Pitfalls and recommendations for event detection of drought to flood transitions

In hydrology, streamflow droughts and floods are typically studied as events that are independent from one another, however, this assumption is not necessarily valid. From a physical perspective, for instance, streamflow can be autocorrelated, with signatures of past flow volumes reflected in present streamflow conditions. From a management perspective, rapid drought to flood transitions can leave strategies designed for one event counter-effective when dealing with the other extreme. Furthermore, impacts of rapid drought to flood transitions have the potential to be highly destructive.

The definition of drought and flood events can unintentionally bias detection of transitions in particular regions or for certain types of hydrological regimes or events. This can potentially alter the attributes of detected events, a problem which in a context of transitions has not yet been addressed. Thus, we aim to improve extreme event detection, with a particular focus on hydrologic transitions. We assess the sensitivity of transitions detection to different methodological choices, and we evaluate their appropriateness for various applications. We use eight global case study catchments to examine how existing methodological and parameter variation choices influence transition detection using the threshold level method. The case studies cover different hydroclimatological regimes ranging from a heavily snow driven catchment in Norway, to a semi-arid catchment in Texas, a flashy sub-alpine catchment in Switzerland, and a monsoonal regime in Australia, among others. We examine the impact of threshold type, its level, data aggregation window, and temporal transition window. 

Using a combination of quantitative and qualitative analyses applied to these case studies, we demonstrate the following. First, the choice of event detection approach and parameters can alter the detection and duration of events, resulting in some methods detecting “transitions” where others will not. For instance, fixed thresholds are more likely to capture dry conditions, while daily varying thresholds are better at identifying anomalous conditions as compared to the normal flow regime These characteristics point to different aspects of drought to flood transitions e.g. changes in the hydrophobicity of soil or context-specific aspects of water management. Second, less extreme drought and flood thresholds than those used in the study of individual events may be appropriate because the probability of transition occurrence within a specified time period can be very low, even if the independent events are probable. This, however, can be highly regime-dependent and careful consideration of what a transition “means”, in context, is essential for meaningful interpretation of hydrologic transitions across regime types. Finally, the selected time lag between the end of a drought and the beginning of a flood event is important for determining the presence of transition periods in the time series of different hydrological regimes. We highlight the potential pitfalls of different threshold level choices to aid future research in this field, representing the first ever set of methodology guidelines for hydrological transitions research.