This thesis research focuses on drought propagation analysis, classification, and prediction of anomalies by applying three new and one already established threshold level approaches. Five selected European catchments were the study areas and drought duration and deficit volume were the drought characteristics investigated using these threshold level approaches.
Investigation of new threshold level approaches towards drought propagation analysis and typology
Threshold level approaches are widely used to identify the threshold below which the flow or other hydrometeorological variable is defined as drought. However, depending on the hydrometeorological behavior of the catchment, these approaches can introduce artifact drought events or provide characteristics that do not represent the actual drought event as it is felt by one or more sectors. In this research variable threshold levels were studied. Four annual variable curves of daily threshold levels have been identified using (1) moving average of monthly quantile (MAM), (2) moving average of daily quantile (MAD), (3) Thirty-days moving window quantile (TMW) and (4) Fast Fourier Transform of daily quantile (FFT). These threshold levels were applied to the time series of hydrometeorological variables that were simulated using the HBV model outcome from five European catchments with contrasting climate conditions and catchment properties. Drought duration and deficit volume were used to investigate the propagation pattern (through the summary statistics of hydrometeorological variables), classification of the drought typologies and prediction of anomalies. We employed neural network using resilient back-propagation algorithm with weight backtracking for the prediction of anomalies in discharge. Despite the weak performance of the FFT annual threshold level approach in Metuje and Upper Sázava catchments, we found that all approaches can be used alternatively to investigate drought propagation patterns and identification of drought typologies. However, TMW threshold approach seems to be a more reliable approach in snow-dominated catchments like the Narsjø catchment in Norway where the flow sharply rises during the snow melt. In addition, we found that artificial neural networks are powerful tools in the prediction of discharge anomalies from mean temperature and precipitation, soil moisture and groundwater storage anomalies determined using MAM and MAD threshold level approaches. The proposed approaches can be successfully applied in water management for early drought warning.