<p>In the real world, uncertainty is a common challenging problem faced by individuals, organizations, and firms. Decision quality is highly impacted by uncertainty because decision makers lack complete information and have to leverage the loss and gain in many possible outcomes or scenarios. This study explores dynamic decision making (with known distributions) and decision learning (with unknown distributions but some samples) in not-for-profit operations and supply chain management. We first study dynamic staffing for paid workers and volunteers with uncertain supply in a nonprofit operation where the optimal policy is too complex to compute and implement. Then, we consider dynamic inventory control and pricing under both supply and demand uncertainties where unmet demand is lost leading to a challenging non-concave dynamic problem. Furthermore, we explore decision learning from limited data of focal system and available data of related but different systems by transfer learning, cross learning, and co-learning utilizing the similarities among related systems.</p>

This dissertation consists of three essays in theoretical and applied microeconomics: the first essay is in cooperative game theory, and the second and third essays relate to forest economics. The first chapter studies a class of cooperative games dubbed ``r-essential games''. Cooperative game theory has proposed different notions of powerful players. For example, big-boss games (Muto et al., 1988) and clan games (Potters et al., 1989) are particular cases of veto games (Bahel, 2016). The first chapter extends these veto games by assuming that there is a given subset of powerful (or essential) players, but only a few (as opposed to all) essential players are required for a coalition to have a positive value. The resulting games, which are called r-essential games, encompass convex games (Shapley, 1971) and veto games. We show that r-essential games have a nonempty core. We give a recursive description of the core. Moreover, it is shown that the core and the bargaining set are equivalent for every r-essential game. An application to networks is provided. The second chapter employs a two-principal, one-agent model to estimate the social cost of fiscal federalism in China's northeast native forests. China's key forested region is located in the northeast and consists of state forest enterprises which manage forest harvesting and reforestation. Deforestation is a major problem there and has resulted in several central government reforms. We develop a framework for assessing the social cost of state forest enterprise deforestation. We first develop a two-principal, one-agent model that fits the federalistic organization of state forests, in that state forest managers make (potentially hidden) decisions under influence of provincial and central government policies. This model is used to quantify the social cost of these hidden actions. We then use panel data from a survey conducted by Peking University to compute social welfare losses and to formally identify the main factors in these costs. A sensitivity analysis shows that, interestingly, command and control through lower harvesting limits and a more accurate monitoring system are more important to lowering social welfare losses than conventional incentives targeting the wages of forest managers. Through regression analysis we also find that the more remote areas with a higher percentage of mature natural forests are the ones that will always have the highest social welfare losses. The third chapter studies the problem of choosing a rotation under uncertain future ecosystem values and timber prices. This problem is nearly as old as the field of forest economics itself. A forest owner faces various uncertainties caused by climate change and market shocks, due to its long-term nature of production and the joint production of interrelated timber and amenity (non-harvesting) benefit streams. The vast literature in stochastic rotation problems simply assumes a known probability distribution for whatever parameter is uncertain, but this type of assumption may lead to misspecification of a rotation decision model if a forest owner has no such information. We study a more relevant question of how to choose rotation ages when there is pure (or Knightian) uncertainty, in that the forest owner does not know distributional features of parameters and further can be averse to this type of information deficit. This chapter is the first to investigate pure uncertainty in amenity benefit streams and is also the first to analytically solve a stochastic rotation problem under pure uncertainty in either amenity streams or market prices. We use robust methods developed in macroeconomics that are particularly suited to forest capital investment problem, but with important differences owing to the nature of forest goods production. The results show that newer models suggesting rotation ages could be longer under volatile parameter distributions do not hold generally when pure uncertainty and forest owner uncertainty aversion is considered. Rather, the earlier literature showing faster or greater harvesting with increases in risk under risk neutrality may actually be a more general result than current literature supposes. In particular, we find that a landowner tends to harvest more when his degree of uncertainty aversion is higher and the model is misspecified by assumption, or when the volatility of an uncertain process is higher. These situations tend to magnify model misspecification costs, especially because the forest manager always assumes the worst case will happen when there is uncertainty. This implies the decision maker is pessimistic in the sense that he or she is always trying to maximize the utility under the worst possible state of nature (the lowest amenity benefit or the lowest timber price). Whether landowners are in fact uncertainty averse and assume the worst case in their decisions remains to be empirically investigated, but our work suggests it is an important question that must be answered. / Doctor of Philosophy / This dissertation consists of three essays in theoretical and applied microeconomics: the first essay is in cooperative game theory, and the second and third essays relate to forest economics. The first chapter studies a class of cooperative games dubbed ``r-essential games''. Cooperative game theory has proposed different notions of powerful players. For example, veto games (Bahel, 2016) have powerful players that are named veto players. Any coalition needs to include all these powerful players to achieve a positive coalition value. The first chapter extends these veto games by assuming that there is a given subset of powerful (or essential) players, but only a few (as opposed to all) essential players are required for a coalition to have a positive value. The resulting games, which are called r-essential games, encompass two classic games, convex games (Shapley, 1971) and veto games. We show that each r-essential game has at least one solution that is an allocation guaranteeing that no coalition can do better on its own. We provide a process allowing to compute this allocation in each r-essential game. An application to networks is provided. The second chapter estimates the damage of deforestation in China's northeast forests. This region consists of state forest enterprises which manage harvesting and reforestation and have represented the most important source of wood supplies since the 1950s. Deforestation is a major problem there. We develop a framework for assessing the damage to the society because of deforestation. We develop a theoretical model to describe the forest management structure, in which state forest managers make (potentially hidden) decisions under influence of provincial and central government policies. This model is used to quantify the damage. We then use data from a survey conducted by Peking University to compute the damage and confirm the main factors in these damages in practice. We find that lower harvesting limits and a more accurate monitoring system are the keys to lowering the damage. These are more important than conventional instruments used by the governments such as the wages for managers that achieve certain targets. We also find that the remote areas with a higher percentage of mature natural forests are the ones that will always have the largest damage. These areas are the hardest to monitor, but our results show they must be a critical focus moving forward. The third chapter studies when should a forest owner harvest under uncertain future ecosystem values and timber prices. A forest owner faces various uncertainties caused by climate change and market shocks, due to its long-term nature of production and the joint production of interrelated timber and non-harvesting benefit streams (such as the recreation value, the biodiversity value and the clean air supported by forests). Previous studies assume a known probability distribution for whatever parameter is uncertain, but this type of assumption may lead to a wrong decision model if a forest owner has no such information. We study a more relevant question of how to choose when to harvest with pure uncertainty, in that the forest owner does not know distributional features of parameters and further can be averse to this type of information deficit. This chapter is the first to investigate pure uncertainty and is also the first to analytically solve a harvest decision making problem under pure uncertainty in either non-harvesting benefit streams or market prices. We use macroeconomics methods that are particularly suited to forest capital investment problem. We find that a landowner tends to harvest more when there is pure uncertainty. Because the forest manager is pessimistic and always thinks the worst case will happen when there is uncertainty.

A análise de modelos de contato na presença de elementos de desordem fixa indica o surgimento de desvios em relação ao comportamento crítico do modelo uniforme subjacente. Nesse trabalho consideramos o efeito da aperiodicidade, que também é capaz de produzir flutuações de natureza geométrica. Utilizamos distri­ buições aperiódicas de probabilidades, definidas através de regras de substituição determinísticas, a fim de analisar o comportamento crítico desses modelos de con­ tato. Realizamos simulações de Monte Carlo para modelos definidos por três regras distintas, caracterizadas por um expoente w, associado à intensidade das flutuações geométricas. Nos modelos A e B, com w = -1 e w = 0, não constatamos qualquer mudança em relação à classe de universalidade crítica da percolação direcionada. Já no Modelo C, com w = 0.6309, as flutuações geométricas alteram a classe de universalidade crítica. / The analysis of contact models in the presence of quenched disorder indicates the onset of deviations with respect to the critical behavior of the underlying uniform system. In the present work, we consider the effects of aperiodicity, which are also known to produce fluctuation of geometric nature. We use aperiodic distributions of probabilities, given by deterministic substitution rules, in order to analyze the critical behavior. We performed Monte Carlo simulations for three different rules, characterized by an exponent w, which gauges the intensity of the geometric fluc­ tuations. For models A and B, with w = -1and w = 0, we have not detected any changes with respect to the universality class of directed percolation. For model C, with w = 0.6309, the geometric fluctuations change the critical universality class.

Algumas abordagens para o problema de Planejamento Ótimo da Operação de Sistemas Hidrotérmicos (POOSH) utilizam modelos estocásticos para representar as vazões afluentes dos reservatórios do sistema. Essas abordagens utilizam, em geral, técnicas de Programação Dinâmica Estocástica (PDE) para resolver o POOSH. Por outro lado, muitos autores têm defendido o uso dos modelos determinísticos ou, particularmente, a Programação Dinâmica Determinística (PDD) por representar de forma individualizada a interação entre as usinas hidroelétricas do sistema. Nesse contexto, esta dissertação tem por objetivo comparar o desempenho da solução do POOSH obtida via PDD com a solução obtida pela PDE, que emprega um modelo Markoviano periódico, com distribuição condicional Log-Normal Truncada para representar as vazões. Além disso, é realizada a análise com abordagem bayesiana, no modelo de vazões, para estimação dos parâmetros e previsões das vazões afluentes. Comparamos as performances simulando a operação das usinas hidroelétricas de Furnas e Sobradinho, considerando séries de vazões geradas artificialmente / Some approaches for problem of Optimal Operation Planning of Hydrothermal Systems (OOPHS) use stochastic models to represent the inflows in the reservoirs that compose the system. These approaches typically use the Stochastic Dynamic Programming (SDP) to solve the OOPHS. On the other hand, many authors defend the use of deterministic models and, particularly, the Deterministic Dynamic Programming (DDP) since it individually represents the interaction between the hydroelectric plants. In this context, this dissertation aims to compare the performance of the OOPHS solution obtained via DDP with the one given by SDP, which employs a periodic Markovian model with conditional Truncated Log-Normal distribution to represent the inflows. Furthermore, it is performed a bayesian approach analysis, in the inflow model, for estimating the parameters and forecasting the inflows. We have compared the performances of the DDP and SDP solutions by simulating the hydroelectric plants of Furnas and Sobradinho, employing artificially generated series

Many real-world systems experience deterioration with usage and age, which often leads to low product quality, high production cost, and low system availability. Most previous maintenance and reliability models in the literature do not incorporate condition monitoring information for decision making, which often results in poor failure prediction for partially observable deteriorating systems. For that reason, the development of fault prediction and control scheme using condition-based maintenance techniques has received considerable attention in recent years. This research presents a new framework for predicting failures of a partially observable deteriorating system using Bayesian control techniques. A time series model is fitted to a vector observation process representing partial information about the system state. Residuals are then calculated using the fitted model, which are indicative of system deterioration. The deterioration process is modeled as a 3-state continuous-time homogeneous Markov process. States 0 and 1 are not observable, representing healthy (good) and unhealthy (warning) system operational conditions, respectively. Only the failure state 2 is assumed to be observable. Preventive maintenance can be carried out at any sampling epoch, and corrective maintenance is carried out upon system failure. The form of the optimal control policy that maximizes the long-run expected average availability per unit time has been investigated. It has been proved that a control limit policy is optimal for decision making. The model parameters have been estimated using the Expectation Maximization (EM) algorithm. The optimal Bayesian fault prediction and control scheme, considering long-run average availability maximization along with a practical statistical constraint, has been proposed and compared with the age-based replacement policy. The optimal control limit and sampling interval are calculated in the semi-Markov decision process (SMDP) framework. Another Bayesian fault prediction and control scheme has been developed based on the average run length (ARL) criterion. Comparisons with traditional control charts are provided. Formulae for the mean residual life and the distribution function of system residual life have been derived in explicit forms as functions of a posterior probability statistic. The advantage of the Bayesian model over the well-known 2-parameter Weibull model in system residual life prediction is shown. The methodologies are illustrated using simulated data, real data obtained from the spectrometric analysis of oil samples collected from transmission units of heavy hauler trucks in the mining industry, and vibration data from a planetary gearbox machinery application.

Many real-world systems experience deterioration with usage and age, which often leads to low product quality, high production cost, and low system availability. Most previous maintenance and reliability models in the literature do not incorporate condition monitoring information for decision making, which often results in poor failure prediction for partially observable deteriorating systems. For that reason, the development of fault prediction and control scheme using condition-based maintenance techniques has received considerable attention in recent years. This research presents a new framework for predicting failures of a partially observable deteriorating system using Bayesian control techniques. A time series model is fitted to a vector observation process representing partial information about the system state. Residuals are then calculated using the fitted model, which are indicative of system deterioration. The deterioration process is modeled as a 3-state continuous-time homogeneous Markov process. States 0 and 1 are not observable, representing healthy (good) and unhealthy (warning) system operational conditions, respectively. Only the failure state 2 is assumed to be observable. Preventive maintenance can be carried out at any sampling epoch, and corrective maintenance is carried out upon system failure. The form of the optimal control policy that maximizes the long-run expected average availability per unit time has been investigated. It has been proved that a control limit policy is optimal for decision making. The model parameters have been estimated using the Expectation Maximization (EM) algorithm. The optimal Bayesian fault prediction and control scheme, considering long-run average availability maximization along with a practical statistical constraint, has been proposed and compared with the age-based replacement policy. The optimal control limit and sampling interval are calculated in the semi-Markov decision process (SMDP) framework. Another Bayesian fault prediction and control scheme has been developed based on the average run length (ARL) criterion. Comparisons with traditional control charts are provided. Formulae for the mean residual life and the distribution function of system residual life have been derived in explicit forms as functions of a posterior probability statistic. The advantage of the Bayesian model over the well-known 2-parameter Weibull model in system residual life prediction is shown. The methodologies are illustrated using simulated data, real data obtained from the spectrometric analysis of oil samples collected from transmission units of heavy hauler trucks in the mining industry, and vibration data from a planetary gearbox machinery application.

A análise de modelos de contato na presença de elementos de desordem fixa indica o surgimento de desvios em relação ao comportamento crítico do modelo uniforme subjacente. Nesse trabalho consideramos o efeito da aperiodicidade, que também é capaz de produzir flutuações de natureza geométrica. Utilizamos distri­ buições aperiódicas de probabilidades, definidas através de regras de substituição determinísticas, a fim de analisar o comportamento crítico desses modelos de con­ tato. Realizamos simulações de Monte Carlo para modelos definidos por três regras distintas, caracterizadas por um expoente w, associado à intensidade das flutuações geométricas. Nos modelos A e B, com w = -1 e w = 0, não constatamos qualquer mudança em relação à classe de universalidade crítica da percolação direcionada. Já no Modelo C, com w = 0.6309, as flutuações geométricas alteram a classe de universalidade crítica. / The analysis of contact models in the presence of quenched disorder indicates the onset of deviations with respect to the critical behavior of the underlying uniform system. In the present work, we consider the effects of aperiodicity, which are also known to produce fluctuation of geometric nature. We use aperiodic distributions of probabilities, given by deterministic substitution rules, in order to analyze the critical behavior. We performed Monte Carlo simulations for three different rules, characterized by an exponent w, which gauges the intensity of the geometric fluc­ tuations. For models A and B, with w = -1and w = 0, we have not detected any changes with respect to the universality class of directed percolation. For model C, with w = 0.6309, the geometric fluctuations change the critical universality class.

Submitted by Renata Lopes (renatasil82@gmail.com) on 2016-07-28T12:04:17Z No. of bitstreams: 1 rafaelbrunodasilvabrandi.pdf: 13228407 bytes, checksum: 1e92e8c2fa686ddcaea1c9ed0d33b278 (MD5) / Approved for entry into archive by Adriana Oliveira (adriana.oliveira@ufjf.edu.br) on 2016-07-28T12:16:14Z (GMT) No. of bitstreams: 1 rafaelbrunodasilvabrandi.pdf: 13228407 bytes, checksum: 1e92e8c2fa686ddcaea1c9ed0d33b278 (MD5) / Made available in DSpace on 2016-07-28T12:16:14Z (GMT). No. of bitstreams: 1 rafaelbrunodasilvabrandi.pdf: 13228407 bytes, checksum: 1e92e8c2fa686ddcaea1c9ed0d33b278 (MD5) Previous issue date: 2016-02-29 / CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / CNPq - Conselho Nacional de Desenvolvimento Científico e Tecnológico / O Sistema Elétrico Brasileiro (SEB) apresenta características peculiares devido às grandes dimensões do país e pelo fato da geração elétrica ser proveniente predominantemente de usinas hidráulicas. Como as afluências a estas usinas possuem comportamento estocástico e grandes reservatórios proporcionam ao sistema a capacidade de uma regularização plurianual, a utilização dos recursos hidráulicos deve ser planejada de forma minuciosa em um horizonte de tamanho considerável. Assim, o planejamento da operação de médio prazo compreende um período de 5 a 10 anos com discretização mensal e é realizado por uma cadeia de modelos computacionais tal que o principal modelo desta cadeia é baseado na técnica da Programação Dinâmica Dual Estocástica (PDDE). O objetivo deste trabalho é obter avanços nas metodologias de programação dinâmica atualmente utilizadas. Partindo-se da utilização da inserção iterativa de cortes, implementa-se um modelo computacional para o planejamento da operação de médio prazo baseado na metodologia de Programação Dinâmica Estocástica (PDE) utilizando uma discretização mais eficiente do espaço de estados (PDEE). Além disso, a metodologia proposta de PDE possui um critério de convergência bem definido para o problema, de forma que a inclusão da medida de risco CVaR não altera o processo de avaliação da convergência de forma significante. Dado que a inclusão desta medida de risco à PDDE convencional dificulta a avaliação da convergência do processo pela dificuldade da estimação de um limite superior válido, o critério de convergência proposto na PDEE é, então, base para um novo critério de convergência para a PDDE tal que pode ser aplicado mesmo na consideração do CVaR e não aumenta o custo computacional envolvido. Adicionalmente, obtém-se um critério de convergência mais detalhado em que as séries utilizadas para amostras de afluência podem ser avaliadas individualmente tais que aquelas que, em certo momento, não contribuam de forma determinante para a convergência podem ser descartadas do processo, diminuindo o tempo computacional, ou ainda serem substituídas por novas séries dentro de uma reamostragem mais seletiva dos cenários utilizados na PDDE. As metodologias propostas foram aplicadas para o cálculo do planejamento de médio prazo do SIN baseando-se em subsistemas equivalentes de energia. Observa-se uma melhoria no algoritmo base utilizado para a PDE e que o critério proposto para convergência da PDDE possui validade mesmo quando CVaR é considerado na modelagem. / The Brazilian National Grid (BNG) presents peculiar characteristics due to its huge territory dimensions and hydro-generation predominancy. As the water inflows to these plants are stochastic and a pluriannual regularization for system storage capacity is provided, the use of hydro-generation must be planned in an accurate manner such that it considersalongplanningperiod. So, thelong-termoperationplanning(LTOP)problemis generallysolvedbyachainofcomputationalmodelsthatconsideraperiodof5to10years ahead such that the primary model of this chain is based on Stochastic Dual Dynamic Programming (SDDP) technique. The main contribution of this thesis is to propose some improvements in Stochastic Dynamic Programming techniques usually settled on solving LTOP problems. In the fashion of an iterative cut selection, it is firstly proposed a LTOP problem solution model that uses an ecient state space discretization for Stochastic Dynamic Programming (SDP), called ESDP. The proposed model of SDP has a welldefined convergence criterion such that including CVaR does not hinder convergence analysis. Due to the lack of good upper bound estimators in SDDP when including CVaR, additional issues are encountered on defining a convergence criterion. So, based on ESDP convergence analysis, a new criterion for SDDP convergence is proposed such that it can be used regardless of CVaR representation with no extra computational burden. Moreover, the proposed convergence criterion for SDDP has a more detailed description such that forward paths can be individually assessed and then be accordingly discarded for computational time reduction, or even define paths to be replaced in a more particular resampling scheme in SDDP. Based on aggregate reservoir representation, the proposed methodsofconvergenceofSDDPandtheESDPwereappliedonLTOPproblemsrelatedto BNG. Results show improvements in SDDP based technique and eectiveness of proposed convergence criterion for SDDP when CVaR is used.

Submitted by Renata Lopes (renatasil82@gmail.com) on 2017-03-20T13:40:45Z No. of bitstreams: 1 wellingtoncarlosdaconceicao.pdf: 4259949 bytes, checksum: 52410bbb422df8d4e80e7f6956efc71e (MD5) / Approved for entry into archive by Adriana Oliveira (adriana.oliveira@ufjf.edu.br) on 2017-03-21T13:12:55Z (GMT) No. of bitstreams: 1 wellingtoncarlosdaconceicao.pdf: 4259949 bytes, checksum: 52410bbb422df8d4e80e7f6956efc71e (MD5) / Made available in DSpace on 2017-03-21T13:12:55Z (GMT). No. of bitstreams: 1 wellingtoncarlosdaconceicao.pdf: 4259949 bytes, checksum: 52410bbb422df8d4e80e7f6956efc71e (MD5) Previous issue date: 2016-12-12 / O sistema de produção de energia elétrica brasileiro é um sistema hidrotérmico de grande porte com forte predominância de usinas hidrelétricas. O planejamento e operação do sistema é realizado considerando diversos fatores, tais como, estocasticidade das afluências, usinas hidrelétricas em cascata e acoplamento temporal da operação. A resolução deste tipodeproblemaéfeitaconsiderandodiversoshorizontesdeplanejamento. Oplanejamento da operação de médio prazo compreende um período de 5 anos de estudo, e este período é discretizado em base mensal. O presente trabalho apresenta uma metodologia alternativa para resolução do problema de planejamento da operação de médio prazo de sistemas hidrotérmicos utilizando a Programação Dinâmica Estocástica (PDE) com discretização dointercâmbiodeenergiaentreossubsistemas(PDE-INT).Alémdisso, utiliza-seatécnica de sistemas equivalentes de energia e o algoritmo de fechos convexos (convex hull) para obtenção da função de custo futuro a partir dos pontos obtidos pela PDE-INT. Nesta abordagem, para cálculo da política energética, os subsistemas são considerados isolados, e desta forma, as variáveis que compõem o espaço de estados que são discretizadas são a energia armazenada e o intercâmbio líquido entre os subsistemas. Inicialmente, para análise e avaliação da metodologia proposta na resolução do problema de planejamento hidrotérmico, criou-se um sistema tutorial, composto por dois subsistemas. Em seguida, a metodologia foi utilizada considerando todo o sistema elétrico brasileiro, representado por quatro subsistemas ou submercados. Os resultados mostraram que com a técnica de separação dos subsistemas há uma redução significativa no tempo computacional quando comparados com as técnicas tradicionais que utilizam programação dinâmica. Desta forma, a metodologia proposta pode ser utilizada para uma análise rápida e inicial do caso em estudo, servindo como base para estudos e refinamentos posteriores. / The Brazilian power production system is a large scale hydrothermal system with a strong predominance of hydroelectric power plants. The electric power system operation planning must take into consideration several factors, such as uncertainty of the water inflows, hydroelectric plants in cascade and temporal coupling. This problem is solved considering different planning horizon. The long-term operation planning problem is generally solved by a chain of computational models that consider a period of 5 years ahead with monthly discretization. This work presents an alternative strategy to solve hydrothermalsystemsoperationplanningbyStochasticDynamicProgramming(SDP)with discretization of energy interchange between subsystems (SDP-INT). Under the presented approach, the hydroelectric plants are grouped into energy equivalent subsystems and the expected operation cost functions are modeled by a piecewise linear approximation, by means of the convex hull algorithm. Also, under this methodology, the subsystems are solved isolated, but the net energy interchange (export – import) between subsystems is set as a state variable of the cost function, together with the energy storage Initially, for the analysis and evaluation of the proposed methodology applied on solving the hydrothermalplanningproblem, themethodologyisusedinatutorialsystem, composedof two subsystems. Next, a simulation with the whole Brazilian electrical system considering four subsystems is presented. The results have shown that this subsystems separation technique reduces significantly the computation time when compared with the traditional techniques, demonstrating the effectiveness of the proposed methodology. Thus, the proposed methodology can be used for a fast and initial analysis of the case study, serving as a basis for further studies.

Dans les régions semi-arides, les systèmes de production agricole dépendent fortement de l'irrigation et font face à des difficultés croissantes (épuisement des ressources naturelles, forte volatilité des prix du marché, hausse des coûts de l'énergie, incertitude sur les changements climatiques). Modéliser ces systèmes agricoles et la façon dont ils s'adaptent est important pour les décideurs politiques afin de mieux évaluer leur flexibilité et leur résilience. Pour comprendre la capacité des systèmes agricoles à s'adapter, il est essentiel de considérer l'ensemble du processus de décision : des décisions sur le long-terme à l'échelle de l'exploitation aux décisions de court-terme à l'échelle de la parcelle. Pour ce faire, cette thèse conçoit un système de production agricole adaptable dans un contexte de diminution de l'eau et de changement climatique. Elle fournit une méthodologie guidant l'acquisition de données, leur analyse et la conception de modèle. Elle présente le modèle de simulation NAMASTE représentant les décisions des agriculteurs, les interactions entre agriculteurs pour l'utilisation des ressources communes et met l'accent sur la rétroaction entre pratiques agricoles et évolution de la nappe phréatique. Le modèle a été initialement développé pour résoudre les problèmes critiques de baisse des eaux souterraines liés aux pratiques agricoles dans un bassin versant du sud-ouest de l'Inde. Sa structure, ses cadres conceptuels et ses formalismes peuvent être utilisés dans d'autres contextes agricoles. / In semi-arid regions, agricultural production systems depend greatly on irrigation and encounter increasing challenges (depletion of natural resources, high volatility in market prices, rise in energy costs, growing uncertainty about climate change). Modeling farming systems and how these systems change and adapt to these challenges is particularly interesting for policy makers to better assess their flexibility and resiliency. To understand the ability of farming systems to adapt, it is essential to consider the entire decision-making process: from long-term decisions at the farm scale to short-term decisions at the plot level. To this end, the thesis conceives a flexible and resilient agricultural production system under a context of water scarcity and climate change. It provides a step-by-step methodology that guides data acquisition and analysis and model design. It proposes a simulation model NAMASTE that simulates the farmers' decisions in different time and space scales, represents the interactions between farmers for resource uses and emphasizes the feedback and retroaction between farming practices and changes in the water table. The model was initially developed to address critical issues of groundwater depletion and farming practices in a watershed in southwestern India. Its structure, frameworks and formalisms can be used in other agricultural contexts.