Utilizing waveform synthesis in harmonic oscillator seasonal trend model for short- and long-term streamflow drought modeling and forecasting

Abstract

This study introduces an improved version of the harmonic oscillator seasonal trend (HOST) model framework to accurately simulate medium- and long-term changes in extreme events, focusing on streamflow droughts in the Mobile River catchment. Performance of the model relative to the initial framework was enhanced through the inclusion of new mathematical models and waveform synthesis. The updated framework successfully captures long-term and seasonal patterns with a Kling–Gupta efficiency exceeding 0.5 for seasonal fluctuations and over 0.9 for trends. The best-fit model explains around 98% of long-term and approximately 55% of seasonal variance. Test sets show slightly lower accuracies, with about 20% of nodes underperforming due to the absence of drought during the test phase resulting in false-positive model forecasts. The newly developed weighted occurrence classification outperforms the binary classification occurrence model. In addition, application of an automatic period multiplier for decomposition using the seasonal trend decomposition using LOESS method improves test dataset performance and reduces false-positive forecasts. The improved framework provides valuable insights for extreme flow distribution, offering potential for improved water management planning, and the combination of the HOST model with physical models can address short-term drivers of extreme events, enhancing drought occurrence forecasting and water resource management strategies.