Soil moisture is an essential variable for the exchange of energy, moisture, and substances between the land surface and the atmosphere. Its effects on temperature and precipitation are diverse and complex. However, the schemes used in climate models to simulate soil moisture, also called soil hydrological schemes, are often very simplified due to the origin of climate models from weather models. In climate models, which compute simulations at coarse resolutions of tens or hundreds of kilometers of edge length, many processes can be neglected. However, the resolution of those models is steadily increasing and now generally has 0.22° in the recently published coordinated project of regional climate models called CORDEX-CORE. As a consequence, higher resolved data and more processes have to be simulated. This is even more true with respect to convection-permitting simulations having edge lengths of a few kilometers. With increasing model resolutions, the complexity and differentiation of questions to be answered by the use of climate models increases as well. This is also the case of the BigData@Geo-project, in which framework this thesis was written. The aim of this project is to provide high-resolution climate information for the Bavarian administrative district of Lower Franconia for stakeholders from agriculture, forestry, and viticulture.
Due to these applied and basic requirements and objectives, there is also the need of model development for the regional climate model REMO (version 2015) used in this work. Thus, the main goal of this thesis is to replace the existing singlelayer soil hydrological scheme by a multilayer one. The advantage of multiple simulated soil layers is that the vertical movement of water, thus percolation and capillary rise, can now be simulated. This is done on the basis of soil hydrological parameters, those value is determined by the water retention curve as a function of soil texture and soil moisture. Various parameterizations have been developed for this curve, whereas the one of Clapp-Hornberger and van Genuchten were used herein. Additionally, the soil moisture can now be simulated to a depth of approximately 10 m or the bedrock's depth, respectively. Thus, in contrast to the previous scheme, which depth is limited to the rooting depth, there is the possibility that water is also available below the root zone. Hence, the absolute amount of water in the root zone is increased. Furthermore, the layering allows evaporation from bare soil based only on the water available in the uppermost layer. Another process, that can be reparameterized due to the layering and the data sets explained subsequently, is infiltration.
To use the new scheme, information on soil hydrological parameters, rooting depth, and the depth to bedrock is required. For this purpose, appropriate data sets have to be prepared and implemented into the model. Regarding the rooting depth, three data sets with different depths, definitions, and resolutions were compared. Finally, the rooting depth from the vegetation module iMOVE, coupled with another REMO version, is used since a coupling between iMOVE and the multilayer soil scheme is planned in the future. With this, the rooting depths are consistent. In addition, the underlying resolution of the data is high and maximum rooting depths are considered, which are particularly important for simulating land surface-atmosphere interactions. These advantages were not provided by the other data sets. In the final model version, SoilGrids data are used for the depth to bedrock and grain size distributions. A comparison with other soil data sets was done in a parallel thesis (Ziegler 2022). For SoilGrids, it should be underlined that the grain size distributions enable the use of continuous pedotransfer functions instead of five discrete texture classes for the soil hydrological parameters. This leads to a much better differentiation of the heterogeneous soil.
For this thesis, 19 simulations were calculated for Europe and an extended German region with resolutions of 0.44° and 0.11°, respectively, covering the period of 2000 to 2018. The implementation of the multilayer soil scheme leads to a decrease in root zone soil moisture compared to the singlelayer scheme. Nevertheless, the absolute amount of soil moisture increases by the consideration of soil below the root zone. Related to the individual layers, the soil moisture is thus underestimated, but in the process of model development an improvement can be achieved compared to ERA5. Furthermore, the new scheme results in a reduction of evapotranspiration that occurs across all model development steps and is especially present during summer. When compared to validation data from ERA5 and GLEAM, this is shown to be an improvement that occurs in space as well as bias and distribution.
The same was found for surface runoff. Schemes implemented for this purpose (Philip, Geen-Ampt), which differ from the defaultly used Improved-Arno scheme by taking hydrlogical parameters into account, were able to show a further improvement in lowlands. In mountainous regions, however, the bias increased due to the not included consideration of slopes. Consequently, the final model version uses the Improved-Arno scheme. Temperature initially increases through the original version of the multilayer scheme, resulting in an overestimation instead of the previous underestimation by the singlelayer soil relative to E-OBS. Although the model development leads to a reduction in temperature, this reduction turns out to be too large, so that the temperature bias is ultimately higher than in the singlelayer model version. However, since evapotranspiration has been significantly improved, this error can possibly be attributed to a temperature overtuning.
The analysis of heat events investigating the summers of 2003 and 2018 has shown that the model development leads to an improved simulation of these events compared to the singlelayer scheme. While this is not true for the spatial behavior of the mean temperature, there is a clear improvement of its temporal one. Additionally, the better simulation of daily extreme temperatures, especially its minimum, leads to an increase of the daily temperature range. This is usually underestimated too much by climate models.
The consideration of vertical water fluxes has shown that there is still enormous potential for model development with regard to (soil) hydrological processes. This is especially true for future simulations with convection-permitting resolution. Thus, subgrid information of the soil and the orography should be considered. On the one hand, this serves to represent existing heterogeneities and, on the other hand, can contribute to the improvement of existing processes, as shown by the example of infiltration schemes. Since the simulated drainage increases due to the multilayer soil scheme to the same extent as the surface runoff decreases, the water is subsequently no longer available to the model. Therefore, groundwater should also be considered in the model. A number of studies have found an added value from integrating this variable and related processes. In the medium term, however, coupling to a hydrological model is generally recommended in order to be able to adequately represent the processes relevant in high-resolution simulations. ParFlow or mHM, for example, are suitable for this purpose.
Overall, it can be noted that the multilayer soil scheme provides an added value because variables like evapotranspiration and surface runoff, that are difficult to simulate and subsequently to be bias adjusted in postprocessing, are modeled much better than using the singlelayer scheme. This is also true for extreme temperatures. Both improvements are caused by the soil layering and associated processes. Regarding the data, it can be seen that the rooting depth, the consideration of SoilGrids, and the vertical soil information is are responsible for the further optimization. In addition, the higher information content available by representing the layered soil moisture can also be classified as an added value.