With increasing evidence of climatic variability, there is a need to improve forecast for hydroclimatic variables i.e., precipitation and streamflow preserving their spatial and temporal variability. Climatologists have identified different oceanic-atmospheric oscillations that seem to influence the behavior of these variables and in turn can be used to extend the forecast lead time. In the absence of a good physical understanding of the linkages between oceanic-atmospheric oscillations and hydrological processes, it is difficult to construct a physical model. An attractive alternative to physically based models are the Artificial Intelligence (AI) type models, also referred to as machine learning or data-driven models. These models do not employ traditional forms of equations common in physically based models, but instead have flexible and adaptive model structures that can extract the relationship from the data.
With this motivation this research focuses on increasing the precipitation and streamflow forecast lead times and enhancing the temporal resolution of precipitation within the Colorado River Basin (CRB). An AI-type data-driven model, Support Vector Machine (SVM), was developed incorporating oceanic-atmospheric oscillations to increase the precipitation and streamflow forecast lead times. The temporal resolution of precipitation was improved using the stochastic nonparametric K-Nearest Neighbor (KNN) approach. The hydrologic data used in the dissertation comprised of climate division precipitation data and naturalized streamflow data for the Colorado River Basin. The interdecadal and interannual Pacific Ocean (Pacific Decadal Oscillation (PDO) and El Niño-Southern Oscillation(ENSO)) and Atlantic Ocean (Atlantic Multidecadal Oscillation (AMO) and North Atlantic Oscillation(NAO)) climatic variability was used in this dissertation.
Initially, the coupled and individual effect of oceanic-atmospheric oscillations in relation to annual precipitation within Colorado River Basin was investigated using the statistical SVM modeling approach. Next, the SVM modeling was used to investigate the coupled and individual effect of oceanic-atmospheric oscillations in relation to annual streamflow volume within Colorado River Basin. Finally, the long-term changes (Trend and Step) in seasonal precipitation within Colorado River Basin were analyzed using nonparametric statistical tests (Mann-Kendall, Spearman’s Rho, and Rank Sum). Additionally, the temporal resolution of precipitation was enhanced from annual (water year) to seasonal precipitation (autumn, winter, spring, and summer) using the nonparametric K-Nearest Neighbor disaggregation approach.
The results indicated that annual precipitation predictions for 1-year lead time for the Upper Colorado River Basin can be successfully obtained using a combination of PDO, NAO, and AMO indices, whereas coupling AMO and ENSO results in improved precipitation predictions for the Lower Colorado River Basin. Satisfactory annual streamflow predictions for 3-year lead time for the Upper Colorado River Basin can be obtained using a combination of NAO and ENSO. The seasonal changes in precipitation indicated a decrease in the Upper Basin and increase in the Lower Basin winter precipitation due to an abrupt step change. KNN disaggregation results indicated satisfactory seasonal precipitation estimates during winter and spring season compared to the autumn and summer season.
The major contributions of this research are threefold. First, this research is the first of its kind that used an AI-type SVM modeling approach to increase precipitation and streamflow forecast lead times using oceanic-atmospheric oscillations for the Colorado River Basin. Second, the results indicated that there is no single climate signal that can be used to explain the hydroclimatology within Colorado River Basin. Coupled response of oceanic-oscillations in relation to precipitation and streamflow is more pronounced in CRB compared to their individual effects. Finally, this is the first study that used a nonparametric KNN disaggregation approach for estimating seasonal precipitation for the Colorado River Basin. Other studies have focused on disaggregating streamflow within CRB from one scale to the other but no other study has attempted to disaggregate precipitation within the Colorado River Basin. Overall, this research improves the understanding of the relationship between climatic variables and hydrology within Colorado River Basin. The long lead time estimates of precipitation and streamflow developed in this research can help water managers in managing the water resources (e.g. reservoir releases, allocation of water contracts etc.) within the Colorado River Basin.