Off-campus UNL users: To download campus access dissertations, please use the following link to log into our proxy server with your NU ID and password. When you are done browsing please remember to return to this page and log out.
Non-UNL users: Please talk to your librarian about requesting this dissertation through interlibrary loan.
Data-Driven Models for Groundwater Management in Irrigated Cropland
The goal of this thesis is to develop a forecast-based framework to support groundwater management in irrigated cropland. To do so, the forecasting capabilities of seven data-driven models (artificial neural networks, support vector machines, random forests, extreme learning machines, genetic programming, autoregressive and naïve) are first tested for different lead-times (one to five months), hydrogeological regimes and water availability conditions. Then, the most accurate among the models is selected and evaluated at the aquifer scale across the High Plains. When non-satisfactory forecasts are obtained, a Multi Model Combination (MuMoC) based on a hybrid of an artificial neural network and instance-based learning method is applied. This alternative model uses forecasts from neighboring wells to improve the accuracy obtained with a single model. Finally, sensitivity analysis is applied to assess the contribution of the observational input uncertainty on the error. Analysis of the results allows the identification of artificial neural network (ANN) as the most accurate among the predictors, with good forecasting skills across lead-times, hydrological regimes and water availability conditions. When ANN is used at the aquifer-scale the results show high average forecasting skills, with metrics of performance illustrating higher error in areas of strong interaction between hydro-meteorological forcings, irrigation, and the aquifer. In those areas, the implementation of MuMoC lead to an increase in forecasting accuracy by about 25%. Sensitivity analysis results shows that modelling error is particularly sensitive to evapotranspiration uncertainties (followed by rainfall), especially during the crop growing season, while inputs as snowmelt and streamflow has a significant effect in modeling performances only for few times steps. We can conclude that the proposed framework can provide water managers with the proper information: (1) to select the most accurate data-driven model; (2) to assess how the model forecasting accuracy changes across hydrogeological regimes and lead times; (3) to determine how input uncertainties affect modeling performance. We therefore believe that an operational implementation of the proposed methodology can support decision making for managing water in irrigated areas.
Bioengineering|Plant sciences|Water Resource Management
Amaranto, Alessandro, "Data-Driven Models for Groundwater Management in Irrigated Cropland" (2019). ETD collection for University of Nebraska - Lincoln. AAI13426686.