Estimation of Vadose-Zone Hydraulic Parameters by Inverse Modelling and Ensemble Smother: a Comparison on a Dataset Collected at the Margin of the Venice Lagoon

The hydraulic characterization of the vadose zone is fundamental to understand soil moisture dynamics in agricultural fields with heterogeneous soil properties. Laboratory techniques for the estimation of water retention curves are time consuming and results are affected by the sample scale. For this reason, inverse methods that allow the calibration of soil hydraulic parameters from field observations are usually preferred. Among those, conventional inverse modeling techniques based on optimization approaches are the most widely used even if the partial consideration of the uncertainties of state and parameters limits their applicability. Ensemble-based data assimilation has the capability to fully account for and quantify the uncertainties in prior knowledge, observed data, model approximation, and boundary conditions. The ensemble Kalman filter (EnKF) has been recently applied for the estimation of hydraulic parameters in vadose zone hydrology, and it consists of prediction and assimilation repeated at each time step of the observation time series. However, hydraulic parameters are usually considered as time independent. Although its applicability in variably saturated soils is still challenging, the ensemble smother (ES) can be an alternative to EnKF as it allows to predict and assimilate at once the available observations. The objective of this work was to test the ES approach and compare it with Hydrus 1D inverse modeling package. To this end, field data were collected from May to September 2011 at three monitoring sites in a heterogeneous maize field located near the Venice Lagoon, Italy. Volumetric water content was hourly monitored at 0.1, 0.3, 0.5 and 0.7 m. Time-variable boundary conditions were obtained from on-site measurements of depth to the water table and weather conditions. Firstly, a synthetic experiment was set-up to test the capability of ES to calibrate the Van Genuchten-Mualem parameters. Then, the observations dataset was split into two periods for calibration and validation by using Hydrus 1D and ES inverse modelling. Results confirmed that parameter calibrations by Hydrus 1D inverse modeling and ES were both effective in the state prediction although Hydrus 1D inverse modeling is more time-consuming and strongly affected by the initial value of the calibrated parameters.