Proceedings of the 2023 Mississippi Water Resources Conference

Year: 2023 Authors: Raczynski K., Dyer J.

Drought is a complex phenomenon that is difficult to predict due to the variety of associated environmental drivers; however, its cyclic reoccurrence provides a basis for time-scale pattern analysis. The time factor reflects a complex structure of meteorological, geological, and surface factors affecting the specific way that drought develops in each region. In terms of the precipitation to evaporation ratio, seasonal dry periods indicate a time of increased probability for hydrological drought to form; however, underground water inertia within a region might affect the process in the opposite way, depending on the aquifer stage and general impact on runoff. The combination of long- and short-term changes in runoff structure might provide valuable insights into drought risk assessment and, especially due to climate change, improve the recognition of temporal patterns of drought occurrence. The objective of this work is to apply the Harmonic Oscillator Seasonality-Trend (HOST) model framework to analysis of hydrologic drought patterns, using a simulated streamflow dataset covering the Southeast US. Daily flow data from the National Water Model retrospective dataset v.2.1 were used for the period Feb. 1979–Dec. 2020, which provides long-term simulated streamflow data at a high spatial resolution across the study region . Droughts are identified using an objective threshold approach and aggregated on monthly scales. The temporal changes are assessed by a set of superimposed harmonic functions calculated for decomposed time series, representing drought occurrences and minimal flows. The HOST model can capture and reflect drought patterns on both short- and long-term time scales, with the first reflecting annual precipitation patterns and the second occurring close to the El Nino-Southern Oscillation cycle. Input data are divided with a 70/30 split for training and testing, respectively. Results are evaluated using contingency statistics and indicate about 80% accuracy for trained models (with an IQR between 74% and 87%) and around 60% accuracy in testing sets (with an IQR between 47% and 76%). In terms of spatial distribution, areas in southern Mississippi and eastern South Carolina are identified as regions with extended function periods, indicating prolongation of dry conditions. The work presents an early stage of model software development and initial results, with current limitations identified and discussed.