Harmonic Oscillator Seasonal Trend (HOST) Model for Hydrological Drought Pattern Identification and Analysis
Raczynski, K., & Dyer, J. (2023). Harmonic Oscillator Seasonal Trend (HOST) Model for Hydrological Drought Pattern Identification and Analysis. Journal of Hydrology. 620(B), 129514. DOI:10.1016/j.jhydrol.2023.129514.
Droughts, due to their complex nature, are difficult to predict; however, research is showing that there are often clearly defined patterns to their occurrence. The temporal aspects are usually more apparent when comparing regions with defined climatic characteristics, such as those with recognized wet or dry seasons, but recognizing these patterns can help in defining and understanding the overall processes influencing drought occurrence. Especially in the context of climate change, it is crucial to recognize temporal patterns and quantify them to help diagnose current and future event probabilities. The objective of this work is to introduce a new tool for assessing temporal patterns of streamflow drought, based on the harmonic oscillator theorem called the harmonic oscillator seasonal trend (HOST) model. The testing dataset consists of monthly minimal flows and drought occurrence information calculated based on daily flows from the National Water Model (NWM) retrospective v.2.1 dataset for the period 1979–2020. The model framework allows for trend and seasonality signal decomposition as well as the extraction of the first five harmonics by means of a Fast Fourier Transform (FFT)-based machine learning algorithm. The model framework enables the extraction and evaluation of five different harmonic models based on short and long-range components as well as linear regression sloped harmonics. The results show that the best-fit models are based on two superimposed sinusoidal functions with an accuracy of around 70–90 % for streamflow drought occurrence, with different spatial parameters reflecting varied environmental conditions. The models were able to capture two patterns: (1) annual precipitation variation and (2) longer-term inertia, the latter of which is likely driven by large-scale circulation with a roughly-eight year period.