Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Os fatores que explicam a distribuição observada em plantas e animais é uma pergunta que intriga naturalistas, biogeógrafos e ecólogos há mais de um século. Ainda nos primórdios da disciplina de ecologia, as tolerâncias ambientais já haviam sido apontadas como as grandes responsáveis pelo padrão observado da distribuição dos seres vivos, o que mais tarde levou à concepção de nicho ecológico das espécies. Nos últimos anos, o estudo das distribuições dos organismos ganhou grande impulso e destaque na literatura. O motivo foi a maior disponibilidade de catálogos de presença de espécies, o desenvolvimento de bancos de variáveis ambientais de todo o planeta e de ferramentas computacionais capazes de projetar mapas de distribuição potencial de um dado organismo. Estes instrumentos, coletivamente chamados de Modelos de Distribuição de Espécies (MDEs) têm sido desde então amplamente utilizados em estudos de diferentes escopos. Um deles é a avaliação de potenciais áreas suscetíveis à invasão de organismos exóticos. Este estudo tem, portanto, o objetivo de compreender, através de MDEs, os fatores subjacentes à distribuição de duas espécies de corais escleractíneos invasores nativos do Oceano Pacífico e ambas invasoras bem sucedidas de diversas partes do Oceano Atlântico, destacadamente o litoral fluminense. Os resultados mostraram que os modelos preditivos da espécie Tubastraea coccinea (LESSON, 1829), cosmopolita amplamente difundida na sua região nativa pelo Indo- Pacífico demonstraram de maneira satisfatória suas áreas de distribuição nas áreas invadidas do Atlântico. Sua distribuição está basicamente associada a regiões com alta disponibilidade de calcita e baixa produtividade fitoplanctônica. Por outro lado, a aplicação de MDEs foi incapaz de predizer a distribuição de T. tagusensis (WELLS,1982) no Atlântico. Essta espécie, ao contrário de sua congênere, tem distribuição bastante restrita em sua região nativa, o arquipélago de Galápagos. Através de análises posteriores foi possível constatar a mudança no nicho observado durante o processo de invasão. Finalmente, o sucesso preditivo para T. coccinea e o fracasso dos modelos para T. tagusensis levantam importantes questões sobre quais os aspectos ecológicos das espécies são mais favoráveis à aplicação de MDEs. Adicionalmente, lança importantes ressalvas na utilização recentemente tão difundida destas ferramentas como forma de previsão de invasões biológicas e em estudos de efeitos de alterações climáticas sobre a distribuição das espécies. / The factors underpinning the observed distribution of plants and animals across time and space are a central question in ecology and has intrigued scientists for over a century. But even back on those early times, the role of climatic tolerances of the species were recognized as one of the main explanations for such distributional patterns. Later, these assumptions gave rise to the concept of niche which triggered several advances in the study of natural history. Recently, these studies were addressed in the light of novel computational techniques capable of providing potential distributional maps for a given species, generically called Species Distribution Models (SDMs). This coupled with the broader availability of species occurrence records and of environmental data from international databases made studies with SDMs very popular and ubiquitous in the literature. One of the main uses of the SDMs approach is the assessment of potentially susceptible areas of invasion by non- indigenous species. Therefore, here we used SDMs to better understand the major factors related to the current distribution of two well established invasive scleractinian coral species in the Atlantic, both from the Pacific Ocean. The results showed that the models were successful in predicting the potentially invaded sites by the cosmopolitan Tubastraea coccinea (LESSON, 1829), broadly distributed throughout the Pacific. This species distribution was basically associated with increasing concentrations of calcite and lower levels of phytoplankton activity. However, the models were incapable of predicting the survival and establishment of T. tagusensis (WELLS, 1982) in the Atlantic. This species, unlike its congener, has a very restricted distribution in its native regions, the Galapagos Islands. A posterior analyzes indeed showed a niche shift during the invasion event of T. tagusensis in the Atlantic. Finally, the good modelling results for T. coccinea contrasted with the failure of modelling T. tagusensis invasion highlight important explanations on methodological procedures in SDMs. It also helps to better understand which ecological aspects of the species are favourable toward good modelling performance. In addition to that, these results calls for precaution when analyzing SDMs results, particularly in invasion and climate change scenarios studies.

Submitted by Cláudia Bueno (claudiamoura18@gmail.com) on 2015-12-09T16:11:54Z No. of bitstreams: 2 Tese - Rodrigo de Mello - 2014.pdf: 3401133 bytes, checksum: c4dce4e1d5f7934dbb6f49fb174f42c6 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Approved for entry into archive by Luciana Ferreira (lucgeral@gmail.com) on 2015-12-10T06:26:56Z (GMT) No. of bitstreams: 2 Tese - Rodrigo de Mello - 2014.pdf: 3401133 bytes, checksum: c4dce4e1d5f7934dbb6f49fb174f42c6 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Made available in DSpace on 2015-12-10T06:26:56Z (GMT). No. of bitstreams: 2 Tese - Rodrigo de Mello - 2014.pdf: 3401133 bytes, checksum: c4dce4e1d5f7934dbb6f49fb174f42c6 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) Previous issue date: 2014-07-28 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / Historical biogeography is currently invigorated due to the molecular genetics revolution in systematics and population genetics, besides the new approaches to test biogeographical hypothesis. Furthermore, new tools for generating ecological niche and palaeoclimate models are shifting the direction of phylogeographic studies by making them more integrated. The Brazilian savanna (Cerrado biome) vegetation covers some 2 million km² of Central Brazil, representing about 23% of the land surface of the country. Although recent studies indicate that the biome harbours a rich, complex and characteristic reptilian fauna, patterns of species distribution and genetic structure remain poorly understood for the Cerrado herpetofauna. The lizard Tropidurus oreadicus has an interesting ecological history related to the open vegetation formations, occuring in open areas of Cerrado and in savanna enclaves in Amazonia rainforest. In order to investigate its evolutionary history, we used a robust geographical and genetic sampling, a multilocus dataset and palaeodistribution modelling based on the projection of current distribution into past environments (6 and 21ky BP) using 11 methods for SDMs and five coupled atmosphere–ocean global circulation models (AOGCMs). Our results provide new insights on the genetic and spatial structure of the species revealing six major mtDNA lineages and divergence dates with both deep (Tertiary) and shallow (Quaternary) patterns of diversification for the species tree. The spatial analysis for paleodistribution modelling showed little retractions in Mid-Holocene and the refugia map suggests a broad stable area for the species. The relative influences of Neogene geomorphological events and Quaternary climatic changes as causal mechanisms on the species diversification are discussed. Lastly, we present some inferences about potential future changes in the species geographical distribution, aiming to contribute with the recent conservation discussions on how to preserve Brazilian biodiversity in public policies of conservation / A biogeografia histórica está atualmente revigorada devido à revolução da genética molecular na sistemática filogenética e genética populacional, além das novas abordagens para se testar hipóteses biogeográficas. Além disso, novas ferramentas para gerar modelos paleoclimáticos e de nicho ecológico estão mudando a direção dos estudos filogeográficos, tornando-os mais integrados. O bioma Cerrado cobre cerca de 2 milhões de km² do Brasil Central, representando cerca de 23% da superfície terrestre do país. Embora novos estudos recentes indiquem que o bioma abriga uma fauna de répteis rica, complexa e diversificada, os padrões de distribuição e estrutura genéticas permanecem pouco conhecidos para a herpetofauna do Cerrado. O lagarto Tropidurus oreadicus tem uma histórica ecológica interessante relacionada às formações vegetacionais abertas, ocorrendo em áreas abertas do Cerrado e em enclaves de savana na floresta Amazônica. A fim de investigar sua história evolutiva, nós utilizamos uma amostragem genética e geográfica robusta, um conjunto de dados multi-lócos e modelagem de paleodistribuição baseada na projeção da distribuição atual para ambientes pretéritos (6 e 21 mil anos atrás) usando 11 métodos de modelagem de distribuição de espécie e cinco modelos acoplados de circulação atmosfera-oceânica. Nossos resultados fornecem novos vislumbres sobre a estrutura genética e espacial da espécie, revelando seis grandes linhagens mitocondriais e tempos de divergências com padrões tanto antigos (Terciário) quanto mais recentes (Quaternário) para a árvore de espécie. As análises espaciais para modelagem de paleodistribuição mostram pequenas retrações durante o Holoceno Médio, e o mapa de refúgios sugere uma ampla área de estabilidade para a espécie durante as flutuações climáticas. As importâncias relativas de eventos geomorfológicos do Neógeno e das mudanças climáticas do Quaternário como mecanismos causais da diversificação da espécie são discutidas. Por fim, inferências sobre potenciais mudanças futuras na distribuição geográfica da espécie são apresentadas, visando contribuir com as recentes discussões conservacionistas para a preservação da biodiversidade brasileira em politicas públicas de conservação.

This thesis addresses the problem of gas distribution modelling for gas monitoring and gas detection. The presented research is particularly focused on the methods that are suitable for uncontrolled environments. In such environments, gas source locations and the physical properties of the environment, such as humidity and temperature may be unknown or only sparse noisy local measurements are available. Example applications include air pollution monitoring, leakage detection, and search and rescue operations. This thesis addresses how to efficiently obtain and compute predictive models that accurately represent spatio-temporal gas distribution. Most statistical gas distribution modelling methods assume that gas dispersion can be modelled as a time-constant random process. While this assumption may hold in some situations, it is necessary to model variations over time in order to enable applications of gas distribution modelling for a wider range of realistic scenarios. This thesis proposes two time-dependent gas distribution modelling methods. In the first method, a temporal (sub-)sampling strategy is introduced. In the second method, a time-dependent gas distribution modelling approach is presented, which introduces a recency weight that relates measurement to prediction time. These contributions are presented and evaluated as an extension of a previously proposed method called Kernel DM+V using several simulation and real-world experiments. The results of comparing the proposed time-dependent gas distribution modelling approaches to the time-independent version Kernel DM+V indicate a consistent improvement in the prediction of unseen measurements, particularly in dynamic scenarios under the condition that there is a sufficient spatial coverage. Dynamic scenarios are often defined as environments where strong fluctuations and gas plume development are present. For mobile robot olfaction, we are interested in sampling strategies that provide accurate gas distribution models given a small number of samples in a limited time span. Correspondingly, this thesis addresses the problem of selecting the most informative locations to acquire the next samples. As a further contribution, this thesis proposes a novel adaptive sensor planning method. This method is based on a modified artificial potential field, which selects the next sampling location based on the currently predicted gas distribution and the spatial distribution of previously collected samples. In particular, three objectives are used that direct the sensor towards areas of (1) high predictive mean and (2) high predictive variance, while (3) maximising the coverage area. The relative weight of these objectives corresponds to a trade-off between exploration and exploitation in the sampling strategy. This thesis discusses the weights or importance factors and evaluates the performance of the proposed sampling strategy. The results of the simulation experiments indicate an improved quality of the gas distribution models when using the proposed sensor planning method compared to commonly used methods, such as random sampling and sampling along a predefined sweeping trajectory. In this thesis, we show that applying a locality constraint on the proposed sampling method decreases the travelling distance, which makes the proposed sensor planning approach suitable for real-world applications where limited resources and time are available. As a real-world use-case, we applied the proposed sensor planning approach on a micro-drone in outdoor experiments. Finally, this thesis discusses the potential of using gas distribution modelling and sensor planning in large-scale outdoor real-world applications. We integrated the proposed methods in a framework for decision-making in hazardous inncidents where gas leakage is involved and applied the gas distribution modelling in two real-world use-cases. Our investigation indicates that the proposed sensor planning and gas distribution modelling approaches can be used to inform experts both about the gas plume and the distribution of gas in order to improve the assessment of an incident.

Small Island Developing States (SIDS) of the Caribbean are one of the world’s ‘hottest’ ‘biodiversity hotspots’. However, this biodiversity continues to be threatened by habitat loss, and now, by climate change. The research reported here investigated the potential of species distribution modelling (SDM) as a plant conservation tool within Caribbean SIDS, using Trinidad as a case study. Prior to the application of SDM, ancillary analyses including: (i) quantification and mapping of forest cover change (1969 to 2007) and deforestation rates, and (ii) assessment of the island’s vegetation community distribution and associated drivers were carried out. Community distribution and commercial importance and global/regional rarity were used to generate a list of species for assessing the potential of SDM within Trinidad. Species occurrence data were used to generate species distribution models for present climate conditions within the SDM algorithm, MaxEnt. These results were assessed through expert appraisal and concurrence with results of ecological analyses. These models were used to forecast suitable species climate space forty years into an SRES A2 future. Present and future models were then combined to produce a ‘collective change map’ which showed projected areas of species’ range expansion, contraction or stability for this group of species with respect to Trinidad’s Protected Areas (PAs) network. Despite the models being indicative rather than accurate, it was concluded that species’ climate space is likely to decrease or disappear across Trinidad. Extended beyond Trinidad into the remainder of the Caribbean region, SDM may be a crucial tool in identifying which PAs within the region (and not individual islands) will facilitate future survival of given target species. Consideration of species conservation from a regional, rather than an individual island perspective, is strongly recommended for aiding the Caribbean SIDS to adapt in response to climate change.

The work describes the processing and analysis of remote sensing time series data for a comparative assessment of changes in different tropical rainforest areas in East Africa. In order to assess the effects of the derived changes in land cover and forest fragmentation, the study made use of spatially explicit modelling approaches within a geographical information system (GIS) to extrapolate sets of biological field findings in space and time. The analysis and modelling results were visualised aiming to consider the requirements of three different user groups. In order to evaluate measures of forest conservation and to derive recommendations for an effective forest management, quantitative landscape-scale assessments of land cover changes and their influence on forest biodiversity patterns are needed. However, few remote sensing studies have accounted for all of the following aspects at the same time: (i) a dense temporal sequence of land cover change/forest fragmentation information, (ii) the coverage of several decades, (iii) the distinction between multiple forest formations and (iv) direct comparisons of different case studies. In regards to linkages of remote sensing with biological field data, no attempts are known that use time series data for quantitative statements of long-term landscape-scale biodiversity changes. The work studies three officially protected forest areas in Eastern Africa: the Kakamega-Nandi forests in western Kenya (focus area) and Mabira Forest in south-eastern Uganda as well as Budongo Forest in western Uganda (for comparison purposes). Landsat imagery of in total eight or seven dates in regular intervals from 1972/73 to 2003 was used. Making use of supervised multispectral image classification procedures, in total, 12 land cover classes (six forest formations) were distinguished for the Kakamega-Nandi forests and for Budongo Forest while for Mabira Forest ten classes could be realised. An accuracy assessment via error matrices revealed overall classification accuracies between 81% and 85%. The Kakamega-Nandi forests show a continuous decrease between 1972/73 and 2001 of 31%, Mabira Forest experienced an abrupt loss of 24% in the late 1970s/early 1980s, while Budongo Forest shows a relatively stable forest cover extent. An assessment of the spatial patterns of forest losses revealed congruence with areas of high population density while a spatially explicit forest fragmentation index indicates a strong correlation of forest fragmentation with forest management regime and forest accessibility by roads. For the Kenyan focus area, three sets of biological field abundance data on keystone species/groups were used for a quantitative assessment of the influence of long-term changes in tropical forests on landscape-scale biodiversity patterns. For this purpose, the time series was extended with another three land cover data sets derived from aerial photography (1965/67, 1948/(52)) and old topographic maps (1912/13). To predict the spatio-temporal distribution of the army ant Dorylus wilverthi and of ant-following birds, GIS operators (i.e. focal and local functions) and statistical tests (i.e. OLS or SAR regression models) were combined into a spatial modelling procedure. Abundance data on three guilds of birds differing in forest dependency were directly extrapolated to five forest cover classes as distinguished in the time series. The results predict declines in species abundances of 56% for D. wilverthi, of 58% for ant-following birds and an overall loss of 47% for the bird habitat guilds, which in all three cases greatly exceed the rate of forest loss (31%). Additional extrapolations on scenarios of deforestation and reforestation confirmed the negative ecological consequences of splitting-up contiguous forest areas but also showed the potential of mixed indigenous forest plantings. The visualisation of the analysis and modelling results produced a mixture of different outcomes. Map series and a matrix of maps both showing species distributions aim to address scientists and decision makers. The results of the land cover change analysis were synthesised in a map of land cover development types for each study area, respectively. These maps are designed mainly for scientists. Additional maps of change, limited to a single class of forest cover and to three dates were generated to ensure an easy-to-grasp communication of the major forest changes to decision makers. Additionally, an easy-to-handle visualisation tool to be used by scientists, decision makers and local people was developed. For the future, an extension of this study towards a more complete assessment including more species/groups and also ecosystem functions and services would be desirable. Combining a framework for land cover simulation with a framework for running empirical extrapolation models in an automated manner could ideally result in a GIS-based, integrated forest ecosystem assessment tool to be used as regional spatial decision support system. / Die Arbeit beschreibt die Prozessierung und Analyse von Fernerkundungs-Zeitreihendaten für eine vergleichende Abschätzung von Veränderungen verschiedener tropischer Waldökosysteme Ostafrikas. Um Effekte der Veränderungen bzgl. Landbedeckung und Waldfragmentierung auf Biodiversitätsmuster abzuschätzen, wurden verschiedene räumlich explizite Modellierungssätze innerhalb eines geographischen Informationssystems (GIS) zur räumlichen und zeitlichen Extrapolation biologischer Felderhebungsdaten benutzt. Die Visualisierung der Analyse- und Modellierungsergebnisse erfolgte unter Berücksichtigung der Bedürfnisse von drei verschiedenen Nutzergruppen. Um Waldschutzmaßnahmen zu evaluieren und Empfehlungen für ein effektives Waldmanagement abzuleiten, sind quantitative Abschätzungen von Landbedeckungsveränderungen sowie von deren Einfluss auf tropische Waldbiodiversitätsmuster nötig. Wenige fernerkundungsbasierte Studien haben jedoch bislang alle der folgenden Faktoren berücksichtigt: (i) Informationen zu Veränderungen von Landbedeckung und Waldfragmentierung in dichter zeitlicher Sequenz, (ii) die Abdeckung mehrerer Jahrzehnte, (iii) die Unterscheidung zwischen mehreren Waldformationen, und (iv) direkte Vergleiche von unterschiedlichen Fallstudien. Hinsichtlich Verknüpfungen von Fernerkundung mit biologischen Felddaten sind bisher keine Studien bekannt, die Zeitreihendaten für quantitative Aussagen zu Langzeitveränderungen von Biodiversität auf Landschaftsebene verwenden. Die Arbeit untersucht drei offiziell geschützte Gebiete: die Kakamega-Nandi forests in Westkenia (Hauptuntersuchungsgebiet) sowie Mabira Forest in Südost-Uganda und Budongo Forest in West-Uganda (zu Vergleichszwecken). Es wurden Landsat-Daten für insgesamt acht bzw. sieben Zeitpunkte zwischen 1972/73 und 2003 in ungefähr gleichen Abständen erworben. Mit Hilfe von überwachten, multispektralen Klassifizierungsverfahren wurden für die Kakamega-Nandi forests und Budongo Forest jeweils 12 Landbedeckungsklassen (sechs Waldformationen) und für Mabira Forest zehn Klassen unterschieden. Eine Genauigkeitsprüfung mit Hilfe von Fehlermatrizen ergab Gesamtklassifizierungsgenauigkeiten zwischen 81% und 85%. Die Kakamega-Nandi forests sind durch eine kontinuierliche Waldabnahme von 31% zwischen 1972/73 und 2001 gekennzeichnet, Mabira Forest zeigt einen abrupten Waldverlust von 24% in den späten 1970ern/frühen 1980ern, während die Ergebnisse für Budongo Forest eine relativ stabile Waldbedeckung ausweisen. Während eine Abschätzung der räumlichen Muster von Waldverlusten eine hohe Deckungsgleichheit mit Gebieten hoher Bevölkerungsdichte ergab, deutet die Anwendung eines räumlich expliziten Waldfragmentierungsindexes auf eine starke Korrelation von Waldfragmentierung mit der Art von Waldmanagement sowie mit der Erreichbarkeit von Wald über Straßen hin. Um den Einfluss von Langzeit-Landbedeckungsveränderungen auf Biodiversitätsmuster auf Landschaftsebene für das kenianische Hauptuntersuchungsgebiet quantitativ abzuschätzen wurden drei Datensätze mit biologischen Felderhebungen zur Abundanz von Schlüsselarten/-gruppen verwendet. Zu diesem Zweck wurde die Zeitreihe zunächst um drei weitere Landbedeckungs-Datensätze ergänzt, die aus Luftbildern (1965/67, 1948/(52)) bzw. alten topographischen Karten (1912/13) gewonnen wurden. Zur Vorhersage der raum-zeitlichen Verteilung der Treiberameise Dorylus wilverthi wurden GIS-Operatoren und statistische Tests (OLS bzw. SAR Regressionsmodelle) in einem räumlichen Modellierungsablauf kombiniert. Abundanzdaten von drei sich hinsichtlich ihrer Abhängigkeit von Wald unterscheidenden Vogelgilden wurden direkt auf fünf Waldbedeckungsklassen hochgerechnet, die in der Zeitreihe unterschieden werden konnten. Die Ergebnisse prognostizieren Abundanzabnahmen von 56% für D. wilverthi, von 58% für Ameisen-folgende Vögel und einen Gesamtverlust von 47% für die Vogelgilden, was in allen drei Fällen eine deutliche Überschreitung der Waldverlustrate von 31% darstellt. Zusätzliche Extrapolationen basierend auf Szenarien bestätigten die negativen ökologischen Konsequenzen der Zerteilung zusammenhängender Waldflächen bzw. zeigten andererseits das Potential von Aufforstungen mit einheimischen Arten auf. Die Visualisierung der Analyse- bzw. Modellierungsergebnisse führte zu unterschiedlichen Darstellungen: mit einer Reihe von nebeneinander positionierten Einzelkarten sowie einer Matrix von Einzelkarten, die jeweils Artenverteilungen zeigen, sollen Wissenschaftler und Entscheidungsträger angesprochen werden. Aus den Ergebnissen der Landbedeckungsanalyse für die drei Untersuchungsgebiete wurden Landbedeckungsveränderungstypen generiert und jeweils in einer synthetischen Karte dargestellt, die hauptsächlich für Wissenschaftler gedacht sind. Um die wesentlichen Waldveränderungen auch auf einfache Weise zu den Entscheidungsträgern zu kommunizieren, wurden zusätzliche Karten erstellt, die nur eine aggregierte Klasse „Waldbedeckung“ zeigen und jeweils auf drei Zeitschritte der Zeitreihen begrenzt sind. Zusätzlich wurde ein leicht zu bedienendes Visualisierungstool entwickelt, das für Wissenschaftler, Entscheidungsträger und die lokale Bevölkerung gedacht ist. Für die Zukunft wäre eine umfassendere Abschätzung unter Berücksichtigung zusätzlicher Arten/-gruppen sowie auch Ökosystemfunktionen und –dienstleistungen wünschenswert. Die Verknüpfung einer Applikation zur Landbedeckungsmodellierung mit einer Applikation zur Ausführung von empirischen Extrapolationsmodellen (in stärkerem Maße automatisiert als in dieser Arbeit) könnte im Idealfall in ein GIS-basiertes Tool zur integrativen Bewertung von Waldökosystemen münden, das dann als räumliches Entscheidungsunterstützungssystem verwendet werden könnte.

Face verification is a challenging pattern recognition problem. The face is a biometric that, we as humans, know can be recognised. However, the face is highly deformable and its appearance alters significantly when the pose, illumination or expression changes. These changes in appearance are most notable for texture images, or two-dimensional (2D) data. But the underlying structure of the face, or three dimensional (3D) data, is not changed by pose or illumination variations. Over the past five years methods have been investigated to combine 2D and 3D face data to improve the accuracy and robustness of face verification. Much of this research has examined the fusion of a 2D verification system and a 3D verification system, known as multi-modal classifier score fusion. These verification systems usually compare two feature vectors (two image representations), a and b, using distance or angular-based similarity measures. However, this does not provide the most complete description of the features being compared as the distances describe at best the covariance of the data, or the second order statistics (for instance Mahalanobis based measures). A more complete description would be obtained by describing the distribution of the feature vectors. However, feature distribution modelling is rarely applied to face verification because a large number of observations is required to train the models. This amount of data is usually unavailable and so this research examines two methods for overcoming this data limitation: 1. the use of holistic difference vectors of the face, and 2. by dividing the 3D face into Free-Parts. The permutations of the holistic difference vectors is formed so that more observations are obtained from a set of holistic features. On the other hand, by dividing the face into parts and considering each part separately many observations are obtained from each face image; this approach is referred to as the Free-Parts approach. The extra observations from both these techniques are used to perform holistic feature distribution modelling and Free-Parts feature distribution modelling respectively. It is shown that the feature distribution modelling of these features leads to an improved 3D face verification system and an effective 2D face verification system. Using these two feature distribution techniques classifier score fusion is then examined. This thesis also examines methods for performing classifier fusion score fusion. Classifier score fusion attempts to combine complementary information from multiple classifiers. This complementary information can be obtained in two ways: by using different algorithms (multi-algorithm fusion) to represent the same face data for instance the 2D face data or by capturing the face data with different sensors (multimodal fusion) for instance capturing 2D and 3D face data. Multi-algorithm fusion is approached as combining verification systems that use holistic features and local features (Free-Parts) and multi-modal fusion examines the combination of 2D and 3D face data using all of the investigated techniques. The results of the fusion experiments show that multi-modal fusion leads to a consistent improvement in performance. This is attributed to the fact that the data being fused is collected by two different sensors, a camera and a laser scanner. In deriving the multi-algorithm and multi-modal algorithms a consistent framework for fusion was developed. The consistent fusion framework, developed from the multi-algorithm and multimodal experiments, is used to combine multiple algorithms across multiple modalities. This fusion method, referred to as hybrid fusion, is shown to provide improved performance over either fusion system on its own. The experiments show that the final hybrid face verification system reduces the False Rejection Rate from 8:59% for the best 2D verification system and 4:48% for the best 3D verification system to 0:59% for the hybrid verification system; at a False Acceptance Rate of 0:1%.

Submitted by Franciele Moreira (francielemoreyra@gmail.com) on 2017-08-17T18:33:27Z No. of bitstreams: 2 Dissertaçao - Lara de Macedo Monteiro - 2017.pdf: 68300760 bytes, checksum: ba7fbce35b9ab3e46180337ae3129580 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) / Approved for entry into archive by Luciana Ferreira (lucgeral@gmail.com) on 2017-08-18T12:02:30Z (GMT) No. of bitstreams: 2 Dissertaçao - Lara de Macedo Monteiro - 2017.pdf: 68300760 bytes, checksum: ba7fbce35b9ab3e46180337ae3129580 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) / Made available in DSpace on 2017-08-18T12:02:30Z (GMT). No. of bitstreams: 2 Dissertaçao - Lara de Macedo Monteiro - 2017.pdf: 68300760 bytes, checksum: ba7fbce35b9ab3e46180337ae3129580 (MD5) license_rdf: 0 bytes, checksum: d41d8cd98f00b204e9800998ecf8427e (MD5) Previous issue date: 2017-03-15 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / Earth is facing the highest species' extinction rates of its history, and humans are the major stressar. Adding up to this biodiversity crisis, species-rich areas, which also coincide with areas highly transformed by humans (e.g. biodiversity hotspots), are poorly covered by protected areas. ln Brazil this reality is not different. Responsible for harbouring a third of all plant species already classified under a threat category (n= 645), the Brazilian Cerrado has only 8.3% of its area legally protected. ln this biorne, the campos rupestres, a mountaintop grassland ecosystem, stands out for its high number of threatened species currently underrepresented in conservation strategies. ln chapter 1, we aimed at indicating priority areas to secure protection of the threatened plant species from the southern Espinhaço mountains, a region that encampasses large areas of campos rupestres. We found that it is possible to protect, on average, more than 25% of the threatened species' ranges, avoiding sites with extensive use for farming and mining and favouring areas with intensive fire frequency by constraining the management to a relatively small area of only 17% of the region. Conservation plans such as these proposed for campos rupestres represent important opportunities to fulfil the gap existent between research and implementation. However, we do not rule out the need for increasing sophisticated tools that account for the consequences of complex processes threatening biodiversity in the near future ( e.g. clima te change and deforestation) and especially the need for predictive and realistic conservation strategies that anticipate and mitigate their negative effects. Unfortunately, until now we have been relying species protection to a residual system of PAs that provide minimal conservation impact. Thus, in chapter 2 we aimed to select spatial conservation priorites that minimize the risk of deforestation while retaining sites with high plant biodiversity value threatened from climate change in the Brazilian Cerrado. We simulated two ways of spacing out priorities for conservation actions ("time-step action" and "acting now''), and two methods of setting priorities: one that minimizes expected habitat conversion and prioritizes high valuable sites to plant biodiversity at risk from climate change (maximum conservation impact) and another that prioritizes sites based only on their value for plant biodiversity at risk from climate change, regardless their vulnerability to land conversion ("usual approach''). We found that although the scenarios that maximize conservation impact avoided higher amounts of vegetation loss, they prevented least species' range loss. Moreover, the acting now scenarios always performed better in terms of range loss avoided compared to the time-step scenarios under the sarne method of prioritization. Finally, we believe that planning for vegetation loss avoidance is a more conservative strategy because vegetation information is less subjective to any source of bias and is a better surrogate for general biodiversity. We also recommend that acting as soon as possible is always the best strategy to guarantee biodiversity conservation in the Cerrado. / A Terra vem enfrentando as maiores taxas de extinção de espécies de sua história, e os humanos são a maior causa disso. Além da crise de biodiversidade, áreas ricas em espécies, que, por sua vez, coincidem com locais sob alta influência de atividades humanas (ex: hotspots de biodiversidade), são pouco representadas por Unidades de Conservação. No Brasil, essa realidade não é diferente. Responsável por abrigar um terço de todas as espécies de plantas já classificadas sob uma das categorias de ameaça (n=645), o Cerrado brasileiro possui somente 8.3% de sua área legalmente protegida. Nesse biorna, o ecossistema de campos rupestres destaca-se pelo seu alto número de espécies ameaçadas atualmente subrepresentadas em estratégias de conservação. No capítulo 1, nosso objetivo foi indicar áreas prioritárias para assegurar a proteção de espécies ameaçadas de plantas da Serra do Espinhaço Meridional, uma região que abrange grandes trechos de campos rupestres. Nós encontramos que é possível proteger, em média, mais de 25% da distribuição das espécies ameaçadas restringindo o manejo a uma área relativamente pequena de apenas 17% da região e evitando locais de uso extensivo do solo para agropecuária e mineração e favorecendo locais com alta ocorrência de queimadas. Planos de conservação como esse proposto para campos rupestres representam importantes oportunidades para preencher a lacuna existente entre pesquisa e implementação. No entanto, nós não descartamos a necessidade de ferramentas mais sofisticadas que considerem as consequências dos complexos processos que ameaçam a biodiversidade em um futuro próximo ( ex: mudanças climáticas e desmatamento) e, especialmente, a necessidade de estratégias de conservação preditivas e realistas que antecipem e mitiguem seus efeitos negativos. Infelizmente, até agora a proteção das espécies tem se restringido a um sistema residual de unidades de conservação de baixo impacto para a conservação. Portanto, no capítulo 2 nosso objetivo foi selecionar espacialmente locais de alto valor para a biodiversidade de plantas ameaçadas em um cenário de mudanças climáticas e ao mesmo tempo minimizar o risco de conversão da vegetação desses locais. Nós simulamos duas formas de particionar as ações de conservação ("ação em intervalos de tempo" e "agir agora") e dois métodos de estabelecer prioridades: um que minimiza a conversão de hábitat esperada e prioriza locais altamente importantes para a biodiversidade de plantas ameaçadas em um cenário de mudanças climáticas ("máximo impacto da conservação") e outro que prioriza locais baseando-se somente no seu valor para a biodiversidade de plantas ameaçadas em um cenário de mudanças climáticas, independentemente de sua vulnerabilidade ao desmatamento ("abordagem habitual''). Nós encontramos que, embora os cenários que maximizem o impacto da conservação tenham evitado maiores perdas de vegetação, eles evitaram uma menor perda no tamanho médio da distribuição das espécies comparado às abordagens habituais. Além disso, constatamos que os cenários "agir agora" tiveram um melhor desempenho em termos de perda de distribuição evitada comparado aos cenários de implementação sequencial de ações considerando um mesmo método de priorização. Finalmente, nós acreditamos que planejar para evitar perda de vegetação é uma estratégia mais segura, porque a informação sobre vegetação é menos sujeita a qualquer viés e é um melhor indicador para biodiversidade em geral. Também recomendamos que agir o quanto antes é sempre a melhor estratégia para garantir a conservação da biodiversidade no Cerrado.

The work describes the processing and analysis of remote sensing time series data for a comparative assessment of changes in different tropical rainforest areas in East Africa. In order to assess the effects of the derived changes in land cover and forest fragmentation, the study made use of spatially explicit modelling approaches within a geographical information system (GIS) to extrapolate sets of biological field findings in space and time. The analysis and modelling results were visualised aiming to consider the requirements of three different user groups. In order to evaluate measures of forest conservation and to derive recommendations for an effective forest management, quantitative landscape-scale assessments of land cover changes and their influence on forest biodiversity patterns are needed. However, few remote sensing studies have accounted for all of the following aspects at the same time: (i) a dense temporal sequence of land cover change/forest fragmentation information, (ii) the coverage of several decades, (iii) the distinction between multiple forest formations and (iv) direct comparisons of different case studies. In regards to linkages of remote sensing with biological field data, no attempts are known that use time series data for quantitative statements of long-term landscape-scale biodiversity changes. The work studies three officially protected forest areas in Eastern Africa: the Kakamega-Nandi forests in western Kenya (focus area) and Mabira Forest in south-eastern Uganda as well as Budongo Forest in western Uganda (for comparison purposes). Landsat imagery of in total eight or seven dates in regular intervals from 1972/73 to 2003 was used. Making use of supervised multispectral image classification procedures, in total, 12 land cover classes (six forest formations) were distinguished for the Kakamega-Nandi forests and for Budongo Forest while for Mabira Forest ten classes could be realised. An accuracy assessment via error matrices revealed overall classification accuracies between 81% and 85%. The Kakamega-Nandi forests show a continuous decrease between 1972/73 and 2001 of 31%, Mabira Forest experienced an abrupt loss of 24% in the late 1970s/early 1980s, while Budongo Forest shows a relatively stable forest cover extent. An assessment of the spatial patterns of forest losses revealed congruence with areas of high population density while a spatially explicit forest fragmentation index indicates a strong correlation of forest fragmentation with forest management regime and forest accessibility by roads. For the Kenyan focus area, three sets of biological field abundance data on keystone species/groups were used for a quantitative assessment of the influence of long-term changes in tropical forests on landscape-scale biodiversity patterns. For this purpose, the time series was extended with another three land cover data sets derived from aerial photography (1965/67, 1948/(52)) and old topographic maps (1912/13). To predict the spatio-temporal distribution of the army ant Dorylus wilverthi and of ant-following birds, GIS operators (i.e. focal and local functions) and statistical tests (i.e. OLS or SAR regression models) were combined into a spatial modelling procedure. Abundance data on three guilds of birds differing in forest dependency were directly extrapolated to five forest cover classes as distinguished in the time series. The results predict declines in species abundances of 56% for D. wilverthi, of 58% for ant-following birds and an overall loss of 47% for the bird habitat guilds, which in all three cases greatly exceed the rate of forest loss (31%). Additional extrapolations on scenarios of deforestation and reforestation confirmed the negative ecological consequences of splitting-up contiguous forest areas but also showed the potential of mixed indigenous forest plantings. The visualisation of the analysis and modelling results produced a mixture of different outcomes. Map series and a matrix of maps both showing species distributions aim to address scientists and decision makers. The results of the land cover change analysis were synthesised in a map of land cover development types for each study area, respectively. These maps are designed mainly for scientists. Additional maps of change, limited to a single class of forest cover and to three dates were generated to ensure an easy-to-grasp communication of the major forest changes to decision makers. Additionally, an easy-to-handle visualisation tool to be used by scientists, decision makers and local people was developed. For the future, an extension of this study towards a more complete assessment including more species/groups and also ecosystem functions and services would be desirable. Combining a framework for land cover simulation with a framework for running empirical extrapolation models in an automated manner could ideally result in a GIS-based, integrated forest ecosystem assessment tool to be used as regional spatial decision support system. / Die Arbeit beschreibt die Prozessierung und Analyse von Fernerkundungs-Zeitreihendaten für eine vergleichende Abschätzung von Veränderungen verschiedener tropischer Waldökosysteme Ostafrikas. Um Effekte der Veränderungen bzgl. Landbedeckung und Waldfragmentierung auf Biodiversitätsmuster abzuschätzen, wurden verschiedene räumlich explizite Modellierungssätze innerhalb eines geographischen Informationssystems (GIS) zur räumlichen und zeitlichen Extrapolation biologischer Felderhebungsdaten benutzt. Die Visualisierung der Analyse- und Modellierungsergebnisse erfolgte unter Berücksichtigung der Bedürfnisse von drei verschiedenen Nutzergruppen. Um Waldschutzmaßnahmen zu evaluieren und Empfehlungen für ein effektives Waldmanagement abzuleiten, sind quantitative Abschätzungen von Landbedeckungsveränderungen sowie von deren Einfluss auf tropische Waldbiodiversitätsmuster nötig. Wenige fernerkundungsbasierte Studien haben jedoch bislang alle der folgenden Faktoren berücksichtigt: (i) Informationen zu Veränderungen von Landbedeckung und Waldfragmentierung in dichter zeitlicher Sequenz, (ii) die Abdeckung mehrerer Jahrzehnte, (iii) die Unterscheidung zwischen mehreren Waldformationen, und (iv) direkte Vergleiche von unterschiedlichen Fallstudien. Hinsichtlich Verknüpfungen von Fernerkundung mit biologischen Felddaten sind bisher keine Studien bekannt, die Zeitreihendaten für quantitative Aussagen zu Langzeitveränderungen von Biodiversität auf Landschaftsebene verwenden. Die Arbeit untersucht drei offiziell geschützte Gebiete: die Kakamega-Nandi forests in Westkenia (Hauptuntersuchungsgebiet) sowie Mabira Forest in Südost-Uganda und Budongo Forest in West-Uganda (zu Vergleichszwecken). Es wurden Landsat-Daten für insgesamt acht bzw. sieben Zeitpunkte zwischen 1972/73 und 2003 in ungefähr gleichen Abständen erworben. Mit Hilfe von überwachten, multispektralen Klassifizierungsverfahren wurden für die Kakamega-Nandi forests und Budongo Forest jeweils 12 Landbedeckungsklassen (sechs Waldformationen) und für Mabira Forest zehn Klassen unterschieden. Eine Genauigkeitsprüfung mit Hilfe von Fehlermatrizen ergab Gesamtklassifizierungsgenauigkeiten zwischen 81% und 85%. Die Kakamega-Nandi forests sind durch eine kontinuierliche Waldabnahme von 31% zwischen 1972/73 und 2001 gekennzeichnet, Mabira Forest zeigt einen abrupten Waldverlust von 24% in den späten 1970ern/frühen 1980ern, während die Ergebnisse für Budongo Forest eine relativ stabile Waldbedeckung ausweisen. Während eine Abschätzung der räumlichen Muster von Waldverlusten eine hohe Deckungsgleichheit mit Gebieten hoher Bevölkerungsdichte ergab, deutet die Anwendung eines räumlich expliziten Waldfragmentierungsindexes auf eine starke Korrelation von Waldfragmentierung mit der Art von Waldmanagement sowie mit der Erreichbarkeit von Wald über Straßen hin. Um den Einfluss von Langzeit-Landbedeckungsveränderungen auf Biodiversitätsmuster auf Landschaftsebene für das kenianische Hauptuntersuchungsgebiet quantitativ abzuschätzen wurden drei Datensätze mit biologischen Felderhebungen zur Abundanz von Schlüsselarten/-gruppen verwendet. Zu diesem Zweck wurde die Zeitreihe zunächst um drei weitere Landbedeckungs-Datensätze ergänzt, die aus Luftbildern (1965/67, 1948/(52)) bzw. alten topographischen Karten (1912/13) gewonnen wurden. Zur Vorhersage der raum-zeitlichen Verteilung der Treiberameise Dorylus wilverthi wurden GIS-Operatoren und statistische Tests (OLS bzw. SAR Regressionsmodelle) in einem räumlichen Modellierungsablauf kombiniert. Abundanzdaten von drei sich hinsichtlich ihrer Abhängigkeit von Wald unterscheidenden Vogelgilden wurden direkt auf fünf Waldbedeckungsklassen hochgerechnet, die in der Zeitreihe unterschieden werden konnten. Die Ergebnisse prognostizieren Abundanzabnahmen von 56% für D. wilverthi, von 58% für Ameisen-folgende Vögel und einen Gesamtverlust von 47% für die Vogelgilden, was in allen drei Fällen eine deutliche Überschreitung der Waldverlustrate von 31% darstellt. Zusätzliche Extrapolationen basierend auf Szenarien bestätigten die negativen ökologischen Konsequenzen der Zerteilung zusammenhängender Waldflächen bzw. zeigten andererseits das Potential von Aufforstungen mit einheimischen Arten auf. Die Visualisierung der Analyse- bzw. Modellierungsergebnisse führte zu unterschiedlichen Darstellungen: mit einer Reihe von nebeneinander positionierten Einzelkarten sowie einer Matrix von Einzelkarten, die jeweils Artenverteilungen zeigen, sollen Wissenschaftler und Entscheidungsträger angesprochen werden. Aus den Ergebnissen der Landbedeckungsanalyse für die drei Untersuchungsgebiete wurden Landbedeckungsveränderungstypen generiert und jeweils in einer synthetischen Karte dargestellt, die hauptsächlich für Wissenschaftler gedacht sind. Um die wesentlichen Waldveränderungen auch auf einfache Weise zu den Entscheidungsträgern zu kommunizieren, wurden zusätzliche Karten erstellt, die nur eine aggregierte Klasse „Waldbedeckung“ zeigen und jeweils auf drei Zeitschritte der Zeitreihen begrenzt sind. Zusätzlich wurde ein leicht zu bedienendes Visualisierungstool entwickelt, das für Wissenschaftler, Entscheidungsträger und die lokale Bevölkerung gedacht ist. Für die Zukunft wäre eine umfassendere Abschätzung unter Berücksichtigung zusätzlicher Arten/-gruppen sowie auch Ökosystemfunktionen und –dienstleistungen wünschenswert. Die Verknüpfung einer Applikation zur Landbedeckungsmodellierung mit einer Applikation zur Ausführung von empirischen Extrapolationsmodellen (in stärkerem Maße automatisiert als in dieser Arbeit) könnte im Idealfall in ein GIS-basiertes Tool zur integrativen Bewertung von Waldökosystemen münden, das dann als räumliches Entscheidungsunterstützungssystem verwendet werden könnte.

La diversité bêta est une mesure essentielle pour décrire l'organisation de la biodiversité dans l'espace. Le calcul des contributions locales à la diversité bêta (LCBD), en particulier, permet d'identifier des sites à forte unicité écologique montrant une diversité exceptionnelle au sein d'une région d'intérêt. Jusqu’à présent, l'utilisation des LCBD s'est restreinte à des échelles locales ou régionales avec un petit nombre de sites. Dans ce mémoire, j'ai examiné si les modèles de répartition d'espèces (SDM) permettent d'évaluer l'unicité écologique sur de plus grandes étendues spatiales. J'ai également étudié l’effet des changements d’échelle sur la quantification de la diversité bêta. Pour ce faire, j'ai utilisé la base de données eBird et des arbres de régression additifs bayésiens pour prédire la répartition des parulines en Amérique du Nord. J'ai ensuite calculé les LCBD sur ces prédictions, ce qui permet de couvrir de plus grandes étendues spatiales et un nombre de sites plus élevé. Mes résultats ont montré que les SDM fournissent des estimations d'unicité fortement corrélées avec les données observées et montrant une association spatiale statistiquement significative. Ils ont également montré que la relation entre la richesse et les LCBD varie selon la région et l'étendue spatiale et qu'elle est influencée par la proportion d'espèces rares dans les communautés. Ainsi, les sites identifiés comme uniques peuvent varier selon les caractéristiques de la région étudiée. Ces résultats montrent que les SDM peuvent être utilisés pour prédire l'unicité écologique, ce qui pourrait permettre d'identifier d'importantes cibles de conservation au sein de régions non échantillonnées. / Beta diversity is an essential measure to describe the organization of biodiversity through space. The calculation of local contributions to beta diversity (LCBD), specifically, allows the identification of sites with high ecological uniqueness and exceptional diversity within a region of interest. To this day, LCBD indices have primarily been used on regional and smaller scales, with relatively few sites. Furthermore, their use is typically restricted to strictly sampled sites with known species composition, leading to gaps in spatial coverage on broad extents. Here, I examined whether species distribution models (SDMs) can be used to assess ecological uniqueness over broader spatial extents and investigated the effect of scale changes on beta diversity quantification. To this aim, I used observations recorded in the eBird database and Bayesian additive regression trees to model warbler species composition in North America, then computed LCBD indices on the predictions, thus covering a broader spatial extent and a higher number of sites. My results showed that SDMs provide uniqueness estimates highly correlated with observed data with a statistically significant spatial association. They also showed that the relationship between richness and LCBD values varies according to the region and the spatial extent and that it is affected by the proportion of rare species in communities. Sites identified as unique may therefore vary according to regional characteristics. These results show that SDMs can be used to predict ecological uniqueness over broad spatial extents, which could help identify beta diversity hotspots and important targets for conservation purposes in unsampled locations.

Lyme disease is an emerging tick-borne illness in Canada, with human case numbers increasing 15- to 20-fold since Lyme disease became nationally notifiable in 2009 until the present. In Ontario, Canada's largest province by population, average Lyme disease incidence across the province is similar to that of national estimates. However, in eastern Ontario, which is near tick endemic regions in the northeastern Unites States, Lyme disease incidence is disproportionately higher compared to the rest of the province. The objectives of this thesis are to identify environmental Lyme disease risk areas in Ontario, to explore spatiotemporal trends in Lyme disease emergence, and to identify neighbourhood-level socioecological risk factors for Lyme disease. In addition, this thesis also aims to assess the risk of other tick-borne illnesses that are transmitted by the blacklegged tick, Ixodes scapularis, which is also the main vector for Lyme disease in Canada. Using maximum entropy species distribution modelling to correlate blacklegged tick occurrence data with environmental variables, predictive risk models for I. scapularis and the Lyme disease pathogen, Borrelia burgdorferi, were developed. The model prediction was used to classify low and high environment risk areas and, using a case-control epidemiological study, we assessed that residence in risk areas was a strong predictor of Lyme disease. However, this relationship was modulated by socioecological factors linked to higher overall rurality of the locality of home residence. Spatial cluster analyses further revealed that human Lyme disease cases clustered in regions with the high numbers of reported B. burgdorferi-infected ticks in the environment. Many individuals residing in large metropolitan regions, like the City of Ottawa, reported tick exposures outside their public health unit of residence; however, local clusters of Lyme disease were also detected in suburban regions near conservation areas, trails, and urban woodlands. The prevalence of other tick-borne pathogens was low, although several pathogens of public health significance including Borrelia miyamotoi and Anaplasma phagocytophilum were detected at multiple sites surveyed for ticks between 2017-2021. Overall, this thesis identify patterns in Lyme disease emergence (and potentially other tick-borne illnesses), defines environmental risk areas for Lyme disease in Ontario, and highlights important socioecological risk factors for Lyme disease in eastern Ontario.