where \(\theta_{wilt,i}\) is the volumetric SM value at wilting point in a given soil layer. \(\theta_{\text{target}}\) and\(\theta_{\text{wilt}}\)
are calculated by inverting the equation for soil matric potential (SMP) (Eq. 7.53 in D Lawrence et al. [2018]) at the respective depth. By default, the SMP parameters \(\psi_{\text{target}}\) and\(\psi_{\text{wilt}}\) are set to -34 and -1500 kPa, considered field capacity and permanent wilting point, respectively.
In addition to \(w_{\text{target}}\), \(w_{\text{wilt}}\),\(f_{\text{thresh}}\), and \(z_{\text{irrig}}\), the user can define the irrigation duration (𝑇𝑖𝑟𝑟𝑖𝑔). Irrigation is applied directly to the ground surface at an intensity equal to\(\frac{D_{\text{irrig}}}{T_{\text{irrig}}}\). Irrigation parameters are not spatially distributed but are defined globally for a given model domain independent of geographic location or crop type.

2.2.4 Irrigation data stream implementation

To study and evaluate the modeling outcomes under specific observed irrigation practices, an irrigation data stream was implemented in CLM5 to enable continuous prescription of irrigation parameters, i.e., irrigation rate, duration, and start time. These parameters are defined separately for one or multiple crop types and for each grid cell. This allows to account for differences in irrigation management depending on crop type and location to accurately reproduce local management practices. In addition, using the data stream, the applied irrigation amount can be easily adjusted, thus creating different irrigation scenarios while maintaining the same irrigation schedule. As irrigation is prescribed, the irrigation SM threshold that is calculated in the standard irrigation routine is not needed for this implementation.

2.3 Model Implementation

2.3.1 Orchard scale simulations

For the simulations of S09 and S10, CLM5-FruitTree was run in single point mode and forced with hourly meteorological data from the two orchards. Fertilizer amount and soil texture were adjusted according to information provided by the farmer and soil samples. The default parameter file was adapted to account for the local climate and orchards characteristics. Crop parameters such as the different phenological stages were adjusted according to observations from the phenocam pictures, harvest information, and communication with the farmer. In the absence of observed bud break dates, parameters for the bud break prediction model were calibrated such that bud break would occur around the estimated date of 15th of March using the available local climate data. The modified crop parameters are listed in Table 3. Additionally, the observed irrigation time series was used as input to the irrigation data stream.
In order to balance ecosystem carbon and nitrogen pools and total water storage in CLM5 [D Lawrence et al. , 2018], a 200 years model spin-up was performed. For this, the CRUNCEPv7 atmospheric forcing data set from 1986 to 2016 [Viovy , 2018] and the parameterized apple plant functional type were used. Using the model state at the end of the spin-up, simulations were then re-initiated from planting in 2013 (S09) and 2015 (S10) using meteorological data from climate station CS1 (2016-2020) and data from the Atmos41 sensors installed in the orchards for the years 2021 and 2022.
Table 3: Local crop parameters for the apple plant functional type.