Modelling soil moisture dynamics of an irrigated agroforestry windbreak system in South Africa

The understanding of soil hydrological processes in agroforestry systems has increased in recent years. However, key aspects determining the successful functionality of agroecosystems (e.g. plant-water availability, nutrient supply) are influenced by many factors and are therefore challenging to generalize. Such information is critical for management and planning strategies. If dominant processes such as evapotranspiration and the related controls on the soil water stock can be represented adequately in hydrological models at the plot scale, they can provide useful insights for practitioners.

Here, we tested whether the physically-based CATFLOW model is capable of reproducing soil moisture dynamics in a South African agroforestry system consisting of windbreaks and irrigated blackberry plants. We initialised the model with matric potential measurements and calibrated it to soil moisture dynamics at two locations with differing vegetation within the test site. After successful calibration, several numerical experiments were performed to shed light on the presence and absence of windbreaks and of different irrigation strategies on the seasonal dynamics of plant available soil water at the test site. The model and observations were also compared in the frequency domain by using empirical mode decomposition as an additional model verification.

The measured soil moisture time series are dominated by a gradual drying of the soil throughout the summer, which is less pronounced in the deeper soil ranges, while several rain events interrupted this general pattern. The model captured the drying as well as the amplitudes of the rain peaks well. However, there was an offset between measured and modelled absolute values due to the initiation based on matric potential. The simulated water balance revealed distinct differences between the windbreak and berry rows due to differences in rain interception, and evapotranspiration or irrigation patterns. Similarities in the frequency spectrum of observed and modelled values were apparent. On the other hand, the modelled time series showed more distinct spectra and less noise. Currently, more detailed analyses are being carried out to extract information from the frequency spectra.