Evaluation of Regionalization Methods for Hourly Continuous Streamflow Simulation Using Distributed Models in Boreal Catchments Teklu T. Hailegeorgis 1 ; Yisak S. Abdella 2 ; Knut Alfredsen 3 ; and Sjur Kolberg 4 Abstract: Regionalization for prediction in ungauged basins at hourly resolution is important for water resources management (e.g., floods and hydropeaking). In the research reported in this paper, calibration of 26 catchments (39–3,090 km 2 ) in mid-Norway was performed using hourly records and three spatially distributed (1 × 1 km 2 ) precipitation–runoff models, as follows: (1) first-order nonlinear system model, (2) Hydrologiska Byråns Vattenballansavdelning (HBV) model, and (3) basic grid model. Four regionalization methods for each model [(1) parameter set yielding maximum regional weighted average performance measures (PMs), (2) regional median of optimal parameters, (3) nearest neighbor (NN), and (4) physical similarity (PS)] were evaluated and compared with three benchmarks. Parameter transfer from best regional donor and from an ideal best arbitrary single-donor, and local calibration (LC) were as benchmarks. The PS attributes include hypsometric curves, land use, drainage density, catchment area, terrain slope, bedrock geology, soil type, and combination of all. Compre- hensive evaluation of single-donors and multidonors, simple benchmarks, and more advanced regionalization methods using multimodels, two PMs, and their statistical evaluation indicate that the identification of regionalization methods is dependent on the models, the PM, and their statistical evaluation. In general, the hypsometric curves, land use, and best regional donor methods performed better for the Nash–Sutcliffe efficiency based on boxplots and regional median values of both the Nash–Sutcliffe efficiency and relative deterioration or improvement of the Nash–Sutcliffe efficiency from the LC due to the regionalization. The methods also performed better for the individual catchments. The terrain slope, regional median of optimal parameters, maximum regional weighted average, and best regional donor methods performed better for the natural-logarithm-transformed streamflow (i.e., regarding the Nash–Sutcliffe efficiency) based on the same evalu- ation criteria. Similar performance to the more advanced regionalization methods of transfer of homogeneous parameter sets across the whole region from the best regional donor for both Nash–Sutcliffe efficiency and natural-logarithm-transformed streamflow (i.e., regarding the Nash–Sutcliffe efficiency) indicate the potential of the simple regionalization approach for predicting runoff response in the region. DOI: 10.1061/(ASCE)HE.1943-5584.0001218. © 2015 American Society of Civil Engineers. Introduction Continuous time precipitation–runoff modeling has been used to represent the hydrological processes and understand the basin re- sponse. The modeling task entails parameter calibration procedures for gauged basins. However, model identification is not always straightforward due to uncertainties in input data (e.g., climate forcing and streamflow), model structure, and potential nonidenti- fiability of some model. Moreover, there are further challenges for prediction in ungauged basins (PUBs; Sivapalan et al. 2003) through transfer of information from the calibrated gauged basins to ungauged sites. The uncertainty in precipitation measurments due to the inability of the existing gauges to properly capture the spatial variability of precipitation is a major source of data uncertainty that affects parameter calibration and model prediction. Application of spatially distributed hydrological modeling has been encouraged due to the availability of spatial data and their potential to simulate streamflow at interior catchments. Distributed model efficiency seems to depend on rainfall and model spatial resolution (Pechlivanidis et al. 2011). When the model can capture the spatial information content of precipitation (e.g., McIntyre and Al-Qurashi 2009), effective parameters that are calibrated for a catchment based on spatially distributed inputs, and computations of fluxes and states may have the potential for better performance when transferred to interior locations within the catchment over a cali- bration based on the lumped counterpart. Hence, spatially distrib- uted modeling and parameter regionalization are important for the PUBs. Regionalization methods have previously been used to transfer knowledge from gauged to ungauged basins (Blöschl and Sivapalan 1995; Oudin et al. 2010), which require evaluation and identification of the methods. Regionalization Methods Several methods for parameter regionalization have been reported. Parajka et al. (2013) categorized parameter regionalization methods into five groups. The first method is based on regional calibration by utilizing data from multisites in the region. Fernandez et al. (2000) implemented a regression-model-based regional calibra- tion and concluded that improved regional relationships between 1 Researcher, Dept. of Hydraulic and Environmental Engineering, Norwegian Univ. of Science and Technology (NTNU), NO-7491 Trondheim, Norway (corresponding author). E-mail: tekhi09@gmail.com 2 Researcher, Dept. of Energy Systems (Water Resources), SINTEF Energi AS, Sem Sælands vei 11, NO-7465 Trondheim, Norway. 3 Professor, Dept. of Hydraulic and Environmental Engineering, Norwegian Univ. of Science and Technology (NTNU), NO-7491 Trondheim, Norway. 4 Researcher, Dept. of Energy Systems (Water Resources), SINTEF Energi AS, Sem Sælands vei 11, NO-7465 Trondheim, Norway. Note. This manuscript was submitted on July 11, 2014; approved on February 25, 2015; published online on April 13, 2015. Discussion period open until September 13, 2015; separate discussions must be submitted for individual papers. This paper is part of the Journal of Hydrologic Engi- neering, © ASCE, ISSN 1084-0699/04015028(20)/$25.00. © ASCE 04015028-1 J. Hydrol. Eng. J. Hydrol. Eng. Downloaded from ascelibrary.org by NTNU on 04/19/15. Copyright ASCE. For personal use only; all rights reserved.