Advances in Quantifying Streamflow Variability Across Continental Scales: 1. Identifying Natural and Anthropogenic Controlling Factors in the USA Using a Spatially Explicit Modeling Method

Abstract

Despite considerable progress in hydrological modeling, challenges remain in the interpretation and accurate transfer of hydrological information across watersheds and scales. In the conterminous United States (CONUS), these limitations are related to spatial inconsistencies and constraints in hydrological model structures, including a lack of spatially explicit process components (streams, reservoirs, and watershed development) and restricted estimation of model parameters across watersheds. Collectively, such limitations can impede identification of the causes of streamflow variations across the diversity of watershed sizes and land uses in the CONUS and contribute to model imprecision and spatial inconsistencies in prediction uncertainties. We addressed these concerns with a new approach, the first hybrid (statistical-mechanistic) SPARROW (SPAtially Referenced Regression On Watershed attributes) model of long-term mean annual streamflow, applied across diverse environmental settings of the CONUS. The hybrid model coupled previous catchment-scale (1 km) water balance predictions of "natural" unit area runoff, which are inclusive of major water cycling processes, with additional explanatory variables (e.g., soils, vegetation, land use, topography, water losses in streams, and reservoirs) that account for the effects of natural and cultural water supply and demand processes that operate over large spatial scales and explain streamflow variability across CONUS river basins. Accounting for these statistically unique effects, including a nonlinear surface area-dependent scaling of water loss in river networks, significantly improved the accuracy of mean streamflow predictions in CONUS basins. Our hybrid modeling approach provides new methods for transferring hydrological information to ungauged locations in river networks, especially those in larger and more culturally diverse CONUS watersheds.

Advances in Quantifying Streamflow Variability Across Continental Scales: 2. Improved Model Regionalization and Prediction Uncertainties Using Hierarchical Bayesian Methods

“…We use a previously developed nonhierarchical SPARROW (Smith et al, ) model for long‐term mean annual streamflow in the CONUS (Alexander et al, ) as the foundational model for this study. The CONUS model employs a unique conceptual approach—one that couples the spatially explicit SPARROW river‐network structure (62,000 reach watersheds) with catchment‐scale (1‐km) predictions of mean annual unit‐area runoff from prior applications of monthly Thornthwaite water balance models in the CONUS (Wolock & McCabe, ).…”

“…The precision of statistical hydrological models (e.g., Vogel et al, , ) is enhanced by their more parsimonious structure and large‐scale applications that incorporate a broader range of spatial variation in the controlling variables. However, the overly simplistic description of catchment processes in statistical models, including the lack of mass balance constraints and spatially distributed watershed properties and the use of spatially constant model coefficients and uncertainties, can contribute to regional prediction biases and imprecision (Alexander et al, ; Preston et al, ; Smith et al, ).…”

“…In other studies, the regionalization procedures have included the simultaneous calibration of models across representative catchments having similar watershed attributes (e.g., Arheimer et al, ) and the use of spatial transfer functions based on the regression of catchment model parameters on watershed characteristics (e.g., Hundecha et al, ; Rakovec et al, ). Despite advances in these methods that have contributed to improved sharing of hydrological information over regional and continental spatial scales (e.g., Archfield et al, ; Bock et al, ; Hundecha et al, ; Rakovec et al, ], developing statistical methods for estimating parameters that are spatially and structurally consistent over large spatial scales remains difficult (Alexander et al, ). Spatial constraints on data sharing and parameter estimation reduce the quantity and information content of the data, which can confound the statistical uniqueness and sensitivity of model parameters and adversely affect model accuracy (Schwarz et al, ).…”

“…To address these limitations, we regionalized the coefficients and uncertainties of a previously developed SPAtially Referenced Regression On Watershed attributes (SPARROW; Alexander et al, ) model of long‐term mean annual streamflow in the CONUS, using a new generation of Bayesian methods ( RStan ; Stan Development Team, ). These methods offer more advanced computational efficiencies than the original framework that significantly advanced the field (e.g., WinBUGS) (Lunn et al, ).…”

“…Monitoring sites for the SPARROW mean annual streamflow model: (a) map of locations for with 1,778 calibration sites and 890 validation sites in the CONUS; boundaries for 16 major water resource regions are derived from major Hydrologic Unit Code (HUC‐2) regions; see Table for definitions and section for explanations; (b) plot of the number of calibration sites in each region versus total drainage area. See the companion study, Alexander et al (), for details on the selection of the calibration and validation sites.…”