A regional model domain, encompassing the entire PHO, was set up at a spatial resolution of 1 ha. This resolution was a compromise between accounting for the diverse, patchy landscape with small field and orchard sizes (from a few 100 m² to some hectares) and a reasonable computational effort. For the land use information, the database of agricultural fields and orchards was combined with the remaining land uses digitized from satellite imagery. Since CLM5 allows to define fractional land use in a single grid cell, the overall area of individual land use classes was still accurately represented.

The slope of the terrain was derived from the EU-DEM. Furthermore, the surface parameter defining the depth to bedrock was adjusted based on the minimum (0.27 m) and maximum (1.3 m) depths available to roots from the ESDB, which were linearly scaled by the slope. In the plain area, the value was set between 10 and 20 m to represent the thick alluvial deposits and prevailing free drainage conditions. Lastly, the maximum fractional saturated area (f_max) that controls runoff generation was set to zero for all grid cells containing crops due to the deep groundwater table, gentle sloping in the plain, and assuming that there are no large saturated areas in the fields and orchards. f_max was set to 0.16 in the remaining areas of the catchment as extracted from the global dataset. The adjusted parameters for apple were used as described in section 2.3.1 while a separate parameter set was used for cherries to account for the earlier start of the growing season and harvest, and lower productivity as compared to apples. For the sake of consistency, parameters for winter wheat and potato were also modified based on Boas et al. [2021] with minor adjustments to growing seasons to account for the local climate [Dercas et al. , 2022; FAO , 2023].

For the model spin-up, the available global GSWP3 v1 atmospheric forcing data set providing data from 1901 to 2010 at a 3-hourly temporal and 0.5° spatial resolution was used [Lange and Büchner , 2020]. The model was spun-up for 720 years until equilibrium for soil carbon and nitrogen pools, soil water storage, and other ecosystem variables was reached for all land uses in the catchment. For the remaining simulations, the model was forced with a 7-year time series obtained from the observational data of meteorological stations CS1, CS2 (2016-2022), and CS3 (2018-2022) in the study area as well as from the two Atmos41 stations in orchard S09 and S10 (2021-2022) (Figure 1). The data was spatially interpolated to the same resolution as the surface data using inverse distance weighting. The interpolation of precipitation and temperature included a weighting factor for elevation variation using a linear correlation between station elevation and mean annual station precipitation and temperature, respectively, as described in Panagoulia [1995]. Another short spin-up period of 3 years was performed as the orchards had just reached their maximum lifespan before orchard rotation is initiated and new seedlings need a couple of years to reach the full productivity level [Olga Dombrowski et al. , 2022].

To assess how well CLM5-FruitTree can represent soil moisture dynamics and crop growth in the study area, 1D simulations were first performed in orchards S09 and S10 for the growing seasons 2021 and 2022. Two model set-ups were tested: the first used the default CLM5 irrigation routine with adapted parameterization to approximate the observed irrigation schedule, while the second was prescribed with the observed irrigation through the irrigation data stream. By directly applying irrigation water to the ground surface, CLM5 assumes an irrigation efficiency of 100 % which is hardly achieved in sprinkler irrigation [Gilley and Watts , 1977]. For the irrigation data stream, we thus assumed that only 75 % of the water volume measured by the hydrometers is reaching the ground surface while the rest is lost through evaporation from leaf surfaces, transpiration of the grass cover in the orchard alleys, and leakages in the piping system. Modeling results were compared to observed SM and tree transpiration at a daily time step as well as crop yield and development. Pearson’s r (r), the root mean square error (RMSE) and the percent bias (%bias) were calculated for statistical model evaluation.

For the regional case, we conducted three simulation experiments to test different irrigation scenarios. Regional data on irrigation outside the instrumented orchards S09 and S10 was not available. Thus, the model was run using the default CLM5 irrigation routine with the same parameterization that was used for the point scale simulations, in the following considered the full irrigation scenario (FI). Based on this scenario, two deficit irrigation scenarios were created for both apple and cherry orchards with 75 and 50 % of full irrigation (DI75 and DI50, respectively) using the irrigation data stream. All scenarios were run over the same 7-year period (2016-2022). To investigate the differences between irrigation scenarios, multi-year averages and seasonal dynamics of irrigation, SM, crop growth or yield, and crop water use efficiency (CWUE) were calculated and compared. In this study, CWUE was defined as the amount of yield produced per unit volume of water consumed [Ibragimov et al. , 2007]: