Figure 11: Seasonal pattern of monthly precipitation, irrigation, and fruit biomass averaged over all apple orchards within the PHO and the period 2016-2022, under full irrigation (FI) and the difference for the 75 % and the 50 % deficit irrigation scenarios (DI75 and DI50).

4 Discussion

4.1 Evaluation of the CLM5 irrigation routine

The direct comparison of simulated SM dynamics to observed SM from a dense sensor network in two irrigated orchards gave valuable insights into model performance. Our findings demonstrate that the standard CLM5 irrigation routine lacks the necessary flexibility to represent specific irrigation practices observed in the orchards. Simulated crop growth and transpiration at the orchard scale were not sensitive to the difference in irrigation amount and timing between the two model runs using the standard irrigation routine and the implemented irrigation data stream respectively. However, as differences between the simulated and actual irrigation practices increase, the effects may become more important especially considering runoff generation or sensible and latent heat fluxes that were not analyzed in this study. Similarly, if the irrigation is limited so that the crop experiences some degree of water stress, the timing of irrigation may become more important. This could be further tested by applying different irrigation schedules under various amounts of irrigation using the irrigation data stream.
Prior studies using the irrigation module in CLM were limited to calibrating the target SM or adjusting the irrigation threshold fraction to match gross irrigation requirements reported at the country or regional level, or performed no calibration at all [Felfelani et al. , 2018; Leng et al. , 2015; Leng et al. , 2013;Zhu et al. , 2020]. The model, however, does not currently consider restrictions on irrigation schedule, over irrigation, or irrigation efficiency that significantly affect gross irrigation requirements as our results revealed. The newly implemented irrigation data stream can be used to overcome some of these limitations by prescribing crop and farmer specific irrigation schedules and amounts. This allows investigating the irrigation-induced effects on e.g., crop yield, SM, or carbon and energy fluxes under observed irrigation practices and can help to identify existing model biases by removing one possible source of uncertainty. While the use of the irrigation data stream at larger scale is currently hampered by the limited availability of precise information on irrigation practices in most areas [Felfelani et al. , 2018], it can serve as a valuable tool to investigate the modeled effect of different irrigation schedules and water availability scenarios. This can offer a basis and direction for further developments of the irrigation routine that are necessary for a more realistic representation of irrigation management practices [Yao et al. , 2022].

4.2 Model uncertainties and limitations of this study

4.2.1 Parametric uncertainty

SM dynamics outside the growing season were well reproduced by CLM5, indicating that the model was able to capture infiltration and soil water redistribution in the studied orchards. However, the significant SM bias in S10 suggests structural and parametric uncertainty in the estimation of soil hydraulic properties, probably due to inappropriate pedotransfer functions implemented in CLM5 [X. Han et al. , 2015]. Gao et al. [2021] found that poor performance of CLM5 in reproducing observed root zone soil moisture was mainly due to uncertainty in porosity estimates. In addition, a high content of rock fragments, which is typical of many Mediterranean soils [Nijland et al. , 2010; Poesen and Lavee , 1994; Zalidis et al. , 2002], can strongly influence the SM regime through non-linearity in soil hydraulic conductivity and by reducing the soils’ effective porosity [Angulo‐Jaramillo et al. , 1997]. For this reason, most pedotransfer functions fail to correctly reproduce the hydraulic properties of stony soils [Nasri et al. , 2015], which likely led to biases in simulated SM in S10. Further investigation of the results would be needed to confirm this hypothesis, e.g. data assimilation of observed soil variables could be used to optimize soil hydraulic parameters [Strebel et al. , 2022]. In both orchards, the simulations showed a lower simulated SM dynamic in 50 cm depth, which could be the result of uncertainties in the rooting distribution and thus root water uptake within the soil profile. The current parameterization of the vertical discretization of root fraction results in a rather shallow profile while deeper roots may still contribute to root water uptake in the studied orchards. Shrestha et al. [2018] encountered a similar issue when analyzing root zone SM on a grassland site using CLM3.5 and were able to improve simulated SM dynamics by increasing the root fraction in deeper layers. This may help to improve the simulated SM dynamics at 50 cm on our study sites.
The sensitivity experiments performed using two parameters of the CLM5 irrigation routine (Figure 5) and the results from the irrigation scenarios revealed relatively low sensitivity of crop yield to reduced irrigation (Figure 9). The new plant hydraulics introduced byKennedy et al. [2019] advanced the physical basis for hydraulic stress in the model, but there is large uncertainty in its parameterization and in capturing the relationship of plant water stress and SM deficit for different crops. To better quantify the model performance and find the most suitable parameters for apple orchards, comparison of simulations to observations from stressed and non-stressed crops would be necessary. Additionally, sensitivity analysis of plant hydraulic parameters, which was out of the scope for this paper, could help to better constrain these model parameters.

4.2.2 Crop representation

The PHO catchment is characterized by a diversity of small-scale farm holders resulting in considerable heterogeneity in management practices, which cannot be fully captured by the model. While simulated yield was close to observations during a “good” year for the point-scale simulations, according to Mattas et al. [2019] average Greek apple production in 2016 was only ~23 t ha-1. This suggests a great variability in orchard productivity, apple cultivars, or type of end product (e.g. apples for direct consumption or for juice) which would necessitate the inclusion of additional crop types and management practices in CLM5. In striving for global applicability, CLM5 and other LSMs face constraints in computational resources and often insufficient observational data to parameterize additional crop types, which results in biases in certain regions, while others are more accurately represented [Lombardozzi et al. , 2020]. In our case, the model demonstrated a strong correlation of yield and irrigation with the climatic gradient induced by the topography in the PHO, indicating a high sensitivity to model forcing data. The large simulated differences in yield between orchards in the plain (~50 ton ha-1) vs. the higher altitudes (as low as 16 ton ha-1) may however be exacerbated, as CLM5 employs a single set of parameters for a given crop across diverse geographies and climates. In reality, various cultivars of the same crop type, along with plant physiological adaptations to their environments, can lead to comparable productivity levels despite variations in climatic conditions. This phenomenon is evident in the cultivation of numerous crops, including apples, across climates on a global scale [Sherman and Beckman , 2002]. The issue has been addressed byLombardozzi et al. [2020] who recommended further developments in CLM5 to improve phenological triggers and agricultural management, and to include different cultivars. In the future, the incorporation of additional satellite-derived crop data, advanced parameterizations, or the use of crop calendars to constrain these models may help reduce some of the biases [Pongratz et al. , 2018; Yao et al. , 2022; Z Zhang et al. , 2020].
At the orchard scale, we found discrepancies between observed and simulated SM during the growing season that suggest limitations specific to the current representation of orchards. As CLM5 does not allow intercropping, the actively growing grass cover in the orchard alleys is not included in the CLM5-FruitTree sub-model [Olga Dombrowski et al. , 2022]. Consequently, our simulations do not account for the additional root water uptake and transpiration as well as interception of the irrigation water from the grasses. The former may explain the smaller simulated decline in SM early in the season compared to the observations, while we considered the latter to some extent by assuming a reduced irrigation efficiency. In doing so, we did however neglect the additional ET flux. Yao et al. [2022] developed and tested different irrigation techniques in CLM5 and found an increase in canopy evaporation through increased interception for their implementation of sprinkler irrigation. However, the overall impact on ET and total applied irrigation remained small compared to the control run using the standard CLM5 irrigation. More importantly, accounting for conveyance and application losses would increase the simulated irrigation amount and could lead to more realistic irrigation values [Yao et al. , 2022].
Despite these limitations, and though we could not validate the simulation of crop yield and irrigation requirements in the PHO catchment due to the lack of observational data, the reasonable modeling results at the orchard scale give some confidence in the robustness of the regional simulations.

4.3 Implications for irrigation management

We studied the relationship between crop yield and water use efficiency, and irrigation at the regional scale, as it is determinant for a reasonable allocation of irrigation water according to crop needs. For most part of the PHO, CWUE and yield were little affected when irrigation was reduced to 75 %, suggesting that this scenario lies closer to the optimal irrigation that maximizes yield while minimizing water consumption as opposed to the FI scenario. These results are similar to a study by Li et al. [2018] who used CLM to schedule irrigation in a citrus orchard in Spain which resulted in 24 % less irrigation compared to the farmers’ practices. This could indicate that farmers irrigate too much when water is available and water prices are low [Latinopoulos , 2005]. Simulated apple yield was sensitive to a reduction of 50 % of the applied irrigation water causing up to 30 % decline in yields. The effect, however, varied with different meteorological conditions and soil types within the PHO. At higher altitudes, cooler temperatures and lower incoming radiation rather than water scarcity limited crop growth. Irrigation in these orchards could thus be greatly reduced without negatively affecting yield. Moreover, under the same climatic conditions, orchards growing on soils with a higher percentage of clay (southeastern part of the catchment) could maintain similar yield and CWUE under 50 % reduction in irrigation water because of the greater water holding capacity of the soil. This will make orchards growing on these soils less prone to experience to water stress. The effect of deficit irrigation on fruit growth and yield varied between years and throughout the growing season. Years with high productivity and greater dependence on irrigation (due to low rainfall) showed greater yield loss under deficit irrigation (Figure 10). At the seasonal scale, fruit growth showed the highest reduction in August followed by July and September (Figure 11). This was mainly an effect of higher temperature, little rainfall, and larger leaf area that resulted in high irrigation requirements in this month. Apples, similarly to other crops, show different susceptibility to drought stress depending on their growth stages whereby flowering and fruit set as well as fruit development and maturation are highly susceptible to drought. The latter stage falls within the period July-September where the model showed the largest reduction in fruit growth. While the simulated plant water stress is currently linked to environmental conditions rather than capturing plant physiological differences in this stage of growth, it suggests that under limited water conditions, irrigation should be prioritized during these months to maintain reasonable yields.

4.4 Perspectives for further application and model development

The analysis performed in this study displays the current ability and potential way forward of applying CLM5 for irrigation and water resources management at various scales. Prospectively, future applications and research studies should focus on the improvement of input datasets, crop and irrigation parameterizations, and process representation. Input related improvements include the creation of high-resolution climate and land use information, especially crop types and the extent and type of irrigation. Our results clearly showed how climate and environmental heterogeneity (e.g. topography, landuse, soil properties) can greatly affect total crop water requirements, emphasizing the need for spatially explicit modeling for large-scale applications. Model investigation at the orchard scale revealed the importance of soil and crop-specific parameterization to correctly represent soil moisture and phenology dynamics, and harvest time. Extending simulations to larger scales will thus require further improving soil hydraulic parameterization through improved pedotransfer functions [Vereecken et al. , 2022] or parametrization of soil hydraulic properties through data assimilation approaches [Xujun Han et al. , 2014]. Furthermore, information on crop management and improved differentiation between different crop varieties and cultivars (e.g. different growing seasons and harvest of cherry compared to apple trees) is necessary, as these can result in distinct irrigation seasons and amounts. Concerning irrigation, this could include either crop-specific or spatially explicit values for irrigation parameters that are currently the same for all irrigated crops, hence not reflecting different management strategies or susceptibilities to water stress. Lastly, some processes could be refined or added to represent irrigation requirements more realistically. These include a parameterization of irrigation efficiency, water availability considerations and more flexible irrigation schedules that can be tailored to represent typical field practices. Conducting parallel testing and assessment of future developments covering greater spatial and temporal scales (e.g., in the form of long-term observatories) will be crucial, especially as more accurate irrigation data becomes available.

5 Conclusions

This study assessed the ability of the CLM5-FruitTree sub-model to represent irrigation practices in fruit orchards in a small Mediterranean catchment and explored the effects of different irrigation scenarios on simulated yield and CWUE. The standard CLM5 irrigation routine could not accurately reproduce observed irrigation practices, which motivated the implementation of an irrigation data stream that directly prescribes measured irrigation data. Using this irrigation data stream, observed SM dynamics in the two studied apple orchards were well captured by the model. We did however find some discrepancies between observed and simulated SM, transpiration, and yield that were related to uncertainties in soil hydraulic parameters and limitations in the crop representation, which does, for instance, not account for the active grass cover growing in the alleys.
To examine the potential to improve regional irrigation management using CLM5, we simulated different irrigation scenarios and analyzed their effect on crop yield and CWUE. The model showed distinct effects of deficit irrigation on yield and CWUE for scenarios with 25 % and 50 % reduction in irrigation (DI75 and DI50, respectively) that were tested using the irrigation data stream. While DI75 had negligible negative effect on yield and CWUE, DI50 notably reduced both yield and CWUE. Based on the modeling results, this would suggest substantial water savings of up to 125 mm year-1 with little to no effect on apple yields and up to 250 mm year-1 when accepting up to 30 % reduction in yield (although potential effects of fruit quality need to be considered as well). These effects varied depending on climatic conditions, soil type, and timing of irrigation. Hence, under limited water availability, irrigation should primarily focus on the summer months July to September and on sandy soils with lower water holding capacity.
The outcomes of this study demonstrate the potential use of CLM5 in irrigation and water resources management research and applications. Future research efforts should focus on improving soil and crop parameterizations, and as well as process representation. Finally, we anticipate that implementing more realistic irrigation schedules in land surface models such as CLM5 will allow for better water resource management at the local and regional level.

Acknowledgments

This research received support from the ATLAS project, funded through the EU’s Horizon 2020 research and innovation program, under grant agreement No. 857125. Further support was received from the German Research Foundation (DFG) through the project 357874777, which is part of the research unit FOR 2694 Cosmic Sense as well as through the PhenoRob project as part of Germany’s Excellence Strategy - EXC 2070–390732324. Additional support was received from TERENO (Terrestrial Environmental Observatories) of the Helmholtz Association of National Research Centers (HGF), Germany. Furthermore, this work was in part supported by the National Center for Atmospheric Research (NCAR), which is a major facility sponsored by the NSF under Cooperative Agreement 1852977. Simulations were conducted using computing resources from the supercomputers JURECA and JUWELS from Jülich Supercomputing Centre (JSC).
The authors would like to thank Nikos and Stamatis Oikonomou for granting access and use of their orchards, providing information, and supporting instrument maintenance. We further thank Bernd Schilling and Ansgar Weuthen for instrument preparation and monitoring of the data transmission, the team at SWRI including Vassilios Pisinaras, Andreas Panagopoulos, Anna Chatzi, and Konstantinos Babakos for their great support in instrument installation and maintenance, and the members of the Soil, Water and Plant Tissue Analysis Laboratory of SWRI for conducting the soil analysis.

Open Research

The CLM5-FruitTree sub-model used in this work is freely available via Zenodo at https://doi.org/10.5281/zenodo.8154390 [O. Dombrowski , 2022] and on Github at https://github.com/odombro/CTSM.git under the branch release-clm5.0_FruitTree. The irrigation data stream implementation is available on Zenodo at https://doi.org/10.5281/zenodo.8290143 [Swenson , 2023] and on Github at https://github.com/swensosc/ctsm.git under the branch irrigation_streams. Climate data from climate stations CS1, CS2, and CS3 from the TERENO sites Agia (TERENO ID: AGIA_K_001, AGIA_K_002, AGIA_CK_003) are freely available via the TERENO data portal [TERENO , 2023]. Data collected from the two apple orchards S09 and S10 is available via the TEODOOR database at https://doi.tereno.net/landingpage/doi/10.34731/e1ss-pc69 [O. Dombrowski and Bogena , 2023]. Data analysis was performed in Python version 3.10.4 [PythonSoftwareFoundation , 2023] available at https://www.python.org/downloads/ and figures were made with Matplotlib version 3.5.2 [Caswell et al. , 2022], available under the Matplotlib license at https://matplotlib.org/. The map overview was created with QGIS version 3.12.3 [Dawson et al. , 2022] available at https://qgis.org/.