts and pieces of manuscripts and texts from this thesis: Ali Birket, Alistair Campbell, Ben Rowan, Daniel Tregidgo, Fábio Frazão, Hannah Griffiths, Jos Barlow, Joe Griffiths, Laís Maia, Natalie Swan, Phillip Donskeley and Ross Thomas: worlds can not describe how greatful I am. Finally, the best outcome from these past 4-5 years is finding my best friend and soul mate. I give a hugely special thanks to Laís Maia. When I met she, I found the best person out there. She has been essential to my life, instilling my confidence and having faith in my intellect and me, even when I could not have it by myself. I truly thank her for being such a non-judgmental person, teaching me so much about life and giving me immensurable patience and unfaltering love during all the stages of this Ph.D. I also thank her for waiting for me, for inspiring me and for all support and encouragement she has given me to achieve my dreams. I also sincerely thank to the Ferreira & Maia’s families for receiving me as part of their family. In particular to Cristina, André and Clara, who I consider as my own family. Conclusively, to all those who directly or indirectly helped in the preparation of this thesis. Thank you very much: most of you have added something on me and I am result of your love and kindness. RESUMO O corte seletivo é um dos maiores agentes de degradação nas florestas tropicais, e o entendimento de seus impactos biológicos é essencial para elaborar estratégias de conservação associadas a produção madeireira. Igualmente importante é examinar se os desenhos experimentais mais usados são confiáveis para revelar os verdadeiros impactos das atividades antrópicas, e assim prover informações confiáveis para efetivar a conservação biológica. Essa tese objetiva tratar dessas lacunas de conhecimento através do uso do desenho experimental ‘before-after-control-impact’ (BACI), que foi usado para quantificar os impactos do corte seletivo em besouros escarabeíneos e suas funções ecológicas. O primeiro capítulo de dados (Capítulo 2) compara até que ponto as abordagens BACI e ‘space-for-time’ (SFT) geram diferentes conclusões sobre a relação entre o aumento da intensidade de corte seletivo e a riqueza e biomassa de besouros escarabeíneos. Esse capítulo mostra que SFT, o desenho amostral mais frequente na literatura, pode levar a subestimação dos impactos da degradação florestal na biodiversidade. O capítulo 3 investiga a ocorrência de pontos críticos de intensidade de corte seletivo influenciando os padrões de diversidade e funções ecológicas mediados pelos escarabeíneos; e como esses padrões são influenciados pela escala espacial que o corte seletivo é mensurado. Como resultado, foi encontrado que respostas biológicas ao corte seletivo podem ser não lineares e dependentes da escala. No capítulo 4 é explorado o papel que a estrutura florestal tem em mediar as respostas dos Scarabaeinae e seus processos ecológicos à realização do corte seletivo. Os resultados mostram que impactos da extração madeireira na estrutura florestal (abertura de dossel) não necessariamente explicam os efeitos negativos e desproporcionalmente mais fortes que esse distúrbio teve na riqueza e biomassa de besouros. Além disso, enquanto ressalta que o corte seletivo não influenciou outras duas variáveis ambientais (serapilheira e proporção de areia no solo) ou o consumo fecal e a bioturbação do solo realizados pelos escarabeíneos; esse capítulo mostra que as interações entre esses quatro componentes foram modificadas após a realização do corte seletivo. Finalmente, no capítulo 5, a quantidade de gordura corporal de três espécies de escarabeíneos foi avaliada para investigar, pela primeira vez na literatura, se o corte seletivo causa efeitos subletais em invertebrados tropicais. Esses resultados mostram que besouros coletados em áreas de floresta com corte seletivo tiveram maior proporção de gordura corporal do que besouros coletados em florestas não perturbadas; o que corrobora com os impactos negativos do corte seletivo observados na abundancia relativa de cada espécie. Dessa forma, essa tese discute sobre os desafios para conservação da biodiversidade em um mundo onde as taxas de degradação florestal aumentam a cada dia. Palavras-chave: Florestas tropicais. Corte seletivo. Floresta Amazônica. Rola-bostas. Processos ecossistêmicos. ABSTRACT Selective logging is one of the main drivers of forest degradation in tropical forests, which makes the understanding of its biological consequences essential to inform conservation strategies associated to timber production in those forests. It is also important to examine whether the most frequently used study designs are likely to reveal the true impacts of human activities, thus provide reliable information to develop effective conservation strategies. This thesis aims to fill these knowledge gaps by using an experimental design known as before-after-control-impact (BACI) to quantify the impacts of selective logging on tropical dung beetles and their ecological functions. The first chapter (Chapter 2) compares to what extent space-for-time (SFT) and before-and-after approaches draw different conclusions regarding the relationship between selective logging intensity and dung beetle species richness and biomass. This chapter shows that SFT studies, the most frequently used approach, may underestimate the impacts of forest degradation on biodiversity. The Chapter 3 investigates the presence of thresholds in dung beetle responses to logging intensity, and whether those would be influenced by the spatial scale at which logging intensity is measured. The results from this chapter show that biological responses to selective logging can be non-linear and scale-dependent. The chapter 4 addresses the role of forest structure in mediating the responses of dung beetles and mediated faecal processes to selective logging occurrence. The results show that changes in the forest structure due to selective logging (here measured as canopy openness) not necessarily explain the negative and disproportionally stronger effects of this disturbance on dung beetle biomass and species richness. Therefore, while highlighting that selective logging did not influence two environmental variables (leaf litter and soil sand content) or dung beetle-mediated faecal consumption and soil bioturbation; this chapter shows that the linkages among these four components were modified after the selective logging. Finally, the last experimental chapter (Chapter 5) examines the amount of body fat of three dung beetle species to investigate for the first time in the literature whether selective logging could induce sublethal effects on tropical invertebrates. The results show that dung beetles sampled within selectively logged forests have a higher proportion of body fat than those from undisturbed forests, which matches with the negative impacts of selective logging on the relative abundance of each examined species. Overall, this thesis discuss about the challenges to conserve the biodiversity and ecosystem functioning in a world where forest degradation rates are increasing every day. Keywords: Tropical forests. Selective logging. Amazon rainforest. Dung beetles. Ecosystem processes. LIST OF FIGURES CHAPTER1: General Introduction Figure 1.1 Selective logging impacts on tropical forest.. ................................... 24 Figure 1.2 Dung beetle nest strategies. ................................................................ 30 Figure 1.3 Map of study area.. ............................................................................ 35 Chapter 2 Figure 2.1 SFT × BA: linear model comparisons................................................ 50 Figure 2.2 Accuracy comparisons. ...................................................................... 51 Figure 2.3 Minumum sampling effort and number of spatial replicates. ............ 52 CHAPTER 3 Figure 3.1 Dung beetle sample-based and individual-based rarefaction curves .. 79 Figure 3.2 Responses of dung beetle community metrics and ecological functions to logging intensification (m3 ha-1) at two scales. ................................... 81 Figure 3.3 Local and broader scale AICc comparisons. ...................................... 82 CHAPTER 4 Figure 4.1 Dung beetle-mediated ‘brown chain’ of faecal detritus-pathways. .. 101 Figure 4.2 Pre- and post-logging differences at control and logging sites for dung beetle species richness, biomass, soil bioturbation rates and canopy openness. .......................................................................................... 111 Figure 4.3 Independent effects of environmental and biological predictors on dung beetle-mediated faecal detritus-pathways. ....................................... 112 Figure 4.4 Flowchart illustrating how selective logging alters the relationships in- between dung beetle-mediated faecal processes and those with the environmental variables ................................................................... 116 CHAPTER 5 Figure 5.1 Sublethal and population effects of selective logging. ..................... 134 CHAPTER 6 Figure 6.1 Summary of the ecological linkages associated with selective logging that are discussed in this thesis. ................................................................. 142 CONTENTS FIRST PART Chapter 1: GENERAL INTRODUCTION .......................................... 18 1.1 Tropical Forests ............................................................................. 19 1.2 Selective logging in tropical forests .............................................. 21 1.3 Study taxa – Dung beetles ............................................................. 28 1.4 Study System – The Brazilian Amazon ........................................ 33 1.5 Research objectives ........................................................................ 36 1.6 Thesis Structure ............................................................................. 38 SECOND PART Chapter 2: WHAT THE EYES DO NOT SEE, THE FOREST DOES FEEL: ARE WE UNDERESTIMATING BIODIVERSITY LOSS IN DISTURBED TROPICAL FORESTS? .................. 39 2.1 Abstract........................................................................................... 40 2.2 Introduction .................................................................................... 41 2.3 Methods........................................................................................... 43 2.4 Results ............................................................................................. 48 2. 5 Discussion ...................................................................................... 53 2.6 Conclusions ..................................................................................... 55 2.7 Supplementary Information ......................................................... 57 Chapter 3: IDENTIFYING THRESHOLDS IN DUNG BEETLE RESPONSES TO LOGGING INTENSITY TO IMPROVE THE SUISTAINABILITY OF TROPICAL FOREST MANAGEMENT ........................................................... 67 3.1 Abstract........................................................................................... 68 3.2 Introduction .................................................................................... 69 3.3 Methods........................................................................................... 72 3.4 Results ............................................................................................. 78 3.5 Discussion ....................................................................................... 82 3.6 Conclusions ..................................................................................... 87 3.7 Supplementary Information ......................................................... 89 Chapter 4: ASSESSING THE INFLUENCE OF FOREST DISTURBANCES ON ‘BROWN WORLD’ ECOSYSTEM PROCESSES ................................................................ 97 4.1 Abstract ......................................................................................... 98 4.2 Introduction ................................................................................... 99 4.3 Methods.......................................................................................... 102 4.4 Results ............................................................................................ 109 4.5 Discussion ...................................................................................... 113 4.6 Conclusions .................................................................................... 117 4.7 Supplementary Information ........................................................ 118 Chapter 5: DOES SELECTIVE LOGGING STRESS TROPICAL FOREST INVERTEBRATES? USING FAT STORAGES TO EXAMINE SUBLETHAL RESPONSES IN DUNG BEETLES 124 5.1 Abstract......................................................................................... 125 5.2 Introduction .................................................................................. 126 5.3 Methods......................................................................................... 128 5.4 Results ........................................................................................... 132 5.5 Discussion ..................................................................................... 135 5.6 Conclusions ................................................................................... 138 Chapter 6: GENERAL DISCUSSION ....................................... 139 6.1 Key findings .................................................................................. 141 6.2 Key implications .......................................................................... 142 6.4. Priorities for future work ............................................................ 146 6.5. Concluding remarks .................................................................... 150 7. References ..................................................................................... 152 8. Appendices – Other Outcomes .................................................... 189 FIRST PART 18 Chapter 1: GENERAL INTRODUCTION 19 1.1 Tropical Forests 1.1.1. What about tropical forests? Tropical forests are extremely diverse biomes, which house two-thirds of the known world terrestrial biodiversity (Gardner, 2010). These forests represent approximately 50% of total forest ecosystems on Earth (Pan et al., 2011) and cover around 1,240 Mha distributed across Southeast Asia, Africa and South America (Kim et al., 2015; Pan et al., 2011). Due to their great diversity, complexity and extension (Kim et al., 2015; Mayaux et al., 2005), tropical forests contribute disproportionately to local, regional and global ecosystem processes (Cardinale et al., 2012a; Foley et al., 2007). As examples, there are global climate regulation (Anderson-Teixeira et al., 2012), carbon sequestration and storage (Berenguer et al., 2015, 2014), disease control (Hahn et al., 2014), and biological diversity conservation (Foley et al., 2007). These forests also have an important economic role, providing timber and non- timber products to over 800 million people living in these ecosystems (Chomitz et al., 2007). Consequently, most tropical forests are now imperilled by the increased forest degradation and land-use conversion rates caused by anthropogenic activities (Gibson et al., 2011; Kim et al., 2015; Newbold et al., 2015). 1.1.2 Threats to tropical forests Anthropogenic disturbances pose a major threat to the world’s forested environments (Newbold et al., 2015; Sala et al., 2000), and primary forests are rapidly disappearing due to land-use change and environmental degradation, which are widely recognized as the greatest drivers of biodiversity loss on Earth (Newbold et al., 2015; Pan et al., 2011). This is particularly so in the tropics, where most primary forests co-occur with the highest forest degradation rates (Kim et al., 2015; Lambin et al., 2003; Pan et al., 2011) and over 230 Mha of forests were lost between 2000 and 2012 20 (Hansen et al., 2013). This trend is expected to continue over the coming years (Newbold et al., 2015; Pereira et al., 2010) and, as a result, a growing number of attempts have focused on summarising the consequences of these disturbances through meta-analyses (Burivalova et al., 2014; Gibson et al., 2011; Newbold et al., 2015, 2012; Pfeifer et al., 2014). For example, a global meta-analysis clearly shows how land-use changes and associated pressures reduce the local terrestrial biodiversity (Newbold et al., 2015), while a pan- tropical meta-analysis provides some hope by highlighting the relatively high biodiversity value of selectively logged forests (Gibson et al., 2011). As a result, increased attention has also been given to policies and financial investments such as REDD+ (where REDD is Reducing Emissions from Deforestation and Forest Degradation; UNFCCC, 2008), which support the sustainable forestry management as a strategy to avoid the consequences of forest degradation in tropical forests (Panfil and Harvey, 2015). 1.1.3 Integrating tropical forest production and conservation A central challenge for sustainability in tropical forests is how to preserve the forest biodiversity and ecosystem services while enhancing timber production. The insatiable global demand for timber products poses severe challenges to tropical forests outside protected areas (Asner et al., 2009; FAO, 2010), thus a key challenge arises for the future: How to integrate timber production to tropical forest conservation? This question is, so far, one of the most recognized and controversial trade-offs related to forest management (Duncker et al., 2012; Lafond et al., 2015) and has triggered debates over the need for REDD+ projects to more explicitly develop conservation efforts linked to anthropogenic activities in tropical forests (Panfil and Harvey, 2015). Yet, despite controversies over the synchronism between biodiversity conservation and the timber production in tropical forests (Zimmerman and Kormos, 2012), studies have suggested the sustainable forest management as a good alternative for the long-term 21 maintenance of environmental, social and economic benefits derived from tropical forests (Gardner, 2010 and references therein). Consequently, sustainable forest management is among the most adopted REDD+ interventions to date (Salvini et al., 2014), and is enshrined as a key aim of international biodiversity targets for 2020 (Convention on Biological Diversity, 2014). 1.2 Selective logging in tropical forests One example of sustainable forestry practice widely adopted in the tropics is the selective logging (Blaser et al., 2011; Salvini et al., 2014), which involves the extraction of particular high-cost commercial tree species with minimum trunk diameter (~40-60 cm), thus leaving a great number of non- commercial species with lower trunk diameter unlogged (Edwards et al., 2014c; Griscom and Cortez, 2013). Selective logging is considered the most widespread disturbance driver throughout tropical forests (Blaser et al., 2011; Edwards et al., 2014c; Koltunov et al., 2009; Putz et al., 2012) and contrasts with clear-cut methods frequently used in temperate and boreal regions. In the tropics, at least 28% of all forests were under industrial timber operations until 2009 (Laurance et al., 2009) and, due to its relevance to many local and national economies, tropical regions are now responsible for around one eighth of the global timber production (Blaser et al., 2011). As a consequence, over 400 Mha of tropical forests are now officially designated as timber concessions (Blaser et al., 2011) and, to date disturbed forests occupy larger extensions than primary ones in most of the tropics (Laurance et al., 2014; Nelson and Chomitz, 2011). The Amazonian Basin is an exception, still retaining two-thirds of primary forests (Laurance et al., 2014). However, this trend is rapidly changing due to the increased acceleration of deforestation in Latin America, which was the strongest among tropical regions from 1990 to 2010 (Kim et al., 2015). 22 1.2.1 Logging methods in tropical forests In the tropics, most forests have been harvested by conventional logging techniques (CL) (FAO, 2010); whereas more recently the reduced- impact logging (RIL) system has been considered to reduce the deleterious environmental damages caused by selective logging (Bicknell et al., 2014b; Putz et al., 2008). The conventional logging practices are characterized by unsustainably high logging rates and poor road design and poor silvicultural practices (Laurance et al., 2009; Putz et al., 2012), which are known to increase the negative impacts on both forest structure and diversity (Martin et al., 2015; Putz et al., 2008). In addition, the common unsupervised tree fellers and lack of tangling liana cutting induces unnecessary damage to residual forest by bringing down the stand adjacent to felled trees (Sist et al., 2003). Consequently, even at low-harvest intensities, forests which are subject to CL lose much of their environmental importance due to increased impacts on soil and canopy cover (Asner et al., 2004a; Newbold et al., 2015; Pereira et al., 2002). Contrastingly, RIL techniques aim to reduce the collateral damage from timber extraction by implementing a series of measures to minimize collateral impacts of selective logging (Pinard and Putz, 1996; Putz et al., 2008; West et al., 2014). Although the application of RIL methods is context- dependent and not uniform, it generally starts with a pre-harvesting plan, which inventories and maps all profitable trees (size, number, species and types) that can be cut (Putz and Pinard, 1993; Sist et al., 2003). This plan is designed to minimize ground disturbances caused by skid trails, roads and log yards (Figure 1.1), which are related to extensive impacts on tropical forests and biodiversity (Laurance et al., 2009; Yamada et al., 2014). Thus, tangling lianas are cut to improve work safety and to avoid damage to neighbouring trees during the tree felling, which is directional and bucking to alleviate collateral damage to other trees in the adjacent logging units or nearby protected areas. Lastly, silvicultural treatments and closing operations 23 generally are applied to assure long-term prospects to increase the forest productivity recovery after the first cutting cycle (Putz and Pinard, 1993; Sist et al., 2003). Whether RIL or CL is more profitable depends on the perspective from which benefits and costs are estimated (Putz et al., 2008). Overall, RIL methods can be more expensive, but costs and forest productivity of logged forests under RIL and CL are influenced by many factors, such as diversity and heterogeneity of the forest, labour compensation practices, marked timber, and also the spatial and temporal scales of the research (Medjibe and Putz, 2012; Putz et al., 2008). Nevertheless, when considering the long-term effects of both CL and RIL methods on logged forests from a broader perspective and including environmental and social impacts, RIL takes the lead as the best alternative (Bicknell et al., 2014b; Medjibe and Putz, 2012; Putz et al., 2008). Additionally, growing evidence has shown the importance of other landscape strategies proposed to meet timber demands through more “biodiversity friendly” logging activities (Edwards et al., 2014a). As result, the debate about land-sparing versus land-sharing strategies, generally applied to agricultural production (Fischer et al., 2008; Phalan et al., 2011), has been applied to the logging context (Edwards et al., 2014a). This literature indicates that the shape of biodiversity-disturbance curves can bring insights into the best strategy for biodiversity conservation in modified environments (Phalan et al., 2011; von Wehrden et al., 2014). Overall, it has been suggested to spare lands for biodiversity conservation within production forests (Edwards et al., 2014a); or in the cases where concave biodiversity-yielding relationships are observed and the biodiversity is more sensitive to the initial presence of a disturbance (Phalan et al., 2011; von Wehrden et al., 2014). In contrast, land- sharing has been recommended as better strategy where the relationships between biodiversity-disturbance are concave and relatively large proportion of the biodiversity can be maintained across a low-disturbed landscape (Phalan et al., 2011; von Wehrden et al., 2014). From the perspective of 24 maintaining higher biodiversity levels within logging concessions, it has been recently suggested that the conservation value of logged forests will be optimized if RIL methods are implemented under the land-sparing strategy (Edwards et al., 2014a). Figure 1.1 Selective logging impact on tropical forest. (A) Secondary-level logging road; (B) Timber transport truck in a main logging road; (C) Physical evidence of timber removal resulting in (D) a forest canopy gap; (E) Forest gap caused by a log yard (timber already removed) and (F) wooden logs of Dinizia excela Ducke (Fabacea, Mimosoideae) timber species (Angelim vermelho) stored in a log yard. All photos were taken by F.F. in the Amazon forest, state of Pará, Brazil. 1.2.2 Ecological impacts of selective logging in tropical forests The ecological impacts of selective logging depend very much on the harvesting intensity, logging methods and pre-logging regional forest aspects 25 (Bicknell et al., 2015, 2014a; Martin et al., 2015; Putz et al., 2012). Logging intensities can be calculated by different metrics (i.e. percentage of tree basal area removal and/or number of removed trees per hectare), but most studies measure the volume of timber extracted per hectare (m3/ha), which ranges between 0.22 and 145.3 m3 ha-1 in the tropics (Laufer et al., 2013; Zimmerman and Kormos, 2012). Although the RIL method can result in halved-damages to the residual stand when compared to CL practices (Sist et al., 1998), this method can fail in terms of sustainability in highly stocked forests, such as those in Southeast Asia (Medjibe and Putz, 2012). In such forests, logging intensities very often exceed 100 m3/ha and damage to residual forests can exceed 50%, therefore threatening future yields, forest biodiversity and functioning (Medjibe and Putz, 2012; Sist et al., 1998). On the other hand, where forests have only a small proportion of tree species with high timber value, logging operations generally remove below 50 m3ha-1 (Burivalova et al., 2014; Sist et al., 1998). Such low-intensity logging tends to be less damaging to forest structure, biodiversity and ecosystem functioning (Edwards et al., 2014c; Imai et al., 2012; Slade et al., 2011). Regardless of the method, as logging intensity increases, so do the forest gaps caused by tree fall, log yards, skid trails and logging roads (Asner et al., 2004a; Laurance et al., 2009). These cause greater damage to residual stands (Gatti et al., 2015; Martin et al., 2015) and forest canopy (Asner et al., 2004a; Pereira et al., 2002), thus leading to changes in the microhabitat conditions due to understory desiccation caused by the higher exposition to sunlight (Yamada et al., 2014), heat and drying winds (Costa et al., 2015; Mazzei et al., 2010). For instance, higher canopy openness in selectively logged forests can influence tree seedling development (Duah-gyamfi et al., 2014), the amount and moisture of ground leaf litter (Chung et al., 2000; but see: Burghouts et al., 1992) and even the impacts of dry seasons, when compared to unlogged forests (Koltunov et al., 2009). The response of animal communities to logging effects diverges among taxonomic groups and continental location (Burivalova et al., 2014), 26 and can also depend on forest seasonality and characteristics (Davis and Sutton, 1998; Davis et al., 2001), logging methods (Edwards et al., 2012b), species traits and environmental tolerance (Burivalova et al., 2015; Schwitzer et al., 2011; Slade et al., 2011). Overall, studies have found both positive, neutral and negative consequences of selective logging on forest fauna populations and communities (Edwards et al., 2012b; Laufer et al., 2015; Schwitzer et al., 2011; Struebig et al., 2013; Wearn et al., 2013). In addition, although still under explored, studies have examined the sublethal stress- induced effects of selective logging on animal communities (i.e. Leshyk et al., 2012; Lucas et al., 2006; Mastromonaco et al., 2014; Rimbach et al., 2013; Suorsa et al., 2003). This body of literature has shown that selective logging leads to increased stress-induced responses in vertebrates from logged forests; however, most studies were carried out in temperate regions (but see: Rimbach et al., 2013) and, to date there are no assessments of sublethal effects caused by selective logging on tropical invertebrates. 1.2.3 Functional value of selectively logged tropical forests The functional importance of selectively logged tropical forests is controversial. Studies have demonstrated that ecosystem processes have considerable resilience to logging operations (Mazzei et al., 2010; Newbold et al., 2015), whereas others suggest that it might take many decades for logged forests to fully recover (Osazuwa-Peters et al., 2015; Yamada et al., 2014). Indeed, selective logging has been suggested as a major driver of tropical forest degradation (Asner et al., 2009; Gatti et al., 2015; Zimmerman and Kormos, 2012), negatively affecting both carbon stocks (Berenguer et al., 2015, 2014) and ecosystem functioning (Edwards et al., 2014c; Foley et al., 2007; Gutiérrez-Granados and Dirzo, 2010; Schleuning et al., 2011; Slade et al., 2011). Nevertheless, selective logging is undoubtedly less environmentally severe than other forms of anthropogenic disturbances like fire, pastures and plantations (i.e. Barlow et al., 2006; Edwards et al., 2012; 27 Gibson et al., 2011; Scheffler, 2005). As such, logged forests and primary forests can retain similar biodiversity and ecosystem processes (Berry et al., 2010; Edwards et al., 2012a, 2012b; Ewers et al., 2015; Mazzei et al., 2010; Yamada et al., 2014), but this will depend on logging intensity (Burivalova et al., 2014) and absence of further logged-forest disturbances (Edwards et al., 2014c; Lee-Cruz et al., 2013; Luke et al., 2014). Undoubtedly, when it comes to preserving tropical biodiversity, there is no better substitute than pristine forests (Gibson et al., 2011). However, the ecological value of logged forests should not be overlooked (Ewers et al., 2015), since these forests provide economic benefits through provision of timber and non-timber products, besides retaining tropical biodiversity and ecosystem processes (Berry et al., 2010; Edwards et al., 2012b; Ewers et al., 2015; Gardner et al., 2009; Gibson et al., 2011). In addition, most of the endemic and threatened species on Earth are found entirely outside the existing reserve network (Laurance et al., 2014; Rodrigues et al., 2004), which likely means that a great part of tropical biodiversity may be within the over 400 Mha of tropical forests officially designated as timber concessions (Blaser et al., 2011). Therefore, because protected areas alone may not be enough to conserve the exceptional biodiversity in tropical forests (Laurance et al., 2012), the forests selectively logged under sustainable forestry management appear as alternative refuges for the persistence of at least part of the forest- species in modified landscapes (Gardner, 2010 and references therein). Yet, the high ecological value of selectively logged forests (Gibson et al., 2011) justifies why conservation efforts should invest in both sustainable forest management (Zimmerman and Kormos, 2012) and protection of large extensions of logged forests to attenuate the threats to tropical biodiversity (Gibson et al., 2013; Reynolds et al., 2011). 28 1.3 Study taxa – Dung beetles Dung beetles (Coleoptera: Scarabaeidae) are a dominant dung feeding group of insects globally distributed on every continent except Antarctica (ScarabNet, 2008). With just over 6,200 described species and 267 genera, dung beetles are one of the more morphologically diverse animal taxa (Tarasov and Génier, 2015). To date, the most commonly used dung beetle classification follows the precedent set by Balthasar (1963), which divides dung beetles into two distinct subfamilies with six tribes each, as follows: Coprinae with tribes Coprini, Dichotomomiini, Oniticellini, Onitini, Onthophagini and Phanaeini; and Scarabaeinae with tribes Canthonini, Eucranii, Eurysternini, Gymnopleurini, Scarabaeini and Sisyphini. Nevertheless, recent phylogenetic studies have called for further research integrating morphological, molecular and fossil date aspects to clarify the present dung beetle phylogeny (Mlambo et al., 2015; Tarasov and Génier, 2015). Although a cosmopolitan group, Scarabaeinae beetles have their highest diversity in the tropical regions, where they are most abundant in savannah and forest environments (Hanski and Cambefort, 1991; Nichols et al., 2007; Philips, 2011). Their high abundance and diversity (Tarasov and Génier, 2015) coupled to their close association with specific vegetation and soil types (Hanski and Cambefort, 1991) make dung beetles an exceptional study model (Nichols and Gardner, 2011). Moreover, due to their high sensitivity to changes in environmental conditions (Bicknell et al., 2014a; Menéndez et al., 2014), inexpensive surveys and ability to predict responses of many other taxa (Edwards et al., 2014b; Gardner et al., 2008a), dung beetles are highly recommended as a cost-effective and responsive taxonomic group for biodiversity monitoring and inventory taking across the tropics (Gardner et al., 2008a; Nichols et al., 2007). As a consequence, a large body of literature has considered dung beetles in the evaluation of the impacts of anthropogenic activities in tropical forests (e.g. Audino et al., 2014; Bicknell et al., 2014; 29 Gardner et al., 2008a; Nichols et al., 2007). In addition, through dung manipulation for feeding and nesting purposes (Hanski and Cambefort, 1991), these detritivore beetles provide a number of ecological functions (reviewed in Nichols et al., 2008), as well as ecosystem services valued at $380 and £367 million for the cattle industry in the US and UK, respectively (Beynon et al., 2015; Losey and Vaughan, 2006). 1.3.1 Dung beetle breeding behaviour Scarabaeinae dung beetles are among the most popular common Coleoptera subfamilies in the world, being among the most cited subfamily of beetles on Google Scholar (Tarasov and Génier, 2015). A large part of this is because of their consumption of faecal matter for feeding and nesting purposes, which allows them to thrive in most ecosystems and to crucially contribute to several key processes on Earth’s ecosystems (Nichols et al., 2008). Although dung beetles exhibit a range of feeding behaviours (i.e. feeding on rotting fruit and fungus, vertebrate carrion and dead invertebrates; Hanski and Cambefort, 1991; Larsen et al., 2006), most species are coprophagous and feed primarily on the rich protein and microbial liquid content of animal excreta (Hanski and Cambefort, 1991; Tixier et al., 2015). As a result, these detritivore beetles are an essential component of the dung- based food webs in the “brown-world” (Nichols and Gardner, 2011; Wu et al., 2011). Although there are some kleptoparasite species, which have lost the dung-collecting behaviour and become nest-parasites of other Scarabaeinae species, most dung beetles can be divided into three broad nesting/feeding strategies: the tunnelers (paracoprids), rollers (telecoprids) and dwellers (endocoprids) (Hanski and Cambefort, 1991). Briefly, tunneler species dig in close proximity to or below the dung pat and transport dung into these vertical tunnels for adult feeding or breeding (Figure 1.2A). The rollers make brood balls that are transported some horizontal distance away before burial beneath 30 soil surface for feeding on or breeding in (Figure 1.2B). Lastly, the dweller species feed and breed within the dung mass itself, or in a pit immediately under dung pats (Figure 1.2C; Halffter and Edmonds, 1982a; Hanski and Cambefort, 1991). These varied patterns of dung consumption and relocation performed by dung beetles lead to a series of ecosystem functions, which are directly relevant to tropical forests (reviewed in Nichols et al., 2008). Figure 1.2 Dung beetle strategies: tunneler or paracoprid species (A); roller or telecoprid species (B); and dweller or endocoprid species (C). Modified from Halffter & Edmonds (1982). 1.3.2 Dung beetle-mediated ecosystem processes 1.3.2.1 Dung removal and consumption Dung beetles are the most important taxa contributing to dung decomposition (Lee and Wall, 2006). Through their manipulation of the dung resources, these insects provide a range of key ecosystem processes that facilitate the transfer of energy and matter through dung-based detrital food webs (Nichols and Gardner, 2011). In tropical forests, dung beetles are attracted to dung within the first few minutes after deposition (F. França, personal observation), and removal can be very quick. In Southeast Asian forests around 22% of the dung was removed within 24 hours (Kudavidanage et al., 2012), whereas in Malaysian Borneo removal ranged from 63% in high intensity-logged forests to 99% in undisturbed and selectively logged forests 31 (780 g of cow dung; Slade et al., 2011). In the Peruvian Andes around 50- 100% of the dung was removed within three days (~100 g of pig dung, Horgan and Fuentes, 2005) and in the Brazilian Amazon the removal rates in 24 hours varied from 100% in primary forests to ~60% and 30% in managed forests and introduced pastures, respectively (70 g of human dung; Braga et al., 2013, 2012). Furthermore, such differences in dung removal rates between forested and deforested habitats may be caused by the increased exposure of dung pats to sunlight and high temperatures in the opened habitats. This leads to crust formation on the surface of the dung and, consequently, a decrease of dung attractiveness to detritivores and reduced dung consumption and decomposition rates (Nadeau et al. 2015). On the other hand, in forested environments, dung pats may remain attractive longer, because their moisture and quality are retained under shade (Horgan et al. 2005). 1.3.2.2 Incidental ecosystem processes While leading to increased dung beetle biomass production for predators (Young, 2015), the dung consumption also instigates incidental effects on edaphic properties with important implications for both green and brown worlds (Nichols et al., 2013b; Wu et al., 2011). Previous research has shown that dung beetle activity increases both dry matter yields and nitrogen content in vegetation (Bang et al., 2005), as well as leaf litter decomposition rates (Tixier et al., 2015). In addition, it has been noted that dung beetle activities decrease the volatilization of NH3 and increase the mineralization and nitrification processes (Yokoyama et al., 1991), which may explain how dung beetles collaborate to the transfer of nitrogen from dung to soil (Yamada et al., 2007). However, these benefits may result not only from dung beetle activities, but also from the indirect effects they have on microorganisms responsible for the incorporation of nutrients into the soil (Slade et al., 2015). As well as promoting the recycling of nutrients, dung beetles promote the soil fertility through bioturbation, “the displacement and mixing of sediment 32 particles” (Nichols et al., 2008). Through tunnelling activities and soil movement, dung beetles improve the soil aeration, water porosity and water absorption capacity, therefore improving soil conditions for root penetration (Miranda, 2006). Although Bang et al., (2005) provide the only empirical evidence that soil bioturbation, as a consequence of tunnel building by tunneler dung beetle species, modifies soil physical properties; a previous study in Costa Rica has shown that soil water retention increased and soil bulk density declined under dung pats in seasonally dry pastures (Herrick and Lal, 1996). In addition, recent evidence has suggested that dung beetle-mediated soil bioturbation modifies the soil microbial diversity and functioning, therefore contributing as a mobile route between decomposition processes above and below ground (Slade et al., 2015). Two other important dung beetle-mediated incidental ecosystem processes, are secondary seed dispersal and parasite suppression. Dung beetles very often influence the fate of dispersed seeds in tropical forests by relocating those, which are dispersed by mammal defecation (Andresen, 2002; Griffiths et al., 2015). In doing so, dung beetles transport seeds both horizontally (by roller species) and vertically (by tunnelers) from the deposition site (Nichols et al., 2008). This can benefit the survival of buried seeds by decreasing seed predation and pathogen mediated mortality (Estrada and Coates-Estrada, 2002), or by dispersing them to more favourable microclimates for germination and growth (Andresen and Levey, 2004) as well as reducing density dependent competition mortality (Lawson et al., 2012; Santos-Heredia and Andresen, 2014). Even though seeds are dung contaminants from a “dung beetle’s perspective” (Nichols et al., 2008), by secondarily dispersing seeds the dung beetles can have far consequences on emergence and survival of seedlings (H. Griffiths J. Louzada, R. Bardgett, and J. Barlow, unpublished manuscript), therefore influencing plant communities and forest regeneration. In addition, a recent review has explored the diverse ways by which coprophagous beetles directly or indirectly contribute (or not) to parasite transmission (Nichols and Gómez, 2014). While showing that linkages 33 between dung beetles and parasites may have divergent consequences on transmission intensity within specific transmission cycles and across transmission cycle types, they bring attention to the diverse mechanisms in which dung beetles can maintain, reduce or increase the mammal parasitic helminth contamination. 1.4 Study System – The Brazilian Amazon Covering approximately 6.5 million km2 and spanning nine countries in South America, the Amazon forest is the largest remaining tropical rainforest in the world. Housing ~25% of terrestrial species (Dirzo and Raven, 2003; Sala et al., 2000) and storing ~ 86 Pg of carbon (Berenguer et al., 2015, 2014; Pan et al., 2011), it clearly plays a vital role in the global, regional and local provision of ecosystem services (Leadley et al., 2014). Nevertheless, in recent years increased human activities have led to growing deforestation (Kim et al., 2015) and, consequently, less carbon storing (Berenguer et al., 2014) and potential modifications in river flow, higher forest fire frequency and large scale changes in rainfall (Leadley et al., 2014). Above all human activities, the selective logging is a major activity occurring in the Amazon, particularly so in the Brazil where the largest fraction of this ecosystem is retained (Foley et al., 2007). As result, over 50 Mha of Amazonian forests are under timber concessions (Macpherson et al., 2010) and between 1990 and 2010 the world’s largest acceleration of annual net forest area loss that occurred in Latin America was dominated by Brazil (Kim et al., 2015). This present project was carried out within the 1.7 Mha Jari Florestal landholding, located in the state of Pará in the north-eastern Brazilian Amazonia (00°27′–01°30′ S, 51°40′–53°20′ W). The region comprises a mosaic of Eucalyptus plantations and regenerating secondary forests within ~1.5 Mha of primary forests subjected to very low levels of disturbance (Barlow et al., 2010; Parry et al., 2009a). These forests are characterized as evergreen dense tropical rainforest (Souza, 2009), often dominated by the 34 timber species Dinizia excela Ducke (Fabacea, Mimosoideae) (Laufer et al., 2015), which corresponds to about 50% of exploited timber in some Amazonian regions (Barbosa, 1990). Within this large landscape about 544,000 ha of native forest is divided in “Annual Operating Planning” (POA) subsets, each one planned to be logged every year (since 2003) under a 30 year cutting cycle. Logging activities are planned following the FAO model code of forest harvesting (Dykstra and Heinrich, 1996) and during the pre- harvest inventory each POA is subdivided into 10 ha (250 x 400 m) units planned to be logged with a specific logging intensity (m3 ha-1). The experiments of this project were established in 34 planned logging units (Figure 1.3), where dung beetles and ecological functions were sampled before and after logging operations at the same sites and following the same methods. 35 Figure 1.3 Map of study area. (A) Brazil; (B) state of Pará; (C) Jari landholding and (D) the experimental design in the Jari region where we sampled within 34 planned logging units. The units that we sampled and were selectively logged after the first dung beetle collection are highlighted in dark grey (1-29), whereas the five control units, which remained unlogged during the course of the study, are clear (30-34). 36 1.5 Research objectives This thesis focuses on understanding the direct and indirect impacts of forest degradation on biodiversity and ecosystem processes. In particular, how selective logging influence tropical dung beetle diversity, physiology and the ecological process they govern. This was addressed in the following four topics: - Chapter 2: What the eyes do not see, the forest does feel: Are we underestimating biodiversity loss in disturbed tropical forests? Understanding the rate and spatial distribution of biodiversity loss due to the human alteration of global environment requires realistic assessments. Most studies about the consequences of forest disturbances on biodiversity adopt the space-for-time substitution (SFT), but when researchers are able to sample prior to the disturbance event, a before-after-control-impact (BACI) design can be used. Yet, it is not clear to what extent a reliance on SFT studies could affect inference about impacts of human activities on biodiversity in terrestrial environments. Therefore, the first research objective was to compare the conclusions drawn by STF and BACI designs regarding the responses of dung beetle species richness and biomass to logging intensification. Chapter 2 research questions: (1) Are we underestimating the biodiversity consequences of tropical forest degradation? (2) What are the pros and cons of SFT and BACI experimental designs? - Chapter 3: Identifying thresholds in dung beetle responses to logging intensity to improve the sustainability of tropical forest management Selective logging is one of the most widespread economic activities in tropical forests. Despite progress made to understand logging consequences on tropical biodiversity and ecosystem processes, to date just few meta-analysis studies have explored logging intensities as a continuous, rather than a categorical metric, to identify thresholds of logging intensity above which multi-taxa biodiversity decreases. Yet, while studies have shown the scale- 37 dependent responses of species richness in logged forests, our knowledge of logging impacts is also limited by uncertainty about the spatial-scale of management recommendations. Thus, the second aim of this research was to empirically examine whether dung beetle diversity and species composition, and associated ecological functions respond scale-dependently and non- linearly to increasing logging intensity. Chapter 3 research questions: (1) What are the effects of increasing the intensity of selective logging on dung beetles and ecological processes they govern? (2) What is the relationship between logging intensification and biological responses? (3) Are these relationships scale-dependent on the spatial range that the logging intensity is measured? - Chapter 4: Assessing the influence of forest disturbance on ‘brown world’ ecosystem processes Despite evidence showing that forest degradation leads to direct and indirect impacts on biodiversity and ecosystem functioning, we have a limited understanding about how forest structure buffers and/or mediates these impacts along detritus-based food webs of the ‘brown world’. Consequently, there is very little empirical evidence exploring the extent to which forest degradation modifies the influence of environmental factors on dung beetle mediated faecal pathways. The third research aim was therefore to address this by investigating how selective logging alters the relative importance of canopy openness, leaf litter weight and soil sand content on dung beetle- mediated consumption, production and incidental processes. Chapter 4 research questions: (1) Do logging impacts on forest structure mediate and/or buffer the negative consequences on dung beetle-mediated detritus-pathways? (2) Does selective logging alter the influence of environmental conditions on post-logging dung beetle-mediated brown processes? (3) Does selective logging generate cascade effects on dung beetle faecal consumption and soil bioturbation? 38 - Chapter 5 – Does selective logging stress tropical forest invertebrates? Using body fat stores to examine sublethal responses in dung beetles Most research on sublethal effects of forest degradation is focused on vertebrates in temperate regions. Consequently, we have a very limited understanding about how environmental disturbances could induce sublethal effects on invertebrates. Therefore, the final research aim of this thesis was to address this knowledge gap by investigating whether selective logging could stress-induce sublethal effects upon the body fat content of three dung beetle species, and how these sublethal responses link to population-scale patterns. Chapter 5 research questions: (1) Does selective logging cause stress- induced sublethal effects on dung beetles? (2) Do sublethal effects match with population-scale responses to selective logging? 1.6 Thesis Structure Each of the experimental chapters of this thesis have been written for publication: Chapter 2 is under peer-review in Journal of Applied Ecology, and I intend to submit Chapters 3, 4 and 5 for review and publication (target journals are Forest Ecology and Management, PLoS ONE, and Conservation Physiology, respectively). The structure of this thesis is therefore made up of stand-alone chapters linked by a common theme of selective logging in tropical forests and the effects on dung beetle diversity, physiology and ecosystem processes they mediate. Chapter 6 provides a summary of the key findings resulting from each data chapter, considering their importance for science and conservation policies in tropical forests under timber production pressure, as well as highlighting future research needs. Lastly, I gathered all the references from each of the above chapters and placed them after the Chapter 6, therefore reducing the thesis length (as well as the unnecessary use of paper) and avoiding the repetition of references cited in more than one chapter. The appendices at the end of the thesis demonstrate publications that have resulted from research I was involved with in addition to my direct doctoral research. 39 Chapter 2: WHAT THE EYES DO NOT SEE, THE FOREST DOES FEEL: ARE WE UNDERESTIMATING BIODIVERSITY LOSS IN DISTURBED TROPICAL FORESTS? Selectively logged forest in the Brazilian Amazon, state of Pará. 40 2.1 ABSTRACT Human alteration of the global environment is leading to a pervasive loss of biodiversity. Most studies evaluating human impacts on biodiversity occur after the disturbance has taken place using spatially distinct sites to determine the undisturbed reference condition. This approach is known as a space-for-time substitution (SFT). However, SFT could be underestimating biodiversity loss if spatial controls fail to provide adequate inferences about pre-disturbance conditions. We compare the SFT substitution with a before-and-after (BA) approach by assessing dung beetles before and after a logging exploration in the Brazilian Amazon. We sampled 34 logging management units, of which 29 were selectively logged with different intensities after our first collection. We used dung beetles species richness, species composition and biomass as our biodiversity response metrics and the gradient of selective logging intensity as our explanatory metric. Only the BA approach consistently demonstrated the negative impacts of logging intensification on all dung beetle community metrics, and it doubled estimates of species loss when compared to SFT. Moreover, the BA approach explained more of the variance in the relationships, and reached the critical significance p-value with a smaller number of spatial replicates than SFT. Our results suggest that SFT substitution may greatly underestimate the consequences on local species diversity and community turnover. These results have important implications for researchers investigating human impacts on biodiversity. Incentivising BA approaches will require longer-term funding, more time spent to gather the data and stronger links between researchers and landowners. However, BA approaches are accompanied by many logistical constraints, making the continued use of SFT studies inevitable in many cases. Finally, we highlight that non-significant results and weak effects should be viewed with caution. Keywords: Before-after-control-impact (BACI). Chronosequences. Dung beetles. Land-use change. Rain forest. Reduced-impact logging. Resampling. Selective logging. 41 2.2 Introduction It is well known that human alteration of the global environment is leading to a pervasive loss of biodiversity (Cardinale et al., 2012b; Newbold et al., 2015). Habitat loss and degradation remain the main causes of biodiversity loss and species extinctions across the world (Krauss et al., 2010; Mantyka-pringle et al., 2012). This is particularly so in the tropics, which contain most of the world’s biodiversity and have some of the highest land- use change rates (Lambin et al., 2003; Romdal et al., 2013). Understanding the rate and spatial distribution of biodiversity loss requires accurate assessments of the impacts of land-use change and land management (Gibson et al., 2011; Romdal et al., 2013). Much ecological research has been directed at this, and there are a growing number of attempts to summarise this in meta-analyses (Bicknell et al., 2014b; Burivalova et al., 2014; Gibson et al., 2011; Newbold et al., 2015, 2012; Pfeifer et al., 2014). For example, a global meta-analysis clearly shows how land-use changes and associated pressures reduce the local terrestrial biodiversity (Newbold et al., 2015), while a pan-tropical meta-analysis provides some hope by highlighting the relatively great biodiversity value from selectively logged forests (Gibson et al., 2011). Despite the obvious appeal of these global syntheses, any such meta-analyses will only ever be as reliable as the design of the many studies that supply the data. It is therefore timely and important to examine whether the most frequently used study designs are likely to reveal the true impacts of human activities, and provide information that can be used for developing effective conservation strategies. One important problem researchers face when evaluating human impacts on biodiversity is that the main disturbance events have already taken place. As a result, studies are forced to use spatial reference sites in nearby regions where the human impact of interest has not yet occurred (e.g. Edwards et al. 2011, 2012b; a; Thomaz et al. 2012; Berenguer et al. 2014). This approach is known as a space-for-time (SFT) substitution and dominates the literature on land-use change. For example, we reviewed the available 42 literature evaluating selective logging impacts on tropical invertebrates, and found that 49 out of 53 publications evaluating these effects were based on space-for-time approaches (see Appendix S2.1). A major issue with SFT approaches is to assume that average changes over space are representative of changes through time (Johnson and Miyanishi, 2008). In an ideal world, when researchers are able to sample prior to the disturbance event, they can use a before/after-control/impact design (BACI) (Smith, 2013) - abbreviated here to before-and-after (BA). Before-and-after designs have been conducted in several experimental landscape manipulations (Chai et al., 2012; Forkner et al., 2006; Kibler et al., 2011) and studies (e.g. see Table S1, Appendix A). While most researchers recognise the potential benefits of a BA design (Bicknell et al., 2015; Kibler et al., 2011), it is not clear to what extent a reliance on SFT studies could be affecting inference about human impacts on biodiversity in terrestrial environments. We address this by using a planned commercial logging operation in the Brazilian Amazon to assess whether space-for-time assessments could result in an underestimation of biodiversity loss in tropical forests. We focus on selective logging as it is one of the most important economic activities in tropical forests (Guariguata et al., 2010; Wilson et al., 2010) and has been suggested as less environmentally damaging compared to other anthropogenic disturbances like fire, agriculture and fragmentation (e.g. Barlow et al. 2006; Gibson et al. 2011; Edwards et al. 2012a; b). We use dung beetle as a model system, since they are considered as a cost-effective and responsive taxonomic group for evaluating the biological impacts of forestry practices (Bicknell et al., 2014a; Edwards et al., 2012b, 2011; Gardner et al., 2008a; Scheffler, 2005; Slade et al., 2011). In particular, we examine to what extent space-for-time and before- and-after approaches yield different conclusions regarding the relationship between selective logging intensity and changes in local dung beetle species richness, species composition and biomass. We focus on richness and composition as they have been frequently used in previous studies on a range 43 of tropical taxa (Barlow et al., 2007; Bicknell et al., 2014b; Burivalova et al., 2014; Edwards et al., 2012b; Gibson et al., 2011; Imai et al., 2012; Socolar et al., 2015). We include biomass as this has been extensively used to assess the impacts of tropical forest disturbance on dung beetles (Nichols et al., 2013a; Scheffler, 2005; Slade et al., 2011). We compare SFT with BA by focusing on the difference in effect size (slope of regression) and proportion of explained variance (R2). Finally, we use a resampling procedure to examine which approach can assess the effects of selective logging with the smallest number of sample units, given that before-and-after approaches require at least two visits to a study region. 2.3 Methods 2.3.1 Site description Sampling was carried out in the Jari Florestal landholding, located at the State of Pará in the Northeastern Brazilian Amazon (0o27’S, 51o40’W; Chapter 1: Figure 1.3). The primary forests in the region are subject to low levels of disturbance from subsistence hunting and extraction of non-timber forest products (Parry, Barlow & Peres 2009). The climate is characterized by hot-humid (Köppen’s classification), with annual average temperature and precipitation of 26 ºC and 2115 mm respectively (Coutinho and Pires, 1996). Reduced-impact commercial logging started in 2003, with plans to log approximately 544,000 ha of native forest over a 30 year cutting cycle. This management is certified by the Forest Stewardship Council (FSC) and is one of the largest certified logging concessions in the Amazon with average annual production of 30,000 m3 of timber (FSC 2014). Logging activities are planned following FAO guidelines (Dykstra and Heinrich, 1996), which included a pre-harvest mapping and measuring of all commercially viable trees with DBH ≥ 45 cm. The harvesting and extraction of timber along skid trails generally take place during the dry season (August to November), and directional felling is used to minimise collateral damage to other trees. During 44 the pre-harvest inventory the logging concession is subdivided into 10 ha (250 x 400 m) planning units. Commercially viable trees are mapped across all of these planning units, and this forms the basis for planning the logging operation in the following year. 2.3.2 Spatial design We used the company’s pre-harvest inventory and operational logging plan to select 34 sample units situated along a gradient of planned logging intensity (see Chapter 1: Figure 1.3). These included five control sites that would not be logged during the course of the study, and 29 logging units which were destined to be logged between July and September 2012 (Appendix S2.2). As logging impacts are related to logging intensity (Burivalova et al., 2014; Picard et al., 2012), we aimed to assess logging impact as a continuous (rather than categorical) effect. We therefore selected logging units along the gradient of planned logging intensities, which resulted in gradient from 0 - 7.9 trees ha-1 (or 0 - 50.31 m3 ha-1) of timber that was eventually extracted (see Table S2.2). The five unlogged control units included in this range were the same size as the logged units, and held dung beetle communities representative of undisturbed primary forests in our study region (see Appendix S2.3; Figure S2.2). They were located approximately 6.5 km from the closest logging operations to ensure sampling independence and to avoid any spillover effects from the logging operation (Block et al., 2001). As such, they are representative of the distance between logged and undisturbed reference sites in many logging studies using space-for-time approaches (Table S2.1). We used the number of removed trees in each 10 ha sampled unit as our measure of logging intensity for all analyses, as a priori we assumed that the number of tree fall events and skidding trails would be the most important predictor of ecological impacts. Moreover, like others we found high co- linearity (n=34, r = 0.91, p<0.001) among number of trees and volume of 45 removed timber by selective logging (c.f. Picard, Gourlet-Fleury & Forni 2012). 2.3.3 Temporal design We carried out two dung beetle collections in all 34 sample units. The first collection gathered pre-logging data and occurred between June and July 2012, approximately 45 days before the logging operation began. The second collection took place in 2013, and gathered post-logging data approximately 10 months after logging activities ended. It also occurred in June and July, to minimize possible effects from seasonal variation. At all sites, dung beetles were sampled in exactly the same locations, and following the same methods, in both sample periods. Sampling locations were relocated based on marking tape, or by GPS when disturbance from logging activities meant this could not be found. 2.3.4 Sampling of dung beetles In both collection periods, dung beetles were sampled in each unit using six pitfall traps spaced 100 meters apart in a 2x3 rectangular grid, so that traps were at least 75 meters from the edge of the logging unit (Appendix S2.1, Figure S2.3). This spacing of traps helped insure independence between them (Silva & Hernández 2015) as well as an even spatial coverage of each logging unit. Pitfall traps were plastic containers (19 cm diameter and 11 cm deep) buried with their opening at ground level, containing approximately 250 ml of a saline solution. A plastic lid was placed above the top as a rain cover. A small plastic cup containing approximately 35 g of pig dung mixed with human dung (4:1 pig to human ratio, Marsh et al. 2013) was attached by a wire above each pitfall. Data from the six pit fall traps in each unit were pooled to get an aggregate value and improve representation. We restricted our sample window to 24 hours in each collection period, as short sample periods are known to be efficient at attracting a 46 representative sample of the local beetle community (Braga et al., 2013; Nichols et al., 2013b). Moreover, longer sample periods would have increased the probability of attracting dung beetles from outside the sample units (Silva & Hernández 2015), and therefore from units with different logging intensities. Finally, evidence from data collected in the same region suggests a 24 hour sampling period as a good predictor of community metrics from longer sampling durations (see Appendix S2.3; Figure S2.5). All dung beetles that fell in pitfall traps were dried and transported to the laboratory where they were identified to species, or morphospecies where this was not possible. We calculated the average biomass of each species from the dry weight of 15 individuals (when possible) using a Shimatzu AY220 balance with precision to 0.0001g. Voucher specimens were added to the Reference Collection of Neotropical Scarabaeinae in the Insect Ecology and Conservation Laboratory, Universidade Federal de Lavras, Brazil. 2.3.5 Data analyses We ran all analyses and statistical models in the R Software version 3.2.0 (R Core Team, 2015). We used generalized linear models (GLMs) to obtain the slope, R2 and p-value of the relationship between logging intensity and the dung beetle species richness, composition and biomass (Figure 2.1). All GLMs were submitted to residual inspection to evaluate the adequacy of error distribution (Crawley, 2002). We outline the two different sets of GLMs below. Before and after (BA): The pre-logging dung beetle community metrics were used as a temporal control/baseline to examine post-logging effects under the BA approach. Thus, we used Δ species richness, Δ species composition and Δ biomass as dependent variables. Δ was based on the difference between total species richness and biomass from post-logging minus pre-logging collection within each sampled unit. The Δ species composition was measured as the pairwise beta-diversity (Socolar et al., 2015) 47 based on the Bray-Curtis similarity index (1 – dissimilarity) among pre- and post-logging collections within each sample unit. Space for time (SFT): We only considered the post-logging biological metrics and analysed the slope and R2 using the actual values of species richness, species composition, and biomass. Species composition was estimated as the average Bray-Curtis similarity between each of the 29 logged units and the five control units. For control units, species composition was considered as the average similarity between each control plot and the other four control units. Species composition was calculated through the vegdist function (vegan package; Oksanen et al. 2015). Lastly, we tested whether our control sites represent typical undisturbed forest communities by comparing them with eleven primary forest sites sampled in the same year across the landscape (see Appendix S2.3). To determine which sample design (BA or SFT) provided a more precise evaluation of the relationship between the number of removed trees and biological metrics, we used a resampling procedure based on 1000 bootstrap samples with replacement in the boot.ci function from boot package (Canty & Ripley 2015; Davison and Hinkley, 1997). This function was also used to estimate frequency distributions, median precision and 95% confidence intervals of regression slopes and R2’s from the SFT and BA linear models (Figure 2.2). To compare the sampling efficiency of the SFT and BA approaches, we checked the minimum sampling effort and number of spatial replicates needed to get a significant regression (at p<0.05). We ran the models 10000 times per sample size with a randomized input variable between five and 34 samples, without replacement, to generate the mean p-values (±SD) for each sample size (Figure 2.3). As adjacent sites may be more similar and naturally hold more closely related biological communities (Kühn and Dormann, 2012; Soininen et al., 2007), we checked for spatial autocorrelation by performing Pearson-based Mantel tests (Legendre and Legendre, 1998) with 1000 permutations in the mantel function (vegan package; Oksanen et al. 2015). We repeated the 48 Mantel tests using both the pre- and post-logging dung beetle data, allowing us to examine whether spatial autocorrelation existed on both sets of analysis. We also repeated these including and removing the five control plots, to examine whether our controls were important in changing patterns. Finally, we plotted the residuals from the GLMs themselves on spatial maps of the sample sites, providing an intuitive visual assessment of the presence of spatial effects in the analysis (Baddeley et al., 2005; Kühn and Dormann, 2012) (see Appendix S2.4 for details of Mantel tests and residual plots). 2.4 Results Across our 34 sample units, we recorded 4846 dung beetles (pre- logging: 3720; post-logging: 1126) from 53 species (pre-logging: 49; post- logging: 40). Irrespective of where or when we sampled, undisturbed forests (i.e. the control sites pre-logging, the control sites post-logging, and the logging units pre-logging) held statistically similar numbers of dung beetle species (Appendix S2.3, Figure S2.2). 2.4.1 Before-and-after (BA) and Space-for-time (SFT) approaches Both BA and SFT approaches showed significant negative effects of increasing logging intensities on dung beetle species richness, although the BA approach was both more significant than SFT and explained more of the variance in the dung beetle richness (Figure 2.1A & D). The BA approach also had a higher proportion of variance explanation than SFT when considering species composition and biomass. Crucially, it was the only method to detect a significant effect of logging intensity at p<0.05 for these metrics (Figure 2.1B-C). The greater statistical power of the BA approach for detecting changes in the local species richness, species composition and biomass was clearly demonstrated using bootstrapping: BA had significantly higher R2 values than SFT (Figure 2.2A-C), and the bootstrapped regression slopes for 49 species richness, composition and biomass were significantly more negative for BA than SFT (Figure 2.2D-F). Our analysis of sampling efficiency showed that the BA approach reaches the critical p-value (0.05) with fewer spatial replicates than SFT, and this occurred for species richness (>15 versus >23, respectively), species composition (>20 versus >35, respectively) and biomass (>24 versus >34, respectively) (Figure 2.3A-C). Results were somewhat different for species richness and biomass when we considered the sampling effort instead of the number of sample replicates, as the BA approach requires both pre- and post- logging collections. Under this scenario, the BA approach required approximately 35 and 50 sample units to reach the p-value line for species richness and biomass, respectively, while STF required over 25 and 35 sample units to reach the same critical p-value (Figure 2.3D & F). For species composition, the BA approach still required less sample units than SFT (Figure 2.3E). The Mantel tests of distances among sampled units with corresponding dung beetle species richness and biomass showed a weak but significant degree of spatial autocorrelation in the pre-logging data (species richness r = 0.18, p = 0.005; biomass r = 0.12, p = 0.035). Importantly, this spatial autocorrelation disappeared in the post-logging collection (species richness r = -0.41, p = 0.999; biomass r = -0.42, p = 0.999), even when control units were excluded from analysis (see Appendix S2.4), and there was no discernible visual association between model residuals and geographical location (Appendix S2.4). 50 Figure 2.1 (A) Δ species richness; (B) Δ species composition; (C) Δ biomass; (D) post-logging species richness; (E) post-logging species composition and (F) post-logging biomass of dung beetles (n = 34) versus increased number of removed trees (n.10 ha-10) in the Amazon forest, Brazil. Black dots represent the 29 logging units with different selective-logging intensities and the five grey dots represent the five unlogged control units. The lines result from fitting the data to the generalized linear models with respective family distribution. 51 Figure 2.2 Accuracy comparison regarding the proportion of explanation (R2) and effect size (slope of fitted regression) from generalized linear models through the before-and-after (BA) and space-for-time (SFT) approaches. Proportion of explanation and effect size comparisons were made for dung beetle species richness (A & D), species composition (B & E) and biomass (C & F) facing the increased number of removed trees by selective logging on Amazon forest, Brazil. Bootstrapped confidence intervals (represented by vertical dashed lines) were created by resampling procedure based on 1000 bootstrap samples with replacement. On the boxplots, the notch area marks the 95% of confidence intervals for the medians (black horizontal lines). The grey and dashed horizontal line marks the zero line and outliers are shown in black dots. 52 Figure 2.3 Minumum sampling effort and number of spatial replicates for BA and SFT approaches reach the critical 0.05 p-value (represented by the black horizontal dashed line). Bootstrapped p-values from GLMs with dung beetle species richness, species composition and biomass facing the increased number of removed trees in the Amazon forest, Brazil. On thex-axis, spatial replicates indicates how many sampled units each approach needs to achieve the siginificance level for (A) species richness, (B) species composition and (C) biomass, whereas sample effort indicates the number of replicates dubling the sampled units for before- and-after approach, since this approach needs to go twice in each spatial replicate, for species richness (C), species composition (D) and biomass (E). GLMs were run 10,000 times, without replacement and with a randomized input variable between 5 and 34 samples. For SFT inference we considered just the post- logging data, while for the BACI approach we ran the models with Δ species richness, Δ species composition and Δ biomass as dependent variables. The grey bounds represent the 95% confidence intervals. 53 2. 5 Discussion Although both before-and-after and space-for-time approaches identified some disturbance effects on dung beetle communities, our comparison provide important evidence that BA approaches highlight more severe consequences of human disturbance on local (α) diversity (species richness) and β-diversity (compositional similarity; Socolar et al. 2015). In particular, BA approaches revealed more than double the number of species lost from the most disturbed plots, as well as significantly higher estimates of changes in dung beetle species composition and biomass. The significantly weaker effects revealed by the SFT approach are of great concern: SFT designs are the most commonly used method for assessing the biological consequences of selective logging on tropical invertebrates (Table S1), and underpin most assessments of biodiversity and ecosystem functioning losses caused by anthropogenic forest disturbances (e.g. Edwards et al. 2011, 2012b; a; Thomaz et al. 2012; Berenguer et al. 2014; Solar et al. 2015). Although our comparison is restricted to a single taxa and a single disturbance event, the magnitude in the scale of effects revealed by BA and SFT approaches for dung beetle α and β diversity suggests that the potential issues of SFT could apply to other anthropogenic disturbances (such as wildfire, hunting or land-use intensification) and other taxa. Furthermore, the robustness of our conclusions was supported by the Mantel test results and spatial residual plots (Appendix S2.4) showing that these patterns were driven by logging intensity and not by any spatial autocorrelation in the data. However, our post-logging collection was conducted about one year after logging operations, when logged sites are in their most disturbed state (West et al., 2014). It would be important to evaluate how BA and SFT studies compare when examining longer-term recovery post-disturbance. Likewise, although providing evidence that BA approach better detect changes in species diversity and composition at local scales (Chai et al., 2012; Kappes et al., 2010), further work is needed to examine how study designs alter effect 54 sizes based on gamma diversity, which often contribute to global or pan- tropical meta-analyses (e.g. Gibson et al. 2011; Newbold et al. 2015). With further investigation across sites and taxa it may be possible to develop a generalizable correction factor to be applied to biodiversity loss assessments relying on space-for-times designs (e.g. Cleary et al. 2009; Edwards et al. 2011, 2012b; a). Despite the advantages of BA studies, there are good reasons why they have not been used with more frequency (Kibler et al., 2011). Even when disturbance events have not yet occurred, it is often impossible to accurately predict where and when they will happen. This makes it particularly hard to apply BA designs to wildfires, illegal logging or land-use change. Moreover, even where activities are planned 2-3 years in advance, as in the case of licenced and certified selective logging, it is necessary to have an effective communication between researchers, decision makers and practitioners (companies, planners, and resource managers) in order for BA studies to take place. The fact that most assessments of the biological impact of selective logging rely on SFT approaches (Appendix S2.1) shows the difficulty of developing these relationships within the time frame of research projects. Our results therefore support calls to close this ‘knowledge-doing’ gap that exists throughout conservation science (Boreux et al., 2009; Habel et al., 2013), and show how effective communication and partnerships between researchers and the private sector could be used to support effective conservation practice (Wu and Hobbs, 2002). These partnerships need to start long before research is undertaken, both to optimise the experimental design and integrate or overcome concerns from researchers and stakeholders. We also highlight an important logistical constraint of BA, in that it needs double the sampling effort (Figure 2.3) compared to just once in the SFT approach. Achieving the pre- and post-disturbance samples inevitably increases both the time and costs required to collect data, but the additional time may be an equally important limiting factor: most research projects, including postgraduate studies, are a maximum of 3-5 years in duration, which 55 limits the data collection phase of projects to just 1-2 years. It is clearly difficult for students and researchers to undertake BA studies in relatively short-term research projects or doctoral theses, which rarely allow time enough for two or more field seasons. This can be resolved by longer-term research partnerships that transcend individual studies. Finally, if the biological baseline as a whole has been shifted by widespread disturbance, then before-and-after approaches themselves risk underestimating biodiversity loss. We were fortunate that Jari landholding has relatively undisturbed primary forests prior to logging operations (Parry, Barlow & Peres 2009). This allowed us to sample both pre- and post-logging, and verify the intactness of our pre-logging controls by comparing them with other sites in undisturbed primary forests (Appendix S2.3). However, where forests have been affected by widespread anthropogenic activities (e.g. fires or hunting), the biota present in the before survey will have been filtered by previous disturbances and will not contain the most disturbance-sensitive species. In these case, before-and-after comparisons risk underestimating biodiversity loss, and need to be interpreted accordingly (Baum and Myers, 2004; Gardner et al., 2009; Kibler et al., 2011). 2.6 Conclusions Our study has broad implications for applied ecology and conservation science, as we show that the most frequently used experimental design may lead us to underestimate the consequences of land-use change and forest disturbances on local species diversity and their turnover. While before- and-after approaches are accompanied by many logistical constraints (e.g. they require a longer time and more sample effort), we believe they should be strongly encouraged in order to re-evaluate human impacts on biodiversity. Finally, although our main aim was to compare methodological approaches, our results also have some important implications for reduced-impact logging which is being planned in timber concessions across 400 Mha of tropical forest 56 (Blaser et al., 2011), as they demonstrate high rates of community turnover as well as sharp losses in species diversity and dung beetle biomass, particularly at high logging intensities (c.f. Burivalova, Şekercioğlu & Koh 2014). This emphasizes the need for careful planning and further research before forest management can be termed sustainable for biodiversity conservation. Acknowledgements: The authors are grateful to Jari Forestal for their logistical support. We thank our field assistants, as well as Lisiane Zanella for helping in Fig. S1; Teotônio de Carvalho and Fábio Frazão for statistical advice; and Ricardo Solar for building the R function for bootstrapping p- values. The manuscript benefited from comments by Luiz Fernando Magnago, Carla Ribas and Alistair Campbell, also from the editors and two anonymous reviewers. Research was supported by grants from the CNPq-PELD site 23, therefore by FAPEMIG, CNPq and CAPES. F.F. was awarded by a CAPES studentship (BEX 5528/13-5). J.B. was supported by CNPq (400640/2012-0). J.M.S was funded by a post-doctoral FAPEMIG fellowship. 57 2.7 Supplementary Information Appendix S2.1 Literature review Using ISI Web of Knowledge (http://webofknowledge.com), we searched for publications on the impacts of selective logging on the tropical invertebrate biodiversity. We performed this search in December 2014 by using the terms “selective logging” AND “tropical forest” AND “invertebrate” AND “insects”. We did not specify where these terms should occur in the articles or the publication date of the articles. Web of Knowledge returned 31 publications, which were then examined and filtered to ensure we considered all studies reporting the impact of selective logging on tropical invertebrate fauna. We assessed the pertinence of each article according to its tittle and abstract: all field-based studies related to the impact of selective logging on biodiversity of invertebrate fauna were retained. Modeling studies and reviews were excluded. In total, 53 articles were considered for final assessment. Table S2.1 List of studies examining the impacts of selective logging on tropical invertebrates. Table shows the authors and year of publication, the assessed taxa, published journal, evaluated experimental design (ED) and distance between control and logging units in kilometers (DC). aSFT = space-for-time; bBA = before-and-after approach; cNA = No information, dAdj = adjacent. Study Taxa Journal ED DC Aguilar-Amuchastegui & Henebry (2007) Dung beetles Forest Ecology and Management SFT a ≤10 Akutsu, Khen & Toda (2007) Different groups Ecology Research SFT NA c Andersen et al. (2009) Ant Forest Ecology and Management SFT ≤2 Basset et al. (2001) Herbivore insects J. of Applied Ecology BA b ≤2 Berry et al. (2010) Ants Biodiversity and Conservation SFT ≤5 Bicknell et al. (2014) Dung beetles Ecological Indicators SFT ≤2 Burghouts et al. (1992) Litter invertebrates Phil. Trans. Royal Soc. of London, Series B SFT NA Cartar (2005) Bumble bees Biodiversity and Conservation BA ≤5 Chung et al. (2000) Beetles Bulletin of Entomological Res. SFT 80 58 Study Taxa Journal ED DC Cleary (2003) Butterflies Oecologia SFT ≤5 Cleary (2004) Butterflies J. of Economic Entomology SFT ≤2 Cleary & Mooers (2006) Butterflies Diversity and Distributions SFT ≤5 Cleary et al. (2005) Butterflies J. of Applied Entomology SFT NA Cleary et al. (2009) Butterflies Basic and Applied Ecology SFT NA Davis (2000) Dung beetles Environmental Entomology SFT >1 Davis et al. (2001) Dung beetles J. of Applied Ecology SFT >1 Davis & Sutton (1998) Dung beetles Diversity and Distributions SFT >1 Dumbrell & Hill (2005) Butterflies Biological Conservation SFT Adj d Dumbrell et al. (2008) Butterflies J. of Applied Ecology SFT Adj Edwards et al. (2010) Dung beetles Proc. Royal Soc. of London B Biol. Science SFT 1-90 Edwards et al. (2012b) Dung beetles and ants Ecological Applications SFT 1-90 Edwards et al. (2012a) Understory invertebrates J. of Insect Conservation SFT Adj Edwards et al. (2014) Dung beetles and ants Global Change Biology SFT 1-90 Eltz (2004) Bees J. of Tropical Ecology SFT >50 Eltz et al. (2002) Bees Oecologia SFT >50 Eltz et al. (2003) Bees Forest Ecology and Management SFT >50 Forkner et al. (2006) Leaf-chewing herbivore insects Conservation Biology BA NA Ghazoul & Ghazoul (2002) Butterflies Biodiversity and Conservation SFT ≤10 Gunawardene, Majer & Edirisinghe (2010) Ant Forest Ecology and Management SFT >1 Hamer et al. (2003) Butterflies J. of Applied Ecology SFT ≤10 Hill (1999) Butterflies J. of Applied Ecology SFT Adj Hill et al. (1995) Butterflies J. of Applied Ecology SFT >1 Hill et al. (2003) Butterflies J. of Applied Ecology SFT NA Jones & Prasetyo (2002) Termites The Raffles Bulletin of Zoology SFT NA Jones et al. (2003) Termites J. of Applied Ecology SFT <5 Kreutzweiser et al (2005) Macroinvertebrates J. North Am. Benth. Society BA <1 Lewis (2001) Butterflies Conservation Biology SFT Adj Lima et al. (2000) Termites Forest Ecology and Management SFT Adj Negrete-Yankelevich et al. (2007) Soil macroinvertebrates Applied Soil Ecology SFT ≤2 Nummelin & Zilihona (2004) Arthropod fauna Forest Ecology and Management SFT NA Ol