Assessing the Potential Robustness of Conceptual Rainfall‐Runoff Models Under a Changing Climate. Issue 7 (28th July 2018)

• Record Type: Journal Article
• Authors: Guo, Danlu; Johnson, Fiona; Marshall, Lucy
• Is Part Of: Water resources research. Volume 54:Issue 7(2018)
• Page Start: 5030
• Page End: 5049
• Publication Date: 2018-07-28
• Subjects: conceptual rainfall‐runoff model -- climate impact assessment -- potential robustness -- variability -- uncertainty -- runoff prediction; Hydrology -- Periodicals; 333.91
• DOI: 10.1029/2018WR022636
• Languages: English
• ISSNs: 0043-1397

Abstract

Abstract: Conceptual rainfall‐runoff (CRR) models are commonly used to assess the potential impact of climate change on water resources systems. However, they are often characterized by poorer performance when used to simulate a different climate compared to that of the calibration period. This is generally referred to as low model robustness, and these issues have been thoroughly explored using historical data. However, the implications of robustness are unknown for a changing climate where models may have to operate under conditions that lie beyond existing observations. This study extends these ideas to evaluate the "potential robustness" of different CRR models in the context of a changing climate. To achieve this aim, we combine a generalized split‐sample test framework with a stochastic weather generator. This allows us to assess the variabilities in runoff predictions obtained from using different calibration periods within each CRR model. We tested the potential robustness on three catchments with contrasting hydroclimatic conditions. We observed a consistent higher potential robustness in all models under drier conditions at all catchments. The three catchments illustrate contrasting patterns in the relative potential robustness of the three CRR models, which are related to both the structures of the CRR models and the unique catchment characteristics, highlighting the need of case‐specific assessment. This study illustrates a transferable empirical testing strategy to understanding variabilities in CRR model predictions. This approach can improve our knowledge of model behavior and thus informs the suitability of alternative models to simulate catchments hydrology under a changing climate. Plain Language Summary: Conceptual rainfall‐runoff (CRR) models are commonly used to understand how climate change may affect river flows, floods, and droughts. These models need to be calibrated to historical data. However, they tend to not perform well when used to simulate wetter or drier conditions. This property is referred to as low model robustness . This study extends this idea to focus on the uncertainty in model predictions under a changing climate, which is defined as potential robustness . By generating synthetic rainfall data, we can represent a range of possible future changes in rainfall, including scenarios that are very different from the historical climate. We tested the potential robustness of three CRR models on three river catchments. There are contrasting patterns in the relative potential robustness amongst models, which are related to both the structure of the CRR models and the unique characteristics of individual catchments. This means that potential robustness cannot be estimated in advance of a modeling study and a result catchment‐specific testing is essential. This study also illustrates a transferable approach to perform such testing. The potential robustness approach can improve our knowledge of model behaviors and help to choose suitable CRR models to assess climate change impacts. Key Points: We studied potential robustness of conceptual rainfall‐runoff models, as variability in predictions from different calibration periods Use of stochastic weather generator allows us to assess potential robustness for climate conditions that extrapolates beyond observations The approach is transferable across case studies to inform model selection and calibration strategies for climate change assessments