Grid-based calibration of the WRF-Hydro with Noah-MP model with improved groundwater and transpiration process equations

Abstract

The physically-based distributed WRF-Hydro modelling system, including the Noah land surface model with multiple parameterization options (Noah-MP) and the hydrological extension of the WRF atmospheric model (Weather Research and Forecasting model), has recently been widely used for water balance investigations, streamflow and coupled land–atmosphere simulations. Despite the multiple available physical parameterizations in the model, equations for simulating particular losses from the water balance are missing, and a grid-based calibration of distributed parameters across multiple watersheds has not been studied. To fill these gaps, this study aims: (i) to analyze the impact of soil, runoff, groundwater and vegetation parameters on water balance components; (ii) to improve baseflow and transpiration equations; and (iii) to test a grid-based calibration approach for distributed model parameters, using streamflow observations. The WRF-Hydro groundwater model was improved through the introduction of a groundwater loss factor and the Jarvis stomatal conductance model was modified to account for nocturnal transpiration. The grid-based calibration was performed for three parameters (infiltration, hydraulic conductivity and percolation) for 19 spatially-distributed classes, with the Parameter Estimation (PEST) software. The study area includes 31 small mountainous watersheds (5–115 km2) in Cyprus, in the Eastern Mediterranean. A two-year period (2011–2013) was used for calibration and a five-year period (2013–2018) for the evaluation. The baseline model set-up overestimated streamflow, on average, by 50 % in 2011–2012 and more than 100 % in 2012–2013. Overall, streamflow and evapotranspiration (ET) could vary by about ±30 % from the baseline simulation, using different model parameters and model options. The simulation of groundwater losses as a function of groundwater level reduced total streamflow, on average, by 30 %. The use of the proposed Jarvis equation for nocturnal transpiration increased the total ET, on average, by 25 %. The grid-based approach facilitated the calibration of the distributed parameters over the area of the 31 watersheds. The median Nash-Sutcliff Efficiency (NSE) was 0.49 during calibration, but 0.02 in the drier evaluation period. The calibrated WRF-Hydro model reproduced the annual variability of ET and the improved groundwater and transpiration equations reduced the substantial streamflow overestimation of WRF-Hydro. The model performance during dry years demonstrated the need for representation of more processes that occur in semi-arid environments with ephemeral streams and are not included in WRF-Hydro and Noah-MP. The grid-based WRF-Hydro parameterization can be applied to the full study area for fully-coupled atmospheric-hydrologic simulations.