A previsão do PIB é um dos principais balizadores para as decisões produtivas de agentes econômicos. Com o objetivo de realizar previsões para o crescimento do PIB trimestral brasileiro, são utilizadas 16 séries mensais financeiras e econômicas como potenciais preditores, abrangendo o período do segundo trimestre de 1996 ao quarto trimestre de 2012. Para isso, aplicou-se as abordagens MIDAS (Mixed Data Sampling) e UMIDAS (Unrestricted Mixed Data Sampling), confrontando seus resultados de previsão fora da amostra com o benchmark ARMA. Foram encontrados erros de previsão menores nessas abordagens, principalmente quando utilizadas informações dentro do trimestre de previsão. Os resultados foram ainda melhores quando empregados múltiplos regressores. / The GDP forecast is an important indicator for production decisions taken by economic agents. In order to make forecasts for the Brazilian quarterly GDP growth, we used 16 monthly financial and economic series as potential predictors, covering the period from the second quarter of 1996 to the fourth quarter of 2012. For this purpose, we applied MIDAS (Mixed Data Sampling) and UMIDAS (Unrestricted Mixed Data Sampling) approaches and compared the out of sample forecasts with the benchmark ones provided by ARMA. MI- DAS and UMIDAS showed smaller prediction errors, especially when information inside the quarter forecast is used. The results were even better when multiple regressors were employed.

This research investigates the moderating effect of cultural differences between countries on the relationship between tenure and analyst accuracy. To investigate this research looks at the expected earnings per shares and the realised earnings per share from the shares included into the AEX, CAC, DAX and FTSE. The dataset consists of 466 analysts and 3.040 observations. The time period observed is 2015, 2016 and 2017. This research shows that there is no significant relationship between tenure and analyst accuracy. The results show that masculinity, individualism and long term orientation have a moderating effect on analyst accuracy. A practical implication is that managers could employ methods to change the work culture to increase analyst accuracy. An academic implication of this research is that culture should be included as a moderating factor in future analyst accuracy research.

Introdução. O estudo e compreensão dos efeitos da poluição atmosférica podem contribuir para o planejamento de estratégias de controle de emissões de poluentes e na tomada de decisões em relação à saúde pública. Modelos de previsão da poluição do ar são importantes, na medida em que podem antecipar precauções e providências de ações públicas. Objetivo. Elaborar e analisar modelos de previsão do ozônio troposférico para a Região Metropolitana de São Paulo (RMSP). Métodos. Foram ajustados modelos de previsão de ozônio utilizando redes neurais artificiais (RNAs), denominadas técnicas de inteligência artificial. Os dados de entrada do modelo foram os meteorológicos, obtidos do CPTEC - Centro de Previsão de Tempo e Estudos Climáticos e do INMET - Instituto Nacional de Meteorologia e os dados do poluente ozônio monitorados pela CETESB - Companhia Ambiental do Estado de São Paulo. Foram considerados, para o ozônio, o padrão nacional de qualidade do ar (1 hora) e o padrão estadual de qualidade do ar (8 horas). Os dados foram distribuídos entre as médias do período da manhã (07h às 12h) e as médias do período da tarde (13h às 18h), obtendo-se como saída as concentrações máximas de ozônio para o período da tarde. O período analisado foi de 2008 a 2014. Resultados. Foram realizados 311 testes distribuídos de acordo com o padrão de qualidade do ar do ozônio (O3-1h ou O3-8h) e a origem dos dados meteorológicos (CPTEC ou INMET). Os valores de ozônio observados e estimados mostraram-se muito bem correlacionados. Para os ajustes usando o banco de dados do CPTEC, os melhores resultados das estatísticas de verificação para O3-1h foram: A=90 por cento ; B=0,41; FAR=47 por cento ; POD=22 por cento ; r=0,60. Sendo A a porcentagem de acertos nas previsões de eventos e não eventos; B indica, na média, se as previsões estão subestimadas ou superestimadas; FAR é a porcentagem de concentrações que foram previstas altas e que não ocorreram; POD é a capacidade de prever altas concentrações ( por cento ) e r é o coeficiente de correlação entre o valor observado e o valor estimado. Para O3-8h: A=96 por cento ; B=0,1; FAR=14 por cento ; POD=6,5 por cento ; r=0,72. Considerando-se o banco de dados do INMET, os melhores resultados para O3-1h foram: A=93 por cento ; B=0,54; FAR=29 por cento ; POD=38 por cento , r=0,84. Para O3-8h: A=95 por cento ; B=0,76; FAR=47 por cento ; POD=40 por cento ; r=0,85. As variáveis temperatura e vento meridional foram as mais importantes nos modelos. Conclusões. No geral, os modelos simulados com os dados meteorológicos do INMET apresentaram melhores resultados que os apresentados pelos dados do CPTEC, tanto para O3-1h, quanto para O3-8h. O modelo simulado com os dados do INMET, considerando O3-8h, apresentou melhor previsibilidade. Os modelos ajustados por redes neurais mostraram-se como boas alternativas de instrumentos para prever a concentração de ozônio na RMSP. / Introduction. The study and understanding of the effects of air pollution can contribute to the planning of pollutant emission control strategies and decision-making in relation to public health. Air pollution forecasting models are important, as they can anticipate precautions and actions of public action. Objetive. Develop and analyze tropospheric ozone forecasting models for the São Paulo Metropolitan Area (SPMA). Methods. Ozone forecasting models were adjusted using artificial neural networks (ANNs), called artificial intelligence techniques. The model input data were the weather, obtained from CPTEC - Weather and Climate Studies Prediction Center and INMET - National Meteorology Institute and the pollutant ozone data monitored by CETESB - São Paulo State Environmental Company. Were considered for ozone, the national standard of air quality (1 hour) and the state standard of air quality (8 hours). Data were distributed among the averages of the morning (07h to 12h) and the average of the afternoon (13h to 18h), obtaining as output the maximum concentrations of ozone to the afternoon. The study period was from 2008 to 2014. Results. Were conducted 311 tests distributed according to the standard of ozone air quality (O3-1h or O3-8h) and the source of meteorological data (CPTEC or INMET). The observed and estimated ozone values were shown to be very well correlated. For the settings using the CPTEC database, the best results of the verification statistics for O3-1h were: A= 90 per cent ; B=0.41; FAR=47 per cent ; POD=22 per cent ; r=0.60. Where A is the percentage of correct answers of forecasts in the events and not events; B indicates, on average, if the predictions are underestimated or overestimated; FAR is the percentage concentrations that were predicted high and that did not occur; POD is the ability to predict high concentrations ( per cent ) and r is the correlation coefficient between the observed value and the estimated value. To O3-8h: A=96 per cent ; B=0.1; FAR=14 per cent ; POD=6.5 per cent ; r=0.72. Considering the INMET database, the best results for O3-1h were: A=93 per cent ; B=0.54; FAR=29 per cent ; POD=38 per cent , r=0.84. To O3-8h: A=95 per cent ; B=0.76; FAR=47 per cent ; POD=40 per cent ; r=0.85. The variables temperature and meridional wind were the most importante in the models. Conclusions. Overall, the simulated models with meteorological INMET data showed better results than those presented by the CPTEC data for both O3-1h, and for O3-8h. The simulated model with INMET data, considering O3-8h, presented better predictability. The models adjusted by neural networks showed up as good instruments to predict the ozone concentration in the SPMA.

De nos jours, les pays sont amenés à prendre des mesures visant à une meilleure rationalisation des ressources en électricité dans une optique de développement durable. Des solutions de comptage communicantes (Smart Meters), sont mises en place et autorisent désormais une lecture fine des consommations. Les données spatio-temporelles massives collectées peuvent ainsi aider à mieux connaitre les habitudes de consommation et pouvoir les prévoir de façon précise. Le but est d'être en mesure d'assurer un usage « intelligent » des ressources pour une meilleure consommation : en réduisant par exemple les pointes de consommations ou en ayant recours à des sources d'énergies renouvelables. Les travaux de thèse se situent dans ce contexte et ont pour ambition de développer des outils de fouille de données en vue de mieux comprendre les habitudes de consommation électrique et de prévoir la production d'énergie solaire, permettant ensuite une gestion intelligente de l'énergie.Le premier volet de la thèse s'intéresse à la classification des comportements types de consommation électrique à l'échelle d'un bâtiment puis d'un territoire. Dans le premier cas, une identification des profils types de consommation électrique journalière a été menée en se basant sur l'algorithme des K-moyennes fonctionnel et sur un modèle de mélange gaussien. A l'échelle d'un territoire et en se plaçant dans un contexte non supervisé, le but est d'identifier des profils de consommation électrique types des usagers résidentiels et de relier ces profils à des variables contextuelles et des métadonnées collectées sur les usagers. Une extension du modèle de mélange gaussien classique a été proposée. Celle-ci permet la prise en compte de variables exogènes telles que le type de jour (samedi, dimanche et jour travaillé,…) dans la classification, conduisant ainsi à un modèle parcimonieux. Le modèle proposé a été comparé à des modèles classiques et appliqué sur une base de données irlandaise incluant à la fois des données de consommations électriques et des enquêtes menées auprès des usagers. Une analyse des résultats sur une période mensuelle a permis d'extraire un ensemble réduit de groupes d'usagers homogènes au sens de leurs habitudes de consommation électrique. Nous nous sommes également attachés à quantifier la régularité des usagers en termes de consommation ainsi que l'évolution temporelle de leurs habitudes de consommation au cours de l'année. Ces deux aspects sont en effet nécessaires à l'évaluation du potentiel de changement de comportement de consommation que requiert une politique d'effacement (décalage des pics de consommations par exemple) mise en place par les fournisseurs d'électricité.Le deuxième volet de la thèse porte sur la prévision de l'irradiance solaire sur deux horizons temporels : à court et moyen termes. Pour ce faire, plusieurs méthodes ont été utilisées parmi lesquelles des méthodes statistiques classiques et des méthodes d'apprentissage automatique. En vue de tirer profit des différents modèles, une approche hybride combinant les différents modèles a été proposée. Une évaluation exhaustive des différents approches a été menée sur une large base de données incluant des paramètres météorologiques mesurés et des prévisions issues des modèles NWP (Numerical Weather Predictions). La grande diversité des jeux de données relatifs à quatre localisations aux climats bien distincts (Carpentras, Brasilia, Pampelune et Ile de la Réunion) a permis de démontrer la pertinence du modèle hybride proposé et ce, pour l'ensemble des localisations / Nowadays, countries are called upon to take measures aimed at a better rationalization of electricity resources with a view to sustainable development. Smart Metering solutions have been implemented and now allow a fine reading of consumption. The massive spatio-temporal data collected can thus help to better understand consumption behaviors, be able to forecast them and manage them precisely. The aim is to be able to ensure "intelligent" use of resources to consume less and consume better, for example by reducing consumption peaks or by using renewable energy sources. The thesis work takes place in this context and aims to develop data mining tools in order to better understand electricity consumption behaviors and to predict solar energy production, then enabling intelligent energy management.The first part of the thesis focuses on the classification of typical electrical consumption behaviors at the scale of a building and then a territory. In the first case, an identification of typical daily power consumption profiles was conducted based on the functional K-means algorithm and a Gaussian mixture model. On a territorial scale and in an unsupervised context, the aim is to identify typical electricity consumption profiles of residential users and to link these profiles to contextual variables and metadata collected on users. An extension of the classical Gaussian mixture model has been proposed. This allows exogenous variables such as the type of day (Saturday, Sunday and working day,...) to be taken into account in the classification, thus leading to a parsimonious model. The proposed model was compared with classical models and applied to an Irish database including both electricity consumption data and user surveys. An analysis of the results over a monthly period made it possible to extract a reduced set of homogeneous user groups in terms of their electricity consumption behaviors. We have also endeavoured to quantify the regularity of users in terms of consumption as well as the temporal evolution of their consumption behaviors during the year. These two aspects are indeed necessary to evaluate the potential for changing consumption behavior that requires a demand response policy (shift in peak consumption, for example) set up by electricity suppliers.The second part of the thesis concerns the forecast of solar irradiance over two time horizons: short and medium term. To do this, several approaches have been developed, including autoregressive statistical approaches for modelling time series and machine learning approaches based on neural networks, random forests and support vector machines. In order to take advantage of the different models, a hybrid model combining the different models was proposed. An exhaustive evaluation of the different approaches was conducted on a large database including four locations (Carpentras, Brasilia, Pamplona and Reunion Island), each characterized by a specific climate as well as weather parameters: measured and predicted using NWP models (Numerical Weather Predictions). The results obtained showed that the hybrid model improves the results of photovoltaic production forecasts for all locations

La génération actuelle des sondeurs infrarouges avancés constitue l’une des sources les plus importantes d’observation dans les systèmes d’assimilation de données dans les modèles de la Prévision Numérique du Temps (PNT). Cependant la richesse d’informations fournies par ce type de capteur avec son grand nombre de canaux et sa couverture globale est loin d’être complètement exploitée. La présence de nuages dans le champ de vision de l’instrument, qui affecte la majorité des observations, est l’une des raisons pour lesquelles les centres de PNT rejettent une grande quantité des observations des sondeurs. Les centres de PNT ont cependant commencé à assimiler au-dessus des océans les radiances affectées par les nuages en utilisant des canaux dont les effets radiatifs nuageux sont modélisés par un modèle de nuage simple. Certains de ces algorithmes de détection sont évalués dans ce manuscrit, et leurs limitations sont explicitées. Afin d’accroître la quantité de données assimilées, il est nécessaire de mieux représenter les nuages et leurs effets radiatifs. Depuis quelques années, des études ont été menées pour mieux représenter leurs effets dans les modèles de transfert radiatif ([Faijan et al., 2012] ; [Martinet et al., 2013]) ; et utiliser dans l’assimilation de nouveaux canaux infrarouges informatifs sur les hydrométéores nuageux. ([Martinet et al., 2014]). Cette thèse se concentre principalement sur ces méthodes de détection de scènes homogènes en consacrant sa majeur partie à l’établissement, l’évaluation et l’amélioration d’algorithme de détection de scènes homogènes en se basant sur la colocalistion d’observation avec d’autres sondeurs. Ces études sont rendus possibles par la prise en compte des champs d’hydrométéores fournis par les schémas convectifs du modèle ARPEGE en entrée du modèle de transfert radiatif nuageux RTTOV-CLD. Une partie validation des simulations est opérée dans cette thèse, en comparant l’apport les forces et faiblesses du schéma convectif en opérationnel ainsi que PCMT. Par la suite, différents tests, ou critères, de détection sont proposés, et en réalisant des expériences d’assimilation et en évaluant l’impact de ces ces critères de sélection proposés sur la qualité des prévisions à longues échéances, un des tests parmi ceux proposés se démarque des autres en conservant une quantité importante d’observation ciel clair et démontre des impacts neutres à légèrement positifs sur les prévisions. Les nouvelles méthodes de sélection de scènes homogènes proposées dans cette études permettent d’envisager une amélioration significative du contrôle de qualité des observation IASI en ciel clair. Cela ouvre ainsi donc la voie à une utilisation, plus maîtrisée, des scènes nuageuses. Nous expliquons dans ce manuscrit pourquoi il serait imprudent de précéder à des assimilation de radiances infrarouge contaminées par la présence de nuages. Pour contourner cette difficulté, une technique d’assimilation en deux étapes déjà utilisé pour l’assimilation des réflectivité radar ([Wattrelot et al., 2014]) dans AROME est évaluée. Cette méthode basée sur l’inversion bayésienne a récemment été adaptée pour les observations microondes satellitaire ([Duruisseau et al., 2018]). Dans cette étude, nous explorons cette technique pour les observations IASI. Plusieurs tests de sensibilité sont effectués sur les différents paramètres de l’algorithme, avec pour objectif de préparer de futurs travaux sur l’assimilation all-sky infrarouges, explicités dans les perspectives de ce manuscrit. / The current generation of advanced infrared sounders is one of the most important sources of observations in data assimilation systems in numerical weather prediction (NWP) models. However, the total amount of information provided by this type of sensor, with its large number of channels and its global coverage, is far from being fully exploited. The presence of clouds in the instrument’s field of view, which affects the majority of observations, is one of the reasons why NWP centers reject a large amount of observations from sounders. NWP centers, however, have begun to assimilate cloud-affected radiances over the oceans using channels whose cloudy radiative effects are modeled by a simple cloud model. Some of these detection algorithms are evaluated in this manuscript, and their limitations are clarified. In order to increase the amount of assimilated data, it is necessary to better represent clouds and their radiative effects in the models. For several years, studies have been conducted to better represent their effects in radiative transfer models ([Faijan et al., 2012] ; [Martinet et al., 2013]) ; and to use new informative infrared channels of cloudy hydrometeors in the assimilation. [Martinet et al., 2014]. This thesis focuses on several approaches for the assimilation of cloudy radiances. In the first part, the characterization of the cloud parameters currently used for the assimilation of cloudy radiances was evaluated in the global and regional scale models. In addition, as part of the "all-sky" assimilation, which considers both clear and cloudy radiances, the evaluation and improvement of homogeneous scene detection algorithms based on the colocation of observations with other imagers was studied. These studies are made possible by taking into account the hydrometeorological fields provided by the convective schemes of the ARPEGE model as the input of the RTTOV-CLD cloud radiative transfer model. Part of this thesis concerns the validation of simulations, by comparing the contribution of the new convective PCMT scheme to the one used in operational applications. Subsequently, different criteria for selecting homogeneous scenes are proposed. By conducting assimilation experiments and evaluating the impact of these proposed selection criteria on the quality of long-term forecasts, one of the proposed tests stands out from the others by keeping a significant amount of clear sky observations and demonstrating neutral to slightly positive impacts on the forecasts. These new methods for selecting homogeneous scenes proposed in this study allows the consideration of improving the quality control of IASI observations in clear sky. To address the issue of all-sky radiance data assimilation, the two-step assimilation technique, already used for radar reflectivity assimilation in AROME ([Wattrelot et al., 2014]), was evaluated for IASI radiances in the ARPEGE model in a case study. This method based on Bayesian inversion has recently been adapted for satellite microwave observations ([Duruisseau et al., 2018]). Several sensitivity tests were carried out on the different parameters of the algorithm, with the objective of preparing for future work on infrared all-sky assimilation, as explained in the perspectives of this manuscript.

This study examines the determinants of the use of uncertain and weak modal words in 10-K filings and the effect of these words on analyst forecast attributes. I find that the use of uncertain and weak modal words in 10-K filings is positively (negatively) associated with firm size, volatility of business and operations (firm age and number of business segments). More importantly, after controlling for readability and management tone, I find that the use of uncertain and weak modal words in 10-K filings is associated with greater analyst following, lower forecast dispersion, greater forecast accuracy, and lower uncertainty in analysts’ overall and common information environment. The results of this study provide more insights into why management uses uncertain and weak modal words in 10-K filings and how these words in 10-K filings affect analysts’ behavior and their forecast outcomes.

Les îles du bassin caribéen figurent parmi les territoires dans le monde ayant la particularité d’être exposés à tous les types de risque naturel à l’exception du risque d’avalanche. Pourtant, le bilan global quant à leur prise en compte effective reste peu satisfaisant : En effet on constate des lacunes tant au niveau de l’information des populations qu’au niveau des moyens « administratifs et techniques » de réponse au risque. Le risque volcanique en particulier, bien que concernant onze territoires dans les petites Antilles, est relativement « éclipsé » par les autres risques dans les agendas gouvernementaux. Or, les « poudrières » de la Caraïbe sont pour la plupart actives. De plus, du fait de leur exiguïté et de la concentration de populations et d’infrastructures aux abords des volcans ; les territoires insulaires ont une vulnérabilité accrue. Des manifestations violentes peuvent entraîner, comme l’ont montré les cas de la Montagne Pelée en Martinique (1902) et plus récemment celui de la Soufrière Hills à Montserrat (1995 à nos jours), un bilan humain particulièrement lourd. Surtout, même en tempérant le risque de perte de vies humaines grâce à la prévision, une crise volcanique majeure reste synonyme de désastre économique. Le caractère exceptionnel des manifestations volcaniques suffit-il à expliquer ce bilan ? Cette interrogation première nous amène à questionner les mécanismes caractérisant la gestion du risque volcanique dans les Caraïbes insulaires. Notre étude porte ainsi sur les deux crises qui ont été les plus débattues en matière de retour d’expérience : celle de la Soufrière de Guadeloupe en 1976 et celle de la Soufrière Hills de Montserrat, qui a connu son pic en 1997. La comparaison des politiques publiques de gestion des crises étudiées nous permet d’identifier les facteurs orientant la stratégie des autorités compétentes en amont et en aval des crises. / The islands of the Caribbean basin are among the territories in the world having the distinction of being exposed to all types of natural hazard except avalanche risk. However, the overall assessment of their effective consideration remains unsatisfactory: Indeed, there are gaps in both the information of the population and the level of "administrative and technical" means of response to risk. Volcanic risk in particular, although affecting eleven territories in the Lesser Antilles, is relatively "overshadowed" by other risks in government agendas. However, the "powder keg" of the Caribbean are mostly active. Moreover, because of their small size and the concentration of populations and infrastructures around volcanoes; island territories have increased vulnerability. Violent demonstrations can lead, as has been shown in the cases of Mount Pelee in Martinique (1902) and more recently that of the Soufrière Hills in Montserrat (1995 to the present day), a particularly heavy human toll. Above all, even with the risk of loss of life due to the forecast, a major volcanic crisis is synonymous with economic disaster.Is the exceptional character of volcanic events enough to explain this assessment? This first interrogation leads us to question the mechanisms characterizing the volcanic risk management in the insular Caribbean.Our study thus focuses on the two crises that have been the most debated in terms of feedback: that of Soufrière Guadeloupe in 1976 and that of Soufrière Hills Montserrat, which peaked in 1997. The comparison of public crisis management policies studied allows us to identify the factors guiding the strategy of the competent authorities upstream and downstream of crises.

Corrosion initiates in reinforced concrete structures exposed to marine environments when the chloride ion concentration at the surface of an embedded steel reinforcing bar exceeds the chloride corrosion threshold (CT) value. The value of CT is generally assumed to have a conservative fixed value ranging from 0.2% to - 0.5 % of chloride ions by weight of cement. However, extensive experimental investigations confirmed that CT is not a fixed value and that the value of CT depends on many variables. Among those, the potential of passive steel embedded in concrete is a key influential factor on the value of CT and has received little attention in the literature. The phenomenon of a potential-dependent threshold (PDT) permits accounting for corrosion macrocell coupling between active and passive steel assembly components in corrosion forecast models, avoiding overly conservative long-term damage projections and leading to more efficient design. The objectives of this investigation was to 1) expand by a systematic experimental assessment the knowledge and data base on how dependent the chloride threshold is on the potential of the steel embedded in concrete and 2) introduce the chloride threshold dependence on steel potential as an integral part of corrosion-related service life prediction of reinforced concrete structures. Experimental assessments on PDT were found in the literature but for a limited set of conditions. Therefore, experiments were conducted with mortar and concrete specimens and exposed to conditions more representative of the field than those previously available. The experimental results confirmed the presence of the PDT effect and provided supporting information to use a value of -550 mV per decade of Cl- for the cathodic prevention slope βCT, a critical quantitative input for implementation in a practical model. A refinement of a previous corrosion initiation-propagation model that incorporated PDT in a partially submerged reinforced concrete column in sea water was developed. Corrosion was assumed to start when the chloride corrosion threshold was reached in an active steel zone of a given size, followed by recalculating the potential distribution and update threshold values over the entire system at each time step. Notably, results of this work indicated that when PDT is ignored, as is the case in present forecasting model practice, the corrosion damage prediction can be overly conservative which could lead to structural overdesign or misguided future damage management planning. Implementation of PDT in next-generation models is therefore highly desirable. However, developing a mathematical model that forecasts the corrosion damage of an entire marine structure with a fully implemented PDT module can result in excessive computational complexity. Hence, a provisional simplified approach for incorporating the effect of PDT was developed. The approach uses a correction function to be applied to projections that have been computed using the traditional procedures.

It has been noticed that throughout the years, immigrant’s skills, knowledge, and experience have declined. In fact, researchers have noticed the presence of cohort effects, where there are differences in quality and skills across the immigrants. Using data from the German Socio-Economic Panel and through an out-of-sample forecast of the employment of recent cohorts based on how earlier newcomers performed, I can confirm that, over time, immigrants see their probability of being employed decrease. For instance, employment decreased from 99% for immigrants that arrived in Germany in 2010 to 92% for those that came in 2015. The linear probability model also highlights that not only human capital influences directly employment levels of immigrants. Undeniably, the region of origin and the immigrants’ duration of residence in Germany also impact the likelihood of finding a paid job. Therefore, cohort effects cannot only be justified by the fact that newly arrived immigrants are very different from those who arrived some years ago.

Hydrologic and climate models predict variables through a simplification of the underlying complex natural processes. Model development involves minimising predictive uncertainty. Predictive uncertainty arises from three broad sources which are measurement error in observed responses, uncertainty of input variables and model structural error. This thesis introduces ways to improve predictive accuracy of hydroclimatic models by considering input and structural uncertainties. The specific methods developed to reduce the uncertainty because of erroneous inputs and model structural errors are outlined below. The uncertainty in hydrological model inputs, if ignored, introduces systematic biases in the parameters estimated. This thesis presents a method, known as simulation extrapolation (SIMEX), to ascertain the extent of parameter bias. SIMEX starts by generating a series of alternate inputs by artificially adding white noise in increasing multiples of the known input error variance. The resulting alternate parameter sets allow formulation of an empirical relationship between their values and the level of noise present. SIMEX is based on the theory that the trend in alternate parameters can be extrapolated back to the notional error free zone. The case study relates to erroneous sea surface temperature anomaly (SSTA) records used as input variables of a linear model to predict the Southern Oscillation Index (SOI). SIMEX achieves a reduction in residual errors from the SOI prediction. Besides, a hydrologic application of SIMEX is demonstrated by a synthetic simulation within a three-parameter conceptual rainfall runoff model. This thesis next advocates reductions of structural uncertainty of any single model by combining multiple alternative model responses. Current approaches for combining hydroclimatic forecasts are generally limited to using combination weights that remain static over time. This research develops a methodology for combining forecasts from multiple models in a dynamic setting as an improvement of over static weight combination. The model responses are mixed on a pair wise basis using mixing weights that vary in time reflecting the persistence of individual model skills. The concept is referred here as the pair wise dynamic weight combination. Two approaches for forecasting the dynamic weights are developed. The first of the two approaches uses a mixture of two basis distributions which are three category ordered logistic regression model and a generalised linear autoregressive model. The second approach uses a modified nearest neighbour approach to forecast the future weights. These alternatives are used to first combine a univariate response forecast, the NINO3.4 SSTA index. This is followed by a similar combination, but for the entire global gridded SSTA forecast field. Results from these applications show significant improvements being achieved due to the dynamic model combination approach. The last application demonstrating the dynamic combination logic, uses the dynamically combined multivariate SSTA forecast field as the basis of developing multi-site flow forecasts in the Namoi River catchment in eastern Australia. To further reduce structural uncertainty in the flow forecasts, three forecast models are formulated and the dynamic combination approach applied again. The study demonstrates that improved SSTA forecast (due to dynamic combination) in turn improves all three flow forecasts, while the dynamic combination of the three flow forecasts results in further improvements.