Alzraiee, Ayman H., authorGarcia, Luis A., advisorGates, Timothy K., advisorBau, Domenico, committee memberButters, Greg, committee member2007-01-032007-01-032012http://hdl.handle.net/10217/67402Sustainability of crop production in the Lower Arkansas River Basin in Colorado is seriously threatened by the continuous degradation of irrigated lands by the dual impact of soil salinization and waterlogging problems. Integration of improved irrigation practices, upgrades to the irrigation systems, and subsurface drainage are essential components of any plan to stop the deterioration of irrigated lands. Numerical simulations of irrigation and drainage systems are necessary to justify the consequent management actions. Despite the uncertainty of their predictions, numerical models are still indispensable decision support tools to investigate the feasibility of irrigation and drainage systems management plans. However, the uncertainties in input parameters to these models create a risk of misleading numerical results. That is beside the fact that the numerical models themselves are conceptual simplifications of the complex reality. The overarching objective of this dissertation is to investigate the impact of parameters uncertainty on the response of simulated irrigation-drainage systems. In the first part of the research, a Global Sensitivity Analysis (GSA) is conducted using a one-dimensional variably saturated problem to prioritize parameters according to their importance with respect to predefined performance indices. A number of GSA methods are employed for this purpose, and their comparative performances are investigated. Results show that only five parameters out of 18 parameters are responsible for around 73% of crop yield uncertainty. The second part introduces a method to reduce the computational requirements of Monte Carlo Simulations. Numerical simulation of variably saturated three-dimensional fields is typically a computationally intensive process, let alone Monte Carlo Simulations of such problems. In order to reduce the number of model evaluations while producing acceptable estimates of the output statistical properties, Cluster Analysis (CA) is used to group the input parameter realizations, e.g. hydraulic conductivity. The potentials of this approach are investigated using different: 1) clustering schemes; 2) clustering configurations, and 3) subsampling schemes. . Results show that response of 400 realizations ensemble can be efficiently approximated using selected 50 realizations. The third part of the research investigates the impact of input parameter uncertainty on the response of irrigation-drainage systems, particularly on crop yield and root zone hydrosalinity. The three-dimensional soil parameters, i.e. hydraulic conductivity, porosity, the pore size distribution (van Genuchten β) parameter, the inverse of the air entry pressure (van Genuchten α) parameter, the residual moisture content parameter, and dispersivity; are treated as spatial random processes. A sequential multivariate Monte Carlo simulation approach is implemented to produce correlated input parameter realizations. Other uncertain parameters that are considered in the study are irrigation application variability, irrigation water salinity, irrigation uniformity, preferential flow fraction, drain conductance coefficient, and crop yield model parameters. Results show that as the crop sensitivity to salinity increases, the crop yield standard deviation increases. The fourth part of the research investigates an approach for optimal sampling of multivariate spatial parameters in order to reduce their uncertainty. The Ensemble Kalman Filter is used as instrumentation to integrate the sampling of the hydraulic conductivity and the water level for a two-dimensional steady state problem. The possibility of combining designs for efficient prediction and for efficient geostatistical parameter estimation is also investigated. Moreover, the effect of relative prices of sampled parameters is also investigated. A multi-objective genetic algorithm is employed to solve the formulated integer optimization problem. Results reveal that the multi-objective genetic algorithm constitutes a convenient framework to integrate designs that are efficient for prediction and for geostatistical parameter estimation.born digitaldoctoral dissertationsengCopyright and other restrictions may apply. User is responsible for compliance with all applicable laws. For information about copyright law, please see https://libguides.colostate.edu/copyright.groundwatermodelingsalinitystochasticunsaturatedwaterloggingText