No abstract

This study investigates the impact of the spatio-temporal accuracy of four different sea surface temperature (SST) datasets on the accuracy of the Weather Research and Forecasting (WRF)-Hydro system to simulate hydrological response during two catastrophic flood events over the Eastern Black Sea (EBS) and the Mediterranean (MED) regions of Turkey. Three time-variant and high spatial resolution external SST products (GHRSST, Medspiration and NCEP-SST) and one coarse-resolution and timeinvariant SST product (ERA5-and GFS-SST for EBS and MED regions, respectively) already embedded in the initial and the boundary conditions datasets of WRF model are used in deriving near-surface atmospheric variables through WRF. After the proper event-based calibration is performed to the WRF-Hydro system using hourly and daily streamflow data in both regions, uncoupled model simulations for independent SST events are conducted to assess the impact of SST-triggered precipitation on simulated extreme runoff. Some localized and temporal differences in the occurrence of the flood events with respect to observations depending on the SST representation are noticeable. SST products represented with higher cross-correlations (GHRSST and Medspiration) revealed significant improvement in flood hydrographs for both regions. The GHRSST dataset shows a substantial improvement in NSE ($70%), RMSE reduction up to 20%, and an increase in correlation from 0.3 to 0.8 with respect to the invariable SST (ERA5) in simulated runoffs over the EBS region.The use of both GHRSST and Medspiration SST data characterized with high spatiotemporal correlation resulted in runoff simulations exactly matching the observed runoff peak of 300 m 3 /s by reducing the overestimation seen in invariable SST (GFS) in the MED region. Improved precipitation simulation skills of the WRF model with the detailed SST representation show that the hydrographs of GHRSST and Medspiration simulations show better performance compared to the simulated hydrographs by observed precipitation.

In this study, the impact of integrating four different sea surface temperatures (SST) datasets on the accuracy of the Weather Research and Forecasting (WRF)-Hydro system to simulate hydrological response during two catastrophic flood events triggered by the changes in SST is investigated. The selected events occurred over Eastern Black Sea (EBS) and Mediterranean (MED) regions of Turkey, where complex geographical characteristics exist and flash flood occurrences are associated with climatic conditions. Three time-varying and high-resolution external SST products (GHRSST, Medspiration, and NCEP-SST) and one coarse-resolution SST product (ECMWF-SST and GFS-SST for EBS and MED regions, respectively) already embedded in the initial and boundary condition dataset of WRF model are used in deriving near-surface weather variables through WRF. Using these meteorological inputs, the flood hydrographs of topographically complex small catchments located over EBS and MED regions are derived by a calibrated WRF-Hydro model coupled one way with WRF 3-km nest domain. After the proper event-based calibration performed to the WRF-Hydro using hourly and daily streamflow data of small catchments in both regions, model simulations for independent SST events are conducted to assess the impact of SST-triggered precipitation on simulated extreme runoff. The calibrated model over both regions revealed significant improvement in flood hydrographs. Some localized and temporal differences in the occurrence of the flood events with respect to observations depending on the SST representation are noticeable. The high-resolution SST dataset cases (Medspiration and GHRSST) show error reduction up to 20% and increase in correlation from 0.3 to 0.8 with respect to the coarse SST in simulated runoffs of the EBS region. The error reduction reached 35% after the calibration. The same high-resolution SST data revealed the exact match with the observed runoff peak after 100 m3/s reductions obtained with calibration in the MED region.

<p>Global numerical weather prediction models (NWP) such as the European Centre for Medium-Range Weather Forecasts (ECMWF) and Global Forecast System (GFS) generate atmospheric data for the entire world. However, these models provide the data at large spatiotemporal resolutions because of computational limitations. Weather Research and Forecasting (WRF) Model is one of the models, which is capable of dynamically downscaling the NWP models’ output. In this study, all combinations of 4 microphysics and 3 cumulus parametrization schemes, 2 planetary boundary layers (PBL), 2 initial and lateral boundary conditions and 2 horizontal grid spacing (i.e., an ensemble consisting of 96 different scenarios) are simulated to measure the sensitivity of WRF-derived precipitation against different model configurations. The sensitivity analyses are performed for 4 separate events. These events are selected among the extreme precipitation events in the Mediterranean (MED) and eastern Black Sea (EBLS) regions. For each region, a summer and an autumn event are chosen. Here, the fundamental aim is to determine the spatiotemporal differences in WRF input parameters that yield better outcomes. A total of 72 hours simulations are started 24 hours before the event day to avoid spin-up time error. The model is adjusted to produce hourly precipitation outputs. The relative performance of scenarios is measured using Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) method considering 5 categorical validation indices and 4 pairwise statistics calculated between the model estimations and the ground-based precipitation observations. According to the TOPSIS results, microphysics scheme, initial and lateral boundary condition, and horizontal grid spacing are substantially influential on WRF precipitation estimates, while cumulus parameterization has a comparatively low effect. The choice of PBL scheme is essential for the summer events, but the results of the autumn events are independent of PBL selection. WRF products are better for the events of the EBLS basin when ERA5 is used as the initial and lateral boundary condition. On the contrary, GFS is superior in the MED region. In terms of spatial resolution, 9 km horizontal grid spacing is commonly preferable for all the events rather than 3 km. Besides, the model underestimates the area-averaged precipitation amounts except for the MED-autumn incident. Still, the model is successful at catching the peak hours of all events. Moreover, the precipitation detection ability of WRF is better for the autumn months. The probability of detection index is higher than 0.5 at 35% of MED stations and 68% of EBLS stations for the autumn events. The local and convective summer events are investigated considering the event centers. Albeit relatively low relationships are defined for the MED-summer event, a statistically significant correlation is obtained between the central station of the EBLS-summer event and the closest grid for the predictions of 52 scenarios (i.e., 54% of the ensemble).</p>

In this study, the impact of spatio-temporal accuracy of four different sea surface temperature (SST) datasets on the accuracy of the Weather Research and Forecasting (WRF)-Hydro system to simulate hydrological response during two catastrophic flood events over Eastern Black Sea (EBS) and Mediterranean (MED) regions of Turkey is investigated. Three time-varying and high spatial resolution external SST products (GHRSST, Medspiration, and NCEP-SST) and one coarse-resolution and invariable SST product (ERA5- and GFS-SST for EBS and MED regions, respectively) already embedded in the initial and boundary condition dataset of WRF model are used in deriving near-surface weather variables through WRF. After the proper event-based calibration performed to the WRF-Hydro using hourly and daily streamflow data of small catchments in both regions, uncoupled model simulations for independent SST events are conducted to assess the impact of SST-triggered precipitation on simulated extreme runoff. Some localized and temporal differences in the occurrence of the flood events with respect to observations depending on the SST representation are noticeable. SST products represented with higher temporal and spatial correlation revealed significant improvement in flood hydrographs for both regions. The higher spatial and temporal correlations of GHRSST dataset show RMSE reduction up to 20% and increase in correlation from 0.3 to 0.8 with respect to the invariable SST (ERA5) in simulated runoffs over the EBS region. The error reduction with GHRSST reached 35% after the calibration of hydrological model parameters compared to not calibrated model. The use of both GHRSST and Medspiration SST data characterized with high spatiotemporal correlation resulted in runoff simulations exactly matching the observed runoff peak of 300 m3/s by reducing the overestimation seen in not calibrated runs over the MED region.