Precipitation Data Harmonizer: Harmonizing radar, nowcast, and forecast precipitation data for hydrological applications

<p><span lang="en-US">Flash floods often lead to significant damages and human suffering. To mitigate this, hydrological forecasting models provide extended warning time and allow for better preparedness in the affected areas.</span></p> <p><span lang="en-US">Among other data, hydrological modeling highly depends on reliable precipitation input. Typically, for the use case of precipitation based hydrological flood forecasting three data product types appear useful: (i) observed data for the near past until now, (ii) nowcast data for the next few hours, and (iii) forecast data for precipitation amounts in the near future. The German Weather Service (DWD) provides a multitude of different products for all three types covering Germany.</span></p> <p><span lang="en-US">Producing a coherent time series containing data of these three types can be challenging because of different file formats, different temporal and spatial resolutions, and even varying spatial representations (e.g. regular grid versus icosahedron). To facilitate hydrological forecast modeling, we present our open source Python package weatherDataHarmonizer. It overcomes the temporal and spatial differences between the data types and provides a harmonized time series of spatially distributed precipitation.</span></p> <p><span lang="en-US">First, the package contains modules for low-level access of the DWD original binary data for quantitative radar composites. This comprises measured radar data, e.g. RADOLAN RW, and nowcasting products like RADVOR RQ and RADOLAN RV. The modules are generic enough to support other products in this binary format. Beyond RADOLAN binaries, the package provides low-level access to data used at DWD for the regional weather forecast modeling, e.g. Icon-D2 and the ensemble forecast model Icon-D2-EPS in grib2 format. Both low-level access modules offer specific data and metadata classes and include functions to give the correct spatial coordinates.</span></p> <p><span lang="en-US">Second, we added high-level support for the following DWD products: RADOLAN RW, RADVOR RQ, RADOLAN RV, Icon-D2, and Icon-D2-EPS. All these classes comprise methods for reading files, regridding via IDW method, cropping, and exporting to netcdf with data and metadata.</span></p> <p><span lang="en-US">Third, the package involves a weather data class that collects all supported data, harmonizes the temporal resolution, and invokes regridding for the same spatial distribution. It results in a coherent time series of precipitation data from the near past to the maximum forecast time. Users can directly use the harmonized data within Python or rely on the export to netcdf functionality.</span></p> <p><span lang="en-US">For quality assurance and reproducibility purposes, the weatherDataHarmonizer is highly modular and extendable for other products. It further includes unittests and standardized docstrings, which describe packages, classes, methods, and</span><span lang="en-US"> functions.</span></p> <p>The weatherDataHarmonizer is developed and used within the project HoWa-PRO to generate ensemble precipation timeseries for flood early warning in small catchments.</p>