• The choice of energy-balance or temperature-index snowmelt modules is often ad-hoc. • Two snowmelt modules under two snow density functions are examined in SWAT model. • Cascade of uncertainty for future projections varies across spatiotemporal scales. • Snow density approach is a major control of snow depth simulation and projection. • Unlike mountains, in plain, snowmelt module uncertainties are scanty but vary in time. Snowmelt is a major driver of the hydrological cycle in cold regions, as such, its accurate representation in hydrological models is key to both regional snow depth and streamflow prediction. The choice of a proper method for snowmelt representation is often improvised; however, a thorough characterization of uncertainty in such process representations particularly in the context of climate change has remained essential. To fill this gap, this study revisits and characterizes performance and uncertainty around the two general approaches to snowmelt representation, namely Energy-Balance Modules (EBMs) and Temperature-Index Modules (TIMs). To account for snow depth simulation and projection, two common Snow Density formulations (SNDs) are implemented that map snow water equivalent (SWE) to snow depth. The major research questions we address are two-fold. First, we examine the dominant controls of uncertainty in snow depth and streamflow simulations across scales and in different climates. Second, we evaluate the cascade of uncertainty of snow depth projections resulting from impact model parameters, greenhouse gas emission scenarios, climate models and their internal variability, and downscaling processes. We enable the Soil and Water Assessment Tool (SWAT) by coupling EBM, TIM, and two SND modules for examination of different snowmelt representation methods, and Analysis of Variance (ANOVA) for uncertainty decomposition and attribution. These analyses are implemented in mountainous, foothill, and plain regions in a large snow-dominated watershed in western Canada. Results show, rather counter-intuitively, that the choice of SND is a major control of performance and uncertainty of snow depth simulation rather than the choice between TIMs and EBMs and of their uncertain parameters. Also, analysis of streamflow simulations suggest that EBMs generally overestimate streamflow on main tributaries. Finally, uncertainty decompositions show that parameter uncertainty related to snowmelt modules dominantly controls uncertainty in future snow depth projections under climate change, particularly in mountainous regions. However, in plain regions, the uncertainty contribution of model parameters becomes more variable with time and less dominant compared with the other sources of uncertainty. Overall, it is shown that the hydro-climatic and topographic conditions of different regions, as well as input data availability, have considerable effect on reproduction of snow depth, snowmelt and resulting streamflow, and on the share of different uncertainty sources when projecting regional snow depth.

• Simulation-optimization techniques are essential but computationally cumbersome. • Classic surrogates that globally emulate response surfaces can be of limited help. • Local surrogate models are proposed using automatic clustering for simulation. • The proposed method is shown to be efficient and robust in groundwater remediation. Simulation-optimization techniques in support of groundwater management are computationally expensive. To tackle such computational burden, a variety of surrogate modeling-frameworks have been proposed, where a cheaper-to-run model referred to as a surrogate is used in lieu of a computationally intensive model. These frameworks are generally based on what referred herein to as ‘global surrogate modelling’ where a single surrogate approximates the underlying response surface of a model. Such classic frameworks, however, are sub-optimal when the response surface is complex and/or high-dimensional. This paper proposes a novel ‘local surrogate modelling’ framework that simultaneously builds and evolves multiple local surrogates, guided by an automatic clustering method. Unlike traditional clustering methods that select the number of clusters a priori, the proposed automatic clustering method concurrently determines the optimum number of clusters and the clustering scheme itself. To serve as the surrogate, Artificial Neural Networks (ANNs) are used. The proposed framework is applied to solve a computationally intensive groundwater remediation optimization problem. This study shows that the proposed automatic clustering-based local surrogate modeling is effective and reliable while reducing at least 60 percent of the computational burden.

• Time-variant variogram analysis reveals significant event scale parameter importance variability. • The type of modeling objectives used influences parameter importance. • Input rainfall intensity and hyetograph shape affects parameters importance. • VARS is an effective and robust approach for identifying key modeling parameters. Runoff and sediment yield predictions using rainfall-runoff modeling systems play a significant role in developing sustainable rangeland and water resource management strategies. To characterize the behavior and predictive uncertainty of the KINEROS2 physically-based distributed hydrologic model, we assessed model parameters importance at the event-scale for small nested semi-arid subwatersheds in southeastern Arizona using the Variogram Analysis of Response Surfaces (VARS) methodology. A two-pronged approach using time-aggregate and time-variant parameter importance analysis was adopted to improve understanding of the control and behavior of models. The time-aggregate analysis looks at several signature responses, including runoff volume, sediment yield, peak runoff, runoff duration, time to peak, lag time, and recession duration, to investigate the influence of parameter and input on the model predictions. The time-variant analysis looks at the dynamical influence of parameters on the simulation of flow and sediment rates at every simulation time step using the different forcing inputs. This investigation was able to address Simpson’s paradox-type issues where the analysis across the different objective functions and full data set vs. its subsets (i.e., different events and/or time steps) could yield inconsistent and potentially misleading results. The results indicated the uncertainties in the flow responses are primarily due to the saturated hydraulic conductivity, the Manning’s coefficient, the soil capillary coefficient, and the cohesion in sediment and flow-related responses. The level of influence of K2 parameters depends on the type of the model response surface, the rainfall, and the watershed size.