Coupling soil moisture and precipitation observations for predicting hourly runoff at small catchment scale Gokmen Tayfur a,⇑ , Graziano Zucco b , Luca Brocca b , Tommaso Moramarco b a Dept. Civil Engineering, Izmir Institute of Technology, Urla, Izmir, Turkey b Research Institute for Geo-Hydrological Protection, CNR, Perugia, Italy article info Article history: Received 28 June 2013 Received in revised form 11 December 2013 Accepted 26 December 2013 Available online 5 January 2014 This manuscript was handled by Peter K. Kitanidis, Editor-in-Chief, with the assistance of Markus Tuller, Associate Editor Keywords: GRNN Soil moisture Rainfall Runoff Prediction Experimental basins summary The importance of soil moisture is recognized in rainfall–runoff processes. This study quantitatively investigates the use of soil moisture measured at 10, 20, and 40 cm soil depths along with rainfall in pre- dicting runoff. For this purpose, two small sub-catchments of Tiber River Basin, in Italy, were instru- mented during periods of October 2002–March 2003 and January–April 2004. Colorso Basin is about 13 km 2 and Niccone basin 137 km 2 . Rainfall plus soil moisture at 10, 20, and 40 cm formed the input vec- tor while the discharge was the target output in the model of generalized regression neural network (GRNN). The model for each basin was calibrated and tested using October 2002–March 2003 data. The calibrated and tested GRNN was then employed to predict runoff for each basin for the period of Jan- uary–April 2004. The model performance was found to be satisfactory with determination coefficient, R 2 , equal to 0.87 and Nash–Sutcliffe efficiency, NS, equal to 0.86 in the validation phase for both catchments. The investigation of effects of soil moisture on runoff prediction revealed that the addition of soil mois- ture data, along with rainfall, tremendously improves the performance of the model. The sensitivity anal- ysis indicated that the use of soil moisture data at different depths allows to preserve the memory of the system thus having a similar effect of employing the past values of rainfall, but with improved GRNN performance. Ó 2014 Elsevier B.V. All rights reserved. 1. Introduction The importance of soil moisture on runoff, infiltration, and evapotranspiration is well recognized in the literature (Goodrich et al., 1994; Merz and Plate, 1997; Scipal et al., 2008; Brocca et al., 2009a, among others). Specific monitoring programs (Merz and Bardossy, 1998; Aubert et al., 2003; Castillo et al., 2003; Brocca et al., 2009a; Matgen et al., 2012; Morbidelli et al., 2012) and mod- elling studies (Gautam et al., 2000; Anctil et al., 2004; Komma et al., 2008; Berthet et al., 2009; Sheikh et al., 2009; Brocca et al., 2012; Tramblay et al., 2012; Van Steenbergen and Willems, 2013) were carried out to investigate the influence of soil moisture on producing runoff hydrographs. Several studies investigated the benefit of using soil moisture observations within rainfall–runoff models. First applications con- sidered soil moisture data for the improvement of the calibration and verification of rainfall–runoff models (Wooldridge et al., 2003; Koren et al., 2008; Parajka et al., 2009). Other studies directly used the observations for the assessment of the antecedent wetness conditions through in situ (Meyles et al., 2003; Huang et al., 2007; Tramblay et al., 2010; Zehe et al., 2010) and remotely sensed (Jacobs et al., 2003; Brocca et al., 2009b; Beck et al., 2009) estimates. For instance, Goodrich et al. (1994) monitored two sub-catchments of 0.044 km 2 and 6.31 km 2 size in Walnut Gulch experimental watershed of the U.S. Department of Agriculture. They pointed out that a basin-wide remotely sensed average initial soil moisture can be sufficient for rainfall–runoff modelling in semiarid regions provided that the spatial–temporal variability of rainfall is accurately known. Brocca et al. (2009a), at five nested catchments (13–137 km 2 ) in central Italy, found that the integra- tion of in situ soil moisture observations into a simple event-based rainfall–runoff model improved the prediction of runoff hydro- graphs with respect to the use of antecedent precipitation and baseflow indices. Grayson and Western (1998) suggested that a network of a limited number of soil moisture sensors can provide reliable estimates of areal mean soil moisture time series that can potentially be used as antecedent conditions data. All these studies suggest that in situ soil moisture data from a small number of locations can provide useful information for improving runoff prediction at the basin scale. Several studies employed artificial intelligence techniques (or data driven techniques) for modelling the rainfall–runoff transfor- mation process (see Elshorbagy et al., 2010 for a recent review of 0022-1694/$ - see front matter Ó 2014 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jhydrol.2013.12.045 ⇑ Corresponding author. Tel.: +90 533 5565339; fax: +90 232 750 6801. E-mail addresses: gokmentayfur@iyte.edu.tr (G. Tayfur), graziano.zucco@irpi.cn- r.it (G. Zucco), luca.brocca@irpi.cnr.it (L. Brocca), t.moramarco@irpi.cnr.it (T. Moramarco). Journal of Hydrology 510 (2014) 363–371 Contents lists available at ScienceDirect Journal of Hydrology journal homepage: www.elsevier.com/locate/jhydrol