Constraining groundwater recharge estimates at the field scale using soil hydrologic modelling and measured root zone soil moisture: how to deal with the vertical sensitivity of cosmic-ray measurements

Root-zone soil moisture is decisive in partitioning the water fluxes at the land surface, e.g. between evapotranspiration and groundwater recharge. Field-scale estimates of these time variant recharge rates can be derived from 1D soil hydrologic models, if the soil moisture product represents the respective scales and dynamics, and lateral fluxes can be neglected. Measured soil moisture time series can be used to calibrate these models by optimizing soil hydrologic properties (SHPs) and in this way increase confidence in simulated downward flux from a soil column (potential groundwater recharge). To obtain indirectly and non-invasively measure soil moisture at the field scale, Cosmic-ray neutron sensing (CRNS) has gained increasing attention in the last years. However, the variable penetration depth of the sensor and its decreasing sensitivity with depth and distance from the sensor complicate the interpretation of the soil moisture product and limit direct comparison to simulated soil moisture.

Within this study a two-layered Hydrus-1D model (up to 1.5 m depth) has been set up at an agricultural field site for one cropping season and calibrated using different soil moisture products to derive potential groundwater recharge estimates. While the use of point soil moisture sensor network data (SN) in the optimization is straightforward, different options to use CRNS data are evaluated: i) the COSMIC operator (simulates neutron count rates) ii) weighting simulated soil moisture according to CRNS vertical sensitivity, iii) applying a soil moisture profile correction on measured CRNS soil moisture before comparison.

Optimizing the SHPs did result in very good model performance for the SN as well as for the CRNS options (KGE > 0.86). While the SN delivers information down to a depth of 90 cm, using CRNS data that considers the vertical sensitivity (option i) and ii)) can result in difficulties informing the bottom layer of the model, which shows in optimized SHPs hitting the previously determined parameter bounds. Compared to that, using CRNS option (iii) leads to slightly reduced performance measures in the optimization but better informs the SHPs of the bottom layer when averaging modeled soil moisture over a fixed depth. For the successful optimizations, regardless of the method, recharge rates vary little and are comparable to independently estimated water flux at the field site.

Results of the study confirm the ability of the profile correction to increase CRNS information content to the main rooting zone and the validity of assuming a fixed integration depth, although this is expected to vary between field sites. This encourages also the use of CRNS soil moisture for non-experts of the method for soil hydrologic and landscape models as well as water balance calculations targeting downward fluxes.