Testing a new channel routing component in JGrass-NewAge model

The paper presents two applications of the JGrass-NewAge model in order to investigate the influence of an explicit channel routing model on the discharge simulation. The semidistributed, component based hydrological model JGrass-NewAge is based on the Object Modeling System version 3.0 (OMS3). OMS3, which facilitates exchange of model components, was used to set up two different model configurations named Hymod and RHymod, respectively. The two models differ only by the new channel routing component. Different basin delineations (one, three and twenty Hydrological Response Units (HRU)) are analyzed for both the model configurations. Simulated discharges in all the cases are compared with measurements from a quantitative point of view by using classical indices of goodness of fit such as index of agreement, percentage bias and Kling-Gupta efficiency. 1. NewAge-JGrass model The model used in the applications presented in this paper is the NewAge-JGrass system (Formetta et al. (2011)): a system for hydrological cycle simulation at the basin scale. It includes different components dealing with estimation of different hydrological processes such as the space-time structure of precipitation, evapotranspiration, runoff production, aggregation and propagation of flows in channel, and automatic calibration of each model with different methods. The system is based on a hillslope-link geometrical partition of the landscape, so the basic unit for the water budget evaluation is the hillslope. Each hillslope drains into a single associated link rather than cells or pixels. This conceptual partition is developed using informatics with vector features for channels and raster data for hillslopes. Each model is a component, according to the definitions of the OMS3 (http://javaforge.com/wiki/57375; David et al. (2010) and can be substituted easily by others at run time without rewriting the whole code. The model requires interpolation of meteorological variables (air temperature, precipitation, relative humidity) as input data for each hillslope. It can be handled by a deterministic (Inverse distance weighted (Cressie (1992), Goovaerts (1997), Lloyd (2005))), geostatistic (Goovaerts (1997)) or detrended Kriging (Garen et al. (1994) and Garen and Marks (2005)) approach. As a result, time series for required meteorological variables are generated for each hillslope. The energy model, Formetta and Rigon (2011), includes both shortwave and longwave radiation components calculations for each hillslope. The shortwave radiation balance (beam and diffuse components) is described in Iqbal (1983), Bird and Hulstrom (1981) and Corripio (2002). The latter implements algorithms that take into account shade and complex topography. Shortwave radiation under generic sky conditions (all-sky) is computed Formetta, David and Rigon 28 according to Helbig et al. (2010) and using different parameterizations choices such as Erbs et al. (1982), Reindl et al. (1990) and Orgill and Hollands (1977). The longwave radiation budget is based on Brutsaert (1982) and Brutsaert (2005). After computing the net radiation for each hillslope, evapotraspiration can be modelled using three different solutions: the Fao-Evapotraspiration model (Allen et al. (1998)), the Penman-Monteith model (Penman (1948); Monteith et al. (1965)) the Priestley-Taylor model (Priestley (1959), Slatyer and McIlroy (1961), Priestley and Taylor (1972))). The user can choose between two different runoff generation models: Duffy's model (Duffy (1996)) and Hymod model (Moore (1985); Boyle (2001)). In both cases the model is applied for each hillslope. Finally, the discharge generated at each hillslope is routed to each associated stream link according to Mantilla and Gupta (2005) and Mandapaka et al. (2009)). All modelling components can be calibrated using one of the calibration algorithms such as Particle Swarm Optimization algorithm (Kennedy and Eberhart (1995), Eberhart and Shi (2001)) and DREAM (Vrugt et al. (2009)). Every component can be connected, parameterized, and executed either using the OMS3 console (OMS 3.1) or the OMS3 scripting mode within the uDig Spatial Toolbox (http://code.google.com/p/jgrasstools/). Different components can be instantiated, initialized and connected in a sequence. In this way the modeler can build a custom hydrological model and solution by selecting different components to simulate the same hydrological processes. Processes will then use the OMS3 implicit parallelism to improve the computational efficiency in multicore or multiprocessor machines. The complete application of the system is presented in Formetta et al. (2011). 2. Test different modeling solutions. As presented in Formetta et al. (2011), the Hymod component (Moore (1985) and Boyle (2001))) is applied for each HRU into which the basin is split. The rationale of using several Hymods, one for each hillslope, instead of a single one for the whole catchment as is usual in literature, was twofold: firstly, to preserve the geometrical and topological structure of the river network, which provide to embed significant information about the shape of discharge hydrograph, (D'Odorico and Rigon (2003)); and secondly, to allow the use, as input, of spatially varying rainfall and evapotranspiration fields. Finally, the runoff production is then propagated in the channel network. In this paper, a new runoff propagation component presented. To investigate the role and possibly the importance of the channel routing component a testing is performed. Two river basins are used for test and modeled in a three different delineations by using one (DL1), three (DL3) and twenty (DL20) HRU's. Two modeling solutions are setup: Hymod and RHymod Illustrated in fig. (1). The modeling solution RHymod includes: Pristley-Taylor component for the evapotranspiration estimate, ordinary kriging algorithm for the rainfall spatialization, hymod model for the runoff production of the hillslope and finally the new channel routing component presented in the next section. The modeling solution Hymod differs from the model solution RHymod just because the channel routing component is turned off and the discharge for each HRU is simply summed going downstream. LUCA (Hay et al. (2006)) algorithm was selected for calibrating component for both the modeling solutions. The obTesting a new channel routing component in JGrass-NewAge model 29 jective function is the Kling-Gupta efficiency (KGE) function presented in Gupta et al., (2009). Figure 1. Modelling solutions: Hymod (in red dashed line) and RHymod (in blue dashed line). 3.1. The flow routing component. As presented in Formetta et al. (2011) the flow generated for each hillslope is kinematically propagated downstream in the channel network by integrating, in each channel link, a non-linear variant of the Saint Venant equation (e.g. Bras and Rodriguez-Iturbe (1994)). For each link, the continuity equation is in fact: dSi (t) dt = Qgen (t)+ Qtrib(t)