Multi-core architectures are present throughout a large selection of computing devices from cell phones to super-computers. Parallel applications running on these devices solve bigger problems in a shorter time. Writing those applications is a difficult task for programmers. They need to deal with low-level parallel mechanisms such as data distribution, inter-processor communication, and task placement. The goal of the Standard Template Adaptive Parallel Library (STAPL) is to provide a generic high-level framework to develop parallel applications. One of the first steps of a parallel application is to partition and distribute the data throughout the system. An important data structure for parallel applications to store large amounts of data and model many types of relations is the graph. A mesh, which is a special type of graph, is often used to model a spatial domain in scientific applications. Graph and mesh partitioning has many applications such as VLSI circuit design, parallel task scheduling, and data distribution. Data distribution, significantly impacts the performance of a parallel application. In this thesis, we introduce the STAPL Parallel Graph Partitioner Framework. This framework provides a generic infrastructure to partition arbitrary graphs and meshes and to build customized partitioners. It includes the state of the art parallel k-way multilevel scheme to partition arbitrary graphs, a parallel mesh partitioner with parameterized partition shape, and a customized partitioner used for discrete ordinates particle transport computations. This framework is also part of a generic library, STAPL, allowing the partitioning of the data and development of the whole parallel application to be done in the same environment. We show the user-friendly interface of the framework and its scalability for partitioning different mesh and graph benchmarks on a Cray XE6 system. We also highlight the performance of our customized unstructured mesh partitioner for a discrete ordinates particle transport code. The developed columnar decompositions significantly reduce the execution time of simultaneous sweeps on unstructured meshes.

Data processing for Big Data plays a vital role for decision-makers in organizations and government, enhances the user experience, and provides quality results in prediction analysis. However, many modern data processing solutions make a significant investment in hardware and maintenance costs, such as Hadoop and Spark, often neglecting the well established and widely used relational database management systems (RDBMS's). In this dissertation, we study three fundamental graph problems in RDBMS. The first problem we tackle is computing shortest paths (SP) from a source to a target in large network graphs. We explore SQL based solutions and leverage the intelligent scheduling that a RDBMS performs when executing set-at-a-time expansions of graph vertices, which is in contrast to vertex-at-a-time expansions in classical SP algorithms. Our algorithms perform orders of magnitude faster than baselines and outperform counterparts in native graph databases. Second, we studied the PageRank problem which is vital in Google Search and social network analysis to determine how to sort search results and identify important nodes in a graph. PageRank is an iterative algorithm which imposes challenges when implementing it over large graphs. We study computing PageRank using RDBMS for very large graphs using a consumer-grade machine and compare the results to a dedicated graph database. We show that our RDBMS solution is able to process graphs of more than a billion edges in few minutes, whereas native graph databases fail to handle graphs of much smaller sizes. Last, we present a carefully engineered RDBMS solution to the problem of triangle enumeration for very large graphs. We show that RDBMS's are suitable tools for enumerating billions of triangles in billion-scale networks on a consumer grade machine. Also, we compare our RDBMS solution's performance to a native graph database and show that our RDBMS solution outperforms by orders of magnitude. / Graduate

L’allègement dans le secteur de l’automobile revête un enjeu important du fait de normes d’émission de CO2 de plus en plus drastiques, de la nécessité de réduire la consommation en carburant des véhicules et d’une aspiration sociétale à une économie « plus verte ». Pour répondre à ces défis et dans un souci de sécurité et de contrôle des coûts, l’industrie automobile étudie actuellement la possibilité de développer et de produire une 3ème génération d'aciers à très haute résistance. Ils résultent de traitements thermomécaniques généralement innovants, possèdent des microstructures complexes et des propriétés mécaniques améliorées. Le procédé de Quenching and Partitioning (Q&P) est le traitement le plus prometteur, il consiste en une trempe sous la température de début de transformation martensitique Ms, puis, d'un réchauffage et d'un maintien au-dessus de la température initiale de trempe (QT). L'étape de maintien est appelée "étape de partition", car un enrichissement en carbone de l'austénite est attendu. Les propriétés mécaniques exceptionnelles des aciers Q&P sont dues à leur microstructure duplexe complexe : de très fins îlots d'austénite résiduelle imbriqués dans une matrice martensitique revenue et/ou fraîche. Bien que les mécanismes d'enrichissement en carbone de l'austénite résiduelle lors de l'étape de partition soient encore débattus dans la littérature, il existe des preuves tangibles qui attestent d’un phénomène de partition du carbone de la martensite vers l’austénite. Cependant, la formation de bainite et de carbures dans la martensite soulève la question de l’influence de ces réactions et de leurs interactions sur les mécanismes et les cinétiques d’enrichissement en carbone de l'austénite. Il s'agit clairement d'un sujet d'intérêt puisque les propriétés mécaniques de ces aciers reposent principalement sur la teneur en carbone des îlots d’austénite.Cette thèse qui repose sur une approche expérimentale multi-échelle couplée à une approche théorique en champ moyen, a pour ambition d’apporter des éléments de réponse aux mécanismes d’enrichissement en carbone de l’austénite dans un aciers duplex Q&P à très haute résistance de composition Fe-0,3 C-2,5Mn-1,5Si... / The need to reduce the fuel consumption of vehicles while increasing safety led the automotive industry to develop a 3rd generation of Advanced High Strength Steels. Such steels combine innovative processing routes, complex microstructures, improved mechanical properties and are a possible response in vehicle lightweighting. The Quenching and Partitioning (Q&P) process is the most promising route and involves quenching below the martensite start temperature followed by a reheatingand ageing above the initial quench temperature (QT). The ageing step is termed “partitioning step” since carbon enrichment in austenite is expected to occur during this stage. The exceptional mechanical properties of Q&P steels come from their complex duplex microstructure: very fine austenite island retained at room temperature embedded in both recovered and fresh martensite. Although the mechanisms of carbon enrichment in retained austenite during the partitioning step are still debated,strong evidences of carbon partitioning from martensite to austenite exist. However, both the formation of bainite and carbides into martensite raise the question of the effects of competitive reactions on the carbon enrichment in austenite. It is clearly a topic of interest since the benefits of such a treatment in terms of improved mechanical properties depends strongly on the austenite stability and thus on the level of carbon enrichment in austenite during the partitioning step.This thesis aims at combining an innovative multiscale experimental methodology with an original theoretical approach providing a unique opportunity to give some clarifications regarding the microstructure evolution and the mechanisms of carbon enrichment into austenite. After having determined the optimum Q&P parameters using dilatometric and XRD measurements, a Q&P treatment at three different QT (260, 230 and 200°C) and at a partitioning temperature of 400°C was applied to a model Fe-0.3 C-2.5Mn-1.5Si steel. The dilatometric data combined with an SEM image analysis study showed that bainite forms during the partitioning step. The presence of bainite was also confirmed by in-situ High Energy X-Ray Diffraction. While bainite was shown to appear as carbide free laths, tempered martensite showed an advanced state of intra-lath precipitation. The combination of atom probe tomography (APT) and TEM technics showed that theses carbides are transitional andboth η and ε carbides were observed. Their carbon content ranged from 20.0 to 27.7 at.%. APT measurements also highlighted carbon segregation on martensite defects during the initial quench and calculation of the evolution of the carbon excess concentration on laths boundaries suggest that desegregation occurs along the Q&P treatment. In-situ HEXRD permitted to follow the austenite lattice parameter evolution and it was shown that austenite is subjected to a sequence of tensile andcompression state induced by the formation of martensite. A model for the coefficient of thermal expansion of austenite taking into account its stress state was successfully developed. The evolutions of carbon content into austenite for the three QT were determined. Surprisingly the carbon enrichment into austenite was shown not to depend on QT. It was also shown that the increase of carbon content in austenite results from both carbon partitioning and bainite contributions. Lastly, an originaltheoretical approach was developed. It was evidenced that bainite continues to form while partitioning process is rapidly completed, thus bainite transformation controls the maximum austenite carbon enrichment at 400°C, independently of QT. The contribution of partitioning from martensite was shown to be larger with decreasing QT. The developed model successfully described the experimentally observed phase transformations and austenite carbon enrichment by taking into account theinteractions between carbon partitioning, bainite transformation and carbide precipitation.

Breast cancer is one of the deadliest cancers for women. In the US, 1 in 8 women will be diagnosed with breast cancer within their lifetimes. Detection and diagnosis play an important role in saving lives. To this end, many classifiers with varying structures have been designed to classify breast cancer histopathological images. However, randomly partitioning data, like many previous works have done, can lead to artificially inflated accuracies and classifiers that do not generalize. Data leakage occurs when researchers assume that every image in a dataset is independent of each other, which is often not the case for medical datasets, where multiple images are taken of each patient. This work focuses on convolutional neural network binary classifiers using the BreakHis dataset. Previous works are reviewed. Classifiers from previous literature are tested with patient partitioning, where individual patients are placed in the training, testing and validation sets so that there is no overlap. A classifier which previously achieved 93% accuracy consistently, only achieved 79% accuracy with the new patient partition. Robust data augmentation, a Sigmoid output layer and a different form of min-max normalization were utilized to achieve an accuracy of 89.38%. These improvements were shown to be effective with the architectures used. Sigmoid Model 1.1 is shown to perform well compared to much deeper architectures found in literature.

Hypergraph partitioning lies at the heart of a number of problems in machine learning as well as other engineering disciplines. While partitioning uniform hypergraphs is often required in computer vision problems that involve multi-way similarities, non-uniform hypergraph partitioning has applications in database systems, circuit design etc. As in the case of graphs, it is known that for given objective and balance constraints, the problem of optimally partitioning a hypergraph is NP-hard. Yet, over the last two decades, several efficient heuristics have been studied in the literature and their empirical success is widely appreciated. In contrast to the extensive studies related to graph partitioning, the theoretical guarantees of hypergraph partitioning approaches have not received much attention in the literature. The purpose of this thesis is to establish the statistical error bounds for certain spectral algorithms for partitioning uniform as well as non-uniform hypergraphs. The mathematical framework considered in this thesis is the following. Let V be a set of n vertices, and ψ : V ->{1,…,k} be a (hidden) partition of V into k classes. A random hypergraph (V,E) is generated according to a planted partition model, i.e., subsets of V are independently added to the edge set E with probabilities depending on the class memberships of the participating vertices. Let ψ' be the partition of V obtained from a certain algorithm acting on a random realization of the hypergraph. We provide an upper bound on the number of disagreements between ψ and ψ'. To be precise, we show that under certain conditions, the asymptotic error is o(n) with probability (1-o(1)). In the existing literature, such error rates are only known in the case of graphs (Rohe et al., Ann. Statist., 2011; Lei \& Rinaldo, Ann. Statist., 2015), where the planted model coincides with the popular stochastic block model. Our results are based on matrix concentration inequalities and perturbation bounds, and the derived bounds can be used to comment on the consistency of spectral hypergraph partitioning algorithms. It is quite common in the literature to resort to a spectral approach when the quantity of interest is a matrix, for instance, the adjacency or Laplacian matrix for graph partitioning. This is certainly not true for hypergraph partitioning as the adjacency relations cannot be encoded into a symmetric matrix as in the case of graphs. However, if one restricts the problem to m-uniform hypergraphs for some m ≥ 2, then a symmetric tensor of order m can be used to express the multi-way interactions or adjacencies. Thus, the use of tensor spectral algorithms, based on the spectral theory of symmetric tensors, is a natural choice in this scenario. We observe that a wide variety of uniform hypergraph partitioning methods studied in the literature can be related to any one of two principle approaches: (1) solving a tensor trace maximization problem, or (2) use of the higher order singular value decomposition of tensors. We derive statistical error bounds to show that both these approaches lead to consistent partitioning algorithms. Our results also hold when the hypergraph under consideration allows weighted edges, a situation that is commonly encountered in computer vision applications such as motion segmentation, image registration etc. In spite of the theoretical guarantees, a tensor spectral approach is not preferable in this setting due to the time and space complexity of computing the weighted adjacency tensor. Keeping this practical scenario in mind, we prove that consistency can still be achieved by incorporating certain tensor sampling strategies. In particular, we show that if the edges are sampled according to certain distribution, then consistent partitioning can be achieved with only few sampled edges. Experiments on benchmark problems demonstrate that such sampled tensor spectral algorithms are indeed useful in practice. While vision tasks mostly involve uniform hypergraphs, in database and electronics applications, one often finds non-uniform hypergraphs with edges of varying sizes. These hypergraphs cannot be expressed in terms of adjacency matrices or tensors, and hence, use of a spectral approach is tricky in this context. The partitioning problem gets more challenging due to the fact that, in practice, these hypergraphs are quite sparse, and hence, provide less information about the partition. We consider spectral algorithms for partitioning clique and star expansions of hypergraphs, and study their consistency under a sparse planted partition model. The results of hypergraph partitioning can be further extended to address the well-known hypergraph vertex coloring problem, where the objective is to color the vertices such that no edge is monochromatic. The hardness of this problem is well established. In fact, even when a hypergraph is bipartite or 2-colorable, it is NP-hard to find a proper 2-coloring for it. We propose a spectral coloring algorithm, and show that if the non-monochromatic subsets of vertices are independently added to the edge set with certain probabilities, then with probability (1-o(1)), our algorithm succeeds in coloring bipartite hypergraphs with only two colors. To the best our knowledge, these are the first known results related to consistency of partitioning general hypergraphs.

Per- and polyfluoroalkyl substances (PFASs) have drawn great attention recently, due to their environmental persistence, potential toxicity and global distribution. PFAS is a large family of substances, characterized by a perflourinated carbon chain and a functional group. All PFASs are synthetic and have been widely used since the 1950s due to their unique properties of being both hydrophobic and oleophobic, making them useful for many industries. To be able to predict the fate of PFASs in the environment and to obtain detailed understanding of the transport processes, their partitioning behavior between soil particles and water depending on a range of parameters must be investigated. The aims of this study was to investigate the effects of pH, cation composition, functional group and perfluorocarbon chain length on sorption of PFASs to soil particles, by batch sorption experiment in laboratory scale. The laboratory-scale experiments were combined with modelling of the net charge to evaluate if net charge is a good predictor for sorption of PFASs to soil particles. 14 PFASs of varying length and functional groups were studied (PFBA, PFPeA, PFHxA, PFHpA, PFOA, PFNA, PFDA, PFUnDA, PFDoDA, PFTeDA, PFBS, PFHxS, PFOS and FOSA). The effect on sorption of Na+, Ca2+ (two different concentrations) and Al3+ were investigated at pH-range 3-6. Modelling of net charge was carried out in the geochemical model Visual MINTEQ. The soil had 45% organic carbon content. The adsorption of PFASs was strongly correlated with perfluorocarbon chain length, showing a stronger adsorption to particles with increasing perfluorocarbon chain length (i.e. more hydrophobic). The relation between sorption (represented by the distribution coefficient log Kd) and perfluorocarbon chain length was linear for all PFSAs and C3 to C10 PFCAs. The PFSAs (sulfonate functional group) sorbed stronger to soil particles than the PFCAs (carboxylic functional group), and FOSA (sulfonamide functional group) sorbed the strongest. For most PFCAs, (C5-C13) there was a trend of decreasing log Kd (i.e. decreased sorption) with increasing pH, due to pH-dependent changes of the soil particle surfaces. For short and intermediate perfluorocarbon chain length PFCAs (C5-C8) and for PFHxS among the PFSAs, cations had a clear effect on sorption. Aluminium ions (trivalent, Al(NO3)3) had the largest effect, followed by calcium (divalent, Ca(NO3)2) where higher concentration resulted in stronger sorption. Sodium (univalent, NaNO3) had the least influence on sorption. The net charge modelled with Visual MINTEQ takes into account many parameters (including pH) that affect the surface charge and sorption of PFASs to soil particles. When comparing log Kd for the different PFASs with pH and net negative charge, net charge was a better predictor of sorption of PFASs to soil particles than solution pH alone. / Per- och polyfluoroalkylsubstanser (PFAS:er) har dragit stor uppmärksamhet till sig på senare tid, på grund av deras persistenta egenskaper, potentiella toxicitet och globala utbredning. PFAS är en stor grupp ämnen, kännetecknad av en perflourinerad kolkedja och en funktionell grupp. Alla PFAS är syntetiska och har använts i stor utsträckning sedan 1950-talet på grund av deras unika egenskaper av att vara både vatten- och fettavstötande, vilket gör dem användbara för många industriella tillämpningar. För att kunna förutsäga var dessa föroreningars hamnar i miljön och få mer detaljerad förstående för transportprocesserna, måste deras fördelningbeteende mellan jordpartiklar och vattenundersökas för en rad olika parametrar. Syftet med denna studie var att undersöka effekterna av förändrat pH, katjonsammansättning, funktionell grupp och perfluorkolkedjelängd på sorption av PFAS:er till jordpartiklar. Detta gjordes med sorptionsexperiment i laboratorieskala. Laboratorieexperimentet kompletterades med modellering av nettoladdning, för att se huruvida detta väl kunde förklara sorptionen till jordpartiklar. 14 PFAS:er av varierande längd och med tre olika funktionella grupper studerades (PFBA, PFPeA, PFHxA, PFHpA, PFOA, PFNA, PFDA, PFUnDA, PFDoDA, PFTeDA, PFBS, PFHxS, PFOS och FOSA). Effekten på sorption av Na+, Ca2+ (två olika koncentrationer) och Al3+ undersöktes vid pH-intervallet 3-6. Modellering av nettoladdning utfördes i den geokemiska modellen Visual MINTEQ. Jorden som användes hade en halt av organiskt kol på 45%. Adsorptionen av PFAS:er var starkt positivt korrelerad med kedjelängden på de perfluorinerade kolkedjan. Ju längre kolkedja (dvs. mer hydrofob), desto starkare adsorption till partiklar. Relationen mellan sorptionen (här uttryckt som partitioneringskofficienten log Kd) och kedjelängd var linjär för alla PFSA och för C3 till C10 för PFCA. PFSA (sulfonat) adsorberade starkare än PFCA (karboxyl), och FOSA (sulfonamid) adsorberades starkast. För de flesta PFCA, (C5-C13) fanns en allmän trend där log Kd (dvs. sorption) minskade med ökande pH, på grund av pH-beroende förändringar på jordpartiklarna. För korta och medellånga PFCA (C5-C8) och för PFHxS hade katjonsammansättningen en tydlig effekt på sorptionen. Aluminiumjoner (trevärd, Al(NO3)3) hade den största effekten, följt av kalcium (tvåvärd, Ca(NO3)2) där den högre koncentrationen resulterade i starkare sorption. Natrium (envärd, NaNO3) hade minst påverkan på sorptionen till jordpartiklar. Visual MINTEQ tar hänsyn till många parametrar (inklusive pH), när nettoladdningen på jordpartiklarnas yta räknas ut. När log Kd för olika PFAS:er jämfördes med endera pH eller negativ nettoladdning, drogs slutsatsen att nettoladdning korrelerade bättre med sorption än pH.

The problem of how sand and mud was distributed downslope, within linked alluvial-brackish water-marine shoreline systems of an extensive clastic wedge is addressed here. The Iles Clastic wedge accumulated over a time period of a few million years (my), and its component high-frequency regressive-transgressive sequences have a duration of a few 100 thousand years (ky). The sediment partitioning study provides insight into where the thickest sandstones and mudstones were located, and generates a model that can be applied to improving the management of hydrocarbons or water resources. A 300 km 2-D study transect across the Iles Clastic Wedge in SW Wyoming and N Colorado included subsurface well log information and outcrop stratigraphic columns. This information was used to correlate high-frequency sequences across several hundred kilometers, characterize depositional processes from proximal to distal reaches, develop a sediment partitioning model, and understand the role of the likely drivers in the development of the wedge and its internal sequences. The main results of this study are: (1) The Iles Clastic Wedge spans 3 my (500 m thick) and is composed internally of 11 sequences of 200-400 ky, each of which have significant regressive-transgressive transits of up to 90 km. Sediment partitioning analysis shows that within the regressive limb of the large wedge, the component regressive compartments tend to thicken basinwards, whereas transgressive compartments thicken landwards. This geometry is driven by preferential erosion in proximal areas during regression, bypassing much sediment to the marine shorelines, and transgressive backfilling into proximal areas previously eroded more deeply. (2) The greatest concentration of sands tends to be located in the proximal fluvial and estuarine facies of the transgressive compartments and within the medial shoreline/deltaic facies of the regressive compartments. (3) As the high-frequency sequences developed, the effectiveness of basinward sand partitioning reaches a maximum value near the peak regression level of the wedge, reflecting stronger erosion and sediment bypass during this times. (4) The development of the Iles Clastic Wedge was influenced by both tectonic and eustatic drivers, with important tectonic control in the upstream reaches. On a 4th-order timescale, the Iles Wedge internal sequences were likely influenced mainly by eustasy. / text

[pt] A presente tese se insere no contexto de previsão de séries temporais. Nesse sentido, embora muitas abordagens tenham sido desenvolvidas, métodos simples como o de amortecimento exponencial costumam gerar resultados extremamente competitivos muitas vezes superando abordagens com maior nível de complexidade. No contexto previsão, papers seminais na área mostraram que a combinação de previsões tem potencial para reduzir de maneira acentuada o erro de previsão. Especificamente, a combinação de previsões geradas por amortecimento exponencial tem sido explorada em papers recentes. Apesar da combinação de previsões utilizando Amortecimento Exponencial poder ser feita de diversas formas, um método proposto recentemente e chamado de Bagged.BLD.MBB.ETS utiliza uma técnica chamada Bootstrap Aggregating (Bagging) em combinação com métodos de amortecimento exponencial para gerar previsões mostrando que a abordagem é capaz de gerar previsões mensais mais precisas que todos os benchmarks analisados. A abordagem era considerada o estado da arte na utilização de Bagging e Amortecimento Exponencial até o desenvolvimento dos resultados obtidos nesta tese. A tese em questão se ocupa de, inicialmente, validar o método Bagged.BLD.MBB.ETS em um conjunto de dados relevante do ponto de vista de uma aplicação real, expandindo assim os campos de aplicação da metodologia. Posteriormente, são identificados motivos relevantes para redução do erro de e é proposta uma nova metodologia que utiliza Bagging, Amortecimento Exponencial e Clusters para tratar o efeito covariância, até então não identificado anteriormente na literatura do método. A abordagem proposta foi testada utilizando diferentes tipo de séries temporais da competição M3, CIF 2016 e M4, bem como utilizando dados simulados. Os resultados empíricos apontam para uma redução substancial na variância e no erro de previsão. / [en] This thesis is inserted in the context of time series forecasting. In this sense, although many approaches have been developed, simple methods such as exponential smoothing usually produce extremely competitive results, often surpassing approaches with a higher level of complexity. Seminal papers in time series forecasting showed that the combination of forecasts has the potential to dramatically reduce the forecast error. Specifically, the combination of forecasts generated by Exponential Smoothing has been explored in recent papers. Although this can be done in many ways, a specific method called Bagged.BLD.MBB.ETS uses a technique called Bootstrap Aggregating (Bagging) in combination with Exponential Smoothing methods to generate forecasts, showing that the approach can generate more accurate monthly forecasts than all the analyzed benchmarks. The approach was considered the state of the art in the use of Bagging and Exponential Smoothing until the development of the results obtained in this thesis. This thesis initially deals with validating Bagged.BLD.MBB.ETS in a data set relevant from the point of view of a real application, thus expanding the fields of application of the methodology. Subsequently, relevant motifs for error reduction are identified and a new methodology using Bagging, Exponential Smoothing and Clusters is proposed to treat the covariance effect, not previously identified in the method s literature. The proposed approach was tested using data from three time series competitions (M3, CIF 2016 and M4), as well as using simulated data. The empirical results point to a substantial reduction in variance and forecast error.

L’objectif principal de ce travail de thèse est d’évaluer l'impact des activités anthropiques sur le fonctionnement des systèmes aquatiques. L’impact des rejets urbains de Fès sur le Sebou, l'une des plus grandes rivières du Maroc, a été choisi pour cette étude. Les eaux usées domestiques et industrielles de la ville de Fès (~1M hab), véhiculées par son affluent l’oued Fès, sont rejetées dans le Sebou quasiment sans traitement. Deux sites du Sebou, en amont et en aval des rejets de la ville de Fès, et un site situé sur l’oued Fès ont été étudiés. Une campagne de prélèvement de carottes de sédiments et onze campagnes mensuelles de prélèvement d’eau et de matières en suspension (MES) ont été effectuées dans le but de quantifier les apports en nutriments, éléments traces métalliques (ETM) et carbone organique afin d’étudier la dynamique de ces polluants. Les résultats obtenus ont mis en évidence une augmentation des concentrations de presque tous les éléments étudiés en aval de la ville de Fès aussi bien dans la colonne d’eau que dans les sédiments. L’étude des carottes sédimentaires a révélé une contamination modérée par les butylétains totaux (ΣBT) avec la prédominance du monobutyétain dans les trois sites d’étude et sur toutes les profondeurs. Les sédiments du Sebou, en amont de la ville de Fès se caractérisent par les teneurs les plus faibles en métaux alors que ceux de l’oued Fès présentent une forte pollution polymétallique, accentuée dans les sédiments de surface, ce qui reflète clairement la signature d'apports anthropiques récents résultant des rejets non traités de la ville de Fès. L’augmentation consécutive des teneurs des métaux dans les sédiments du Sebou en aval de la confluence Fès-Sebou traduit l'influence significative des particules polluées de l’oued Fès. Dans la colonne d’eau, les concentrations mesurées dans le Sebou en amont de la confluence Fès-Sebou sont proches des rivières naturelles, à l’exception de Cl-, Cr, Na+ et NO3- dont les concentrations traduisent des pollutions agricoles et/ou domestiques. En revanche, la signature anthropique des rejets de la ville de Fès apparait évidente dans les eaux de l’oued Fès qui montrent des concentrations très élevées en ETM dissous et particulaires. La majorité des ETM provenant des apports anthropiques subissent des changements importants de leur coefficient de distribution Kd et présentent un comportement non conservatif dans le mélange entre les eaux du Sebou et celles de l’oued Fès. Ces résultats ont été confirmés par le modèle WHAM qui permet de prédire correctement le fractionnement dissous/particulaire, du Cu, Pb et Zn. La spéciation chimique de ces éléments ainsi que leur répartition dissous/particulaire apparaissent significativement influencées par les conditions particulières (anoxie, forte contamination, teneurs importantes en matière organique) des eaux de l’oued Fès. Ces conditions sont aussi à l’origine du comportement non conservatif de la majorité des éléments lors d'un mélange Sebou/Fès, observé aussi bien dans la colonne d’eau que dans les sédiments. / The main goal of this PhD was to evaluate the impact of the anthropogenic activities (urban, industrial or agricultural activities) on aquatic systems. The Fez metropolitan area and its impacts on the Sebou River, the main Moroccan river, were chosen as a case study. The Fez agglomeration (~1M hab), is surrounded by the Fez River, receiving the wastewaters of this developing city and then flowing into the Sebou. This work was performed on two sites along the Sebou River (upstream and downstream from the confluence with Fez River), and one site in the Fez River, downstream from the Fez city. One sampling campaigns of sediment cores and eleven sampling campaigns were performed during low flow conditions to quantify nutrients, organic carbon and trace metals and to study the dynamic and the transport of trace elements in the Sebou and Fez rivers. The obtained results revealed a global increase of studied elements in the Sebou River from upstream to downstream sampling site, in both sediments and water column. A moderate level of contamination by butyltins was observed, with monobutyltin being the dominant species across all sites and depths. The lowest level of metal pollution was identified in the Sebou's sediments upstream Fez city. Whilst the Fez' sediments were heavily polluted and exhibited bottom-up accumulation trends, which clearly evidence recent inputs from the untreated wastewaters of Fez city. The increase of metal levels in Sebou downstream sediments reflects a significant contribution of polluted particles from the Fez River. Dissolved and particulate concentrations of most of the analyzed elements in the Sebou, upstream from Fez city, are close to the natural rivers, except Cl-, Cr, Na+ et NO3- probably due to untreated urban inputs from several small/medium-size towns located upstream Fez. In the opposite On the contrary, high dissolved and particulate trace metals concentrations in the Fez River clearly indicated strong anthropogenic inputs from Fez city. Most of the studied trace metals originated from anthropogenic sources, underwent significant changes of Kd and behaved non-conservatively in the Sebou/Fez waters mixing due to drastic changes of the waters quality (anoxic condition, high concentration, high organic matter content). Dissolved/particulate partitioning were correctly assessed by WHAM-VII modeling for Cu, Pb and Zn, depicting significant differences in chemical speciation in Fez River when compared to Sebou one.

Water is a vital resource for species inhabiting arid and semi-arid regions and can shape the biotic communities that we observe. Because water is considered a limiting resource for many species in desert environments, there is the potential for competitive interactions between species to occur at or around water sources. For this dissertation I tested hypotheses related to resource competition among different species of wildlife in the Great Basin and Mojave Deserts of western Utah. Chapter one evaluated the influence of feral horses (Equus caballus) on patterns of water use by communities of native birds and mammals. Chapter two determined if feral horses competed with pronghorn (Antilocapra americana) and mule deer (Odocoileus hemionus) for access to water. In chapters one and two, we found evidence that horses compete with native wildlife for water. In chapter one, horses were associated with decreased richness and diversity of native species at water sources. Native species also had fewer visits and spent less time at water sources frequented by horses. In chapter two, we found that pronghorn and mule deer used water sources less often where horse activity was high. There were also significant differences in temporal activity for pronghorn, but not mule deer, at horse-occupied sites versus sites where horses were absent or uncommon. Our results indicated that horses spatially and temporally displaced other species at water sources providing evidence of a negative influence on how communities of native wildlife access a limited resource in an arid environment. Chapter three assessed whether dominant carnivores (coyote (Canis latrans) and bobcat (Lynx rufus)) negatively influenced the spatial use of water sources by the subordinate kit fox (Vulpes macrotis). Our results did not reveal strong negative associations between kit fox visits to water sources and visits by dominant carnivores; in fact, dominant carnivores contributed very little to the use of water by kit foxes. Instead, kit fox visits were more closely associated with habitat features at water sources. Our findings indicate that dominant carnivores are not the primary driver of use of water sources by subordinate carnivores. Chapter four evaluated whether a simulated loss of water due to climate change/increased human use would differentially affect desert bats based on flight morphology and maneuverability. When we experimentally reduced surface area of water sources, larger, less-maneuverable bats experienced a 69% decrease in drinking success and increased competition with smaller, maneuverable bats. Anticipated reductions in the sizes of water sources due to climate change may lead to species with less maneuverability being unable to access water efficiently and facing increased competition from more agile bats.