Time-varying network-based approach for capturing hydrological extremes under climate change with application on drought

Abstract

Hydrologic extremes often lead to droughts and floods that adversely affect the socio-economic development. Change in the characteristics and causes of hydrologic extremes due to climate variability and climate change poses a challenge for its reliable prediction. We propose a time-varying approach to capture such temporal changes, often gradual, in hydrologic extremes through temporal networks (a series of network structures). Graphical Modelling (GM) based networks are developed through Bayesian Model Averaging (BMA) to deal with the complexity between the causal variables and extreme events. A demonstration of the proposed time-varying approach is shown for 1-month and 3-month ahead hydrological drought prediction in terms of Standardized Streamflow Anomaly Index (SSAI), at basin scale, that has notably changed in the recent years in terms of its frequency and severity. The frequency and severity of below-normal flow events has increased, particularly during the monsoon season (high flow months). We hypothesize that time-varying cause-effect relationship is important to capture such gradual change in the characteristics of hydrologic extremes. The results indicate that SSAI values for the low flow months are strongly associated with streamflow whereas for the high flow months the dominant predictors are rainfall, precipitable water and relative humidity. Furthermore, the cause-effect relationship between hydroclimatic variables and extreme events needs to be updated every 2 years for high flow and 3 years for low flow months. The proposed model very well captures the above and below-normal flow events and can be used as a remedial measure to handle similar cases through a proper assessment of time-varying cause-effect relationship between hydroclimatic variables and extreme events.