UNIVERSIDADE DE LISBOA FACULDADE DE CIÊNCIAS DEPARTAMENTO DE BIOLOGIA ANIMAL

Modelling the distribution of São Tomé bird species: Ecological determinants and conservation prioritization

Filipa Macedo Coutinho de Oliveira Soares

Mestrado em Biologia da Conservação

Dissertação orientada por: Doutor Ricardo Faustino de Lima Professor Doutor Jorge Palmeirim

2017

AGRADECIMENTOS

Quero começar por agradecer aos meus orientadores por todo o apoio incansável ao longo deste ano. Este trabalho não seria possível sem todos os “brainstormings” durante as extensas reuniões ao longo de várias semanas. Obrigada por me terem sempre incentivado a dar o meu melhor. Ricardo quero agradecer-te toda a ajuda, logo desde o início quando esta tese era nada mais do que uma pequena ideia. Não poderia ter pedido mais ou melhor orientação, obrigada pela tua infinita disponibilidade (eu sei o quão “chata” eu consigo ser!). O meu obrigado também ao Professor Jorge Palmeirim, a sua ajuda foi indispensável. Este trabalho não seria possível sem a incrível ajuda de ambos, o meu mais sincero obrigado! Este trabalho não teria sido possível sem os dados recolhidos no âmbito da tese de doutoramento “Landuse management and the conservation of endemic species in the island of São Tomé” de Ricardo Faustino de Lima, e da “BirdLife International São Tomé and Príncipe Initiative”. A tese de doutoramento foi financiada pela FCT - Fundação para a Ciência e Tecnologia, através de uma bolsa de doutoramento cedida pelo Governo Português (Ref.: SFRH/BD/36812/2007), e pela “Rufford Small Grant for Nature Conservation”, que forneceu financiamento adicional para o trabalho de campo (“The impact of changing agricultural practices on the endemic birds of Sao Tome” - Ref.: 50.04.09). A “BirdLife International São Tomé and Príncipe Initiative” foi financiada pela “BirdLife’s Preventing Extinctions Programme”, através da família Prentice no âmbito da “BirdLife’s Species Champion Programme”, pela “Royal Society for the Protection of Birds”, pela “Disney Worldwide Conservation Fund”, pela “U.S. Fish and Wildlife Service Critically Endangered Animals Conservation Fund” (AFR1411 - F14AP00529), pela “Mohammed bin Zayed Species Conservation Fund” (Project number 13256311) e pela “Waterbird Society Kushlan Research Grant”. Quero ainda agradecer a toda a equipa de trabalho de campo da Associação Monte Pico que esteve envolvida na recolha de dados, nomeadamente Gabriel Cabinda, Ricardo Fonseca, Gabriel Oquiongo, Joel Oquiongo, Sedney Samba, Aristides Santana, Estevão Soares, Nelson Solé e Leonel Viegas. Este trabalho não teria sido possível sem a coordenação do Hugo Sampaio, da Sociedade Portuguesa para o Estudo das Aves (SPEA), ou sem o apoio institucional e empenho pessoal de Luís Costa (SPEA) e de Alice Ward-Francis (“Royal Society for the Protection of Birds” - RSPB), a quem agradecemos igualmente a disponibilização de dados. Finalmente, um agradecimento especial a Graeme M. Buchanan, pelas orientações e pelo apoio no planeamento experimental deste trabalho. Agradeço também à Associação Monte Pico, pelo alojamento durante a minha estadia em São Tomé. Gostaria também de agradecer a todos os que contribuíram para o “Plano de acção internacional para a conservação das espécies de aves Criticamente em Perigo de São Tomé”, especialmente à Direção Geral do Ambiente, ao Parque Natural do Obô de São Tomé, à Direção das Florestas, à Associação dos Biólogos Santomenses e à associação MARAPA. Queria ainda agradecer em especial ao Eng. Arlindo Carvalho, Diretor Geral do Ambiente por apoiar as nossas atividades em São Tomé. O trabalho de campo não teria sido possível sem a ajuda de Silvino Dias, José Malé, Filipe Santiago, Lidiney e inúmeros outros santomenses. Uma dedicação especial para "Dakubala". Agradecemos a António Alberto, Nuno Barros, Mariana Carvalho, Martin Dallimer, Hugulay Maia, Stuart Marsden, Martim Melo, Fábio Olmos e Longtong Turshak por partilharem todas as suas observações. Quero agradecer a Teotónio Soares pela disponibilidade e ajuda na construção dos loops para o script dos modelos lineares generalizados. II

Não posso deixar de agradecer a todas as pessoas que conheci em São Tomé. Obrigada Nity e Estevão por terem sido os melhores ajudantes de campo. Aos dois, obrigada por terem respondido às minhas perguntas, por terem sempre confiado em mim atrás do volante do nosso táxi (nem eu confiaria!), por terem esperado sempre por mim em todas as nossas escaladas intermináveis. Obrigada por me terem dado a conhecer todas as paisagens incríveis de São Tomé. Obrigada Lucy por nos teres recebido em tua casa, por nos tratares praticamente como filhas quando não era tua obrigação, por teres sido para mim a minha família longe de casa. Nunca conseguirei agradecer-te o suficiente tudo o que fizeste. Obrigada Gégé por todas as conversas, por todos os risos e gargalhadas, por todos os cafés e bolachas, por todas as caminhadas e passeios pela cidade. Obrigada por teres sido um grande amigo quando eu mais precisava. Obrigada Adilécio por toda a ajuda com o carro, por vires sempre ao meu auxílio, ou porque o carro não andava, ou porque andava pouco, ou porque a mala não fechava (acho que praticamente tudo aconteceu àquele carro!). Obrigado Octávio por nos teres recebido em tua casa, ainda hoje consigo lembrar-me dos teus famosos cozinhados. Obrigado Filipe e Fica por me terem recebido de braços abertos e terem sempre mil e uma histórias para contar. Obrigada Mito e Sá também por me terem acolhido, por me mostrarem Emolve e por todos os jantares à luz das velas cheios de gargalhadas e boa disposição. Obrigada Juary, Gabi, Leonel, Catoninho, Lito, Lau, e todos os outros que me ajudaram e tornaram a minha estadia em São Tomé uma das melhores experiências que até hoje vivi. Quero agradecer aos meus pais, à minha irmã Rita e, também, às minhas duas avós por o apoio e companhia ao longo deste ano (particularmente difícil!). Também, quero agradecer ao Afonso por ter estado sempre lá, por ter aturado todas as minhas longas conversas sobre “bichos” (mesmo quando já não conseguia ouvir mais!). Obrigada por seres quem és e por acreditares sempre em mim, mesmo quando já nem eu acredito. Obrigada a todos os meus companheiros e amigos pertencentes à “team cócós”. Obrigada Rita (e Zeus, o melhor cão do mundo!), Manel, Catarina Vegy, Cátia, Catarina Vet, Marvel por toda as aventuras ao longo deste ano (e que aventuras…desde atolar carros a perseguir assassinos em série!). Em especial, quero agradecer ao Professor Francisco Petrucci-Fonseca, protagonista de grande parte das nossas aventuras, por me ter dado a oportunidade de conhecer o que são talvez as serras mais bonitas de Portugal! Obrigada a todos os meus amigos e colegas que me ajudaram e apoiaram ao longo deste ano. Em especial, um grande obrigado à Martina e à Bárbara por toda a companhia durante este longo ano e, principalmente, durante a nossa aventura de dois meses em São Tomé. Foi difícil mas não a trocava por nada, ou escolhia outras pessoas para irem comigo!

III

RESUMO ALARGADO

O Homem tem vindo a alterar a ecologia do planeta, influenciando a distribuição das espécies e o funcionamento dos ecossistemas. A comunidade científica tem dedicado muita atenção ao estudo do impacto das atividades humanas na biodiversidade, uma vez que estas são largamente tidas como responsáveis pela atual crise da perda de biodiversidade. Apesar da dificuldade em determinar com exatidão os processos envolvidos, sabe-se que o aumento da população humana tem tido diversos impactos negativos sobre os ecossistemas naturais. Há então necessidade de definir prioridades globais de conservação, começando pela identificação das principais ameaças, como a alteração antropogénica dos usos do solo. As florestas estão entre os ecossistemas terrestres mais ricos e também mais ameaçados, sendo que nas últimas décadas a pressão humana tem vindo a aumentar sobretudo nas florestas tropicais, estando muitas das suas espécies entre as mais ameaçadas do mundo. A ocupação pelo Homem é sinónimo de fortes alterações na paisagem, tanto nos continentes como em ilhas. No entanto, as ilhas tendem a possuir ecossistemas mais sensíveis, ricos em espécies endémicas, que são particularmente vulneráveis à extinção. Posto isto, assumem uma elevada importância na preservação da biodiversidade, principalmente dada a taxa de alteração do uso do solo ser mais elevada nas ilhas do que nos continentes. São Tomé é uma pequena ilha oceânica situada no Golfo da Guiné, África Central, a cerca de 255 km do continente. De origem vulcânica, possui uma topografia acidentada constituída por encostas de declive acentuado e vales encaixados, com rios pontuados por grandes cascatas. Nas zonas costeiras ocorrem estuários e mangais. Esta topografia explica o gradiente climático, caracterizado por elevados níveis de humidade e chuvas frequentes trazidas pelos ventos fortes do sudoeste da ilha, que contrastam com o nordeste semiárido. O forte gradiente climático tem vindo a moldar a distribuição dos ecossistemas da ilha, mas a paisagem originalmente dominada por floresta tem sofrido alterações desde a colonização humana, que teve início no final do século XV pelos Portugueses. As zonas planas de baixa altitude são as mais intervencionadas, sendo constituídas maioritariamente por áreas não florestadas, tais como savanas e áreas cultivadas. As florestas de baixa altitude foram substituídas por plantações de sombra com árvores exóticas, como cafeeiro, cacaueiro e palmeiras. A floresta nativa mais bem preservada está hoje restrita às áreas montanhosas no centro e sudoeste da ilha, rodeada por floresta secundária, que resultou sobretudo da regeneração de plantações de sombra abandonadas. Apesar da paisagem humanizada, São Tomé mantem uma flora e fauna muito diversas com um número muito elevado de endemismos. As suas florestas têm um enorme interesse para a conservação, tendo sido identificadas como as terceiras mais importantes no mundo para a conservação de espécies de aves florestais. Esta tese está dividida em dois capítulos com objetivos distintos, ambos relacionados com a diversidade das aves de São Tomé. No primeiro capítulo, o objetivo principal é compreender como se distribuem as aves ao longo da ilha, tendo como objetivos específicos: (1) identificar os principais determinantes da distribuição das espécies de aves de São Tomé; (2) compreender como se relaciona o endemismo com as respostas das espécies às variáveis ambientais; (3) analisar a relação entre as guildes tróficas e a resposta das espécies às variáveis ambientais. No final, explorámos a relação entre as respostas das espécies e os fatores determinantes da sua distribuição, dando um foco especial às espécies endémicas e ameaçadas. Neste estudo foram realizados pontos de contagem de aves com duração de 10 minutos, onde foram registadas todas as espécies de aves. O período de amostragem foi de Janeiro a Março de 2017, tendo sido a amostragem direcionada para as zonas não florestadas e de plantação de sombra, bem como algumas zonas de floresta secundária. Estas observações foram agrupadas com observações ocasionais e sistemáticas de estudos anteriores, que se tinham focado sobretudo nas áreas IV

florestais, atingindo um total de 3056 pontos amostrados em toda a ilha, onde foram registadas de forma inequívoca 34 espécies de aves terrestres. Algumas variáveis ambientais, tais como o tipo de uso do solo, a topografia, a precipitação, o declive, a altitude, a acessibilidade e a distância à costa, foram mapeadas e utilizadas na construção dos modelos lineares generalizados para cada espécie. A ordenação dos melhores modelos de distribuição potencial de cada espécie permitiu explorar a resposta de cada espécie às variáveis ambientais. Uma análise de correspondência detrended foi realizada para avaliar a relação entre endemismo, guildes tróficas e variáveis ambientais. O tipo de uso do solo foi identificado como a variável mais importante para explicar a presença das espécies: as espécies endémicas tendem a ocorrer preferencialmente na floresta, em zonas mais remotas, de elevada altitude e precipitação, por sua vez as não endémicas preferem zonas não florestadas e mais humanizadas. A paisagem altamente florestada de São Tomé permite, de uma forma geral, que haja uma dominância das espécies endémicas na ilha. Muitas espécies endémicas estão ameaçadas, o que salienta a necessidade de proteger os habitats florestais. Como tal, propomos um incremento da matriz florestal na paisagem, através da proteção da floresta nativa remanescente e da expansão da floresta secundária, para a conservação das aves de São Tomé. No segundo capítulo, o objetivo principal é avaliar se o Parque Natural do Obô (PNO) inclui uma representação adequada da diversidade de aves da ilha. Como tal, foi modelada a riqueza específica e a composição das aves, dando especial atenção à distribuição de espécies endémicas e não endémicas. A distribuição da diversidade de aves foi comparada com os limites da área protegida. Foi construída uma base de dados com os pontos de contagem de aves de estudos anteriores, que foi complementada por pontos adicionais realizados entre Janeiro e Março de 2017. Os pontos de contagem pertencentes à mesma quadrícula de 1x1 km foram agrupados, criando conjuntos de cinco pontos de contagem por quadrícula num total de 187 quadrículas, onde 36 espécies de aves terrestres foram registadas. Foram utilizadas seis variáveis ambientais, tendo sido excluídas a rugosidade e a acessibilidade, para modelar e mapear a riqueza específica total, das espécies endémicas e não endémicas, bem como a composição da comunidade. Os resultados mostram que o número de espécies endémicas diminui nos habitats mais humanizados, onde aumenta o de espécies não endémicas. O PNO não está a proteger as comunidades mais ricas em aves, mas aquelas que têm mais aves endémicas, que ocorrem nas florestas mais bem preservadas. Definidos com base na distribuição dos habitats e da população humana, os limites do parque permitem a proteção das espécies endémicas ameaçadas, indiscutivelmente as de maior interesse conservacionista. As florestas secundárias atuam como zona de transição para as zonas mais humanizadas, protegendo as espécies endémicas das diversas ameaças antropogénicas. Deve ser realizada uma revisão do zonamento do parque, de modo a integrar o atual conhecimento da distribuição das espécies. Este estudo permitiu aumentar o conhecimento atual sobre a distribuição das aves de São Tomé, salientando a importância do tipo de uso do solo para a ocorrência das espécies e dando, pela primeira vez, uma perspetiva sobre a distribuição da riqueza e da composição das comunidades de aves ao longo da ilha. Esta informação deve ser utilizada na definição de estratégias de conservação e monitorização. No entanto, é necessário aprofundar o conhecimento sobre a distribuição de cada espécie, ao longo do ano e a escalas espaciais mais pormenorizadas, por forma a compreender melhor a resposta de cada espécie à degradação florestal. Destacamos ainda a importância de quantificar o impacto de outras ameaças, como a caça e as espécies introduzidas. Toda esta informação irá permitir definir ações prioritárias de conservação para espécies-alvo, adequadas às necessidades ecológicas de cada espécie, o que é especialmente importante no caso das espécies mais ameaçadas como a galinhola Bostrychia bocagei, o picanço Lanius newtoni e o anjoló Neospiza concolor. Palavras-chave: endemismo; guilde trófica; Parque Natural do Obô; espécies ameaçadas; riqueza específica V

ABSTRACT

Human actions are rapidly changing ecosystems all over the world. Anthropogenic land use change affects the structure and functioning of ecosystems, leading to irreversible biodiversity losses. Understanding how human actions influence biodiversity is therefore key to prevent further biodiversity loss. Tropical forests are among the most diverse and threatened ecosystems, and the increasing human pressure, high number of threatened species and major habitat loss calls for conservation actions. São Tomé is a small oceanic island located in the Gulf of Guinea, Central Africa. Despite the human-dominated landscape, this island maintains a high biodiversity, rich in endemic species, and its forests are of great conservation value. This study has the main goals of: Understanding how bird species are distributed throughout the island. Occasional and systematic observations were gathered from previous studies and complemented by additional 10minute point counts. A total of 3056 sampling locations were used to understand the distribution of 34 terrestrial bird species. Species-specific generalized linear models and detrended correspondence analysis based on presence-absence, were used to explore the links between endemism, feeding guilds and environmental variables. Land use was the most important variable to explain bird species occurrence. The endemics tended to prefer forests located in remote, wetter areas, on rugged terrain and at higher altitudes, while the non-endemics favoured the drier flat lowlands, in more accessible locations and devoid of forest. The change in bird species assemblage from forest endemics to open habitat nonendemic granivores is clearly a result of the land use intensification gradient. The current overall dominance of endemic species across the island is maintained by São Tomé’s forest-dominated landscape. The dependency of endemics on forest highlights the urgent need for their protection. Based on these results, we suggest that protecting remaining native forests and expanding secondary forests will improve landscape matrix and contribute to the survival of the endemic-rich island avifauna worldwide. Assessing how the São Tomé Obô Natural Park (STONP) represents the avifauna of the island. The boundaries of the STONP were defined in 2006, based on the distribution of native ecosystems and of the human population. We compared them to the distribution of bird diversity, by modelling species richness and composition. We used systematic observations from previous studies supplemented by additional bird counts. A total of 187 1x1 km quadrats were sampled by five 10-minute point counts each. Thirty-six terrestrial bird species were identified unambiguously and considered for analyses. The proportion of endemic species decreased along the land use intensification gradient. The STONP did not protect the most species-rich bird assemblages, but included those that were richest in endemics, the best-preserved forests. Thus, the STONP is focusing on the protection of endemic threatened birds, which arguably have the highest global conservation interest. The secondary forests surrounding the park act as a transition zone to areas with more intensive land use types, hence protecting it from pervasive threats. We suggest the zonation of STONP is revised, using the same factors considered for the delimitation of the protected area and the current knowledge on species distribution. This study reveals that protecting well-preserved natural areas with low human density might be a good proxy to identify areas of high conservation interest, when there is little information on the distribution of the multiple components of biodiversity.

Keyword: community ecology; species distribution modelling; endemism; feeding guild; threatened species VI

TABLE OF CONTENTS

GENERAL INTRODUCTION ............................................................................................................... 1 CHAPTER 1: The role of natural gradients and ecosystem humanization in determining the distribution of bird species in São Tomé ................................................................................................. 5 INTRODUCTION ............................................................................................................................... 5 METHODS.......................................................................................................................................... 6 Study Area ....................................................................................................................................... 6 Data Compilation ............................................................................................................................ 7 Field Methods .................................................................................................................................. 7 Sampling design .............................................................................................................................. 7 Bird sampling .................................................................................................................................. 7 Characterizing environmental variables ......................................................................................... 8 Data Analysis .................................................................................................................................. 9 Exploratory analysis........................................................................................................................ 9 Generalized linear models............................................................................................................... 9 Relative variable importance ........................................................................................................ 10 Response to environmental variables ............................................................................................ 10 RESULTS.......................................................................................................................................... 10 Relative variable importance ......................................................................................................... 10 Response of endemic and non-endemic species to environmental variables ................................ 11 Feeding guilds response to environmental variables ..................................................................... 14 Species land use type preferences ................................................................................................. 16 DISCUSSION ................................................................................................................................... 17 Determinants of bird species distribution ...................................................................................... 17 Differential response of endemic and non-endemic bird species .................................................. 17 Differential response of bird species based on feeding guilds ...................................................... 18 Consequences of land use intensification to the endemic-rich avifauna of São Tomé ................. 18 CHAPTER 2: Is the existing protected network adequate for the conservation of the endemic-rich avifauna of São Tomé Island? ............................................................................................................... 20 INTRODUCTION ............................................................................................................................. 20 METHODS........................................................................................................................................ 21 Study Area ..................................................................................................................................... 21 Data Compilation .......................................................................................................................... 22 Field Methods ................................................................................................................................ 22

VII

Sampling design ............................................................................................................................ 22 Bird sampling ................................................................................................................................ 22 Characterizing environmental variables ....................................................................................... 23 Data Analysis ................................................................................................................................ 24 Exploratory analysis...................................................................................................................... 24 Generalized linear models............................................................................................................. 24 Generalized dissimilarity modelling.............................................................................................. 25 Generalized dissimilarity model categorization ............................................................................ 25 Assessing the adequacy of the Obô Natural Park to represent São Tomé bird diversity .............. 26 RESULTS.......................................................................................................................................... 26 Modelling bird species richness .................................................................................................... 26 Bird species compositional dissimilarity ....................................................................................... 28 Is the São Tomé Obô Natural Park adequate to protect the island’s avifauna? ............................. 29 DISCUSSION ................................................................................................................................... 31 Contrasting responses of endemic and non-endemic species to the environment ......................... 31 Species assemblages vary mostly in response to habitat humanization ........................................ 32 Is the São Tomé Obô Natural Park adequate to protect the island’s bird diversity? ..................... 33 Final remarks ................................................................................................................................. 33 FINAL CONSIDERATIONS................................................................................................................ 35 REFERENCES ...................................................................................................................................... 36 SUPPLEMENTARY MATERIALS ..................................................................................................... 45 SECTION I: Environmental Variables.............................................................................................. 45 SECTION II: São Tomé Bird Species ............................................................................................... 60 SECTION III: Binomial Generalized Linear Models ........................................................................ 61 SECTION IV: Proportion of species occurrence per land use type .................................................. 67 SECTION V: Exploratory analysis for species richness and composition modelling....................... 68 SECTION VI: Poisson Generalized Linear Models .......................................................................... 70 SECTION VII: Generalized Dissimilarity Modelling ....................................................................... 73 SECTION VIII: R scripts .................................................................................................................. 77

VIII

LIST OF TABLES Table 1.1. Response of endemic (E) and non-endemic (N), and of distinct feeding guilds (omnivores - O, granivores - G, frugivores – F, and carnivores – C) to environmental variables………………………………………………12 Table 2.1. Species richness and endemic species richness estimated for each average point inside 1x1 quadrats, called, respectively, species and endemic richness point estimate……………………………………………….30 Table S1. (Section I – Supp. Materials) Environmental variables description…………………………………….45 Table S2. (Section I – Supp. Materials) Environmental raster’s characteristics…………………………………...46 Table S3. (Section II – Supp. Materials) Bird species’ characteristics……………………………………………60 Table S4. (Section III – Supp. Materials) Validation of the best multivariable model……………………………..61 Table S5. (Section III – Supp. Materials) Relative variable importance (RVI)………………………………...…62 Table S6. (Section III – Supp. Materials) Single-variable model coefficients. …………………………………...63 Table S7. (Section III – Supp. Materials) Kruskal-Wallis rank test to analyse the difference in environmental variables between endemic and non-endemic species, as well as among feeding guilds. ………………………….64 Table S8. (Section IV – Supp. Materials) Proportion of species occurrence per land use type and topography class………………………………………………………………………………………………………......67 Table S9. (Section V – Supp. Materials) Bird species’ characteristics…………………………………………...68 Table S10. (Section VI – Supp. Materials) Validation of the best model…………………………………………70 Table S11. (Section VI – Supp. Materials) Species richness and environmental variables………………………...72 Table S12. (Section VII – Supp. Materials) Significance test of GDM model……………………………………73 Table S13. (Section VII – Supp. Materials) Significance test for each variable in GDM model…………………....74 Table S14. (Section VII – Supp. Materials) Importance of each predictor variable……………………………….75

LIST OF FIGURES Figure 1.1. Location of sampling point counts and occasional observations (n = 3056) in São Tomé Island………...8 Figure 1.2. Relative variable importance (RVI) of each environmental variable for each bird species generalized linear model…………………………………………………………………………………………………..11 Figure 1.3. Response of endemic (E) and non-endemic (N) species to environmental variables…………………13 Figure 1.4. Detrended Correspondence Analysis (DCA) showing the relationship between endemism, feeding guilds and environmental variables…………………………………………………………………………………...14 Figure 1.5. Feeding guild (omnivores - O, granivores - G, frugivores – F, and carnivores - C) response to environmental variables……………………………………………………………………………………….15 Figure 1.6. Proportion of occurrence of each species by land use types…………………………………………..16 Figure 2.1. São Tomé Island sampling locations………………………………………………………………..23 Figure 2.2. Predictive maps of (a) total species richness, (b) endemic species richness and (c) non-endemic species richness, shown in contrast to the boundaries of the Obô Natural Park and buffer zone……………………………27 Figure 2.3. (a) Continuous and (b) categorical composition dissimilarity maps, as obtained from generalized dissimilarity modelling (GDM)………………………………………………………………………………..28 Figure 2.4. Total, endemic and non-endemic species richness inside (In) and outside (Out) Obô Natural Park…….29 Figure 2.5. Proportion of endemic species and frequency of endemic species for each GDM class (1 to 5)………...30 Figure S1. (Section I – Supp. Materials) Altitude in meters……………………………………………………...47 Figure S2. (Section I – Supp. Materials) Ruggedness…………………………………………………………...48 Figure S3. (Section I – Supp. Materials) Slope in degrees……………………………………………………….49 Figure S4. (Section I – Supp. Materials) Distance to coast line in degrees………………………………………..50 Figure S5. (Section I – Supp. Materials) Separation of flat plain areas and middle slope areas…………………….51 Figure S6. (Section I – Supp. Materials) Transforming continuous Topographic Position Index in a categorical variable……………………………………………………………………………………………………….52 Figure S7. (Section I – Supp. Materials) Topography Position Index……………………………………………53 IX

Figure S8. (Section I – Supp. Materials) Building remoteness index…………………………………………….54 Figure S9. (Section I – Supp. Materials) Remoteness Index…………………………………………………….55 Figure S10. (Section I – Supp. Materials) Rainfall in millimetres………………………………………………..56 Figure S11. (Section I – Supp. Materials) Land use map created by S. Mikulane (resolution of 10x10 meters)…….57 Figure S12. (Section I – Supp. Materials) Land use……………………………………………………………..58 Figure S13. (Section I – Supp. Materials) Correlogram between environmental variables .....................................................................................................................................................................................................59 Figure S14. (Section III – Supp. Materials) Relative variable importance (RVI) of each continuous environmental variable……………………………………………………………………………………………………….65 Figure S15. (Section III – Supp. Materials) Relative variable importance (RVI) of each continuous environmental variable in endemic and non-endemic species…………………………………………………………………..65 Figure S16. (Section III – Supp. Materials) Relative variable importance (RVI) of each continuous environmental variable in every feeding guild species group…………………………………………………………………...66 Figure S17. (Section V – Supp. Materials) Correlogram between environmental variables and response variables.....................................................................................................................................................................................69 Figure S18. (Section VI – Supp. Materials) Pearson and Deviance Residuals……………………………………71 Figure S19. (Section VII – Supp. Materials) Overall model fit in explaining the observed dissimilarities………….73 Figure S20. (Section VII – Supp. Materials) K-fold cross-validation of GDM…………………………………...74 Figure S21. (Section VII – Supp. Materials) Response curves of each predictor variable…………………………76

LIST OF ABBREVIATIONS AND ACRONYMS E N O G F C NF SF SP NFA F V M, Middle U, Upper R Amaboc Ananew Bosboc Colmal Coltho Dretho Lannew Neocon Oricra Otuhar Plogra

Endemics Non-endemics Omnivores Granivores Frugivores Carnivores Native forest Secondary forest Shade plantation Non-forested areas Flat areas Valleys and deep valleys Middle slope areas Upper slope areas Ridges Amaurocichla bocagei, São Tomé Short-tail Anabathmis newtonii, São Tomé Sunbird Bostrychia bocagei, Dwarf Ibis Columba malherbii, São Tomé Bronze-napped Pigeon Columba thomensis, São Tomé Maroon Pigeon Dreptes thomensis, Giant Sunbird Lanius newtoni, São Tomé Fiscal Neospiza concolor, São Tomé Grosbeak Oriolus crassirostris, São Tomé Oriole Otus hartlaubi, São Tomé Scops Owl Ploceus grandis, Giant Weaver

X

Plosan Primol Serruf Teratr Tresan Turoli Zosfea Zoslug Agapul Bubibi Chrcup Collar Cotdel Estast Eupalb Eupaur Euphor Loncuc Milmig Onyful Strsen Uraang Vidmac DistCoast SR ESR NSR STONP PNO

Ploceus sanctithomae, São Tomé Weaver Prinia molleri, São Tomé Prinia Serinus rufobrunneus, (São Tomé) Príncipe Seed-eater Terpsiphone atrochalybeia, São Tomé Paradise Flycatcher Treron sanctithomae, São Tomé Green Pigeon Turdus olivaceofuscus, São Tomé Thrush Zosterops feae, São Tomé White-eye Zosterops lugubris, São Tomé Speirops Agapornis pullaria, Red-headed Lovebird Bubulcus ibis, Cattle Egret Chrysococcyx cupreus, Emerald Cuckoo Columba larvata, São Tomé Cinnamon Dove Coturnix delegorguei, Harlequin Quail Estrilda astrild, Common Waxbill Euplectes albonotatus, White-winged Widowbird Euplectes aureus, Golden-backed Bishop Euplectes hordeaceus, Fire-crowned Bishop Lonchura cucullata, Bronze Mannikin Milvus migrans, Yellow-billed Kite Onychognathus fulgidus, São Tomé Chestnut-winged Starling Streptopelia senegalensis, Laughing Dove Uraeginthus angolensis, Southern Cordon-bleu Vidua macroura, Pin-tailed Whydah Distance to coast Species richness Endemic species richness Non-endemic species richness São Tomé Obô Natural Park Parque Natural do Obô

XI

GENERAL INTRODUCTION

Human population is a major threat to biodiversity Humans have been shaping the environment all over the planet, influencing the distribution of species and functioning of ecosystems. Many studies have associated human activities to the current crisis of biodiversity loss (Balmford & Bond 2005). Defining and measuring biodiversity is a complex and difficult task, therefore studying how human actions affect biodiversity is a major challenge. Additionally, biodiversity threats are unevenly distributed throughout the world, making it difficult to allocate conservation efforts. The urgent need to establish global conservation priorities has been a hot topic between conservationists (Brooks et al. 2006). Myers et al. (2000) identified 25 “biodiversity hotspots”, characterized by having a high concentration of endemic species and also great levels of habitat loss. Anthropogenic land use change is considered a main threat to species across all taxonomic groups (Luck 2007). Tropical forests, known to have both high species diversity and human pressure, are rapidly being converted for agriculture, timber production and other uses, generating humandominated landscapes and leading to forest degradation and destruction (Gardner et al. 2009). Habitat loss is considered to be one of the main reasons for the extinction of many species in the past decades (Sodhi et al. 2004; Stork 2010; Szabo et al. 2012). Many extinct species were island-endemics and because the projected rate for land-cover changes in islands is expected to increase, these fragile ecosystems are a growing global concern for conservationists (Manne et al. 1999). Many believe that given their conservation risks, smaller areas and high endemic species richness, islands could offer high returns for species conservation efforts, and therefore should be a high priority in global biodiversity conservation (Johnson & Stattersfield 1990). São Tomé Island as a study case São Tomé is an oceanic island, which is an excellent model to study the factors influencing species distribution, as well the adequacy of protected areas to represent biodiversity. It is an 857 km2 island, holding a remarkable biodiversity with many endemic species and a wide gradient of land use intensification. Together with Príncipe, it constitutes the Democratic Republic of São Tomé and Príncipe, which is in the Gulf of Guinea, Central Africa. At about 255 km from mainland Africa, São Tomé is of volcanic origin, which explains its rugged topography composed of steep slopes, deep valleys and high ridges, up to 2024 meters above sea level at the São Tomé Peak (Salgueiro & Carvalho 2001). Rivers are intrinsically associated with these narrow valleys, creating multiple waterfalls. The water slows down near the ocean creating small estuaries occasionally with mangroves. In the north-east, the terrain is flatter, especially if compared to the centre and west of the island. This diverse topography explains the incredibly varied climate found in São Tomé. The high mountains are a barrier to the strong winds, bringing heavy rains and coming from the south-west of the island. Thus, the south-west is characterized by high levels of humidity, having an almost permanent cloud cover, frequent rains and an annual rainfall of over 6000 mm, while the north-east is much drier, some areas receiving less than 600 mm of rain each year (Tenreiro 1961). São Tomé’s climate is characterized by a wet season, which occurs for most of the year, and two drier seasons. The longer dry season, called “gravana”, starts in May and ends in September, being more evident in the north of the island, and corresponding to the coldest months of the year. The shorter dry season, the “gravanito”, lasts for a few weeks in January and February. The

1

strong altitudinal gradient influences the mean annual temperature; coastal areas can reach maximum mean annual temperatures of 25.5º C, while at higher altitudes it might be as low as 9º C (Silva 1958). The strong climatic gradient has shaped the distribution of ecosystems throughout São Tomé. Having highly diverse landscapes with many different ecosystems, four land use types are usually recognized: non-forested areas, shade plantations, secondary forests and native forests (Jones & Tye 2006). Native forests are characterized by having a high density of native flora and few exotic species (e.g. Elaeis guineensis). Mangroves established along the lowest parts of the rivers and coastal lagoons can also be considered native forests. Exell (1944) defined three distinct rainforest types following the altitudinal gradient: lowland forests (up to 800 meters a.s.l), montane forests (between 800 and 1400 meters a.s.l.) and mist forests (above the 1400 meters a.s.l., along ridges of the central mountain range). Secondary forests appeared with the regeneration of abandoned shade plantations and with the intensive exploitation of timber, holding an assemblage poorer in forest species and with shade and fruit trees (e.g. breadfruit Artocarpus altilis, African nutmeg Pycnanthus angolensis). Shade plantations initially created as intensive monocultures by the Portuguese replaced most of the lower altitude forests. It is an agroforestry system composed mostly of exotic trees, such as cocoa Theobroma cacao, coffee Coffea sp. and coral trees Erythrina sp. (Salgueiro & Carvalho 2001). Nowadays, shade plantations have become more varied and produce many other crops, mostly for the internal market (banana Musa sp., cocoyam Colocasia sculenta and Xanthosoma sp., oil palm Elaeis guineensis, avocado Persea americana, papaya Carica papaya). Non-forested land uses include active and resting agricultural areas with different systems, such as monocultures of sugar cane Saccharum sp, coconut Cocos nucifera or oil palm, and artificial savannahs and smallholder horticultures (Diniz et al. 2002). The human occupation of São Tomé started in the late 15th century, after the Portuguese discovered the island, allegedly uninhabited and entirely covered by forest. Since then, the dried coastal lowland forests have suffered the most, being first cleared for sugar cane (Tenreiro 1961). During the 19th and 20th century, extensive cocoa and coffee plantations were grown in shade plantations, in large agricultural plantation systems, known as “roças”, further decreasing the area covered by native forests (Oliveira 1993; Frynas 2003). Nowadays, many shade plantations rely on medium and smallholdings that produce many subsistence products besides the main export crops. Swidden agriculture appeared to meet the demand for horticultural foods, expanding in forest borders and replacing abandoned shade plantations, being therefore included in non-forested land uses (Eyzaguirre 1986; Albuquerque et al. 2008). In the centre and south-west of the island a large patch of well-preserved native forest remains, nowadays enclosed by secondary forest, which in turn is surrounded by active shade plantations mixed with several non-forested land uses (Jones et al. 1991; Diniz et al. 2002). São Tomé has an incredible diverse flora and fauna. The right amount of isolation allowed many species to evolve in environments distinct from those found in the mainland (Miller et al. 2012). São Tomé and Príncipe hold 28 endemic bird species in an area little over 1000 km2 (Melo 2006). Out of 45 resident terrestrial species, São Tomé alone has 17 single-island endemics, 3 endemics to the Gulf of the Guinea oceanic islands (Annobón, São Tomé and Príncipe) and 8 widespread species represented in the island by an endemic subspecies (Jones & Tye 2006). As is often the case in other islands, some species are larger than their mainland relatives. That is the case of the Giant Sunbird Dreptes thomensis, the Giant Weaver Ploceus grandis, the São Tomé Grosbeak Neospiza concolor, the São Tomé Speirops Zosterops lugubris and the São Tomé Thrush Turdus olivaceofuscus. However, a few species, like the Dwarf Ibis Bostrychia bocagei, become smaller (Melo 2006; Melo et al. 2017). The lack of natural predators also made some species tame, such as the São Tomé Green Pigeon Treron sanctithomae, the São Tomé Maroon Pigeon Columba thomensis and the Dwarf Ibis. São Tomé is in a “biodiversity hotspot” and about 23.3% of its territory is included in Important Bird Areas (Myers et al. 2000; Fishpool & Evans 2001). Its forests are of great conservation interest, belonging to one of Earth’s biological ecoregions, named Gulf of Guinea Islands (Olson & Dinerstein 2

1998). Also, the forests were identified as the third most important in the world for forest bird species conservation (Buchanan et al. 2011). The long history of human occupation has led to habitat destruction and degradation, especially in the lower altitude forests, which were mostly converted to shade plantations. Endemic species have a long relationship with native forest, and many are dependent on these habitats (Rocha 2008; de Lima 2012). This way, the destruction or transformation of these forests might make them into unsuitable habitats. Apart from land use change, the introduction of species and direct exploitation are the main threats to São Tomé avifauna (Jones et al. 1991; de Lima 2012). Like in many oceanic islands, free of native predators of birds, introduced land mammals like rats, mice, dogs, cats, pigs, among others, become a serious threat to native bird species (Johnson & Stattersfield 1990; Dutton 1994; Blackburn et al. 2004). Three endemic bird species are considered Critically Endangered, the Dwarf Ibis, the São Tomé Fiscal Lanius newtoni and the São Tomé Grosbeak, the São Tomé Maroon Pigeon is Endangered, while six other endemic bird species are Vulnerable, two are Near Threatened and eight are Low Concern (IUCN 2017). To protect both native fauna and flora species, as well as their natural habitats, from human activities, the São Tomé Obô Natural Park (STONP) was created in 2006, covering 295 km2 (Direcção Geral do Ambiente 2006). This protected area was born under the umbrella of the “Ecosystemes Forestiers en Afrique Centrale” (ECOFAC) program, which started in 1992, funded by the European Commission, to encourage the conservation and sustainable use of forests in Central Africa. A buffer zone was also envisaged, but never official. The STONP action and management plan were first created in 2008, and revised in 2014 (Albuquerque et al. 2008), but implementation remains weak (de Lima et al. 2015). Thesis scope This thesis has two main goals, both related to understanding the bird diversity in São Tomé. In the first chapter, we explore bird species distribution and their responses to several environmental variables, using generalized linear models (GLMs) and paying close attention to the differences between endemic and non-endemic species, as well as between feeding guilds. Predictive distribution models are used to understand where species occur, which is essential to understand ecological requirements, as well as for conservation and population management (Guisan & Zimmermann 2000; Rushton et al. 2004). Logistic regressions are frequently used by ecologists to model species distribution, having gained a certain appeal because presence-absence data is easy to collect in the field. We considered vegetation, topographic, climatic and anthropogenic variables as potential predictors in logistic models, improving our understanding of which factors condition species occurrence (Seoane et al. 2003; Thuiller et al. 2004). In the second chapter, we model bird species richness and composition patterns to assess if the STONP adequately covers the island’s diverse avifauna. Three generalized linear models with poisson distribution were created to explain total, endemic and non-endemic species richness (Guisan & Zimmermann 2000), while generalized dissimilarity modelling (GDM) was used to map composition patterns (Ferrier et al. 2007). GDM is a novel statistical technique that analyzes and predicts spatial patterns of turnover in community composition (beta diversity). Being an extension of matrix regression, it is designed specifically to accommodate two types of nonlinearity commonly encountered in largescaled ecological data sets: (1) the curvilinear relationship between increasing ecological distance, and observed compositional dissimilarity, between sites; and (2) the variation in the rate of compositional turnover at different positions along environmental gradients (Ferrier et al. 2007; Arponen et al. 2008). In short, this approach compares community composition and environmental variables at pairs of sites to predict compositional difference as a function of environmental difference, extrapolating the prediction beyond surveyed sites. The resulting models give a spatially continuous prediction of 3

turnover, and thus of the spatial structure of diversity (Fitzpatrick et al. 2013; Brown et al. 2014). Predictive distribution maps are used nowadays to design protected areas, evaluate human impacts on biodiversity and test biogeographical hypotheses (Seoane et al. 2004). In this study, maps describing species richness and composition patterns were built to evaluate if the STONP is covering relevant components of the bird assemblage in São Tomé.

4

CHAPTER 1: The role of natural gradients and ecosystem humanization in determining the distribution of bird species in São Tomé

Abstract: Anthropogenic land use change is the main driver of the ongoing biodiversity crisis. Understanding how species respond to land use changes is thus key to minimize the current species extinction rate. São Tomé is a small oceanic island, where forest degradation is a main threat to the endemic-rich avifauna. To preserve this invaluable avifauna, we tried to understand how bird species are distributed throughout the island. We gathered occasional and systematic observations from previous studies, which were later combined with additional 10-minute point counts, adding to a total of 2398 bird point counts and 658 occasional observations. Thirty-four terrestrial bird species were unambiguously identified and considered in subsequent analyses. Species-specific generalized linear models and detrended correspondence analysis based on presence-absence, were used to explore the links between endemism, feeding guilds and environmental variables. Land use was the most important variable to explain bird species occurrence. The endemics tended to prefer forests in wetter, rugged, higher altitude, and remote areas, while the non-endemics favoured flat lowland non-forested areas and shade plantations. São Tomé’s forest-dominated landscape ensures an overall dominance of endemic species, but a change in bird species assemblage from forest endemic species to open habitat nonendemic granivore species was found to be a result of the land use intensification gradient. Many of the forest endemics are threatened, highlighting the urgent need to protected forested habitats. We suggest landscape matrix improvement, through the protection of the remaining native forest and the expansion of secondary forest, as the most important conservation measure to ensure the future of the endemicrich avifauna of the islands. Keyword: endemism; feeding guild; generalized linear model; land use types; threatened species

INTRODUCTION Understanding how animals and plants are distributed on Earth, in both space and time, is a challenging task, especially in our constantly changing planet. A wide range of factors, such as food availability, shelter, environmental abiotic factors (e.g. temperature, humidity), biotic interactions (e.g. competition, predation, mutualism, host-parasite interactions, facilitation), physical barriers (e.g. rivers, mountains), climate (e.g. global climate change), disturbances (e.g. fires, floods, pathogens), among many others, are listed to influence species distribution (Brown 1984; Lawton 1999; Mackey & Lindenmayer 2001; Thomas et al. 2004). All these factors interact at different spatial and temporal scales, imposing limits on species distribution which are expressed from local to global spatial scales. Our understanding of species distribution started with qualitative analyses: observing and recording the relationship between species distributions and the physical environment. Today, numerical techniques are widely used for describing species distribution patterns and making predictions (Elith & Leathwick 2009). For example, species distribution models (SDMs), that combine observations of species occurrence or abundance with environmental variables, allow the prediction of species distributions across the landscape (Guisan & Zimmermann 2000; Rushton et al. 2004). Human activities have been shaping ecosystems across the globe, especially by land use change that is known to alter ecosystem patterns and processes, as well as species distributions (Blair 1996; Cincotta et al. 2000). Anthropogenic land use changes have been considered a major driver of the 5

ongoing biodiversity crisis (Myers et al. 2000). Therefore, understanding species response to humaninduced land use change is essential to guide conservation actions (Maestas et al. 2003; Benton et al. 2003; Chacea & Walsh 2006). Agricultural demand is by far the main cause for land use change (Phalan et al. 2011). Urban sprawling is also promoting the conversion of natural and even agricultural land, further reducing the availability of habitats for wildlife (Assandri et al. 2017). Both are predicted to continue growing in the nearby future. Land use change has consistently reduced overall habitat quality, increased ecosystems fragmentation, isolation and degradation, and promoted the introduction of exotic species (Cadenasso & Pickett 2001; Foley 2005; McKinney 2006; Stork 2010). A study conducted in the north-eastern Brazilian Amazonia showed plantations had a relatively impoverished amphibian and lizard communities, a frequently discussed consequence of land use change (Gardner et al. 2007). In tropical forests, where the species diversity and human pressure is higher, land use change is expected to cause great habitat loss (Sodhi et al. 2004; Walter et al. 2007; Gardner et al. 2009; Stork 2010; Szabo et al. 2012). Local extinction of birds and mammals have been described as a consequence of anthropogenic land use change (Brooks et al. 1999; Sodhi et al. 2004; IUCN 2017). Extinctions have been far more frequent on islands than on continents (Manne et al. 1999). The unique flora and fauna found on insular ecosystems are extremely vulnerable to human actions, and with the increasing rate of land use change, these fragile ecosystems are becoming a growing global concern among conservationists. This main goal of this study is to understand how bird species are distributed in response to natural and anthropogenic factors, using São Tomé, an endemic-rich oceanic island with a known land use intensification gradient, as an example (Melo 2006; Miller et al. 2012; de Lima et al. 2015). We focus on three specific goals: (1) identifying the key determinants of the distribution of bird species; (2) understanding how endemism relates to the response of bird species to environmental variables; and (3) analyse the relationship between feeding guilds and bird species response to environmental variables. We will also explore the relationship between key determinants and species response, paying special attention to endemic and threatened species.

METHODS Study Area São Tomé, together with the neighbouring island of Príncipe, form the Democratic Republic of São Tomé and Príncipe, located in the Gulf of Guinea, Central Africa. This oceanic island is just north of the Equator and about 255 km west of the African Continent. For an 857 km2 island, it has a remarkably unique avifauna (Stattersfield et al. 1990; Peet & Atkinson 1994; Leventis & Olmos 2009). Out of 45 resident terrestrial species, 17 are single-island endemics, 3 are endemic to the Gulf of the Guinea oceanic islands (Annobón, São Tomé and Príncipe) and 8 are widespread species represented in the island by an endemic subspecies (Jones & Tye 2006). The high endemism rate is associated with its location in relation to the African continent: close enough to allow migration, and far enough to allow speciation by isolation (Melo 2006). This island is considered a “biodiversity hotspot” and, recently, its lowland forest belong to one of Earth’s biological ecoregions, the Gulf of Guinea Islands (Olson & Dinerstein 1998; Myers et al. 2000). Also, these forests were identified as the third most important in the world for forest bird species conservation (Buchanan et al. 2011). As in many other oceanic islands, human occupation in São Tomé led to the introduction of several species, namely several bird species, most of which native from the African Continent. Before human intervention, the island was almost entirely covered by forest and the topography was responsible for

6

the strong climatic gradients that shaped the distribution of ecosystems. Human colonization, resulted in much of the lowland forests and some montane forests being replaced by plantations (Jones et al. 1991). Only the inaccessible rugged wet areas in the south-west and centre of the island remain covered by native forest, which is currently surrounded by secondary forest, resulting from logging and plantation abandonment. Enclosing this land use type are extensive areas of active shade coffee and cocoa plantations, a type of agroforestry, which is mixed with non-forested areas, such as oil palm monocultures, horticultures and open savannahs (Exell 1944; Tenreiro 1961; Jones et al. 1991). Despite the long history of intensive conversion to anthropogenic land use, São Tomé’s landscape is still dominated by forested ecosystems. The native forest is almost entirely classified as São Tomé Obô Natural Park (STONP), which covers almost one third of the island (Albuquerque et al. 2008). Unfortunately, the protection and conservation efforts have not been effective and in the last decades human pressure on natural resources has been increasing fast, and the area covered by native forest and shade plantations has decreased, while secondary forest and non-forested areas have been expanding (Salgueiro & Carvalho 2001). Data Compilation In this study, we gathered all records from a single observer, obtained in 2009 and in 2010, for a total of 300 point counts (de Lima 2012), plus 1653 point counts and 677 occasional from BirdLife International São Tomé and Príncipe Initiative (BISTPI), collected between 2013 and 2015 (de Lima et al. 2017). In both studies, point counts were separated by at least 200 meters, to ensure independence, and all birds detected during 10 minutes were registered, regardless of the distance. This information was compiled in a single bird species occurrence database, which had a GIS component. Field Methods Sampling design To identify under-sampled areas in previous studies from which we compiled data, we overimposed the bird occurrence database on the map of São Tomé. The island was then divided in 1x1 km quadrats, grouped in groups of four to form 2x2 km quadrats (de Lima et al. 2017). All 2x2 km quadrats that had more than half of their area occupied by the ocean were excluded. We considered sampled all the 2x2 km quadrats that had at least one 1x1 km quadrat with five 10-minute point counts sampled. Between January and March 2017, we sampled 91 out of the remaining unsampled 96 2x2 km quadrats, located mostly in non-forested low-altitude areas across the island. Bird sampling Each of the 2x2 quadrats were sampled by performing five bird point counts in a randomly selected 1x1 km quadrat (Fig. 1.1), largely following the BISTPI methodology (de Lima et al. 2017). The location of the point counts was chosen to ensure a distance of at least 200 meters between point counts, thereby ensuring independence and that the environmental variability inside each quadrat was sampled in the approximate proportion in which they occurred in the quadrat. In each point count, all bird species detected visually and aurally were registered by an experienced observer, during a 10 minute period, and regardless of the distance. To maximize the number of sampled points during our short sampling period, counts were made throughout the day, from approximately 6 am until 5 pm.

7

Figure 1.1. Location of sampling point counts and occasional observations (n = 3056) in São Tomé Island. The lines in the background represent the 100 m elevation isolines.

Characterizing environmental variables To model the distribution of bird species, we obtained geographically explicit information on altitude, ruggedness, slope, distance to the coast, topography, remoteness, rainfall and land use across São Tomé, using Quantum GIS v. 2.8.3 and v. 2.14.8 (Quantum GIS Development Team 2009a; Table S1 & S2). The variable altitude was derived in meters from a 90 meters resolution Digital Elevation Model (DEM) (Silva 1958; Salgueiro & Carvalho 2001; NASA Jet Propulsion Laboratory 2016; Fig. S1). The ruggedness and slope were also calculated from the DEM raster, using the “raster terrain analysis” QGIS plugin (Quantum GIS Development Team 2009b; Fig. S2 & S3). Slope was primarily calculated in decimal degrees and then transformed to percentage. Distance to the coast was calculated as the minimum linear distance in decimal degrees between each pixel and the nearest point on the coast line, using the DEM and the QGIS “distance matrix” tool (Quantum GIS Development Team 2009a; Fig. S4). The topography was represented using a Topography Position Index (TPI) which allows comparing of each cell’s elevation to the mean elevation of a specified neighbourhood (Jenness 2007). The TPI was calculated using the DEM and the “topography position index” tool in the QGIS GDAL algorithm provider (Quantum GIS Development Team 2009c) and a 0.05º radius neighbourhood, which allows for a good representation of terrain ruggedness and elevation in São Tomé. TPI was later transformed in a five-category discrete variable: flat areas, valleys, middle slopes, upper slopes and ridges (Fig. S5, S6 & S7). Remoteness is expressed as an index that translates the difficulty of movement through the landscape, and it was created using the “accumulated cost” QGIS GDAL algorithm provider (Quantum 8

GIS Development Team 2009d). This index is a cost accumulated surface based on a friction surface derived from slope and weighted by the human population density (Tobler 1993; Instituto Nacional de Estatística 2016; Fig. S8 & S9). Rainfall was obtained by digitizing a map with the island’s mean annual precipitation in millimetres (Silva 1958; Fig. S10). The land use map (Fig. S12) was created mostly by visual interpretation of 2014 satellite images (Google Earth 2017), supplemented by 2009-2017 field land cover information (de Lima 2012; de Lima et al. 2012), 1970 land use map (de Carvalho Rodrigues 1974), military maps (Missão Hidrográfica de Angola e S. Tomé 1958), a 2011-13 preliminary land use map (S. Mikulane, unpublished data - see Fig. S11) and expert knowledge. All variables were standardised to a common in raster grid, using the nearest neighbour sampling method and the TPI raster as a geometric reference. This standardization was made using QGIS “align rasters” tool (Quantum GIS Development Team 2009a), and resulted in a pixel’s size of 0.000833º x 0.000833º and a raster with 359 x 471 cells. Each point count was characterized for each environmental variable using the “point sampling tool” QGIS plugin (Quantum GIS Development Team 2009e). Data Analysis All statistical analyses were made in R v. 3.3.2 using RStudio v. 1.0.143 (R Development Core Team 2017). Exploratory analysis All bird data was compiled in a single database of 2408 point counts and 677 occasional observations. We excluded all species that are aquatic, difficult to identify or had less than 20 presences (Table S3), obtaining a total of 34 species that was considered for subsequent analyses. Point counts with no record of these species or that had inconsistencies between the field land cover classification and the 2014 land use map were also removed, leading to a final of 2398 point counts, plus 658 occasional observations. Multicollinearity was tested using Spearman’s rank correlation coefficient, and visualized in a correlogram built using the “corrgram” package (Wright 2016; Part I, Section VIII). Ruggedness was excluded, since its correlation coefficient with slope was higher than 0.8 (Fig. S13). Variance homogeneity and no outliers were identified by the boxplots drawn for each environmental variable using the “vegan” package (Oksanen 2015). Generalized linear models The data were divided in training and testing sets, using the “caTools” package: 70% of the points were used to create binomial generalized linear model (GLM) to explain species presence (Rushton et al. 2004), while the remaining 30% were used to validate the models (Tuszynski 2014; Part II, Section VIII). We used Variance Inflation Factors (VIF) to double-check multicollinearity, and, once again, ruggedness was chosen to be excluded from all species models for being the only predictor variable having VIFs larger than 10. For each species, we generated all possible models based on the different combinations of explanatory variables, and ranked based on the Akaike Information Criterion corrected for small sample sizes (AICc), using the “dredge” function from the “MuMIn” package (Barton 2016). The goodness of fit was analysed with the McFadden’s index in the “pscl” package (Jackman et al. 2015). We validated the predicted values and calculated the receiving operating characteristic (ROC) curve. The area under the curve (AUC) was calculated to examine the model’s performance with the “ROCR” package (Sing et al. 2015; Table S4).

9

Relative variable importance To identify which variables best explain the presence of each species, we ran the “model averaging” function of the “MuMIn” package to obtain relative variable importance (RVI). Bird species were separated in endemic and non-endemic species, and by feeding guild: carnivore (including insectivore), frugivore, granivore and omnivore (Jones & Tye 2006; HBW Alive 2017; Table S3). For each explanatory variable, we used Kruskal-Wallis rank tests to evaluate the difference in RVI values between endemic and non-endemic, and between feeding guilds (Table S5). To perform post hoc pairwise comparisons between feeding guilds we used Dunn-tests with Benjamini-Hochberg corrections (Thissen et al. 2002). These analyses were done using the “stats” and “FSA” packages (Ogle 2017; Part V, Section VIII). Response to environmental variables To analyse the response of each species to continuous variables, single-variable logistic regression models were created to explain species’ presence and obtain coefficient values (Table S6). The proportion of occurrence in each land use type and in each TPI class was calculated for every species, correcting for sampling effort. Then, it was calculated for each group of species: endemics, nonendemics, carnivores, frugivores, granivores and omnivores. To evaluate the differences between endemic and non-endemic species, and between feeding guilds, among each land use type and topography class, Kruskal-Wallis rank tests were performed using the “stats” package. As previously, Dunn-tests with the Benjamini-Hochberg correction for multiple comparisons were run to analyse differences between feeding guilds (Part V, Section VIII). Both these tests were also used to evaluate the differences in coefficient values between endemic and non-endemic species, and between feeding guilds. To visualize the links between endemism, feeding guilds and environmental variables, a detrended correspondence analysis (DCA) was made. The proportion of occurrence of each species in each land use type was also explored graphically, to gain a better understanding of how endemism and threat status relate to land use types.

RESULTS Only 658 out of the 3056 final data points referred to occasional observations. On average, each species appeared in 24.4% of the systematic point counts, ranging from 88.4% for the São Tomé Sunbird Anabathmis newtonii to 0.6% for the São Tomé Grosbeak Neospiza concolor (Table S3). Relative variable importance The most important variable to explain the occurrence of bird species in São Tomé was land use, followed by rainfall and remoteness (Fig. 1.2 & S14). Distance to coast, altitude and topography had intermediate importance, while slope was the least important. When looking at the species individual responses to environmental variables, it is clear that land use was more important to the endemic species than to the non-endemic. On the other hand, rainfall was more important to non-endemic species. Topography seemed relevant to endemic species distribution, but was the least important variable to non-endemic species.

10

E N

Figure 1.2. Relative variable importance (RVI) of each environmental variable for each bird species generalized linear model. The RVI is represented by a colour gradient, in which: darker cells indicate higher values. The RVI values range from 0 to 1. Endemic (E) and non-endemic (N) species are grouped together and separated by a black line.

When comparing the RVI of endemic and non-endemic species, only land use (H = 6.19, df = 1, p = 0.013) and topography (H = 5.674, df = 1, p = 0.017) had significant differences, and both were more important to the endemics (Table S7 & Fig. S15). Among feeding guilds, altitude (H = 8.3603, df = 3, p = 0.039) and slope (H = 10.373, df = 3, p = 0.016) were the only variables having significantly different RVI values. Altitude was more important to explain the presence of carnivores than that of omnivores, while slope was less important to frugivores than to any other feeding guild (Table S7 & Fig. S16). Response of endemic and non-endemic species to environmental variables The endemic species tended to have significantly higher values for all continuous environmental variables, when compared to the non-endemic (rainfall: H = 14.295, df = 1, p = 0.0002; remoteness: H = 12.765, df = 1, p = 0.0004; distance to coast: H = 11.555, df = 1, p = 0.0007; altitude: H = 12.032, df = 1, p = 0.0005; slope: H = 13.519, df = 1, p = 0.0002; Table 1.1, Fig. 1.3 & 1.4). The proportion of occurrence in almost all land use types was significantly different between endemic and non-endemic species. Endemics tended to occur preferentially in native (H = 17.794, df =

11

1, p = 2.461 x 10-5; Table 1.1 & S8, Fig. 1.3 & 1.4) and secondary forest (H = 11.672, df = 1, p = 0.0006), while non-endemic species preferred non-forested areas (H = 17.206, df = 1, p = 3.355 x 10-5). Endemic and non-endemic species occurrence among each topography class was also almost significantly different for all classes (Table 1.1, Fig. 1.3 & 1.4). Endemics tended to occur mostly in valleys, middle and upper slope areas, and also ridges (valleys: H = 13.911, df = 1, p = 0.0002; middle slope: H = 16.328, df = 1, p = 5.328 x 10-5; upper slope: H = 16.609, df = 1, p = 4.593 x 10-5; ridges: H = 18.115, df = 1, p = 2.08 x 10-5), while the non-endemic species occur in a bigger proportion in flat areas (H = 17.468, df = 1, p = 2.922 x 10-5).

Table 1.1. Response of endemic (E) and non-endemic (N), and of distinct feeding guilds (omnivores - O, granivores - G, frugivores – F, and carnivores – C) to environmental variables. For continuous variables, the differences between E and N coefficients were assessed using Kruskal-Wallis rank tests (KW), while between feeding guild coefficients were assessed using Dunn-tests with Benjamini-Hochberg correction. For categorical variables, land use and TPI, Kruskal-Wallis rank tests were used to analyse differences between endemic and non-endemic species, while between feeding guilds Dunn-tests with Benjamini-Hochberg correction were used. Only p-value < 0.05 are shown.

Variables Rainfall Remoteness Distance to Coast Altitude

Endemism (KW test) 0.0002 E >>> N 0.0004 E >>> N 0.0007 E >>> N 0.0005 E >>> N 0.0002

E >>> N

Native Forest

2.461 x 10-5

E >>> N

Secondary Forest

0.0006

E >>> N

Shade Plantation

-

-

Non-Forested Areas

3.355 x 10-5

E > G 0.0052 C >> G 0.0185 C>G 0.0240 C>G 0.0420 O>G

Land Use 0.020 0.023 0.021 0.015 0.038 0.038

C>G F>G F>G O>G CG F>G

Topography -5

Flat Plain Areas

2.922 x 10

E > N

Middle Slope

5.328 x 10-5

E >>> N

Upper Slope

4.593 x 10-5

E >>> N

Ridges

2.08 x 10-5

E >>> N

12

13

Figure 1.3. Response of endemic (E) and non-endemic (N) species to environmental variables. The boxplots represent the continuous variables coefficients obtained from singlevariable models: the thick line shows the median, the box the first and third quartiles, the whiskers the extremes, and the dots the outliers. The bar-plots represent the standardized proportion of occurrence in every land use type (native forest - NF, secondary forest - SF, shade plantation – SP and non-forest areas - NFA) and topography class (flat areas – F, valleys - V, middle slope areas - M, upper slope areas - U, ridges – R).

Figure 1.4. Detrended Correspondence Analysis (DCA) showing the relationship between endemism, feeding guilds and environmental variables. Each point represents a species, which is identified by the corresponding acronym (See List of Abbreviations and Acronyms, pages IX to X). The black dots represent the endemic and the grey the non-endemic species. The shape of the points represents the feeding guilds (F - frugivores, G- granivores, O - omnivores, and C – carnivores). The panel on the top right corner shows how are environmental variables related to the DCA axes: land use type (NF - native forest, SF secondary forest, SP - shade plantation, and NFA - non-forested areas), TPI (Flat - flat areas, Valleys - valleys, Middle intermediate slope areas, Upper - upper slope areas, Ridges – ridges), Slope, Altitude, Rainfall, Distance to coast (DistCoast), and Remoteness.

Feeding guilds response to environmental variables The feeding guilds showed significant differences in all coefficients obtained from single-variable models (rainfall: H = 8.706, df = 3, p = 0.033; remoteness: H = 11.232, df = 3, p = 0.011; distance to coast: H = 11.96, df = 3, p = 0.008; altitude: H = 8.764, df = 3, p = 0.033; slope: H = 8.714, df = 3, p = 0.033; Table 1.1, Fig. 1.4 & 1.5). The granivores tended to have lower values for all continuous environmental variables. These differences were always significant, when comparing to carnivores (Z = 3.331 for distance to coast and Z = 3.351 for remoteness with p < 0.01, and Z = 2.879 for slope, Z = 2.959 for altitude and Z = 2.901 for rainfall with p < 0.05), and also when comparing to omnivores for slope (Z = 2.456, p = 0.042). Granivores had the most distinct land use type and topography preferences. They tended to use less native forest than carnivores (Z = 2.928, p = 0.020) and frugivores (Z = 2.660, p = 0.023), less secondary forest than frugivores (Z = 2.699, p = 0.021) and omnivores (Z = -3.021, p = 0.015), and more non-forested areas than carnivores (Z = -2.733, p = 0.038) and frugivores (Z = -2.495, p = 0.038; Table 1.1, Fig. 1.4 & 1.5). They were also clearly associated with flat areas (Z = -2.873 for carnivores, Z = -2.731 for frugivores, Z = 2.269 for omnivores, with p < 0.05).

14

15

Figure 1.5. Feeding guild (omnivores - O, granivores - G, frugivores – F, and carnivores - C) response to environmental variables. The boxplots represent the continuous variables coefficients obtained from single-variables models: The thick line shows the median, the box the first and third quartiles, the whiskers the extremes, and the dots the outliers. The bar-plots represent the standardized proportion of occurrence in every land use type (native forest - NF, secondary forest - SF, shade plantation - SP, non-forest areas - NFA) and topography class (flat areas - F, valleys - V, middle slope areas M, upper slope areas - U, ridges - R).

Species land use type preferences Most of the 19 endemic species clearly preferred forested land use types. Nine had more than 75% of their presences in forest, seven had more than 50% in native forest and four had more than 75% in native forest (Fig. 1.6). Some endemic species like the Green Pigeon Treron sanctithomae, the Scops Owl Otus hartlaubi and the Oriole Oriolus crassirostris are also frequently found inside secondary forests. A few endemic species, like the São Tomé Thrush Turdus olivaceofuscus, the São Tomé Prinia Prinia molleri or the São Tomé Sunbird, were almost evenly distributed among all land use types. The Giant Weaver Ploceus grandis, on the other hand, is an exception inside endemic species, being an omnivorous, can be commonly found inside plantations, such as palm plantations (Atkinson et al. 1991). In contrast with the endemics, the 15 non-endemics were clearly associated with non-forested areas, and avoided forests. Ten had more than half of their presences in non-forested areas, while only four even occurred in native forest. The Pin-tailed Whydah Vidua macroura, the Southern Cordon-bleu Uraeginthus angolensis and the Bronze Mannikin Lonchura cucullata are examples of non-endemic granivores found mostly in nonforested areas.

LC

N E

NT VU

EN CR

Figure 1.6. Proportion of occurrence of each species by land use types. Species are grouped by endemism (E – endemic; N – non-endemic), and by conservation status (CR – critically endangered; EN – endangered; VU – vulnerable; NT – near threatened; LC – least concern). Within each group, species are ranked according land use type preferences (native forest – black, secondary forest – dark grey, shade plantation – light grey, and non-forested areas – white).

16

The Cinnamon Dove Columba larvata and the Chestnut-winged Starling Onychognathus fulgidus were the non-endemic species that clustered with the endemic (de Lima et al. 2012). These two species are represented in São Tomé by endemic subspecies that are fairly different from the continental ones, and that might warrant being classified as distinction species (Peet & Atkinson 1994; Leventis & Olmos 2009; Pereira 2013). Our results show the Emerald Cuckoo Chrysococcyx cupreus, also represented in São Tomé by another endemic subspecies, in a similar position (Fig. 1.4 & 1.6). All other non-endemics cluster away from the endemics (Fig. 1.4) and clearly avoid forested land use types (Fig. 1.6), including the endemic subspecies of Harlequin Quail Coturnix delegorguei. All threatened species were endemic, and species with higher threat status tended to have stronger links to native forest, except for the São Tomé White-eye Zosterops feae, which had less than 25% of its presences in this land use type.

DISCUSSION We identified land use as the most important environmental variable to model the distribution of 34 bird species in São Tomé. Determinants of bird species distribution Considering all São Tomé bird species, land use was without a doubt the most important variable to explain their distribution, followed by rainfall and remoteness (Fig. 1.2). All three variables are related to each other, and with the topography of the island. Early studies had already suggested that land use was an important determinant of São Tomé bird species distribution (Jones & Tye 2006), but our results suggest it is actually the most important. Worldwide, habitat has also been repeatedly identified as the primary determinant of species distribution and abundance (Seoane et al. 2004; Tejeda-Cruz & Sutherland 2004; Dallimer & King 2007; Rocha et al. 2015). Flora composition and structure, characteristics clearly dependent on land use, have been considered important factors to explain the distribution and abundance of many passerine species (Maestas et al. 2003). Differential response of endemic and non-endemic bird species The endemics were clearly associated with forested land uses, usually located in remote areas, away from the coast, and where the rainfall is higher. They also tended to prefer higher altitudes and steeper slopes, namely valleys and ridges (Table 1.1, Fig. 1.3 & 1.4). On the contrary, the non-endemics preferred more intensive land uses, such as shade plantations and non-forested areas. Notably they were associated with drier regions of the island, in the accessible lowlands near the coast. The species response to land use change is congruent with previous work, which had already observed a rise in non-endemics and a decrease in endemics along the land use intensification gradient (de Lima et al. 2012). A pattern that also makes sense, considering that the native endemic-rich avifauna of São Tomé evolved in a forest-dominated landscape (Atkinson et al. 1991). Shade plantations were the only land use type where there was no clear preferences associated with endemism (Table 1.1). This agroforestry system usually consists of several agricultural crops shaded by high canopy trees. Despite being almost entirely composed by introduced plant species, it provides ecosystems with intermediate environmental conditions that are both suitable for endemic and nonendemic species (Rocha 2008; de Lima et al. 2014). These findings coincide with studies performed across the globe, showing that shade plantations and other agroforestry systems support a depleted 17

proportion of the native biodiversity, often mixed with introduced species (Thiollay 1999; Waltert et al. 2004; Foley 2005; de Lima et al. 2014). Differential response of bird species based on feeding guilds Being almost entirely composed of endemic species, the frugivores also had a strong association with forested land uses (Table 1.1, Fig. 1.4 & 1.5). They preferred remote areas, far from the coast, with high levels of rainfall and steeper slopes, such as valleys and ridges. Most carnivores are also endemic species, meaning their response resembled that of the endemics. Out of 13 omnivores, eight are endemic species, and so their response was the sum of different species response, having no clear pattern linked to endemism. On the contrary, all granivores are non-endemic species, and therefore were associated with the more intensive land uses in the drier, lowlands of the island. Having evolved in a forest-dominated landscape, most endemic species are frugivores and carnivores that rely on forest resources, and therefore might not be capable of adapting to land use intensification. The endemic species tended to avoid the non-forested areas and shade plantations, where the lack of suitable habitat and other resource limitations, e.g. food, were responsible for their disappearance (de Lima et al. 2012). In contrast, the non-endemics, especially granivores that are open habitat specialists, occurred preferentially in more intensive land uses. Other authors had too stated that primarily frugivorous and insectivorous forest specialists were less likely to occur and less abundant in more intensively used habitats, where habitat generalists thrive (Newbold et al. 2013) As in other studies, the granivore species response and apparent avoidance of forested land uses suggested that non-endemic species were introduced during the colonization, quickly occupying the more intensively managed habitats (Atkinson et al. 1991; Jones & Tye 2006; Rocha 2008; de Lima et al. 2012). The low occurrence of granivore non-endemics inside forested land uses reinforces the idea of no direct competition with forest endemic species, also stated in a different study on avian community responses (Thiollay 1999). More intensive land uses tended to have a higher human pressure, which negatively impacts and conditions the endemic species occurrence (Rocha 2006; de Lima et al. 2012; Andren 1994). Previous authors stated that hunting might be an important threat, especially to frugivore endemic birds, like the two most favoured quarry species, the São Tomé Maroon Pigeon Columba thomensis and the São Tomé Green Pigeon (Carvalho 2015; Margarido 2015). Mammals, such as pigs Sus domesticus, cats Felis catus, black and brown rats Rattus sp., mona monkeys Cercopithecus mona, amongst others, were also brought to the island during colonization (Dutton 1994). The introduction of mammal species in insular ecosystems is considered a great threat to the native avifauna (Johnson & Stattersfield 1990; Blackburn et al. 2004; Szabo et al. 2012). In São Tomé, it is thought the introduced mammal species have a wide distribution among all land use types and thus have an overall negative impact on endemic bird species. Consequences of land use intensification to the endemic-rich avifauna of São Tomé Endemic species were clearly associated with São Tomé forested landscape, declining towards the more intensive land uses, where on the contrary the non-endemic species found suitable conditions. Other studies had already observed a decay in the number of endemic species with greater land use intensification (Rocha 2006; de Lima et al. 2012). Since the colonization, lowland forests and some montane forests were progressively replaced by coffee and cocoa plantations, leaving only the more inaccessible, wet areas of the southwest and central of the island covered by relatively undisturbed and well-preserved forest. We believe that with the discovery of offshore oil reserves (Frynas et al. 2003) and the rapid human population growth (Instituto

18

Nacional de Estatística 2016), the pressure on forest habitats will continue to rise. Our results suggest that the increasing land use intensification, whether by converting São Tomé forests into intensively managed land uses, or by promoting forest degradation, will compromise the long-term persistence of endemic species (Ndang’ang’a et al. 2014; de Lima et al. 2017). As found in similar studies, the gradient of land use intensification is the main responsible for the changes in bird species assemblages, from forest endemic species to non-native open habitat specialists (Hughes et al. 2002; Naidoo 2004; Waltert et al. 2005). Most forest endemic species are frugivores and carnivores, therefore the more intensive land uses, such as non-forested areas and shade plantations, lack suitable conditions essential for these species survival (e.g. habitat, food availability, among others). Other studies also found that insectivores were associated with reduced resilience to habitat conversion (Thiollay 1995; Waltert et al. 2005). Land use intensification had strong negative impacts on São Tomé endemic-rich avifauna. The endemic species, highly dependent on the forested habitats, have been replaced by the non-endemic species inside the intensively managed land uses (Pardini et al. 2010). Non-endemic species were able to colonize these disturbed areas, being mostly granivore and omnivore, open habitat species (Naidoo 2004; Tejeda-Cruz & Sutherland 2004), which suggest they were introduced to the island (Jones & Tye 2006). This change from endemic to non-endemic species also suggests the gradient of land use intensification is acting as a facilitator of the spread of non-native species (Didham et al. 2007). The most threatened endemic species in São Tomé are also the ones with the higher association to the native forest, thus rising even more their already high conservation value (Margarido 2015; de Lima et al. 2017). The Fiscal Lanius newtoni, the Grosbeak and the Dwarf Ibis Bostrychia bocagei, all occurred almost uniquely inside native forests, being considered critically endangered by IUCN (IUCN 2017). In order to protect São Tomé threatened endemic species and their forested habitats, we urge the need to reduce and ultimately cease land use intensification, thus preventing further conversion and degradation of forested land uses. At last, São Tomé provides a good example of how a strong gradient of land use intensification, inside small historical forest-dominated islands, can rapidly reduce the proportion of forested land uses, while simultaneously acting as a facilitator of the spread of non-native species. Given our findings, we suggest focusing first on the full understanding of the native threatened species response to land use intensification, and just then, define specific conservation measures to protect indigenous forests and restore already degraded land uses. This strategy promotes the maintenance of an endemic-rich avifauna, while preventing the spread of non-native species facilitated by land use intensification.

19

CHAPTER 2: Is the existing protected network adequate for the conservation of the endemic-rich avifauna of São Tomé Island?

Abstract: Tropical forests are some of the most diverse and threatened terrestrial ecosystems. The increasing human pressure, high number of threatened species and major habitat loss forces conservation action prioritization. São Tomé is a small oceanic island with an endemic-rich avifauna. It has a single protected area: the São Tomé Obô Natural Park (STONP), whose boundaries were defined in 2006, based on ecosystem and human population distribution. We compared the distribution of bird diversity with the boundaries of the park to assess how it represented the island’s avifauna. Systematic observations from previous studies were gathered and supplemented by additional bird counts. Five 10minute point counts were grouped in 1x1 km quadrats (n = 187). Thirty-six terrestrial bird species were identified unambiguously and considered for analyses. The proportion of endemic bird species decreases along the land use intensification gradient: forest endemics decline towards humanized habitats, where non-endemic granivores are most abundant. The STONP did not protect the most species-rich bird assemblages, but covered most of the best-preserved forests, which are the richest in endemic species. The STONP boundaries are well located for the protection of endemic threatened birds, arguably those of higher global conservation interest. Secondary forests act as a transition zone to humanized areas, and protect the park from pervasive threats. The zonation of the STONP should be revised, using the same factors considered for the delimitation of the protected area and the current knowledge on species distribution. This study suggests that protecting well-preserved natural areas with low human density might be a good proxy to identify areas of high conservation interest, when there is little information on the distribution of the multiple components of biodiversity. Keywords: São Tomé Obô Natural Park; species richness; generalized dissimilarity modelling; species distribution modelling; conservation planning

INTRODUCTION Human activities are causing a biodiversity crisis (Brooks et al. 2006), through the transformation and sometimes complete destruction of natural habitats (Stork 2010). Temperate forests are a living proof of the devastating impact of humans (Pimm & Askins 1995), but few species have been considered extinct in continental tropical forests. Tropical forests include some of the most diverse terrestrial ecosystems but, in recent decades, also some of the most threatened (Myers et al. 2000), due to the increasing human pressure, which is expected to rise in upcoming years with the growing human population (Cincotta et al. 2000; Luck 2007). Nowadays many tropical species are threatened by habitat loss and degradation (IUCN 2017). The rise of extinction rates in tropical forests is therefore likely to occur in the near future (Brooks et al. 1999). The high number of threatened species, the great diversity of threats and the limited funding force conservationists to establish priorities. Twenty-five “biodiversity hotspots” have been identified by exceptional concentrations of endemic species and habitat loss, containing 44% of the Earth's plant species and 35% of its vertebrates in just 1.4% of its land surface (Myers et al. 2000). These hotspots are the focus of many conservation programs, aiming to reduce the current rate of biodiversity loss (Cincotta et al. 2000).

20

Protected areas are one of the main conservation actions to safeguard threatened species and their habitats. About 38% of the “biodiversity hotspots” are already protected in parks and reserves, which range from highly restrictive areas where all human activities are excluded to more inclusive management strategies involving local communities (Schwartzman et al. 2000). São Tomé is an oceanic island located in the Gulf of Guinea. It is included in a “biodiversity hotspot” (Myers et al. 2000), and the high concentration of avian endemism contributes to its unique biodiversity (Melo 2006; Miller et al. 2012; de Lima et al. 2015). Its forests, together with Príncipe and Equatorial Guinea, belong to the Earth’s biological ecoregions named Gulf of Guinea Islands forests, which has a critical/endangered conservation status (Olson & Dinerstein 1998). Most recently, its lowland forests were identified as the third most important in the world for the conservation of forest bird species (Buchanan et al. 2011). All this incredible biodiversity urged the creation of a protected area. In August 2006, the São Tomé Obô Natural Park (STONP) became official, covering almost one third of the island. A buffer zone surrounding the park was later added for further protection (Direcção Geral do Ambiente 1999). Due to the lack of resources and enforcement capacity, illegal activities are still a regular sight within the protected area (Albuquerque et al. 2008). Our main goal is to assess how the STONP represents the island’s avifauna. We will start by modelling bird species richness and composition, in order to capture its spatial patterns, while paying special attention to the distribution of endemic and non-endemic species. Then, we will compare the distribution of bird diversity with the boundaries of the STONP to assess if the protected area includes an adequate representation of the multiple aspects of the island’s bird diversity.

METHODS Study Area São Tomé Island is in the Gulf of Guinea, Central Africa, and together with Príncipe Island forms the Democratic Republic of São Tomé and Príncipe. It is a small oceanic island, lying just north of the Equator and about 255 km west of Gabon. Covering only 857 km2, it has an incredible unique avifauna (Peet & Atkinson 1994; Leventis & Olmos 2009): out of 45 resident terrestrial species, 17 are singleisland endemics, 3 are endemic to the Gulf of the Guinea oceanic islands (Annobón, São Tomé and Príncipe) and 8 are endemic subspecies of widespread species (Jones & Tye 2006). The high endemism results from the island’s location relative to the African continent: close enough to allow frequent migration, but far enough to allow speciation by isolation (Melo 2006). The mountainous topography is responsible for strong environmental gradients, which still shape the distribution of natural and anthropogenic ecosystems. The island was almost entirely covered by forest when Portuguese navigators first discovered it and started its occupation in the late 15 th century. Nowadays, most lowland areas and some montane regions have been converted to plantations, while the best-preserved patches of native forest occur mostly in the rugged rainy areas in the south-west and centre of the island. This forest is surrounded by large extents of secondary forest, which result mostly from agricultural abandonment and logging activities. This forest is in turn enclosed by active shade plantations of coffee and cocoa, mixed with non-forested areas, such as oil palm monocultures, horticultures and savannahs (Exell 1944; Tenreiro 1961; Jones et al. 1991). Despite the increasingly humanized landscape, São Tomé is still dominated by forested ecosystems. The native forest is almost entirely included in the STONP, which covers approximately one third of the island (Albuquerque et al. 2008). There is growing awareness at local and international levels about

21

the value of the unique biodiversity of the island, and about the urgent need for effective conservation efforts. However, human pressure on natural resources is increasing fast, as shown by the decreasing area of native forest (Salgueiro & Carvalho 2001), and much conservation work is still needed. Data Compilation We gathered systematic observations of São Tomé bird species obtained in 300 point counts, during 2009 and 2010 (de Lima 2012) and in 1653 point counts sampled between 2013 and 2015 (BirdLife International São Tomé and Príncipe Initiative – BISTPI) (de Lima et al. 2017). All records were obtained during a 10-minute sampling sessions, in which an experienced observer recorded all birds detected aurally and visually, regardless of distance. A minimum distance of 200 meters was kept between point counts. All information was compiled in a bird species occurrence database, which included a GIS component. Field Methods Sampling design To identify under-sampled areas in previous studies from which we compiled data, we overimposed the bird occurrence database on the map of São Tomé Island, which was divided in 2x2 km quadrats. Each of these quadrats was subdivided in four 1x1 km quadrats, following the BISTPI methodology (de Lima et al. 2017). We eliminated all 2x2 km quadrats that had more than half of their area occupied by the ocean, and identified all quadrats that had at least five point counts sampled. The remaining 96 2x2 km quadrats, located mostly in non-forested low-altitude areas, were identified for surveying. Subsequently, we randomly ranked each of the 1x1 km quadrats in each of the larger quadrats for sampling, to determine sampling priority (de Lima et al. 2017). Bird sampling To complement the previously compiled bird database, we sampled 91 out of the previously identified 96 quadrats, between January and March 2017. Following previous work (de Lima 2012; de Lima et al. 2017), the quadrats were sampled by conducting five 10-minute point counts. The points were at least 200 meters apart, to ensure independence and that the environmental variability inside each quadrat was sampled in the proportion that they occurred within the quadrat. In each point count, we registered all bird species detected visually and aurally. To maximize the number of sampled points during our short sampling period, counts were made throughout the day. All bird data were compiled in a database totalling 263 1x1 km quadrats. All species that are aquatic or difficult to identify were excluded (Table S9), leaving a total of 36 species for the analyses. Point counts with zero presences for these species or with inconsistencies between the field land cover classification and the 2014 land use map were also excluded. Only five point counts in each sampled quadrat were considered to ensure a balanced sampling effort between quadrats and a good spatial distribution of sampling effort throughout the year (n = 187; Fig. 2.1). For each 1x1 km quadrat, an average point count was calculated based on the average of the coordinates of all five point counts. All bird species records found in each five point counts were considered for the average point count and later transformed into presence/absence data. Total species richness, endemic species richness and nonendemic species richness was calculated for every average point count.

22

Figure 2.1. São Tomé Island sampling locations. The lines in the background represent the 100 m elevation isolines. Each dot corresponds to the average point count for every 1x1 km quadrat sampled (n = 187).

Characterizing environmental variables To model and map species richness and compositional dissimilarity, we assembled geographically explicit information on altitude, ruggedness, slope, distance to the coast, topography, remoteness, rainfall and land use across São Tomé, using Quantum GIS v. 2.8.3 and v. 2.14.8 (Quantum GIS Development Team 2009a; Table S1 & S2). The variable altitude was derived from a 90 meters resolution Digital Elevation Model (DEM) (Salgueiro & Carvalho 2001; NASA Jet Propulsion Laboratory 2016; Fig. S1). Ruggedness and slope were also calculated from this DEM raster, using the “raster terrain analysis” QGIS plugin (Quantum GIS Development Team 2009b; Fig. S2 & S3). The slope was initially calculated in decimal degrees and then transformed to percentage. Distance to the coast was calculated as the minimum linear distance in decimal degrees between each pixel and the nearest coast line point, using the DEM and the QGIS “distance matrix” tool (Quantum GIS Development Team 2009a; Fig. S4). Topography was represented using a Topography Position Index (TPI) which allows comparing of each cell’s elevation to the mean elevation of a specified neighbourhood (Jenness 2007). TPI was calculated using the DEM and the “topography position index” tool in the QGIS GDAL algorithm provider (Quantum GIS Development Team 2009c) and a 0.05º radius neighbourhood, which allows for a good representation of terrain ruggedness and elevation in São Tomé. The continuous TPI thus obtained was transformed in a fivecategory discrete variable: flat areas (1), valleys (2), middle slopes (3), upper slopes (4) and ridges (5) (Fig. S5, S6 & S7). Still, given the nature of further analyses, the TPI variable was considered 23

continuous, reflecting an altitudinal gradient with 2 being lower than the referential flat areas (1) and 5 the highest situation. Remoteness is expressed as an index that translates the difficulty of movement through the landscape, and it was created using the “accumulated cost” QGIS GDAL algorithm provider (Quantum GIS Development Team 2009d). This index is a cost accumulated surface based on a friction surface derived from slope and weighted by the human population density (Tobler 1993; Instituto Nacional de Estatística 2016; Fig. S8 & S9). Rainfall was obtained by digitizing the island’s mean annual precipitation map in millimetres (Silva 1958; Fig. S10). The land use map was created based on 2014 satellite images (Google Earth 2017), supplemented by 2009-2017 field information (de Lima 2012; de Lima et al. 2017), 1970 land use map (de Carvalho Rodrigues 1974), military maps (Missão Hidrográfica de Angola e S. Tomé 1958), a 2011-13 preliminary land use map (S. Mikulane, unpublished data - see Fig. S11) and expert knowledge (Fig. S12). First, it was considered a four-category discrete variable: native forest (1), secondary forest (2), shade plantations (3) and non-forested areas (4). Later, this same variable was transformed in a continuous variable reflecting a gradient of habitat degradation: 1 being the pristine habitat and 4 the habitat with highest level of humanization. All variables were considered continuous and standardised to a common in raster grid, using the nearest neighbour sampling method and the TPI raster as a reference. This standardization was made using QGIS “align rasters” tool (Quantum GIS Development Team 2009a), and resulted in a pixel’s size of 0.000833º x 0.000833º and a raster with 359 x 471 cells. The average point count of each 1x1 km quadrat was characterized for each environmental variable using the “point sampling tool” QGIS plugin (Quantum GIS Development Team 2009e). Data Analysis All statistical analyses were made using R v. 3.3.2 in RStudio v. 1.0.143 (R Development Core Team 2017). Exploratory analysis Multicollinearity was tested using Spearman’s rank correlation coefficient, and visualized in a correlogram built using the “corrgram” package (Wright 2016; Part I, Section VIII). Remoteness index and ruggedness were excluded from further analyses, having correlation coefficients with land use and slope, respectively, equal to or higher than 0.8 (Fig. S17). To identify potential outliers and analyse variance homogeneity, boxplots were drawn for each environmental variable, using the “vegan” package (Oksanen 2015). No outliers were removed from the analysis. Under-dispersion was tested for species richness, endemic species richness and non-endemic species richness, using the “AER” package (Kleiber & Zeileis 2017). The data were divided into a training and a testing set, using the “caTools” package (Tuszynski 2014): 70% of the quadrats were used to create the models and 30% to validate them. Generalized linear models Three generalized linear models (GLMs) with poisson distribution were created to explain total, endemic and non-endemic species richness, respectively (Part III, Section VIII). For each GLM, all possible combinations of explanatory variables were ranked based on the Akaike Information Criterion corrected for small sample sizes (AICc), using the “dredge” function from “MuMIn” package (Barton 2016). The models were validated using the testing data. Goodness of fit was analysed with the McFadden’s index in the “pscl” package and with the Residual Deviance (Jackman et al. 2015). Validation was also explored by plotting the Pearson and Deviance residuals

24

against the predicted values, using the “stats” package (R Development Core Team 2017; Table S10 & Fig. S18). To identify which variables best explain the species richness models, we ran the “model averaging” function from the “MuMIn” package to obtain relative variable importance (RVI). To evaluate the response of total, endemic and non-endemic species richness to each continuous variables, we calculated the Spearman’s rank correlation coefficient (Table S11). Finally, a map with predictions from each of the three fitted models was generated, using the “raster” package and the environmental variables in raster format (Hijmans et al. 2016). Generalized dissimilarity modelling Generalized dissimilarity modelling (GDM) was used to map beta diversity using the “gdm” package (Manion et al. 2017; Part IV, Section VIII). GDM compares community composition and environmental variables at pairs of sites to predict compositional difference as a function of environmental difference, extrapolating the prediction beyond surveyed sites. The resulting models give a spatially continuous prediction of turnover, and thus of the spatial structure of diversity. To quantify the compositional dissimilarity between different sites, a dissimilarity matrix was calculated using the Bray–Curtis dissimilarity statistics. The model fit was examined by the total deviance explained by the model and by plotting the observed dissimilarities against the predicted values (Fig. S19). To assess the model significance of each variable a significance test was made using 100 permutations. The significance testing in the “gdm” package is still in the early phase of development, and it is therefore rather computationally intensive. The variable importance was measured as the percent change in deviance explained by the full model and the deviance explained by a model fit with that variable permuted. The significance was estimated using the bootstrapped p-value when the variable was permuted (Table S12 & S13). A robust assessment of model’s capacity to generate predictions was made by validating the independent testing set. A k-fold cross-validation was used to test the predictive accuracy of the model, using 100 permutations. The output of the cross-validation was the correlation between the observed and predicted compositional dissimilarities, for the testing set of sites (Fig. S20). To generate spatially explicit GDM model predictions for São Tomé Island, we created transformed environmental layers for each predictor using the spline functions from the fitted model. A principal components analysis (PCA) was made on the dissimilarities between classes to reduce dimensionality and assign the first three components to an RGB colour palette (red, green and blue). This way, similar colours represent a similar avifauna composition. The output was a raster image composed of three single rasters representing the three ordination axes. The relative importance of each predictor variable was determined by summing the coefficients of the I-splines from the fitted generalized dissimilarity model (Table S14). The response curves were used to evaluate the response of predicted compositional dissimilarity to each predictor variable (Fig. S21). Generalized dissimilarity model categorization An unsupervised classification method was applied to the continuous GDM, using modified kmeans classification in the Whitebox Geospatial Analysis Tools v. 3.4.0 “Image Classification” menu (Lindsay 2016; Fuss et al. 2016). The algorithm was limited to Euclidian distances smaller than 75, a value that ensured the creation of robust composition categories. We allowed for a maximum of 50 iterations, a 2% pixel class change threshold and a 500 minimum number of pixels per class. The initial cluster centres were generated randomly.

25

Assessing the adequacy of the Obô Natural Park to represent São Tomé bird diversity We assessed how the STONP represented two distinct aspects of the island’s bird diversity: species richness and composition. To explore the differences in species richness inside and outside the STONP, we used the “random points” QGIS tool in “vector” menu (Quantum GIS Development Team 2009a) to sample 3996 random points from the total, endemic and non-endemic species richness maps previously created. Total and endemic species richness were calculated for each GDM class. Total and endemic average species richness were calculated for each quadrat. These were used to calculate the proportion of endemic species (number of endemic species / total number of species) and the frequency of endemic species (number of endemic species detections / total number of detections) for each quadrat. Both median and quartiles were plotted in a single scatterplot to explore the relation between endemic species proportion and detection rate (Part V, Section VIII). Finally, the percentage of each GDM classes included inside the STONP was calculated, using “count raster cells” QGIS plugin.

RESULTS Bird data used to map total, endemic and non-endemic species richness was under dispersed (total: z = -14.375, p = 2.2 x 10-16; endemic: z = -13.498, p = 2.2 x 10-16; non-endemic: z = -15.311, p = 2.2 x 10-16). Modelling bird species richness Total species richness was highest in the centre south of São Tomé Island. These particularly rich areas were located inside the STONP. Some of the poorest areas were also found inside the park, coinciding with higher altitudes and steeper slopes (Fig. 2.2). Endemic species richness pattern was clear: richer areas located inside the protected area with the number of species declining with greater proximity to the coast. Non-endemic species followed the opposite pattern: areas with a lower number of species were found inside the park which progressively increased with humanization and towards the coast. In the south-east of the island, an area can be identified in all three predictive maps, characterized by a smaller number of species than surrounding areas, and it corresponds to a large oil palm plantation. In the total species richness model, none of the environmental variables was statistically significant and relative variable importance (RVI) was always smaller than 0.50. To model endemic species richness, land use was the most important variable. On the other hand, several environmental variables were significant and important to the distribution of non-endemic species richness. The most important variable was rainfall, followed by altitude and land use. Endemic species responded negatively to more intensive land uses, but positively to forested habitats, like native and secondary forests. Whereas non-endemic species had an opposite response and therefore a strong connection to non-forested habitats and humanized landscapes (Table S11).

26

Figure 2.2. Predictive maps of (a) total species richness, (b) endemic species richness and (c) non-endemic species richness, shown in contrast to the boundaries of the Obô Natural Park and buffer zone.

27

Bird species compositional dissimilarity Generalized dissimilarity modelling (GDM) was used to identify areas with similar avifauna composition (Fig. 2.3). The GDM allowed explaining 43.6% of the deviance. The most important environmental predictor was land use, followed by rainfall and altitude (Table S14). A larger rate of species turnover was found for high values of land use, meaning that the biggest changes in bird community composition occurred in humanized habitats, like shade plantations and non-forested areas. In forested habitats, the species composition was similar (Fig. S21). Smaller values were associated to bigger species composition turnover rates for slope, altitude, rainfall and TPI.

c)

d)

3

Class

2 1 5 4

0.0

100.0 Euclidean distance

200.0

Figure 2.3. (a) Continuous and (b) categorical composition dissimilarity maps, as obtained from generalized dissimilarity modelling (GDM). (c) Links and (d) distances between the five GDM classes obtained using modified k-means classification. Class 1 corresponds to the large oil palm monoculture, class 2 to the open areas surrounded by agro-forest habitats in slightly wetter regions, class 3 to the most humanized habitats in the driest parts of the island, class 4 to mixed of forested habitats like shade plantations and secondary forest in the north-east and class 5 to secondary and native forests in the centre and south

28

From the continuous GDM, a categorical map was produced and five classes were identified. The first class to be separated was class 1, suggesting the existence of a very distinctive bird species assemblage in the oil palm plantation, previously identified in all species richness maps (Fig. 2.3). Subsequently, there was also an obvious separation between bird assemblages that inhabit more forested habitats (classes 5 and 4) and those living in non-forested areas (classes 2 and 3). Is the São Tomé Obô Natural Park adequate to protect the island’s avifauna? Of the 3996 random points generated to assess the number of species, both total, endemic and nonendemic, 1097 were located in the park. The predicted number of species was similar inside and outside

Figure 2.4. Total, endemic and non-endemic species richness inside (In) and outside (Out) Obô Natural Park. The boxplots represent the median (thick line), the first and third quartiles (box), the extremes (whiskers) and the outliers (dots).

29

STONP (inside STONP: x̄ = 13.6; outside STONP: x̄ = 13.8). Even so, a bigger range of values was found inside the park (Fig. 2.4), suggesting a wider variety of areas inside the protected area. Endemic species richness had higher values inside the STONP (inside STONP: x̄ = 11.4; outside STONP: x̄ = 9.3), while non-endemic species richness presented the opposite pattern (inside STONP: x̄ = 2.3; outside STONP: x̄ = 4.5). There were no major differences in average species richness between all five GDM classes (Table 2.1). However, there were several differences in average endemic species richness: class 3 showing the lowest value (4.9), followed by classes 1 (8.5) and 2 (9), and the remaining having similar values. There were also differences in terms of total number of species and total number of endemic species. Classes 1 and 2 had identical values, namely the lowest total number of species (20) and an intermediate total number of endemics (12). Class 3 had an intermediate total number of species (23), but the lowest total number of endemics (8). Classes 4 and 5 had the highest total number of species (28), but class 5 had a higher total number of endemics (19 against 15). The class 5 of the GDM had by far the largest area included inside the STONP, having more than half of its area protected (54.9%). The remaining four classes had only 3% or less of their area protected. Table 2.1. Species richness and endemic species richness estimated for each average point inside 1x1 quadrats, called, respectively, species and endemic richness point estimate. Species richness and endemic species richness calculated for each GDM class (1 to 5). Percentage of class included in Obô Natural Park.

1 2 8.5 9 13.3 12.7 12 12 20 20 1.9 1.9

Endemic richness point estimate Species richness point estimate Endemic species richness Species richness % Class Protected

Class 3 4 5 4.9 10.1 10.7 13.2 13.5 14.2 8 15 19 23 28 28 2.7 3.1 54.9

5 4 2 1

3

Figure 2.5. Proportion of endemic species and frequency of endemic species for each GDM class (1 to 5). The bars represent the first and third quartiles of the median values estimated for each quadrat.

30

Regarding endemic species, four different groups of classes can be found (Fig. 2.5): class 3 was clearly distinct from other classes, by having fewer endemic species; classes 1 and 2 had identical intermediate proportions and frequencies of endemic species; class 4 had a proportion of endemics only slightly higher than the previous group, but significantly higher frequencies; class 5 had the highest proportion and frequency of endemics.

DISCUSSION We modelled and mapped bird species richness and composition to understand if the STONP represented the island’s bird diversity. We found that the STONP did not protect necessarily the richest assemblages, but did protect those that were richest in endemic species. Contrasting responses of endemic and non-endemic species to the environment Bird species richness presented a narrow gradient that was not evenly distributed throughout São Tomé and a complex pattern (Fig. 2.2 a), which was not strongly related to any of the environmental variables used in the modelling. The STONP included the richest, but also the poorest areas for avifauna in the island. The highest values of species richness were found in the centre-south of the island, inside native forest, and were almost entirely included in the STONP. Right next to them, two large speciespoor areas can be identified, also mostly included inside the park: the São Tomé Peak and surrounding high altitude areas, and the Cabumbé Peak and the nearby Quija and Xufexufe river valleys. Both are located in the heart of São Tomé’s rainforest, and represent remote mountainous landscapes. Endemic species richness was clearly associated with the best-preserved forest in São Tomé (Fig. 2.2 b). This result coincides with previous findings, indicating that endemic species are associated with forest-dominated habitats and avoid humanized landscapes (de Lima et al. 2012). The highest values of endemic species richness also tended to occur further away from the coast line. These endemic-rich forests were almost entirely inside STONP (Albuquerque et al. 2008). Secondary forests are found mostly around native forests, both inside the STONP and in the buffer zone. Although they shelter less endemic species than native forests, they seem to be acting as a transition zone to more humanized areas (Atkinson et al. 1991), protecting the STONP from pervasive threats (Dallimer et al. 2009). The greatest number of non-endemic species was found in the more humanized habitats near the coast in the north-east of the island (Fig. 2.2 c). A pattern that is rather contrasting to that of the endemic species richness. The northern exclave of the STONP is the only protected area including areas rich in non-endemic bird species. Most of these species are small granivores assumed to have been introduced to the island, well-known for being associated with non-forested areas under strong anthropogenic influence (Jones & Tye 2006). Since non-endemic birds tend to avoid forested areas and to use distinct food resources, they do not seem to be competing with the endemics. Instead, the gradient between endemic and non-endemic dominated bird assemblages seems to be facilitated by the gradient of native forest degradation (Didham et al. 2007). Comparing the distribution of total, endemic and non-endemic species richness (Fig. 2.2), a distinct area can be seen in all maps, located in the south-east of the island. This area corresponds to a large oil palm plantation, characterized by having few bird species, and notably fewer endemics and, proportionally, more non-endemics than the surrounding landscape. Most endemics rely on complex forest environments and do not find the required resources to subsist in these monocultures (Turner et al. 2008; Nájera & Simonetti 2010). Being mostly granivores, the non-endemics also struggle to persist in these plantations, due to the severely impoverished vegetation. Moreover, the extremely wet conditions are not favourable to the production of grains on which they often rely. Studies suggest that 31

the spontaneous development of understory vegetation should be allowed in these oil palm plantations, to function as corridors between natural ecosystems and to promote the appearance of a more varied avifauna, namely of insectivore birds that contribute to pest control (Savilaakso et al. 2014). The maps of total, endemic and non-endemic species richness (Fig. 2.2) also show that an apparent lack of overall obvious pattern in total bird species richness is concealed by the contrasting distribution patterns of endemic and non-endemic species richness. Furthermore, the contrasting response of endemic and non-endemic species richness to land use, also obscures the importance of this environmental variable in explaining patterns of bird diversity in São Tomé (de Lima et al. 2012; Table S11). Species assemblages vary mostly in response to habitat humanization Modelling species composition dissimilarity revealed that bird assemblages were strongly determined by the same humanization gradient that had been identified when analysing species richness: the bird community associated with lowland humanized areas changes progressively towards the forestdominated landscapes, culminating in the native forest (Fig. 2.3 a). This pattern can be seen in the north and south regions of the island, both of which hold rather distinctive species assemblages linked to a wide rainfall gradient. The large oil palm plantation in the south-east once more reveals a very distinctive species assemblages. Land use was again considered the most important variable, followed by rainfall and altitude, which are also intrinsically linked to the distribution of land use in São Tomé Island (Peet & Atkinson 1994; Table S14). The analyses showed a bigger species composition turnover within non-forested habitats (Fig. 2.3 c & S21). Composition response curve to land use suggested that bird assemblages were more distinct within humanized than in natural habitats, as already indicated by previous studies (de Lima et al. 2012). This pattern has been associated with stronger differences between intensive agricultural areas, holding a simplified vegetation, compared to natural ecosystems (Waltert et al. 2004; Rocha 2006). The categorical GDM (Fig. 2.3 b & 2.3 c) separates forested habitats (classes 4 and 5) and nonforested habitats (classes 1, 2 and 3), further supporting that land use is vital in the differentiation of bird species assemblages (de Lima et al. 2014). Class 1 represents the most distinctive bird community to be isolated, and corresponds to the large oil palm plantation already identified in the species richness maps. Although located in the south, where rainfall is much higher, this class is closer to classes 2 and 3, all of which corresponding to non-forested habitats, where the non-endemic species prevail. Classes 2 and 3 represent lowland non-forested areas, where non-endemic species are frequent. Class 3 includes the most humanized habitats, in the driest parts of the island, while the similar class 2 appears in open areas surrounded by agro-forest habitats, in slightly wetter regions. Class 4 is a mixed of forested habitats like shade plantations and secondary forest in the north-east of São Tomé. Class 5 covers secondary and native forests in the centre and south, and holds without a doubt the community with the highest proportion of endemic species. There is an obvious species turnover from forests, where the endemics are clearly dominant, to more open habitats, where non-endemics become more numerous (Lima et al. 2012). Most nonendemics are small granivores, believed to have been introduced (Leventis & Olmos 2009), which suggests that land use change might be promoting the spread of non-native species (Rocha 2006). Islands are known to have a limited pool of species available to colonize disturbed areas (Atkinsons et al. 1991). In São Tomé, urban and non-forested agricultural fields are the most transformed and humanized areas. They have been widely colonized by introduced granivore species, since these are better adapted to nonforested habitats than the native, mostly endemic avifauna (de Lima et al. 2012; Ndang’ang’a et al. 2014). On the other hand, the introduced granivores seem to be much less frequent in forested habitats, including the cocoa and coffee shade plantations, even though the vegetation of these agroforestry

32

systems is almost exclusively composed by introduced plant species (de Lima et al. 2014). Our results seem to provide further support for the hypothesis that the landscape being dominated by forested habitats is involved in maintaining and ensuring the overall dominance of the endemic avifauna (de Lima et al. 2012, 2017). Other factors, such as hunting and the introduction of non-avian forest species might be affecting the avifauna. Hunting has been shown to affect the distribution of birds, and notably large frugivores (Carvalho et al. 2015a; Carvalho et al. 2015b). The introduction of non-avian vertebrates, such as feral pigs Sus domesticus and cats Felis catus, rats Rattus sp., and the mona monkeys Cercopithecus mona, have also been implied in having negative impacts in the endemic-rich native avifauna, namely through predation and habitat changes (Atkinson et al. 1991; Dutton 1994). Despite little empirical evidence, both of these threats, are linked to the land use degradation gradient, which is, without a doubt the key determinant of bird diversity in São Tomé. Is the São Tomé Obô Natural Park adequate to protect the island’s bird diversity? The boundaries of the STONP were established, mostly based in a habitat field survey, and our work represents the first assessment of its adequacy to protect the island’s biodiversity. To do so, we evaluated if bird species richness and assemblage composition was well represented within the boundaries of the protected area, paying special attention to the endemic and non-endemic components of avifauna. The STONP covered some of the highest values of total species richness, but also some of the lowest, resulting in no significant differences when compared with areas outside the park (Fig. 2.4). However, endemic species richness was clearly higher inside the STONP, and non-endemic richness higher outside. These results show that using species richness on its own can be misleading as an indicator of conservation value and that it should be used in combination with other metrics (Le Saout et al. 2013). These results are also encouraging, since the park limits seem to be well established for the protection of the endemic species, which are the most threatened (IUCN 2017) and the most interesting species, in terms of global conservation goals (de Lima et al. 2017). The STONP is almost entirely composed by areas covering the class 5 we identified by GDM, which represents the richest bird assemblage, having the highest number of species and being mostly composed of endemic (Atkinson et al. 1991; Fig. 2.5). This class includes almost all native forest and is the bird assemblage best represented inside the STONP (54.9%; Table 2.1). All other classes have a poor representation inside the protected area, regardless of how many endemics they hold. This is of little concern in terms of global species protection, since all endemic and threatened species are included in class 5. Final remarks The STONP did not represent the diversity of São Tomé avifauna very well, but focused on the endemic and threatened species, arguably those of higher global conservation interest. The boundaries of the STONP were primarily defined based on native forest distribution, natural barriers and small levels of human pressure, but coincide with the distribution of the bird assemblages that are richest in endemics (Albuquerque et al. 2008). This match is due to the key determinants of bird diversity patterns being the same environmental factors that were used to define STONP boundaries (Rocha 2006; de Lima et al. 2012; Chapter I). Our work also highlighted the importance of secondary forests for the avifauna of São Tomé, holding a high proportion of endemic species and providing a valuable buffer zone for many of the small-ranged endemics. Therefore helping to mitigate many negative impacts of human activities, like hunting and logging (Atkinson et al. 1991; de Lima et al. 2017). 33

We advocate STONP zonation should be revised, taking into account the same factors used to define the boundaries of the protected area, and also the current knowledge about bird species distribution, especially that of those of higher conservation interest. This way, key STONP will gain a higher level of protection, contributing to the conservation of threatened small-ranged endemic species, like the Dwarf Ibis Bostrychia bocagei (Dallimer et al. 2009; Leventis & Olmos 2009; Ndang’ang’a et al. 2014; de Lima et al. 2017). At last, STONP provides a good example that areas of higher conservation interest can be identified using the distribution of natural habitats and human population. Protected areas should prioritize natural ecosystems supporting high species richness and high proportions of endemic and threatened species. However, in most cases this information is not available when the boundaries are being defined. Our results suggest focusing first on identifying key natural ecosystems, and then zoning based on the distribution of the different biodiversity components, when these become better known, eventually extending the initial boundaries. This strategy allows for assessing if protected areas are still achieving their key conservation goals, and adjust them while knowledge on their biodiversity increases.

34

FINAL CONSIDERATIONS

Anthropogenic land use change is considered the biggest threat to global biodiversity (Foley et al. 2005; Stork 2009; Szabo et al. 2012). Understanding how human actions affect biodiversity is therefore a first step to minimize and prevent further impacts on species and ecosystems. Human population is expected to grow exponentially within upcoming years, making it crucial to learn how to coexist and share world ecosystems and natural resources (Cincotta et al. 2000; Luck 2007). Our study exemplifies how human occupation can influence species distribution. São Tomé is a small, highly forested island with strong natural and anthropogenic gradients (Salgueiro & Carvalho 2001; Jones & Tye 2006), both of which shape the distribution of species and ecosystems. We have shown that the strong gradient of land use intensification is the main responsible for the changes found in bird species assemblages: from forest endemic species, extremely associated with best-preserved forests, to open habitat non-endemic species, commonly found in more intensively managed land uses (Rocha 2008; de Lima et al. 2012). In São Tomé, the forest-dominated landscape ensures and maintains the overall dominance of endemic avifauna (de Lima et al. 2012). Given these results, forested patches are vital for the persistence of endemic birds inside a landscape increasingly dominated by intensive land uses. Thus, we recommend the protection of the remaining native forest and the expansion or improvement of secondary forest, to provide a landscape matrix more suitable for the endemic species. Non-native birds will have the opposite response, since they tend to avoid forested habitats (Atkinson et al. 1991). Therefore, increasing the forest cover will have the additional benefit of preventing the spread of introduced birds throughout the island. The establishment of protected areas is one the most important and common conservation measures (Myers et al. 2000; Buchanan et al. 2011; Le Saout et al. 2013). The STONP was created based on native forest distribution, natural barriers and small levels of human pressure (Albuquerque et al. 2008). With our study, we concluded that most of the areas with endemic-rich assemblages are well represented inside the park, even though these do not necessarily correspond to the richest bird assemblages. Many endemic species are considered threatened and their reliance on forested habitats is a growing concern, given the increasing destruction and degradation of São Tomé native forests (Ndang’ang’a et al. 2014; IUCN 2017). The conservation of São Tomé endemic bird species relies on the protection and preservation of the remaining native forest. Given the limited resources in São Tomé, STONP is not receiving the active conservation management required, also because most environmental laws do not come with legal force (Albuquerque et al. 2008). We emphasize the need to transform these environmental laws into active conservation actions on the field, setting up a monitoring program to stop or at least minimize the still ongoing threats inside and nearby the park, e.g. uncontrolled hunting, forest burning and intensive logging (Peet & Atkinson 1994; Dallimer et al. 2009; Carvalho et al. 2015a; Carvalho et al. 2015b; de Lima et al. 2017). The expansion and management of secondary forests for conservation could improve the quality of ecosystems in the STONP buffer zone, which has an important role in the conservation of endemic bird species, helping to minimize possible human impacts inside the park and surrounding areas, while providing additional habitat to many of the endemics. The current study is an important basis for future studies, and to establish specific monitoring activities and conservation strategies. However, further research is needed to gain a more detailed knowledge about the distribution of each bird species, namely regarding seasonality and single species response to forest degradation. That information would enable us to define target species actions, more adequate to each species ecological requirements, which is especially important for the most threatened, such as the Dwarf Ibis, the São Tomé Fiscal and the São Tomé Grosbeak. We also highlight the need to gain a better understanding of the impact of other threats, such as hunting and introduced species. 35

REFERENCES (Following the citation rules of Conservation Biology) Albuquerque C, Cesarini D, Tagliabue LC. 2008. Plano de manejo do Parque Nacional Obô de São Tomé 2009/2014. ECOFAC IV, São Tomé. Andren H. 1994. Effects of habitat fragmentation on birds and mammals in landscapes with different proportions of suitable habitat: a review. Oikos 71:355-366. Arponen A, Moilanen A, Ferrier S. 2008. A successful community-level strategy for conservation prioritization. Journal of Applied Ecology 45:1436-1445. Assandri G, Bogliani G, Pedrini P, Brambilla M. 2017. Land use and bird occurrence at the urban margins in the Italian Alps: implications for planning and conservation. North-Western Journal of Zoology 13:77-84. Atkinson P, Peet N, Alexander J. 1991. The status and conservation of the endemic bird species of São Tomé and Príncipe, West Africa. Bird Conservation International 1:255-282. Balmford A, Bond W. 2005. Trends in the state of nature and their implications for human well-being. Ecology Letters 8:1218-1234. Barton K. 2016. MuMIn: Multi-model inference. R package version 1.15.6. Available from https://cran.rproject.org/web/packages/MuMIn/MuMIn.pdf (accessed May 2017). Benton TG, Vickery JA, Wilson JD. 2003. Farmland biodiversity: is habitat heterogeneity the key? Trends in Ecology and Evolution 18:182-188. Blackburn TM, Cassey P, Duncan RP, Evans KL, Gaston KJ. 2004. Avian Extinction and Mammalian Introductions on Oceanic Islands. Science 305:1955-1958. Blair RB. 1996. Land use and avian species diversity along an urban gradient. Ecological Applications 6:506519. Brooks TM, Pimm SL, Oyugi JO. 1999. Time Lag between Deforestation and Bird Extinction in Tropical Forest Fragments. Conservation Biology 13:1140-1150. Brooks TM, Mittermeier RA, Fonseca GAB, Gerlach J, Hoffmann M, Lamoreux JF, Mittermeier CG, Pilgrim JD, Rodrigues ASL. 2006. Global Biodiversity Conservation Priorities. Science 313:58-61. Brown JH. 1984. On the relationship between abundance and distribution of species. The American Naturalist 124:255-279. Brown JL, Cameron A, Yoder AD, Vences M. 2014. A necessarily complex model to explain the biogeography of the amphibians and reptiles of Madagascar. Nature Communications DOI:10.1038/ncomms6046.

36

Buchanan GM, Donald PF, Butchart SHM. 2011. Identifying Priority Areas for Conservation: A Global Assessment for Forest-Dependent Birds. PLoS ONE (e29080) DOI:10.1371/journal.pone.0029080 Cadenasso ML, Pickett STA. 2001. Effect of Edge Structure on the Flux of Species into Forest Interiors. Conservation Biology 15:91-97. Carvalho M, Palmeirim J, Rego FC, Sole N, Santana A, Fa J. 2015a. What motivates hunters to target exotic or endemic species on the island of São Tomé, Gulf of Guinea? Oryx 49:278-286. Carvalho M, Rego F, Palmeirim J, Fa J. 2015b. Wild meat consumption on São Tomé Island, West Africa: implications for conservation and local livelihoods. Ecology and Society 20:27. Carvalho M. 2015. Hunting and Conservation of Forest Pigeons in São Tomé (West Africa). PhD dissertation. Instituto Superior de Agronomia da Universidade Técnica de Lisboa. Chacea JF, Walsh JJ. 2006. Urban effects on native avifauna: A review. Landscape and Urban Planning 74:4669. Cincotta RP, Wisnewski J, Engelman R. 2000. Human population in the biodiversity hotspots. Nature 404:990-992. Dallimer M, King T. 2007. Habitat preferences of the forest birds on the island of Prıíncipe, Gulf of Guinea. African Journal of Ecology 46:258-266. Dallimer M, King T, Atkinson RJ. 2009. Pervasive threats within a protected area: conserving the endemic birds of São Tomé, West Africa. Animal Conservation 12:209-219. de Carvalho Rodrigues FM. 1974. S. Tomé e Príncipe sob o ponto de vista agrícola. Ministério do Ultramar, Junta de Investigações Científicas do Ultramar, Lisboa. de Lima RF. 2012. Land use management and the conservation of endemic species in the island of São Tomé. PhD dissertation. Lancaster University. de Lima RF, Dallimer M, Atkinson PW, Barlow J. 2012. Biodiversity and land use change: understanding the complex responses of an endemic-rich bird assemblage. Diversity and Distributions 19:411-422. de Lima RF, Viegas L, Sole N, Soares E, Dallimer M, Atkinson PW, Barlow J. 2014. Can Management Improve the Value of Shade Plantations for the Endemic Species of São Tomé Island? Biotropica 46:238-247. de Lima RF, Maloney E, Simison WB, Drewes R. 2015. Reassessing the conservation status of the shrew Crocidura thomensis, endemic to São Tomé Island. Oryx 50:360-363. de Lima RF, Dunn JC, Ward-Francis A, Buchanan GM. 2017. Distribution and habitat associations of the critically endangered bird species of São Tomé Island (Gulf of Guinea). Bird Conservation International DOI:10.1017/S0959270916000241. Didham RK, Tylianakis JM, Gemmell NJ, Rand TA, Ewers RM. 2007. Interactive effects of habitat modification and species invasion on native species decline. Trends in Ecology and Evolution 22:489-496. 37

Diniz MA, Fernandes R, Martins ES, Moreira I, Paiva J. 2002. Carta de zonagem agro-ecológica e da vegetação de São Tomé e Principe. Garcia da Orta 15:1-72. Direcção Geral do Ambiente. 1999. Regulamento sobre o processo de avaliação do impacto ambiental (Decreto n.37/1999). Ministério dos Recursos Naturais e Ambiente, São Tomé e Príncipe. Direcção Geral do Ambiente. 2006. Lei do Parque Natural do Obô (Lei n.6/2006). Ministério dos Recursos Naturais e Ambiente, São Tomé e Príncipe. Dutton J. 1994. Introduced mammals in São Tomé and Príncipe: possible threats to biodiversity. Biodiversity and Conservation 3:927-938. Elith J, Leathwick RJ. 2009. Species Distribution Models: Ecological Explanation and Prediction Across Space and Time. Annual Review of Ecology, Evolution, and Systematics 40:677-697. Exell AW. 1944. Catalogue of the vascular plants of S. Tomé (with Príncipe and Annobon). Trustees of the British museum. London. Eyzaguirre PB. 1986. The Ecology of Swidden Agriculture and Agrarian History in São Tomé. Cahiers d'Études Africaines 26:113-129. Foley JA. 2005. Global Consequences of Land use. Science 309:570-574. Ferrier S, Manion G, Elith J, Richardson K. 2007. Using generalized dissimilarity modelling to analyse and predict patterns of beta diversity in regional biodiversity assessment. Diversity and Distributions 13: 252-264. Fishpool LDC, Evans MI. 2001. Important Bird Areas in Africa and associated islands: Priority sites for conservation. Pisces Publications and BirdLife International, Newbury, United Kingdom. Fitzpatrick MC, Sanders NJ, Normand S, Svenning JC, Ferrier S, Gove AD, Dunn RR. 2013. Environmental and historical imprints on beta diversity: insights from variation in rates of species turnover along gradients. Proceedings of the Royal Society DOI:10.1098/rspb.2013.1201. Frynas JG, Wood G, de Oliveira RMSS. 2003. Business and Politics in São Tomé e Príncipe: From Cocoa Monoculture to Petro-State. African Affairs 102:51-80. Fuss CE, Berg AA, Lindsay JB. 2016. DEM fusion using a modified k-means clustering algorithm. International Journal of Digital Earth 9:1242-1255. Gardner TA, Ribeiro-Junior AR, Barlow J, Ávila-Pires TCS, Hoogmoed MS, Peres CA. 2007. The Value of Primary, Secondary, and Plantation Forests for a Neotropical Herpetofauna. Conservation Biology 21:775787. Gardner TA, Barlow J, Chazdon R, Ewers RM, Harvey CA, Peres CA, Sodhi NS. 2009. Prospects for tropical forest biodiversity in a human-modified world. Ecology Letters 12:561:582.

38

Google Earth. 2017. Google Earth. Available from https://earth.google.com/web/@25.00247296,20.30159305,6487.15542149a,11993103.11916947d,35y,56.66 304802h,22.67374979t,-0r (accessed May 2017). Guisan A, Zimmermann NE. 2000. Predictive habitat distribution models in ecology. Ecological Modelling 135:147-186. HBW Alive. 2017. Handbook of the Birds of the World Alive. Available from http://www.hbw.com/ (accessed September 2016). Hijmans RJ, van Etten J, Cheng J, Mattiuzzi M, Sumner M, Greenberg JA, Perpinan Lamigueiro O, Bevan A, Racine EB, Shortridge A. 2016. raster: Geographic Data Analysis and Modeling. R package version 2.5.8. Available from https://cran.r-project.org/web/packages/raster/raster.pdf (accessed May 2017). Hughes JB, Daily GC, Ehrlich PR. 2002. Conservation of tropical forest birds in countryside habitats. Ecology Letters 5:121-129. Instituto Nacional de Estatística. 2016. Instituto Nacional de Estatística. Available from http://www.ine.st/ (accessed November 2016). IUCN. 2017. The IUCN Red List of Threatened Species version 2017-1. IUCN (International Union for the Conservation of Nature). Available from http://www.iucnredlist.org/ (accessed April 2017). Jackman S, Tahk A, Zeileis A, Maimone C, Fearon J. 2015. pscl: Political Science Computational Laboratory, Stanford University. R package version 1.4.9. Available from https://cran.rproject.org/web/packages/pscl/pscl.pdf (accessed May 2017). Jenness JS. 2007. Some Thoughts on Analyzing Topographic Habitat Characteristics. Jenness Enterprises, Flagstaff, United States of America. Available from http://www.jennessent.com/downloads/topographic_analysis_online.pdf (accessed October 2016). Johnson TH, Stattersfield AJ. 1990. A global review of island endemic birds. Ibis 132:167-180. Jones PJ, Burlison JP, Tye A. 1991. Conservação dos ecossistemas florestais na República Democrática de São Tomé e Príncipe. International Union for the Conservation of Nature and Natural Resources, Gland, Switzerland. Jones P, Tye A. 2006. The birds of São Tomé and Príncipe, with Annobón: islands of the Gulf of Guinea. British Ornithologists’ Union, Oxford. Kleiber C, Zeileis A. 2017. AER: Applied Econometrics with R. R package version 1.2.5. Available from https://cran.r-project.org/web/packages/AER/AER.pdf (accessed May 2017). Lawton JH. 1999. Are There General Laws in Ecology? Oikos 84:177-192.

39

Le Saout S, Hoffmann M, Shi Y, Hughes A, Bernard C, Brooks TM, Bertzky B, Butchart SHM, Stuart SN, Badman T, Rodrigues ASL. 2013. Protected Areas and Effective Biodiversity Conservation. Science 342:803-805. Leventis A, Olmos F. 2009. As aves de São Tomé e Príncipe: Um guia fotográfico - The birds of São Tomé e Príncipe: A photoguide. Aves e Fotos Editora, São Paulo. Lindsay JB. 2016. Whitebox GAT: A case study in geomorphometric analysis. Computers & Geosciences 95:75-84. Luck GW. 2007. A review of the relationships between human population density and biodiversity. Biological Reviews 82:607:645. Mackey BG, Lindenmayer DB. 2001. Towards a hierarchical framework for modelling the spatial distribution of animals. Journal of Biogeography 28:1147-1166. Maestas JD, Knight RL, Gilbert WC. 2003. Biodiversity across a Rural Land use Gradient. Conservation Biology 17:1425-1434. Manion G, Lisk M, Ferrier S, Nieto-Lugilde D, Mokany K, Fitzpatrick MC. 2017. gdm: Generalized Dissimilarity Modeling. R package version 1.3.3. Available from https://cran.rproject.org/web/packages/gdm/gdm.pdf (accessed May 2017). Manne LL, Brooks T, Pimm SL. 1999. Relative risk of extinction of passerine birds on continents and islands. Nature 399:258-261. Margarido N. 2015. Seleção de habitat pela galinhola Bostrychia bocagei, ave criticamente ameaçada e endémica da ilha de São Tomé. MSc dissertation. Escola de Ciências e Tecnologia da Universidade de Évora. McKinney ML. 2006. Urbanization as a major cause of biotic homogenization. Biological Conservation 127:247-260. Melo M. 2006. Bird speciation in the Gulf of Guinea. PhD dissertation. University of Edinburgh. Melo M, Stervander M, Hansson B, Jones PJ. 2017. The endangered São Tomé Grosbeak Neospiza concolor is the world’s largest canary. Ibis 159:673-679. Miller EC, Sellas AB, Drewes RC. 2012. A new species of Hemidactylus (Squamata: Gekkonidae) from Príncipe Island, Gulf of Guinea, West Africa with comments on the African-Atlantic clade of Hemidactylus geckos. African Journal of Herpetology 61:40-57. Missão Hidrográfica de Angola e S. Tomé. 1958. Levantamento aerofotogramétrico: Carta de S. Tomé. Ministério do Ultramar, Junta de Investigações Científicas do Ultramar, Lisboa. Myers N, Mittermeier RA, Mittermeier CG, Fonseca GAB, Kent J. 2000. Biodiversity hotspots for conservation priorities. Nature 493:853-858.

40

NASA Jet Propulsion Laboratory. 2016. Shuttle Radar Topography Mission. NASA. Available from http://www2.jpl.nasa.gov/srtm/ (accessed September 2016). Naidoo R. 2004. Species richness and community composition of songbirds in tropical forest-agricultural landscape. Animal Conservation 7:93-105. Nájera A, Simonetti JA. 2010. Can oil palm plantations become bird friendly? Agroforestry Systems 80:203209. Ndang’ang’a PK, Ward-Francis A, Costa L, de Lima RF, Palmeirim J, Tavares J, Buchanan G, Carvalho M, Melo M, Dallimer M, Valle S. 2014. International action plan for the conservation of critically endangered birds on São Tomé: 2014-2018. BirdLife International. Available from http://www.birdlife.org/sites/default/files/SAP%20SaoTomeCRs%20Jan2014.pdf (accessed September 2017). Newbold T, Scharlemann JPW, Butchart SHM, Sekercioglu CH, Alkemade R, Booth H, Purves DW. 2013. Ecological traits affect the response of tropical forest bird species to land-use intensity. Proceedings of the Royal Society B DOI:10.1098/rspb.2012.2131. Ogle DH. 2017. FSA: Simple Fisheries Stock Assessment Methods. R package version 0.8.13. Available from https://cran.r-project.org/web/packages/FSA/FSA.pdf (accessed May 2017). Oksanen J. 2015. Multivariate Analysis of Ecological Communities in R: vegan tutorial. R package version 2.3.0. Available from http://cc.oulu.fi/~jarioksa/opetus/metodi/vegantutor.pdf (accessed May 2017). Oliveira JEDC. 1993. A economia de S. Tomé e Príncipe. Instituto para a Cooperação Económica & Instituto de Investigação Científica Tropical, Lisboa. Olson DM, Dinerstein E. 1998. The global 200: A representation approach to conserving the Earth's most biologically valuable ecoregions. Conservation Biology 12:502-515. Pardini R, Bueno AdA, Gardner TA, Prado PI, Metzger JP. 2010. Beyond the Fragmentation Threshold Hypothesis: Regime Shifts in Biodiversity Across Fragmented Landscapes. PLoS ONE (e13666) DOI:10.1371/journal.pone.0013666. Peet N, Atkinson P. 1994. The biodiversity and conservation of the birds of São Tomé and Príncipe. Biodiversity and Conservation 3:851-867. Pereira H. 2013. Conservation Genetics of the Endemic Pigeons of São Tomé e Príncipe. MSc dissertation. Faculdade de Ciências da Universitudade do Porto. Phalan B, Onial M, Balmford A, Green RE. 2011. Reconciling Food Production and Biodiversity Conservation: Land Sharing and Land Sparing Compared. Science 333:1289-1291. Pimm SL, Askins RA. 1995. Forest losses predict bird extinctions in eastern North America. Proceedings of the National Academy of Sciences 92:9343-9347. Quantum Gis Development Team. 2009a. Quantum GIS Geographic Information System. Open Source Geospatial Foundation Project. Available from http://qgis.osgeo.org (accessed September 2016). 41

Quantum Gis Development Team. 2009b. Raster Terrain Analysis Plugin. Open Source Geospatial Foundation Project. Available from https://docs.qgis.org/1.8/en/docs/user_manual/plugins/plugins_raster_terrain.html (accessed September 2016). Quantum Gis Development Team. 2009c. Topographic position index (tpi). Open Source Geospatial Foundation Project. Available from http://docs.qgis.org/2.6/pt_PT/docs/user_manual/processing_algs/saga/terrain_analysis_morphometry/topogra phicpositionindextpi.html (accessed September 2016). Quantum Gis Development Team. 2009d. Accumulated cost (isotropic). Open Source Geospatial Foundation Project. Available from https://docs.qgis.org/2.6/en/docs/user_manual/processing_algs/saga/grid_analysis/accumulatedcostisotropic.ht ml (accessed September 2016). Quantum Gis Development Team. 2009e. Point sampling tool. Open Source Geospatial Foundation Project. Available from https://plugins.qgis.org/plugins/pointsamplingtool/ (accessed September 2016). Thissen D, Steinberg L, Kuang D. 2002. Quick and Easy Implementation of the Benjamini-Hochberg Procedure for Controlling the False Positive Rate in Multiple Comparisons. Journal of Educational and Behavioral Statistics 27:77-83. R Development Core Team. 2017. R: A language and environment for statistical computing. R Development Core Team, Vienna, Austria. Available from http://www.r-project.org/ (accessed May 2017). Rocha R. 2008. Birds in humanized landscapes: São Tomé endemic birds’ response to agricultural intensification. MSc dissertation. Imperial College. Rocha R, Virtanen T, Cabeza M. 2015. Bird assemblages in a Malagasy forest-agricultural frontier: effects of habitat structure and forest cover. Tropical Conservation Science 8:681-710. Rushton SP, Ormerod SJ, Kerby G. 2004. New paradigms for modelling species distributions? Journal of Applied Ecology 41:193-200. Salgueiro A, Carvalho S. 2001. Proposta de Plano Nacional de Desenvolvimento Florestal 2003-2007. ECOFAC, AGRECO, CIRAD Forêt, São Tomé. Savilaakso S, Garcia C, Garcia-Ulloa J, Ghazoul J, Groom M, Guariguata MR, Laumonier Y, Nasi R, Petrokofsky G, Snaddon J, Zrust M. 2014. Systematic review of effects on biodiversity from oil palm production. Environmental Evidence DOI:10.1186/2047-2382-3-4. Schwartzman S, Moreira A, Nepstad D. 2000. Rethinking Tropical Forest Conservation: Perils in Parks. Conservation Biology DOI:10.1046/j.1523-1739.2000.99329.x. Seoane J, Vinuela J, Díaz-Delgado R, Bustamantea J. 2003. The effects of land use and climate on red kite distribution in the Iberian peninsula. Biological Conservation 111:401-414. Seoane J, Bustamante J, Díaz-Delgado R. 2004. Competing roles for landscape, vegetation, topography and climate in predictive models of bird distribution. Ecological Modelling 171:209-222.

42

Silva HLE. 1958. Esboço da carta de aptidão agrícola de São Tomé e Príncipe. Garcia de Orta 6:61-86. Tenreiro F. 1961. A ilha de São Tomé. Junta de Investigações Científicas do Ultramar, Lisboa. Sing T, Sander O, Beerenwinkel N, Lengauer T. 2015. ROCR: Visualizing the Performance of Scoring Classifiers. R package version 1.0.7. Available from https://cran.rproject.org/web/packages/ROCR/ROCR.pdf (accessed May 2017). Sodhi NS, Liow LH, Bazzaz FA. 2004. Avian Extinctions from tropical and subtropical forests. Annual Review of Ecology, Evolution, and Systematics 35:323-345. Stork N. 2010. Re-assessing current extinction rates. Biodiversity and Conservation DOI: 10.1007/s10531009-9761-9. Szabo KJ, Khwaja N, Garnett ST, Butchart SHM. 2012. Global Patterns and Drivers of Avian Extinctions at the Species and Subspecies Level. PLoS ONE (e47080) DOI:10.1371/journal.pone.0047080. Tejeda-Cruz C, Sutherland WJ. 2004. Bird responses to shade coffee production. Animal Conservation 7:169179. Thiollay JM. 1995. The Role of Traditional Agroforests in the Conservation of Rain Forest Bird Diversity in Sumatra. Conservation Biology 9:335-353. Thiollay JM. 1999. Responses of an avian community to rain forest degradation. Biodiversity and Conservation 8:513-534. Thomas CD, Cameron A, Green RE, Bakkenes M, Beaumont LJ, Collingham YC, Erasmus BFN, de Siqueira MF, Grainger A, Hannah L, Hughes L, Huntley B, van Jaarsveld AS, Midgley GF, Miles L, OrtegaHuerta MA, Peterson AT, Phillips OL, Williams SE. 2004. Extinction risk from climate change. Nature 427:145-148. Thuiller W, Araújo MB, Lavorel S. 2004. Do we need land-cover data to model species distributions in Europe? Journal of Biogeography 31:353–361. Tobler W. 1993. Three Presentations on Geographical Analysis and Modelling. Technical report. Nacional Center for Geographic Information and Analysis, University of California. Turner EC, Snaddon JL, Fayle TM, Foster WA. 2008. Oil Palm Research in Context: Identifying the Need for Biodiversity Assessment. PLoS ONE (e1572) DOI:10.1371/journal.pone.0001572 Tuszynski J. 2014. caTools: Tools: moving window statistics, GIF, Base64, ROC AUC, etc. R package version 1.17.1. Available from https://cran.r-project.org/web/packages/caTools/caTools.pdf (accessed May 2017). Waltert M, Mardiastuti A, Muhlenberg M. 2004. Effects of Land use on Bird Species Richness in Sulawesi, Indonesia. Conservation Biology 18:1339-1346. Waltert M, Bobo KS, Sainge MN, Fermon H, Mϋhenberg M. 2005. From forest to farmland: Habitat effects on Afrotropical forest bird diversity. Ecological Applications 15:1351-1366. 43

Wright K. 2016. corrgram: Plot a Correlogram. R package version 1.12. Available from https://cran.rproject.org/web/packages/corrgram/corrgram.pdf (accessed May 2017).

44

SUPPLEMENTARY MATERIALS SECTION I: Environmental Variables Table S1. Environmental variables description. List of environmental variables used to model each species potential distribution, species richness and species compositional dissimilarity. All variables were built in Quantum Gis program.

Variables Altitude

Topography Position Index

Ruggedness Slope

Land use

Mean Annual Precipitation

Distance to Coast

Remoteness Index

Description Digital Elevation Model based on NASA's Shuttle Radar Topography Mission (SRTM) with 90 meters of horizontal resolution Index representing the position of each pixel regarding the mean elevation of a neighbourhood within a 0.05º radius (Fig. S5 & S6) Ruggedness Index calculated from the Digital Elevation Model with 90 meters of resolution Slope calculated from the Digital Elevation Model

Land use map built from satellite images, field information, 1970 historical land use map and military maps

Vectorised map obtained from a map with 30 years compiled data throughout the island, later smoothed with a circular filter of 20 pixels radius Minimum linear distance between each pixel and the nearest point in coast line Cost accumulated surface created with a friction surface derived from slope and weighted by the population density

Type

Units

Continuous

Meters

Class 1- Flat Plain Areas Class 2 - Valleys Categorical Class 3 - Middle Slope Class 4 - Upper Slope Class 5 - Ridges Continuous

-

Continuous

Decimal Degrees

Categorical

Class 1 – Native Forest Class 2 – Secondary Forest Class 3 – Shade Plantation Class 4 – NonForested Areas

Continuous

Millimetres

Continuous

Decimal Degrees

Continuous

-

45

Table S2. Environmental raster’s characteristics. All variables are in raster format and projected in the same coordinate reference system, WGS 84 (EPSG 4326). Pixel size is 0.000833º x 0.000833º. Dimensions are 471 x 359 cells (rows x columns).

Variable Minimum value Mean value Maximum value Altitude 1 1962 345.372 Topography Position Index 1 5 Ruggedness 1.414 451.537 67.647 Slope 0 65.093 14.383 Land use 1 4 Mean Annual Precipitation 700 7000 3133.940 Distance to Coast 1.606 x 10-5 0.114 0.040 Remoteness Index 0 6.138 2.042

46

Figure S1. Altitude in meters. Altitude is projected in WGS 84 (EPSG 4326). Pixel size is 0.000833º x 0.000833º. Dimensions are 471 x 359 cells (rows x columns).

47

Figure S2. Ruggedness. Ruggedness is projected in WGS 84 (EPSG 4326). Pixel size is 0.000833º x 0.000833º. Dimensions are 471 x 359 cells (rows x columns).

48

Figure S3. Slope in degrees. Slope is projected in WGS 84 (EPSG 4326). Pixel size is 0.000833º x 0.000833º. Dimensions are 471 x 359 cells (rows x columns).

49

Figure S4. Distance to coast line in degrees. Distance to coast is projected in WGS84 (EPSG 4326). Pixel size is 0.000833º x 0.000833º. Dimensions are 471 x 359 cells (rows x columns).

50

TPI continuous

Mask TPI 0/1

Mask TPI 0/1

x

TPI > -0.5

TPI < 0.5

Mask TPI 0/1 -0.5 < TPI < 0.5 x TPI continuous

TPI continuous -0.5 < TPI < 0.5

Mask Slope 0/1 Slope 5º

-0.5 < TPI < 0.5

-0.5 < TPI < 0.5

Slope 5º

Flat Plain Areas

Middle Slope Areas

Figure S5. Separation of flat plain areas and middle slope areas. Both flat and middle slope areas have topography index values comprised between -0.5 and 0.5. Flat areas are characterized by having slope values smaller or equal to 5º degrees and middle slope areas values bigger than 5º degrees. These rasters are projected in WGS 84 (EPSG 4326), have a pixel size of 0.000833º x 0.000833º and dimensions of 471 x 359 cells (rows x columns).

51

-0.5 < TPI < 0.5

TPI continuous

Slope N 0.017* E>N -

-

Feeding Guild Dunn-test 0.031* C>O 0.013* F