An innovative method for determining hydrological calibration parameters for the WRF-Hydro model in arid regions M. Silver a, * , A. Karnieli a , H. Ginat c , E. Meiri c , E. Fredj b, ** a The Jacob Blaustein Institutes for Desert Research, Ben-Gurion University of the Negev, Sede Boker Campus, 8499000, Israel b Lev Academic Center, Havaad Haleumi 21 St., Givat Mordechai, Jerusalem, Israel c Dead Sea and Arava Science Center, Tamar Regional Council, 86910, Israel article info Article history: Received 6 July 2016 Received in revised form 12 January 2017 Accepted 12 January 2017 Keywords: Hydrology Flood forecasting Calibration Landsat classification GIS abstract The techniques presented herein allow to directly determine certain crucial calibration parameters for the WRF-Hydro flood forecasting model. Typically, calibrations are chosen by an iterative, empirical, trial and error procedure. We suggest a more systematic methodology to arrive at a usable calibration. Our method is based on physical soil properties and does not depend on observed runoff from certain basins during specific storm events. Three specific calibration variables that most strongly affect the runoff predictions are addressed: topographic slope, saturated hydraulic conductivity, and infiltration. We outline a procedure for creating spatially distributed values for each of the three variables. Simulation runs are performed covering several storm events with calculated calibrations, with default values, and with an expert calibration. We show that our calibration, derived solely from soil physical properties, achieves forecast skill better than the default calibration and at least as good as an expert based calibration. © 2017 Elsevier Ltd. All rights reserved. 1. Introduction 1.1. Background Properly calibrated hydrological models can predict runoff rates and intensities with reasonable accuracy (Givati et al., 2012; Pennelly et al., 2014; Yucel et al., 2015)), thus aiding drainage au- thority personnel to prevent damages. Over the past decade, damage due to flooding has gained increasing attention (Foody et al., 2004; Gheith and Sultan, 2002) with the focus on fore- casting as a mean to mitigate those damages. Recently integrated modeling and display frameworks have been reported (Fredj et al., 2015; Akbar et al., 2013). These systems publish forecast and model outputs in a fashion easily understood by laymen, on publicly accessible web sites. Thus early warning systems, based on cali- brated forecasting models are being adopted to save lives and property damage. However the calibration process usually continues for years, with many iterations of parameters being applied to certain storm events in certain basins, and with many simulation reruns to find the optimal calibration values. The current research attempts to substantially shorten the calibration process and improve the spatial resolution for three calibration parameters that most influence the runoff forecasts from the WRF-Hydro model: slope, hydraulic conductivity and infiltration. Unlike earlier calibration procedures, we attempt to determine spatially distributed parameter values using physical soil and terrain properties, without requiring repeated trial and error attempts and without focusing on flood events in a particular basin. We employ several Geographic Information System (GIS) ana- lyses to arrive at highly detailed datasets for the three parameters SLOPECAT, REFDK and REFKDF (referring to topographic slope, saturated hydraulic conductivity, and infiltration, respectively) specifically in arid regions. These raster layers, after conversion to NetCDF format, are merged into the Land Surface Model (LSM), a collection of spatial variables such as topography, land cover, al- bedo, soil moisture, all of which take part in solving the hydro- meteorological energy balance equation. Then simulations are run over several storm events, focusing on certain basins, using three calibrations: the WRF-Hydro default, an expert calibration and these calculated calibration values. Resulting forecast outputs * Corresponding author. ** Corresponding author. E-mail addresses: micha@arava.co.il (M. Silver), karnieli@bgu.ac.il (A. Karnieli), hananginat@adssc.org (H. Ginat), eranmeiri@yahoo.com (E. Meiri), fredj@jct.ac.il (E. Fredj). Contents lists available at ScienceDirect Environmental Modelling & Software journal homepage: www.elsevier.com/locate/envsoft http://dx.doi.org/10.1016/j.envsoft.2017.01.010 1364-8152/© 2017 Elsevier Ltd. All rights reserved. Environmental Modelling & Software 91 (2017) 47e69