Global catchment modelling using World-Wide HYPE (WWH), open data, and stepwise parameter estimation

Abstract

Abstract. Recent advancements in catchment hydrology (such as understanding catchment similarity, accessing new data sources, and refining methods for parameter constraints) make it possible to apply catchment models for ungauged basins over large domains. Here we present a cutting-edge case study applying catchment-modelling techniques with evaluation against river flow at the global scale for the first time. The modelling procedure was challenging but doable, and even the first model version showed better performance than traditional gridded global models of river flow. We used the open-source code of the HYPE model and applied it for >130 000 catchments (with an average resolution of 1000 km2), delineated to cover the Earth's landmass (except Antarctica). The catchments were characterized using 20 open databases on physiographical variables, to account for spatial and temporal variability of the global freshwater resources, based on exchange with the atmosphere (e.g. precipitation and evapotranspiration) and related budgets in all compartments of the land (e.g. soil, rivers, lakes, glaciers, and floodplains), including water stocks, residence times, and the pathways between various compartments. Global parameter values were estimated using a stepwise approach for groups of parameters regulating specific processes and catchment characteristics in representative gauged catchments. Daily and monthly time series (>10 years) from 5338 gauges of river flow across the globe were used for model evaluation (half for calibration and half for independent validation), resulting in a median monthly KGE of 0.4. However, the World-Wide HYPE (WWH) model shows large variation in model performance, both between geographical domains and between various flow signatures. The model performs best (KGE >0.6) in the eastern USA, Europe, South-East Asia, and Japan, as well as in parts of Russia, Canada, and South America. The model shows overall good potential to capture flow signatures of monthly high flows, spatial variability of high flows, duration of low flows, and constancy of daily flow. Nevertheless, there remains large potential for model improvements, and we suggest both redoing the parameter estimation and reconsidering parts of the model structure for the next WWH version. This first model version clearly indicates challenges in large-scale modelling, usefulness of open data, and current gaps in process understanding. However, we also found that catchment modelling techniques can contribute to advance global hydrological predictions. Setting up a global catchment model has to be a long-term commitment as it demands many iterations; this paper shows a first version, which will be subjected to continuous model refinements in the future. WWH is currently shared with regional/local modellers to appreciate local knowledge.

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

“…The streamflow model errors are assumed to be spatially constant across all watersheds, and thus, the SPARROW model uncertainties are generally overestimated and underestimated in eastern and western regions, respectively. Our observation of major East‐West differences in model prediction errors is generally consistent with that from previous statistical models of annual streamflow in HUC‐2 hydrological CONUS regions (Vogel et al, , ), in which regional model errors were reported to be positively correlated with a dryness/aridity index; similar regional differences in model performance have been previously reported for the CONUS in applications of mechanistic hydrological models (Bock et al, ; Newman et al, ; Arheimer et al, ; Beck et al, ). However, it is unknown whether measurement errors in the observations of mean annual streamflow at the monitoring stations may partially explain regional variations in model uncertainties.…”

“…Applications of physically based models can be especially challenging, leading to overparameterizations that complicate inferences about process effects and contribute to large prediction uncertainties when the calibrated models are transferred to ungauged locations (e.g., Xia et al, ; Beven, ; McDonnell et al, ; Jakeman & Hornberger, ). In contrast, many continental‐scale applications have employed parsimonious mechanistic model structures with fewer controlling parameters (Archfield et al, ), but these simplified models have had mixed success at improving prediction accuracy (e.g., Bock et al, ; Arheimer et al, ). The methods for upscaling and spatial interpolation have included the following: the regionalization of catchment model parameters (e.g., Bock et al, ; Beck et al, ; Livneh & Lettenmaier, ) and measures of hydrological variability (e.g., Jehn et al, ; Addor et al, ) based on geographic proximity and similarities in hydrological and climatic conditions; the simultaneous calibration of models across representative catchments having similar watershed attributes (e.g., Arheimer et al, ); 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, ); and the aggregate use of model outcomes across large scales from independent calibrations in individual watersheds (e.g., Newman et al, ; Weiskel et al, ; Wolock & McCabe, ).…”

“…Despite advances that have contributed to improved spatial sharing of hydrological information across continental scales (e.g., Bock et al, ; Hundecha et al, ; Rakovec et al, ), a persistent challenge over large scales is developing statistical methods for estimating parameters that are spatially and structurally consistent. Many of the approaches, especially in the CONUS, have entailed a more piecemeal sharing of hydrological information across the continental domain and the associated process components of the models (e.g., Bock et al, ; Livneh & Lettenmaier, ; Arheimer et al, ). The imposed spatial restrictions on the areal extent of the model calibrations and information sharing are partially influenced by the computational demands associated with the high temporal resolution of the models (e.g., daily, hourly) that add considerable complexity to the parameter estimation.…”

“…For example, mechanistic models (Wolock & McCabe, ; Xia et al, ; Weiskel et al, ; Bock et al, ; Beck et al, ; Newman et al, ) have typically generated streamflow from uniform landscape components (e.g., “hydrological response units” partitioned on the basis of uniform soils, land cover, and topography), with river networks used primarily to facilitate flow routing without accounting for water loss processes (e.g., evaporative losses, recharge, and water use) or estimating river loss processes using manual “trial‐and‐error” methods (Weiskel et al, ), which may be susceptible to large uncertainties. Models that include river and reservoir evaporative processes (e.g., Arheimer et al, ; Rakovec et al, ) have not consistently estimated the river network parameters simultaneously with catchment and regionalized parameters. Moreover, statistical models typically lack mass balance constraints and represent hydrological processes in a simplistic manner, using nonspatially distributed, averaged properties across watersheds (Vogel et al, , ; Vogel & Sankarasubramanian, ).…”

“…Whereas the focus on undeveloped catchments is necessary to accurately describe natural water cycling processes, an explicit accounting of the effects of anthropogenic activities on streamflow is needed to advance process understanding across large scales and to improve the management utility of hydrologic models. Although land use has been included in large‐scale hydrological models with regionalized parameters in Europe (e.g., Hundecha et al, ; Rakovec et al, ) and in global hydrological models (e.g., Arheimer et al, ), these have been typically characterized using coarser spatial resolutions. Collectively, these limitations have contributed to uncertainties at continental spatial scales in both identifying the causes of streamflow responses and predicting streamflow in more developed watersheds and river basins (e.g., Bock et al, ; Rakovec et al, ).…”