Creating a Near Real-Time 1-km Hourly Mean Temperature Gridded Dataset for Operational Flood Forecasting in Germany

Hydrological models often require gridded atmospheric fields, yet such datasets, particularly in high-resolution and near real-time, are often unavailable. Al- though precipitation is the most dominant variable in hydrological processes, temperature can influence river flow by influencing snowmelt, leading to snowmelt floods. Daily temperature data are often insufficient to capture floods, highlight- ing the importance of hourly temporal resolution. Currently, there is a lack of reliable real-time, hourly gridded temperature data for Germany. The DWD provides historical hourly temperature that is part of the HOSTRADA product [1]. However, the data are updated monthly by including the data from the pre- vious month. Near real-time hourly station records for temperature are freely available from the DWD. We assume that the average hourly temperature has smooth spatial and temporal distributions, facilitating a reliable interpolation with only 512 active stations.

This study investigates the interpolation of hourly station temperature data to generate a high-resolution historical and near-real-time gridded temperature dataset. The methods explored include ordinary kriging (OK) and external drift kriging (EDK) with topographical elevation as the drift variable. Various variogram models were considered for both methods. Cross-validation [2] was used to select the best interpolation method and determine an optimal distance for interpolation. The performance of the interpolated dataset, particularly EDK with an exponential variogram, was also validated against HOSTRADA from 1995 to 2023. The comparison yielded a root mean square error of 0.8◦C, demonstrating the robustness of the method. Based on this evaluation, a near real-time gridded temperature dataset is generated to serve as input to the hourly configuration of mHM for an operational flood prediction within HI- CAM II project.

In practice, the interpolation was performed using EDK software developed first by Samaniego et al. [3] which is implemented in Fortran and supports parallel computation. Its robustness and efficiency make it well-suited for pro- cessing and interpolating station data in large domains and in operational set- tings, ensuring timely and reliable outputs for hydrological application such as operational flood impact-based forecasting.

 

References

• [1]  S Kr ̈ahenmann, A Walter, S Brienen, F Imbery, and A Matzarakis. High- resolution grids of hourly meteorological variables for germany. Theoretical and Applied Climatology, 131:899–926, 2018.

• [2]  Steffen Zacharias, Heye Bogena, Luis Samaniego, Matthias Mauder, Roland Fuß, Thomas Pu ̈tz, Mark Frenzel, Mike Schwank, Cornelia Baessler, Klaus Butterbach-Bahl, et al. A network of terrestrial environmental observatories in germany. Vadose zone journal, 10(3):955–973, 2011.

• [3]  Luis Samaniego, Rohini Kumar, and Conrad Jackisch. Predictions in a data-sparse region using a regionalized grid-based hydrologic model driven by remotely sensed data. Hydrology Research, 42(5):338–355, 2011.