406 / 2024-03-14 08:18:51

Diagnosing non-stationarity of hydrological model structure using Bayesian model averaging: whether saturation-excess or infiltration-excess dominates runoff generation

Non-stationarity; Runoff generation; Saturation-excess; Infiltration-excess; Bayesian model averaging

Subtheme 5: Climate Change: Adaptation, Mitigation and Resilience > Topic 1: Hydrology and Water Resources response to climate change

The stationarity of hydrological models is increasingly questioned under the impact of climate change and natural variability. Specifically, it is challenging to identify whether the intrinsic structure of the hydrological model varies with different climatic conditions. To address this issue, the presented study focuses on diagnosing the non-stationarity of runoff generation by considering two typical structures, i.e., saturation-excess and infiltration-excess models. First, the historical data were divided into wet and dry periods based on annual anomaly rainfall. For each period, saturation-excess and infiltration-excess models were then calibrated, respectively. Finally, the Bayesian model averaging method was used to combine these two models by setting their weights. The data used in this study came from three catchments in southeastern Australia, which experienced a decadal-long millennium drought. Preliminary analysis indicates that the saturation-excess changes to infiltration-excess after the millennium drought because the main influence factor of the flood volume is from rainfall amount to its intensity. Results also show that the weights of the infiltration-excess model are 0.45, 0.44, 0.33 in wet periods for the three catchments, while they are changed to 0.58, 0.54, 0.45 in dry periods, respectively. The incremental weights show that the infiltration-excess model is more appropriate in drier climatic conditions, demonstrating the non-stationarity of runoff generation after extreme drought. It is indicated that the Bayesian model averaging method is helpful to diagnose the non-stationarity of runoff generation under a changing climate.

The stationarity of hydrological models is increasingly questioned under the impact of climate change and natural variability. Specifically, it is challenging to identify whether the intrinsic structure of the hydrological model varies with different climatic conditions. To address this issue, the presented study focuses on diagnosing the non-stationarity of runoff generation by considering two typical structures, i.e., saturation-excess and infiltration-excess models. First, the historical data were divided into wet and dry periods based on annual anomaly rainfall. For each period, saturation-excess and infiltration-excess models were then calibrated, respectively. Finally, the Bayesian model averaging method was used to combine these two models by setting their weights. The data used in this study came from three catchments in southeastern Australia, which experienced a decadal-long millennium drought. Preliminary analysis indicates that the saturation-excess changes to infiltration-excess after the millennium drought because the main influence factor of the flood volume is from rainfall amount to its intensity. Results also show that the weights of the infiltration-excess model are 0.45, 0.44, 0.33 in wet periods for the three catchments, while they are changed to 0.58, 0.54, 0.45 in dry periods, respectively. The incremental weights show that the infiltration-excess model is more appropriate in drier climatic conditions, demonstrating the non-stationarity of runoff generation after extreme drought. It is indicated that the Bayesian model averaging method is helpful to diagnose the non-stationarity of runoff generation under a changing climate.