Abstract. In the past decade, machine learning methods for empirical rainfall–runoff modeling have seen extensive development and been proposed as a useful complement to physical hydrologic models, particularly in basins where data to support process-based models are limited. However, the majority of research has focused on a small number of methods, such as artificial neural networks, despite the development of multiple other approaches for non-parametric regression in recent years. Furthermore, this work has often evaluated model performance based on predictive accuracy alone, while not considering broader objectives, such as model interpretability and uncertainty, that are important if such methods are to be used for planning and management decisions. In this paper, we use multiple regression and machine learning approaches (including generalized additive models, multivariate adaptive regression splines, artificial neural networks, random forests, and M5 cubist models) to simulate monthly streamflow in five highly seasonal rivers in the highlands of Ethiopia and compare their performance in terms of predictive accuracy, error structure and bias, model interpretability, and uncertainty when faced with extreme climate conditions. While the relative predictive performance of models differed across basins, data-driven approaches were able to achieve reduced errors when compared to physical models developed for the region. Methods such as random forests and generalized additive models may have advantages in terms of visualization and interpretation of model structure, which can be useful in providing insights into physical watershed function. However, the uncertainty associated with model predictions under extreme climate conditions should be carefully evaluated, since certain models (especially generalized additive models and multivariate adaptive regression splines) become highly variable when faced with high temperatures.

There is increasing concern over deep uncertainty in the risk analysis field as probabilistic models of uncertainty cannot always be confidently determined or agreed upon for many of our most pressing contemporary risk challenges. This is particularly true in the climate change adaptation field, and has prompted the development of a number of frameworks aiming to characterize system vulnerabilities and identify robust alternatives. One such methodology is robust decision making (RDM), which uses simulation models to assess how strategies perform over many plausible conditions and then identifies and characterizes those where the strategy fails in a process termed scenario discovery. While many of the problems to which RDM has been applied are characterized by multiple objectives, research to date has provided little insight into how treatment of multiple criteria impacts the failure scenarios identified. In this research, we compare different methods for incorporating multiple objectives into the scenario discovery process to evaluate how they impact the resulting failure scenarios. We use the Lake Tana basin in Ethiopia as a case study, where climatic and environmental uncertainties could impact multiple planned water infrastructure projects, and find that failure scenarios may vary depending on the method used to aggregate multiple criteria. Common methods used to convert multiple attributes into a single utility score can obscure connections between failure scenarios and system performance, limiting the information provided to support decision making. Applying scenario discovery over each performance metric separately provides more nuanced information regarding the relative sensitivity of the objectives to different uncertain parameters, leading to clearer insights on measures that could be taken to improve system robustness and areas where additional research might prove useful.

No abstract

No abstract

No abstract