The evaluation of simulated discharge of coupled surface scheme and river routing in numerical weather prediction WRF (Case study Karoon river)



Land surface parameterization schemes are one of the most important components in GCM and NWP models. These schemes calculate the exchanges of momentum, mass and energy between land surface and the atmosphere. Runoff is one of the important components in the water cycle of the land surface scheme whose parameterization is difficult because of complexity in the processes governing the runoff generation and its strong dependence on time and space. A coarse resolution land surface model cannot explicitly model the complexities of runoff generation within a catchment; instead, it aims to represent the major processes via subgrid scale parameterizations. A popular solution involves the use of probability density functions to represent subgrid variability.
In this paper, the OSU land surface scheme in version 3 of the Weather Research and Forecasting (WRF) model is studied in which runoff is parameterized as a probability density function of the maximum infiltration. Because the river is being studied it needs to couple the land surface scheme (OSU) with a river routing model, then Total Runoff Integrating Pathway (TRIP) is considered. In this paper, only the treatment of runoff in the model was considered, hence some of the errors in simulations could be the result of deficiencies in other parameterizations.
In this paper, the Karoon River is divided into three subbasins including Farsiat, Harmaleh and Soosan located in the south, west and east of the Karoon respectively by using ARCGIS and ARCHYDRO softwares.
The WRF model was run in a one-way method which needs two domains. The simulations are conducted for December 2005 with 5×5 km grid spacing over an internal domain having 106×115 grid points along altitude and longitude respectively and with 15×15 km grid spacing over a parent domain having 67×69 grid points along altitude and longitude respectively and centered on the Karoon at 50?E and 32?N. The initial and boundary conditions are derived from the GFS data. The modeled discharge (OSU-WRF) in the three subbasins was evaluated with the coupled TRIP river routing and OSU-WRF, using daily observation.
The daily study of the simulated discharge between the coupled model and OSU-WRF, indicates slight difference in all the three subbasins. This slight difference is related to the lag time involved in the calculating surface and ground water storages.
The comparison between the coupled modeled and observed discharge shows that the coupled model generally underestimates total runoff during December 2005 and there were high model bias and Mean Absolute Error (MAE) in all of the three subbasins, especially Farsiat and Harmaleh. This is due to great differences between the monthly mean discharge of the coupled modeled and that observed. Also, the subsurface runoff dominates in most of the studied time and the coupled modeled generally underestimates subsurface runoff. This is related to the poor simulation subsurface soil moisture in the lowest soil layer.
The evaluation of the simulated discharge of the land surface scheme in WRF coupled with the river routing shows negative model efficiency in all of the three subbasins, especially Farsiat and Harmaleh. This means that the model is not successful in the discharge simulation and it cannot even indicate the reality of the stream flow as good as the application of average of the observation. In Soosan subbasin, the simulation of discharge is better than other subbasins because of the higher efficiency and lower model bias and mean absolute error.
On the other hand, the coupled model usually underestimates runoff, though the model overestimates precipitation. This can be related to error in the surface runoff parameterization and so in calculating maximum infiltration.
The study of the correlation coefficient between the simulations and observations shows the correlation coefficient is higher for precipitation (0.66 and 0.88) than runoff (0.50 and 0.54) in Harmaleh and Soosan subbasins. While the correlation coefficient for runoff is relatively high (0.6) in Farsiat, that for precipitation is very small (~0.02). The comparison between normalized standard deviation of rainfall and runoff in all of the three subbasins shows the modeled rainfall has higher variability than observation, especially in Farsiat, but the modeled runoff has lower variability than observations.
The error of the WRF model in the rainfall prediction and the error of the OSU land surface scheme in the rainfall-runoff model or the error which existed in the surface parameters used in the performance of the model, especially parameters related to probably density distributed function of the soil infiltration are effective in the error of the estimated river's discharge.
The comparison between the observed and modeled discharge shows the error in the initial conditions used in this paper, especially initial conditions of surface water and ground water storages, could be another source of the error in the simulated discharge.