Does vadose zone flow forecasting depend on the type of calibration data?

Abstract

Unsaturated subsurface water flow is often described by a flow model which is calibrated on either observed soil water content or tensiometric pressure head measurements. For a given model structure the calibration on one data type may lead to significant errors in predictions of the other data type. These errors are difficult to quantify since simultaneous measurements of pressure head and water content are generally not available. Independent vadose zone data of both types were recorded at an intensively investigated experimental field site in the Lake Taupo catchment, New Zealand. A numerical flow model was set up and calibrated (i) using tensiometric pressure head observations, (ii) using soil water content (TDR) observations, and (iii) using both tensiometric and TDR data. The global multi-method search algorithm AMALGAM was used to estimate five soil hydraulic parameters in five model layers, totaling 25 optimized parameters. In the cases (i) and (ii), a single aggregated objective function was defined to fit measurements from four different depths in the vadose zone profile. The third model calibration was placed in a multi-objective context to include the two different data types simultaneously. The trade-off pattern between the fit to the water content and pressure head observations was investigated. Parameter sets from the three calibrations were then used for predicting pressure heads and water content in the vadose zone for independent data, not previously used in the calibration process. The results suggest that predictions of tensiometric pressure head and volumetric water content significantly depend on the type of data used for model calibration. Large differences in the model predictions occur when calibrating to one data type and predicting the other. This demonstrates the need to inform the model about the required prediction data type in the calibration process. This is a prerequisite to make reliable forecasts of vadose zone water flow and to determine realistic uncertainty bounds in vadose zone flow modeling.