This work addresses management of the scarce water resource for irrigation in arid regions where significant delays between the time of order and the time of delivery present major difficulties. Motivated by improvements to water management that will be facilitated by an ability to predict water demand, this work employs a data-driven approach to developing canal flow prediction models using the Relevance Vector Machine (RVM), a probabilistic kernel-based learning machine. Beyond the RVM learning process, which establishes the set of relevant vectors from the training data, a search is performed across model attributes including input set, kernel scale parameter, and model update scheme for models providing superior prediction capability. Models are developed for two canals in the Sevier River Basin of southern Utah for prediction horizons of up to five days. Appendices provide the RVM derivation in detail.

The Sevier River Basin situated in south central Utah is characterized by its semiarid climate, snowmelt-driven runoff, and high dependency on agricultural economy. High evapotranspiration and low precipitation make agricultural production challenging, but naturally stored water in the snowpack in the mountains alleviates water stresses during high water demand seasons. The snowmelt-driven river flow along the main channel is highly exploited for irrigation for farms near the Sevier River. Reservoir operations and river diversions result in heavily regulated flows from the upper to the lower basins. The return flows of over-irrigated water in the upper basin increase salinity of surface water. Long-term applications of salinity water in agriculture eventually produce high soil salinity in the agricultural areas near Delta in the lower basin, which deteriorated farmers’ crop productivity. Farmers cropping near Delta struggle with both water and salinity stresses. Indeed, crop prices and yields are always their concerns. For them, efficient water management can be achieved with consideration of hydrologic, agronomic, and economic aspects of water resources. The overall goal of this research was to develop a decision supporting framework for efficient water and land allocations that considered hydrologic processes, crop response to water in salinity-affected farms, and farmers’ profit and financial risk. This research introduces a methodology for predicting water availability in a given cropping year from the snowpack in the mountains, and agronomic simulations with satellite images follow for quantifying crop response to water. The hydrologic predictions and the agronomic simulations are finally incorporated into an economic analysis that provides efficient water and land allocations with multiple crop selections. In a rural river basin, data limitation is a common concern for water resources engineers; thus simple but robust methodologies are proposed for hydrologic prediction. In the same context, satellite images are used for the estimation of crop yields in individual farms near Delta with no prior crop experimental plots. Historical records of crop prices are used for the economic analysis. The methodologies developed in this research provide a comprehensive decision analysis framework for efficient water management where water is scare and available from snowmelt only, the economy depends on agriculture only, and salinity is present in both soil and water due to long-term irrigation. The case study is for the agricultural area near Delta in the Sevier River Basin, but its applicability is not limited and is flexibly applicable to other agricultural regions.