Дисертації з теми "Multi-Model ensembles"

The representation of physical processes by a climate model depends on its structure, numerical schemes, physical parameterizations and resolution, with initial conditions and future emission scenarios further affecting the output. The extent to which climate models agree is therefore of great interest, often with greater confidence in robust results across models. This has led to climate model output being analysed as ensembles rather than in isolation, and quantifying the sources of variation across these ensembles is the aim of many recent studies. Statistical attempts to do this include the use of variants of the mixed-effects analysis of variance or covariance (mixed-effects ANOVA/ANCOVA). This work usually focuses on identifying variation in a variable of interest that is due to differences in model structure, carbon emissions scenario, etc. Quantifying such variation is important in determining where models agree or disagree, but further statistical approaches can be used to diagnose the reasons behind the agreements and disagreements by representing the physical processes within the climate models. A process-based approach is presented that uses simulation with statistical models to perform a global sensitivity analysis and quantify the sources of variation in multi-model ensembles. This approach is a general framework that can be used with any generalised linear mixed model (GLMM), which makes it applicable to use with statistical models designed to represent (sometimes complex) physical relationships within different climate models. The method decomposes the variation in the response variable into variation due to 1) temporal variation in the driving variables, 2) variation across ensemble members in the distributions of the driving variables, 3) variation across ensemble members in the relationship between the response and the driving variables, and 4) variation unexplained by the driving variables. The method is used to quantify the extent to which, and diagnose why, precipitation varies across and within the members of two different climate model ensembles on various different spatial and temporal scales. Change in temperature in response to increased CO2 is related to change in global-mean annual-mean precipitation in a multi-model ensemble of general circulation models (GCMs). A total of 46% of the variation in the change in precipitation in the ensemble is found to be due to the differences between the GCMs, largely because the distribution of the changes in temperature varies greatly across different GCMs. The total variation in the annual-mean change in precipitation that is due to the differences between the GCMs depends on the area over which the precipitation is averaged, and can be as high as 63%. The second climate model ensemble is a perturbed physics ensemble using a regional climate model (RCM). This ensemble is used for three different applications. Firstly, by using lapse rate, saturation specific humidity and relative humidity as drivers of daily-total summer convective precipitation at the grid-point level over southern Britain, up to 8% of the variation in the convective precipitation is found to be due to the uncertainty in RCM parameters. This is largely because given atmospheric conditions lead to different rates of precipitation in different ensemble members. This could not be detected by analysing only the variation across the ensemble members in mean precipitation rate (precipitation bias). Secondly, summer-total precipitation at the grid-point level over the British Isles is used to show how the values of the RCM parameters can be incorporated into a GLMM to quantify the variation in precipitation due to perturbing each individual RCM parameter. Substantial spatial variation is found in the effect on precipitation of perturbing different RCM parameters. Thirdly, the method is extended to focus on extreme events, and the simulation of extreme winter pentad (five-day mean) precipitation events averaged over the British Isles is found to be robust to the uncertainty in RCM parameters.

Appropriate and defensible statistical frameworks are required in order to make credible inferences about future climate based on projections derived from multiple climate models. It is shown that a two-way analysis of variance framework can be used to estimate the response of the actual climate, if all the climate models in an ensemble simulate the same response. The maximum likelihood estimate of the expected response provides a set of weights for combining projections from multiple climate models. Statistical F tests are used to show that the differences between the climate response of the North Atlantic storm track simulated by a large ensemble of climate models cannot be distinguished from internal variability. When climate models simulate different responses, the differences between the re- sponses represent an additional source of uncertainty. Projections simulated by climate models that share common components cannot be considered independent. Ensemble thinning is advocated in order to obtain a subset of climate models whose outputs are judged to be exchangeable and can be modelled as a random sample. It is shown that the agreement between models on the climate response in the North Atlantic storm track is overestimated due to model dependence. Correlations between the climate responses and historical climates simulated by cli- mate models can be used to constrain projections of future climate. It is shown that the estimate of any such emergent relationship will be biased, if internal variability is large compared to the model uncertainty about the historical climate. A Bayesian hierarchical framework is proposed that is able to separate model uncertainty from internal variability, and to estimate emergent constraints without bias. Conditional cross-validation is used to show that an apparent emergent relationship in the North Atlantic storm track is not robust. The uncertain relationship between an ensemble of climate models and the actual climate can be represented by a random discrepancy. It is shown that identical inferences are obtained whether the climate models are treated as predictors for the actual climate or vice versa, provided that the discrepancy is assumed to be sym- metric. Emergent relationships are reinterpreted as constraints on the discrepancy between the expected response of the ensemble and the actual climate response, onditional on observations of the recent climate. A simple method is proposed for estimating observation uncertainty from reanalysis data. It is estimated that natural variability accounts for 30-45% of the spread in projections of the climate response in the North Atlantic storm track.

Le climat terrestre connaît actuellement des changements rapides et généralisés. Les activités humaines depuis l'ère industrielle, en particulier les émissions de CO2 dans l'atmosphère dues à la combustion de combustibles fossiles, ont intensifié l'effet de serre. Cela a entraîné une augmentation de la température moyenne de l'air à la surface du globe de 1.1°C en 2011-2020 par rapport à 1850-1900. Une autre conséquence majeure est le réchauffement des océans mondiaux, qui ont absorbé plus de 90% de l'énergie excédentaire accumulée dans le système climatique en raison de l'augmentation du forçage radiatif. L'absorption de chaleur par l'océan mondial (OHU) est un processus climatique clé qui joue un double rôle dans le changement climatique d'origine anthropique. D'une part, l'OHU constitue en soi une mesure clé du changement climatique, qui est directement associée à des impacts négatifs tels que l'élévation du niveau de la mer et l'augmentation de la fréquence des événements extrêmes dans l'océan. D'autre part, l'OHU fournit un service climatique essentiel en épargnant l'atmosphère de grandes quantités de chaleur, sans lequel le réchauffement atmosphérique serait bien plus marqué que celui que nous observons actuellement. Malgré leur importance, de nombreux processus physiques qui contrôlent l'OHU restent mal compris, même dans les modèles climatiques numériques utilisés dans les évaluations internationales du changement climatique. Dans cette thèse, nous avançons sur ce problème en nous appuyant sur des simulations climatiques issues de modèles participant au Projet d'intercomparaison des modèles couplés (CMIP). Dans une première étude, nous produisons des estimations améliorées de l'OHU global d'ici à la fin du XXIe siècle en identifiant une relation émergente dans un ensemble de modèles CMIP, qui relie l'état climatique présent de l'hémisphère sud à l'OHU futur. En combinant cette relation avec des données d'observation, nous obtenons des projections mieux contraintes qui montrent que l'OHU futur pourrait être plus important qu'estimé précédemment. Dans une deuxième étude, nous clarifions les processus à l'origine de l'efficacité d'absorption de la chaleur océanique (OHUE), qui quantifie la quantité d'OHU par degré de réchauffement de la surface terrestre. Nous réconcilions plusieurs tentatives antérieures d'explication des facteurs influençant l'OHUE, et montrons que la stratification de l'océan Austral supérieur est une propriété clé qui contrôle l'OHUE dans les modèles climatiques CMIP. Enfin, nous présentons une analyse exploratoire combinant les approches de ces deux études, et menons une analyse statistique des simulations d'un grand ensemble multi-modèle dans le but de contraindre l'OHUE. Au-delà de ces résultats concrets concernant l'OHU global, nous discutons également de certaines questions méthodologiques liées à l'interprétation des incertitudes découlant des ensembles multi-modèles de manière plus générale
The Earth's climate is currently undergoing rapid and widespread changes. Human activities in the industrial era, in particular the emission of CO2 into the atmosphere through the burning of fossil fuels, have led to an enhanced greenhouse effect which has caused an increase in the global average surface air temperature of 1.1°C in 2011-2020 relative to 1850-1900. A further consequence is the warming of the global ocean: it has absorbed over 90% of the excess energy stored in the Earth system due to the increased radiative forcing. This global ocean heat uptake (OHU) is a critical climate process and plays a dual role for anthropogenic climate change. On the one hand, OHU is a measure of the cumulative effects of transient climate change, and scales with negative impacts such as sea level rise and the frequency of oceanic extreme events. On the other hand, OHU provides a crucial service by shielding the atmosphere from large amounts of heat that would otherwise cause much greater global warming than currently observed. Despite their importance, many of the physical processes controlling OHU are still poorly understood, including in state-of-the-art numerical climate models used for international climate change assessments. In this thesis, we address this problem using climate simulations of models participating in the Coupled Model Intercomparison Project (CMIP). In a first study, we provide improved future projections of global OHU by the end of the 21st century by identifying an emergent relationship across an ensemble of CMIP models linking the simulated baseline climate state of the Southern Hemisphere to future global OHU. By combining this relationship with observational data, we obtain constrained projections showing that future OHU is likely larger than previously thought. In a second study, we clarify the processes involved in setting the ocean heat uptake efficiency (OHUE) which quantifies the amount of OHU per degree of global surface warming. We reconcile a number of previous attempts at explaining controls on OHUE, and show that the upper ocean stratification in the Southern Ocean is a key property setting its value in CMIP climate models. Last, we present an exploratory analysis combining the approaches of these two studies, and perform a statistical analysis of simulations from a large multi-model ensemble with the goal of constraining OHUE. Beyond these concrete results concerning global OHU, we also discuss some of the methodological issues related to the interpretation of uncertainties arising from multi-model ensembles more generally

Cette thèse propose des solutions de commande basées sur la planification optimale de trajectoires pour des systèmes dynamiques multi-agents fonctionnant dans un environnement variable (avec obstacles statiques ou mobiles et des perturbations variables dans le temps).Cette planification de trajectoires repose sur l'utilisation combinée de la théorie des ensembles (en particulier des ensembles convexes bornés), de la commande prédictive non-linéaire (NMPC), du calcul de champs de potentiel et des méthodes basées sur des graphes. Elle se base sur la construction de champs de potentiel répulsifs associés à des fonctions de barrière marche-arrêt (on-off barrier functions) qui décrivent et activent ou désactivent les trajectoires libres (sans collision) calculées au préalable par une commande de type NMPC distribuée. Ces constructions sont ensuite utilisées pour maintenir la connectivité dans le groupe d'agents, tout en assurant le suivi du chemin pré-généré. En outre, un observateur pour l'estimation de perturbations non linéaires est intégré dans le schéma de commande afin de les rejeter.Les résultats théoriques obtenus sont validés en simulation, par des comparaisons avec des approches utilisant la programmation mixte en nombres entiers, à l'aide de données numériques réelles provenant d'une plateforme de navigation sécurisée pour les véhicules de surface non habités dans le fjord de Trondheim (Norvège)
This thesis proposes optimization-based control solutions for the motion planning of multi-agent dynamical systems operating in a variable environment (with static/mobile obstacles and time-varying environmental disturbances).Collision-free paths are planned for the agents through the combined use of set theory (particularly, bounded convex sets), non(-linear) Model Predictive Control (MPC), Potential Field (PF) and graph-based methods. The contributions build on the proposal of repulsive potential field constructions together with on-off barrier functions which describe and, respectively, activate/deactivate the collision-free conditions introduced in a distributed NMPC framework. These constructions are further used for connectivity maintenance conditions among the group of agents while ensuring the tracking of the a priori generated path. Furthermore, a nonlinear disturbance observer is integrated within the control scheme for environmental disturbance rejection.Finally, the results are validated in simulation through comparisons with mixed-integer approaches and over a benchmark for the safe navigation of Unmanned Surface Vehicles (USVs) in the Trondheim fjord, Norway, using real numerical data

Le milieu hospitalier est un monde à la fois sensible et complexe, sensible parce que la vie humaine est en jeu et complexe parce que les équipements médicaux augmentent en nombre et en complexité technique. Ainsi, afin de préserver le bon état de fonctionnement de ces équipements et à un niveau élevé de disponibilité, leur entretien est devenu l'une des préoccupations majeures des responsables de l’hôpital. L’objectif de cette thèse est de proposer, aux responsables de maintenance biomédicale dans les établissements de soins, des outils d’aide à la décision qui permettent une meilleure maitrise des coûts. Ceci en assurant la sécurité des patients et des utilisateurs et en maintenant des performances optimales de l’ensemble des équipements médicaux. Tout d’abord, une heuristique a été proposée pour le choix de l’internalisation ou de l’externalisation de la maintenance et pour la sélection du contrat adéquat. La sélection du contrat est basée sur un ensemble de critères tout en considérant la contrainte du budget disponible. Ensuite, afin d’améliorer la procédure proposée, nous avons proposé des outils d’aide à la décision multicritère pour le choix adéquat d’une stratégie de maintenance. Pour l’étude de la criticité des équipements médicaux et le choix de la maintenance, sept critères ont été étudiés en proposant un couplage de l’approche AHP « Analytical Hierarchy Process » à la technique TOPSIS « Technique for Order Performance by Similarity to Ideal Solution ». Comme les experts du service de maintenance présentaient une certaine incertitude dans leurs jugements, nous avons intégré l’évaluation linguistique floue dans l’étude de la criticité des équipements et dans la sélection de la stratégie de maintenance (Fuzzy AHP couplée avec Fuzzy TOPSIS). Un modèle mathématique MILP a été développé pour la définition des limites de la criticité afin de caractériser les trois stratégies de maintenance. Le bon choix de ces limites permet d’optimiser le coût de la maintenance en respectant le budget disponible. Enfin, un deuxième modèle mathématique MILP a été développé en se basant sur l’heuristique proposée. Ce modèle permet de sélectionner pour chaque équipement, la stratégie de maintenance, internaliser ou externaliser la maintenance et le type du contrat tout en considérant le budget disponible et la charge/capacité du service maintenance
The hospital is a world that is both sensitive and complex, sensitive because the human life is involved and complex because medical facilities are growing in number and in technical complexity. Then, the problem of the medical equipment maintenance in order to keep them in safe, reliable and with high level of availability has become a major preoccupation of the hospital. The objective of this thesis is to provide tools to help the biomedical maintenance service of the hospital to make decisions that allow a better control of costs, while ensuring patient and user safety and maintaining optimal performance of medical equipment. First, a heuristic has been proposed for the choice of internalization or outsourcing maintenance and for the selection of the appropriate contract. The selection of the contract is based on a set of criteria while considering the available budget constraint. Then, to improve the proposed procedure, we proposed multi-criteria decision-making tools to select the appropriate maintenance strategies. Seven criteria have been designed to study the criticality of medical equipment and the choice of maintenance by providing a coupling of the AHP approach "Analytical Hierarchy Process" with TOPSIS technique "Technique for Order Performance by Similarity to Ideal Solution." As the expert judgments of the maintenance department presented some uncertainty, we integrated the fuzzy language assessment of the criticality of the equipment and the selection of the maintenance strategy (Fuzzy AHP coupled with Fuzzy TOPSIS). A mixed integer linear programming model (MILP) was developed to define thresholds of criticality to characterize the three maintenance strategies. According to these thresholds, maintenance cost can be optimized within the available budget. Finally, a second mixed integer linear programming model (MILP) was developed based on the proposed heuristic. This model allows selecting for each equipment, the maintenance strategy, the internalization or the outsourcing of the maintenance and the type of contract while considering the available budget and the workload / capacity of the maintenance department

Space weather can have a negative impact on a number of radio frequency (RF) systems, with mitigation by ionospheric and thermospheric modelling one approach to improving system performance. However, before a model can be adopted operationally its performance must be quantified. Taylor diagrams, which show a model’s standard deviation and correlation, have been extended to further illustrate the model’s bias, standard deviation of error and mean square error in comparison to observational data. By normalising the statistics, multiple parameters can be shown simultaneously for a number of models. Using these modified Taylor diagrams, the first known long term (one month) comparison of three model types – empirical, physics and data assimilation - has been performed. The data assimilation models performed best, offering a statistically significant improvement in performance. One physics model performed sufficiently well that it is a viable background model option in future data assimilation schemes. Finally, multi-model thermospheric ensembles (MMEs) have been constructed from which the thermospheric forecasts exhibited a reduced root mean square error compared to non-ensemble approaches. Using an equally weighted MME the reduction was 55% and using a mean square error weighted approach the reduction was 48%.

This research investigates the impact of climate change on water resources using a multi-model ensemble approach through rainfall-runoff projection for A2 and B1 emission scenarios of IPCC (AR4) for mid (2046-2065) and late (2081-2100) century in selected catchments of Western Australia. A bias correction method is also developed to correct projected runoff and a framework for extended hydrologic prediction (EHP) system is outlined. The findings are expected to be beneficial for planning future water resources.

Motivé par l'étude de tissus biologiques, ce travail contribue aux développements d'outils numériques permettant de prédire la réponse mécanique de matériaux hétérogènes non linéaires dans lesquels les énergies d'interfaces deviennent prépondérantes. Ainsi, une méthode d'homogénéisation multi échelle combinée à une technique de réduction de modèle basée sur la décomposition orthogonale aux valeurs propres est proposée dans un cadre thermique et hyperélastique. Les énergies d'interfaces entre les différentes phases des composites sont décrites par un modèle d'interface cohérent et prises en compte numériquement par une approche liant la méthode des éléments finis étendus et la méthode level-set. Une étude de l'étalement d'une cellule vivante entre deux lamelles fixes est ensuite réalisée. Les deux modèles utilisés pour les simulations montrent que l'assemblage cortex d'actine-membrane plasmique ne joue qu'un rôle minime dans la réponse mécanique cellulaire

In questo studio, un multi-model ensemble è stato implementato e verificato, seguendo una delle priorità di ricerca del Subseasonal to Seasonal Prediction Project (S2S). Una regressione lineare è stata applicata ad un insieme di previsioni di ensemble su date passate, prodotte dai centri di previsione mensile del CNR-ISAC e ECMWF-IFS. Ognuna di queste contiene un membro di controllo e quattro elementi perturbati. Le variabili scelte per l'analisi sono l'altezza geopotenziale a 500 hPa, la temperatura a 850 hPa e la temperatura a 2 metri, la griglia spaziale ha risoluzione 1 ◦ × 1 ◦ lat-lon e sono stati utilizzati gli inverni dal 1990 al 2010. Le rianalisi di ERA-Interim sono utilizzate sia per realizzare la regressione, sia nella validazione dei risultati, mediante stimatori nonprobabilistici come lo scarto quadratico medio (RMSE) e la correlazione delle anomalie. Successivamente, tecniche di Model Output Statistics (MOS) e Direct Model Output (DMO) sono applicate al multi-model ensemble per ottenere previsioni probabilistiche per la media settimanale delle anomalie di temperatura a 2 metri. I metodi MOS utilizzati sono la regressione logistica e la regressione Gaussiana non-omogenea, mentre quelli DMO sono il democratic voting e il Tukey plotting position. Queste tecniche sono applicate anche ai singoli modelli in modo da effettuare confronti basati su stimatori probabilistici, come il ranked probability skill score, il discrete ranked probability skill score e il reliability diagram. Entrambe le tipologie di stimatori mostrano come il multi-model abbia migliori performance rispetto ai singoli modelli. Inoltre, i valori più alti di stimatori probabilistici sono ottenuti usando una regressione logistica sulla sola media di ensemble. Applicando la regressione a dataset di dimensione ridotta, abbiamo realizzato una curva di apprendimento che mostra come un aumento del numero di date nella fase di addestramento non produrrebbe ulteriori miglioramenti.

Examples of cooperative behavior in complex networks with inherent or created imperfections are in the focus of study in this thesis. The complex networks represent real world systems where a number of entities cooperate through their interconnections. All entities and their interactions have some unique characteristics that determine the overall dynamics of the system. Therefore, the main topic of the thesis is the study of some of the effects of the uniqueness of the individual entities on the dynamical behavior of real systems, the effects of the neglecting of some physical aspects during modeling and approaches of combining the distinct qualities of individual models of nonlinear systems in developing better representation of the system of interest. The analyses show that model imperfections should lead to appropriate adaptation of the learning procedures used for state and parameter estimation in modeling. Furthermore, approaches of interactively combining imperfect models can lead to improvements in nonlinear modeling of real world phenomena. The study of synchronization and consensus in complex networks revealed some of the effects of the uniqueness of entities and their interactions, which can alter the convergence rate or introduce some new interesting behaviors.

Situata al centro del Mediterraneo, uno degli “hot-spot” del cambiamento climatico, l’Italia è considerata una delle aree più suscettibili al cambiamento climatico globale. L’unicità della posizione geografica e l’elevata eterogeneità climatica rendono difficoltoso elaborare uno scenario climatico univoco per i differenti pattern climatici. In questo contesto, disporre di informazioni climatiche da scala regionale a locale diventa di strategica importanza. La ricerca si basa su questa essenziale transizione, definendo due scenari climatici per il 21° secolo a differente scala spaziale. Uno scenario a scala regionale, riferito alla penisola italiana e uno a scala locale che considera stazioni della regione Marche. Gli scenari climatici si basano su simulazioni numeriche di modelli climatici a differente risoluzione. Le simulazioni sono state post-processate con una tecnica di correzione statistica quantile mapping (QM). Entrambi gli scenari indicano un forte incremento delle temperature in tutte le stagioni specialmente in estate. Le piogge sono attese diminuire in estate e moderatamente aumentare nel nord-Italia d’inverno. Per le Marche, l’ultima generazione di modelli climatici è concorde nell’indicare un considerevole incremento delle temperature e diminuzione delle precipitazione durante l’estate. Le piogge autunnali e invernali sono però attese in incremento. Seppur applicato con diverse configurazioni, l’effetto del QM sul segnale climatico è del tutto simile nei due esperimenti. La ricerca vuole apportare nuovi elementi al dibattito scientifico relativo all’effetto di questa tecnica, sul segnale climatico. Ci si interroga sull’eventuale revisione degli attuali scenari climatici basati su simulazioni non soggette a correzione e quindi influenzate da intrinseco errore. La ricerca fornisce infine dei data set di simulazioni climatiche validate e statisticamente corrette direttamente utilizzabili per la riduzione del rischio climatico e dei suoi impatti.
Lying at the center of Mediterranean basin, one of the most sensitive area to anthropogenic climate change, Italy is expected to be particularly susceptible to global climate change. Unique geographical position and heterogeneous climatic features make difficult defining a comprehensive climate scenario. In such context, establishing regional to local climate information is of strategical importance for all-level society. Doctoral research is based on this conceptual and methodological transition. Two climate scenarios, one at regional and one at local scale are defined. The regional climate scenario considers an area roughly covering Italy, and a local scenario focuses over Marche region stations. Climate scenarios rely on two different-resolution climate model ensemble simulations. Numerical simulations were post-processed according to the quantile mapping (QM) bias correction technique. Original and bias-corrected climate simulations were employed to define 21st century climate change signal (CCS) over principal climate variables. Both scenarios agreed on identifying a severe increase of temperature in all the seasons, especially in summer. Precipitation are projected strongly decrease in summer and increase in winter only over north-Italy. Concerning Marche region stations, newest generation of climate models agree on the severe temperature increase and precipitation reduction in summer but an equivalent increase of autumn-winter precipitation was found. Albeit adopting different configurations, QM coherently affected original CCS in both experiments. Research offers elements to scientific debate on the effect of a common post-processing practice on the CCS. Should we reconsider climate scenarios only relying on original climate model projections? Moreover, following climate services principles, outputs of this research provide comprehensive climate information directly usable by professionals involved in climate risk and impacts research.

This dissertation focused in the comprehensive analysis of the capabilities of some non-tested types of Artificial Neural Networks, specifically: the Probabilistic Neural Networks (PNN) and the Multi-Layer Perceptron (MLP) Ensembles. The analysis of the capabilities of these techniques was performed using the native brown trout (Salmo trutta; Linnaeus, 1758), the bermejuela (Achondrostoma arcasii; Robalo, Almada, Levy & Doadrio, 2006) and the redfin barbel (Barbus haasi; Mertens, 1925) as target species. The analyses focused in the predictive capabilities, the interpretability of the models and the effect of the excess of zeros in the training datasets, which for presence-absence models is directly related to the concept of data prevalence (i.e. proportion of presence instances in the training dataset). Finally, the effect of the spatial scale (i.e. micro-scale or microhabitat scale and meso-scale) in the habitat suitability models and consequently in the e-flow assessment was studied in the last chapter.
Esta tesis se centra en el análisis comprensivo de las capacidades de algunos tipos de Red Neuronal Artificial aún no testados: las Redes Neuronales Probabilísticas (PNN) y los Conjuntos de Perceptrones Multicapa (MLP Ensembles). Los análisis sobre las capacidades de estas técnicas se desarrollaron utilizando la trucha común (Salmo trutta; Linnaeus, 1758), la bermejuela (Achondrostoma arcasii; Robalo, Almada, Levy & Doadrio, 2006) y el barbo colirrojo (Barbus haasi; Mertens, 1925) como especies nativas objetivo. Los análisis se centraron en la capacidad de predicción, la interpretabilidad de los modelos y el efecto del exceso de ceros en las bases de datos de entrenamiento, la así llamada prevalencia de los datos (i.e. la proporción de casos de presencia sobre el conjunto total). Finalmente, el efecto de la escala (micro-escala o escala de microhábitat y meso-escala) en los modelos de idoneidad del hábitat y consecuentemente en la evaluación de caudales ambientales se estudió en el último capítulo.
Aquesta tesis se centra en l'anàlisi comprensiu de les capacitats d'alguns tipus de Xarxa Neuronal Artificial que encara no han estat testats: les Xarxes Neuronal Probabilístiques (PNN) i els Conjunts de Perceptrons Multicapa (MLP Ensembles). Les anàlisis sobre les capacitats d'aquestes tècniques es varen desenvolupar emprant la truita comuna (Salmo trutta; Linnaeus, 1758), la madrilla roja (Achondrostoma arcasii; Robalo, Almada, Levy & Doadrio, 2006) i el barb cua-roig (Barbus haasi; Mertens, 1925) com a especies objecte d'estudi. Les anàlisi se centraren en la capacitat predictiva, interpretabilitat dels models i en l'efecte de l'excés de zeros a la base de dades d'entrenament, l'anomenada prevalença de les dades (i.e. la proporció de casos de presència sobre el conjunt total). Finalment, l'efecte de la escala (micro-escala o microhàbitat i meso-escala) en els models d'idoneïtat de l'hàbitat i conseqüentment en l'avaluació de cabals ambientals es va estudiar a l'últim capítol.
Muñoz Mas, R. (2016). Multivariate approaches in species distribution modelling: Application to native fish species in Mediterranean Rivers [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/76168
TESIS

Most studies of gustatory coding have been performed in either anesthetized or awake, passively stimulated rats. In this dissertation the influences of behavioral state on gustatory processing in awake rats are described. In the first set of experiments, the effects of non-contingent tastant delivery on the chemical tuning of single neurons were explored. Tastants were delivered non-contingently through intra-oral cannulas to restrained, non water-deprived rats while single unit responses were recorded from the gustatory cortex (GC). As the subjects' behavior progressed from acceptance to rejection of the tastants, the chemical tuning of the neurons changed as well. This suggests that the subjects' behavioral state powerfully influences gustatory processing. In the second set of experiments, rats were trained to lick for fluid reinforcement on an FR5 schedule while single unit activity was recorded from GC. In this case, the chemical tuning was much more stable. Under this paradigm, chemosensory responses were rapid (~ 150 ms) and broadly tuned. In the third study, it was found that ensembles of GC neurons could discriminate between tastants and their concentrations on a single trial basis, and such discrimination was accomplished with a combination of rate and temporal coding. Ensembles of GC neurons also anticipated the identity of the upcoming stimulus when the tastant delivery was predictable. Finally, it was found that ensembles of GC neurons could discriminate between the bitter stimuli nicotine and quinine. Nicotine is both a bitter tastant and a trigeminal stimulant, and when the acetylcholine receptors in the lingual epithelium were blocked with mecamylamine, the ensembles failed to discriminate nicotine from quinine.

碩士
國立臺北科技大學
土木與防災研究所
103
In the field of disaster prevention usually use forecasting to estimate the effect of disaster to archive risk management. The process of forecasting can use multi-model ensemble strategy to improve the performance of calculation’s reliability. However, in the past of integration in different model experience, the system must be customized to match the need of each project and model. Once the case was changed, these integrated systems might crash and then affect the destination of using technical way to prevent disaster. The aim of this study is proposing a multi-model integration’s common procedures to provide a standard integrations reference for the future integration and maintenance. It is employed by the real-time flood forecasting system, FEWS_Taiwan, which is developed by our Water Resources Agency and Netherland’s Deltares Hydraulics, as the model-integrating framework. Furthermore, it can summarizes common procedures of FEWS_Taiwan system setup during on-site integrated multi-model and apply these common procedures to National Science and Technology Center for Disaster Reduction (NCDR) to apply hydraulic model. Then this multi-model ensemble framework will examine with a lot of historical and real-time data for testing the system operation. Moreover, the application of this framework can be used in different temporal and spatial distribution and keep working until the system break down. From the research experiment ,this study develops a process of multi-model integration standard in order to archive the differenet temaporal and spatial multi-model integrations.

Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics
The purpose of this thesis is to conduct comparative research between Genetic Programming (GP) and Geometric Semantic Genetic Programming (GSGP), with different initialization (RHH and EDDA) and selection (Tournament and Epsilon-Lexicase) strategies, in the context of a model-ensemble in order to solve regression optimization problems. A model-ensemble is a combination of base learners used in different ways to solve a problem. The most common ensemble is the mean, where the base learners are combined in a linear fashion, all having the same weights. However, more sophisticated ensembles can be inferred, providing higher generalization ability. GSGP is a variant of GP using different genetic operators. No previous research has been conducted to see if GSGP can perform better than GP in model-ensemble learning. The evolutionary process of GP and GSGP should allow us to learn about the strength of each of those base models to provide a more accurate and robust solution. The base-models used for this analysis were Linear Regression, Random Forest, Support Vector Machine and Multi-Layer Perceptron. This analysis has been conducted using 7 different optimization problems and 4 real-world datasets. The results obtained with GSGP are statistically significantly better than GP for most cases.
O objetivo desta tese é realizar pesquisas comparativas entre Programação Genética (GP) e Programação Genética Semântica Geométrica (GSGP), com diferentes estratégias de inicialização (RHH e EDDA) e seleção (Tournament e Epsilon-Lexicase), no contexto de um conjunto de modelos, a fim de resolver problemas de otimização de regressão. Um conjunto de modelos é uma combinação de alunos de base usados de diferentes maneiras para resolver um problema. O conjunto mais comum é a média, na qual os alunos da base são combinados de maneira linear, todos com os mesmos pesos. No entanto, conjuntos mais sofisticados podem ser inferidos, proporcionando maior capacidade de generalização. O GSGP é uma variante do GP usando diferentes operadores genéticos. Nenhuma pesquisa anterior foi realizada para verificar se o GSGP pode ter um desempenho melhor que o GP no aprendizado de modelos. O processo evolutivo do GP e GSGP deve permitir-nos aprender sobre a força de cada um desses modelos de base para fornecer uma solução mais precisa e robusta. Os modelos de base utilizados para esta análise foram: Regressão Linear, Floresta Aleatória, Máquina de Vetor de Suporte e Perceptron de Camadas Múltiplas. Essa análise foi realizada usando 7 problemas de otimização diferentes e 4 conjuntos de dados do mundo real. Os resultados obtidos com o GSGP são estatisticamente significativamente melhores que o GP na maioria dos casos.

Predicting the location and timing of rainfall events has important social and economic impacts. It is also important to have the ability to predict the amount of rainfall accurately. In operational centres forecasters use deterministic model output data as guidance for a subjective probabilistic rainfall forecast. The aim of this research is to determine the skill in an objective multi-model, multi-institute objective probabilistic forecast system. This was done by obtaining the rainfall forecast of two high-resolution regional models operational in South Africa. The first model is the Unified Model (UM) which is operational at the South African Weather Service. The UM contributed three members which differ in physics, data assimilation techniques and horisontal resolution. The second model is the Conformal-Cubic Atmospheric Model (CCAM) which is operational at the Council for Scientific and Industrial Research which in turn contributed two members to the ensemble system differing in horisontal resolution. A single-model ensemble was constructed for the UM and CCAM models respectively with each of the individual members having equal weights. The UM and CCAM single-model ensemble prediction models have been used in turn to construct a multi-model ensemble prediction system, using simple un-weighted averaging. The multi-model system was used to predict the 24-hour rainfall totals for three austral summer half-year seasons of 2006/07 to 2008/09. The forecast of this system was rigorously tested using observed rainfall data for the same period. From the multi-model system it has been found that the probabilistic forecast has good significant skill in predicting rainfall. The multi-model system proved to have skill and shows discrimination between events and non-events. This study has shown that it is possible to make an objective probabilistic rainfall forecast by constructing a multi-model, multi-institute system with high resolution regional models currently operational in South Africa. Thus, probabilistic rainfall forecasts with usable skill can be made with the use of a multi-model short-range ensemble prediction system over the South African domain. Such a system is not currently operational in South Africa. Copyright
Dissertation (MSc)--University of Pretoria, 2012.
Geography, Geoinformatics and Meteorology
Unrestricted

Agricultural production is highly sensitive to climate and weather perturbations. Maize is the main crop cultivated in South Africa and production is predominantly rain-fed. South Africa’s climate, especially rainfall, is extremely variable which influences the water available for agriculture and makes rain-fed cropping very risky. In the aim to reduce the uncertainty in the climate of the forthcoming season, this study investigates whether seasonal climate forecasts can be used to predict maize yields for South Africa with a usable level of skill. Maize yield, under rain-fed conditions, is simulated for each of the magisterial districts in the primary maize producing region of South Africa for the period from 1979 to 1999. The ability of the CERES-Maize model to simulate South African maize yields is established by forcing the CERES-Maize model with observed weather data. The simulated maize yields obtained by forcing the CERES-Maize model with observed weather data set the target skill level for the simulation systems that incorporate Global Circulation Models (GCMs). Two GCMs produced the simulated fields for this study, they are the Conformal Cubic Atmospheric Model (CCAM) and the ECHAM4.5 model. CCAM ran a 5 and ECHAM4.5 a 6- member ensemble of simulations on horizontal grids of 2.1° x 2.1° and 2.8° x 2.8° respectively. Both models were forced with observed sea-surface temperatures for the period 1979 to 2003. The CERES-Maize model is forced with each ensemble member of the CCAM-simulated fields and with each ensemble member of the ECHAM4.5-simulated fields. The CERES-CCAM simulated maize yields and CERES-ECHAM4.5 simulated maize yields are combined to form a Multi-Model maize yield ensemble system. The simulated yields are verified against actual maize yields. The CERES-Maize model shows significant skill in simulating South Africa maize yields. CERES-Maize model simulations using the CCAM-simulated fields produced skill levels comparable to the target skill, while the CERES-ECHAM4.5 simulation system illustrated poor skill. The Multi-Model system presented here could therefore not outscore the skill of the best single-model simulation system (CERES-CCAM). Notwithstanding, the CERES-Maize model has the potential to be used in an operational environment to predict South African maize yields, provided that the GCM forecast fields used to force the model are adequately skilful. Such a yield prediction system does not currently exist in South Africa.
Dissertation (MSc)--University of Pretoria, 2009.
Geography, Geoinformatics and Meteorology
Unrestricted

碩士
國立中央大學
大氣科學學系
105
In this study, we use monthly data from the multi-model ensemble (MME) of Coupled Model Intercomparison Project Phase 5 (CMIP5) decadal prediction experiments to assess interannual prediction skills for several atmospheric and oceanic variables in Tropics (30°S-30°N). First, we applied pattern stability analyses to extract persistent empirical orthogonal functions (EOFs) from observations-based data as reference spatial patterns. By projecting CMIP5 MME predictions to the extracted EOFs, then we compared these associated time series to assess the MME prediction skills. Finally, we applied linear regression and rank histogram to calibrate the associated time series of MME predictions. In the meantime, this study also evaluates the grid-point scale prediction capability in Tropics by EOF reconstructed fields. Pattern stability analyses of the observations-based data indicated that at least 4 persistent EOFs can be found in each examined variable field. The first EOF (EOF1) mainly corresponds to the mean state of the given field, while the second EOF and beyond correspond to more and more localized spatial structures. Except for the third EOF (EOF3) of sea surface temperature (SST) field that has close relation to the El Nino Southern Oscillation (ENSO), most of our efforts focused on the study of interannual prediction skill associated with EOF1. Results indicated that, except for near surface air temperature (SAT) and SST fields, most variable fields did not have any interannual prediction skill. Furthermore, the apparent prediction skill that SAT and SST fields possessed may largely come from the warming trend observed in the last half of the 20th century. As for the ENSO related prediction skill, the EOF3 related time series showed certain prediction skill. This skill may be related to the capability of climate models to better synchronize with ENSO evolution through the adoption of yearly initialization procedure. Additionally, the results of the calibrated MME predicted time series showed that both linear regression and rank histogram calibration methods could effectively reduce the prediction errors and the MME uncertainty. Furthermore, the use of EOF reconstruction reduced MME prediction errors on extensive continent and coastal regions.

Considering its social, economic and natural conditions the Mediterranean Area is a highly vulnerable region by designated affections of climate change. Furthermore, its climatic characteristics are subordinated to high natural variability and are steered by various elements, leading to strong seasonal alterations. Additionally, General Circulation Models project compelling trends in specific climate variables within this region. These circumstances recommend this region for the scientific analyses conducted within this study. Based on the data of the CMIP3 database, the fundamental aim of this study is a detailed investigation of the total variability and the accompanied uncertainty, which superpose these trends, in the projections of temperature, precipitation and sea-level pressure by GCMs and their specific realizations. Special focus in the whole study is dedicated to the German model ECHAM5/MPI-OM. Following this ambition detailed trends and mean values are calculated and displayed for meaningful time periods and compared to reanalysis data of ERA40 and NCEP. To provide quantitative comparison the mentioned data are interpolated to a common 3x3° grid. The total amount of variability is separated in its contributors by the application of an Analysis of Variance (ANOVA). For individual GCMs and their ensemble-members this is done with the application of a 1-way ANOVA, separating a treatment common to all ensemble-members and variability perturbating the signal given by different initial conditions. With the 2-way ANOVA the projections of numerous models and their realizations are analysed and the total amount of variability is separated into a common treatment effect, a linear bias between the models, an interaction coefficient and the residuals. By doing this, the study is fulfilled in a very detailed approach, by considering yearly and seasonal variations in various reasonable time periods of 1961-2000 to match up with the reanalysis data, from 1961-2050 to provide a transient time period, 2001-2098 with exclusive regard on future simulations and 1901-2098 to comprise a time period of maximum length. The statistical analyses are conducted for regional-averages on the one hand and with respect to individual grid-cells on the other hand. For each of these applications the SRES scenarios of A1B, A2 and B1 are utilized. Furthermore, the spatial approach of the ANOVA is substituted by a temporal approach detecting the temporal development of individual variables. Additionally, an attempt is made to enlarge the signal by applying selected statistical methods. In the detailed investigation it becomes evident, that the different parameters (i.e. length of temporal period, geographic location, climate variable, season, scenarios, models, etc…) have compelling impact on the results, either in enforcing or weakening them by different combinations. This holds on the one hand for the means and trends but also on the other hand for the contributions of the variabilities affecting the uncertainty and the signal. While temperature is a climate variable showing strong signals across these parameters, for precipitation mainly the noise comes to the fore, while for sea-level pressure a more differentiated result manifests. In turn, this recommends the distinguished consideration of the individual parameters in climate impact studies and processes in model generation, as the affecting parameters also provide information about the linkage within the system. Finally, an investigation of extreme precipitation is conducted, implementing the variables of the total amount of heavy precipitation, the frequency of heavy-precipitation events, the percentage of this heavy precipitation to overall precipitation and the mean daily intensity from events of heavy precipitation. Each time heavy precipitation is defined to exceed the 95th percentile of overall precipitation. Consecutively mean values of these variables are displayed for ECHAM5/MPI-OM and the multi-model mean and climate sensitivities, by means of their difference between their average of the past period of 1981-2000 and the average of one of the future periods of 2046-2065 or 2081-2100. Following this investigation again an ANOVA is conducted providing a quantitative measurement of the severity of change of trends in heavy precipitation across several GCMs. Besides it is a difficult task to account for extreme precipitation by GCMs, it is noteworthy that the investigated models differ highly in their projections, resulting partially in a more smoothed and meaningful multi-model mean. Seasonal alterations of the strength of this behaviour are quantitatively supported by the ANOVA
Bezüglich seiner sozialen, wirtschaftlichen und natürlichen Gegebenheiten ist der Mittelmeerraum eine Region, die in Anbetracht des zu erwartenden Klimawandels äußerst anfällig ist. Seine klimatischen Eigenschaften sind hoher natürlicher Variabilität, unterschiedlichen Antriebsmechanismen, sowie einer starken saisonalen Schwankung unterworfen. Zudem projizieren Globale Zirkulationsmodelle für diese Region aussagekräftige Trends für ausgewählte Klimavariablen. Diese Vorraussetzungen machen den Mittelmeerraum zu einem hervorragenden Studienobjekt für diese wissenschaftliche Arbeit. Auf der Basis der CMIP3 Datenbank ist das zu Grunde liegende Ziel dieser Arbeit eine detaillierte Untersuchung der Gesamtvariabilität und der damit einhergehenden Unsicherheit, die in den Projektionen der Globalen Zirkulationsmodelle und deren einzelnen Realisationen die Trends der Variablen Temperatur, Niederschlag und Druck überlagert. Besonderes Augenmerk liegt dabei auf dem deutschen Modell ECHAM5/MPI-OM. Für dieses Ziel werden Trends und Mittelwerte für aussagekräftige Zeitperioden berechnet und graphisch den Reanalysedatensätzen NCEP und ERA40 gegenübergestellt. Um quantitative Vergleiche zu ermöglichen werden die angesprochenen Datensätze auf ein gemeinsames geographisches Gitter von 3x3° interpoliert. Der Gesamtanteil der Variabilität wird in seine Entstehungsquellen durch die Anwendung einer Varianzanalyse (ANOVA) aufgeteilt. Dies wird mit einer 1-Wege Varianzanalyse für einzelne Globale Zirkulationsmodelle und ihre Realisationen durchgeführt, wobei ein Anteil dem Signal entspricht, das in allen Realisationen vorhanden ist und ein Anteil dem Rauschen zugeordnet werden kann, das das Signal überlagert und unterschiedlichen Anfangsbedingungen des Modells geschuldet ist. Durch eine 2-Wege Varianzanalyse werden die unterschiedlichen Realisationen mehrerer Klimamodelle in eine Analyse eingebunden, wobei der Anteil der Gesamtvariabilität wiederum in einen gemeinsamen Signalanteil, einem Anteil des linearen Unterschieds zwischen den verschiedenen Klimamodellen, einem Interaktionskoeffizient und dem Rauschen aufgeteilt werden. Die Anwendung dieser Verfahren wird detailliert ausgeführt, indem die Analysen auf jährlicher und saisonaler Grundlage für unterschiedliche Zeitperioden, nämlich 1961-1990 für den Vergleich mit den Reanalysedatensätzen, 1961-2050 als eine Übergangsperiode zwischen den Szenarien, 2001-2098 als reinen zukünftigen Betrachtungszeitraum und 1901-2098 um eine maximal umfassende Zeitperiode zu erhalten, betrachtet werden. Die statistischen Verfahren werden sowohl für regionale Mittelwerte als auch für einzelne Gitterpunkte berechnet. Für jede dieser Berechnungen werden die SRES Szenarien A1B, A2 und B1 herangezogen. Zudem wird der räumliche Ansatz der ANOVA ebenso durch einen zeitlichen ersetzt, wodurch die zeitliche Entwicklung der einzelnen Variabilitäten dargestellt wird. Des Weiteren wird durch gezielte statistische Methoden versucht, künstlich verstärkte Signale zu detektieren. Durch die detaillierte Untersuchung wird offenkundig, dass die unterschiedlichen Randbedingungen (hier die Länge der Zeitperiode, der geographische Ort, die Bezugsvariable, die Saison, das Szenario, das Modell, etc…) eine entscheidende Rolle für das Ergebnis spielen, indem sie einerseits durch deren unterschiedlicher Kombination es sowohl verstärken als auch glätten können. Dies gilt sowohl für die Mittelwerte und die Trends als auch für die unterschiedlichen Partitionen der Variabilitäten, die wiederum die Unsicherheiten und das Signal beeinflussen. Während Temperatur starke Signale über alle dieser Randbedingungen aufweist, so zeigt sich für Niederschlag hauptsächlich ein starkes Rauschen, während für Druck eine sehr ambivalente Verteilung hervortritt. Dies wiederum beweist, dass dieser differenzierte Ansatz bezüglich der Betrachtung der Abhängigkeit dieser Randebedingungen unabdinglich in Klimafolgestudien und der Modellentwicklung ist, da diese Bedingungen auch Informationen über die Wechselbeziehungen im System beinhalten. Schließlich wird noch die Entwicklung von Extremereignissen hinsichtlich der Niederschlagsmengen von Extremereignissen, der Häufigkeit der Ereignisse von extremen Niederschlagsmengen, dem prozentualen Anteil der Niederschlagsmenge aus Extremereignissen zu der Gesamtniederschlagsmenge und der mittleren täglichen Intensität von Niederschlagsextremereignissen untersucht. Hierbei wird ein Extremereignis als ein Ereignis definiert, das in seiner Menge oberhalb des 95.Perzentils der Menge der Gesamtereignisse liegt. So werden Mittelwerte dieser Variablen für ECHAM5/MPI-OM und über alle Modelle sowie deren Veränderungen zwischen ihren Mittelwerten aus den Zeiträumen der Vergangenheit 1981-2000 und den zukünftigen Perioden von 2046-2065 oder 2081-2100 gezeigt. Der Struktur dieser Studie folgend, wird wiederum eine ANOVA angewendet um eine quantitative Ermessung der Stärke der Veränderung im Erscheinungsbild von Extremniederschlagsereignissen über eine Vielzahl verschiedener Zirkulationsmodelle zu gewinnen. Ungeachtet der schwierigen Tatsache, Extremniederschlagsereignisse aus GCMs abzuleiten, ist es erwähnenswert, dass die betrachteten Modelle stark in ihren Projektionen abweichen, was wiederum zu einem in einem gemäßigten und aussagekräftigerem Multi-Modell Mittelwert führt. Saisonale Unterschiede in diesem Verhalten können durch die ANOVA quantitativ belegt werden