Graduation Semester and Year




Document Type


Degree Name

Master of Science in Earth and Environmental Science


Earth and Environmental Sciences

First Advisor

Arne Winguth

Second Advisor

Elizabeth M Griffith


The predictive skill of hydrologic variables such as streamflow and soil moisture, in North Central Texas, has improved substantially in the recent decades. However, substantial model-data biases are still present during extreme climate events, such as droughts and flash floods. In this study, we have optimized the Hydraulic Ensemble Forecasting System (HEFS) through development of a conditional ensemble streamflow system, as well as forecast reservoir hydrometeorological conditions (e.g. drought indices) with an artificial neural network (ANN) model. Improving prediction of these reservoir conditions enables more effective reservoir management in terms of water resource and energy efficiency during regional weather and climate anomalies. In order to improve HEFS, the strength of the regional climatology teleconnections to global climate indices (e.g. the Atlantic Multidecadal Oscillation, AMO, and the Bivariate El Niño Southern Oscillation, ENSO) was evaluated through Pearson product correlation, singular spectrum analysis, and the evaluation of the regional precipitation probability density and cumulative distributions functions in regards to changes in climate phases. These results showed the greatest change in regional precipitation base state occurred during changes in the AMO phases, except in the case of an El Niño or La Niña event. This suggests that a conditional ensemble streamflow system could be constructed based on AMO phases to improve HEFS under regular conditions, and based on ENSO conditions (El Niño or La Niña events). In the pursuit of forecasting hydrometeorological conditions, multiple ANN models of different network architectures were trained and tested utilizing data from 1915-2012; 70% of the available data, from 1915 to 1982, were used for model training and the remaining for validation. The network architecture that produced the smallest prediction error was applied further in this study. The input data comprised regional climate variability observations of minimum and maximum temperature, total precipitation, average wind speed, evapotranspiration, potential evapotranspiration, and the monthly drought index value. The global climate indices investigated included dominant interannual and decadal oscillations. These indices were used to evaluate their respective ability to improve predictive skill during climate anomaly extremes, e.g., El Niño and La Niña conditions. The choice of climate indices were varied as input into retrained ANN models of the same network architecture, so that the improvement due to each climate index could be ranked and less-influential climate indices could be excluded. The selected ANN model architecture and input data mentioned above were then applied to produce 6 month-ahead predictions of monthly drought indices in order to evaluate the overall predictive skill of the generated ANN models. The ANN model was able to skillfully forecast drought conditions with 2-3 months lead time, with the evaporation variables generating the greatest increase in forecasting skill. The use of global climate indices did not exhibit any increase in the ANN models’ forecasting skill of North Central Texas regional hydrometeorological conditions most likely because the local observations consist of a regional signal that is superimposed by the global variations.


North Central Texas, Hydrometeorology, Artificial neural networks, Drought, Climate Indices, Atlantic Multidecadal Oscillation, El Niño Southern Oscillation, Standardized precipitation evapotranspiration index


Earth Sciences | Physical Sciences and Mathematics


Degree granted by The University of Texas at Arlington