Characterization of soil water content variability at the catchment scale using sensor network and stochastic modelling

Wireless sensor network technology has recently been used for high spatial and temporal resolution soil water content measurements to facilitate better understanding of hydrological processes in catchment scale. Its performance strongly depends on the quality of the sensors and the number of sensor nodes. In the first paper, the newly developed SPADE soil water content sensor was calibrated using a two-step laboratory-based procedure using dielectric reference liquids. The sensor accuracy was evaluated in terms of sensor-to-sensor variability and temperature effect. Using sensor-specific calibration significantly improved the estimation of apparent dielectric permittivity as compared to using a universal calibration function. The transferability of the temperature correction function from reference liquids to soils was successful and has been verified with undisturbed soil samples. A site-specific petrophysical model (complex refraction index model, CRIM) was used to convert apparent dielectric permittivity into soil water content using 15 soil samples from the Rollesbroich catchment, with RMSE values of 0.028, 0.025, and 0.022 cm3cm-3 for 5, 20, and 50 cm, respectively. In the second paper, a two-year time series in-situ soil water content from a wireless sensor network deployed in the Rollesbroich catchment was analyzed in terms of spatial variability using the mean relative difference (MRD) of the soil water content and saturation degree. The MRDs were also used to explore the potential controls of hydraulic properties on the spatial variability of soil water content at the catchment scale. To this end, hydraulic properties were estimated by inverse modeling using the physically-based soil water model Hydrus-1D and the global optimization algorithm SCE. Correlations between van Genuchten-Mualem (VGM) parameters were used as prior information for the parameter optimization. These hydraulic properties were derived from texture information and the Rosetta pedotransfer function. Soil texture was determined from soil samples taken in the Rollesbroich catchment using standard laboratory procedures. The inverse Hydrus-1D model was able to reproduce the observed time series of soil water content at 41 locations and three depths with RMSE smaller than 0.08 cm3cm-3 and R2 larger than 0.75. The MRDs of soil water content and saturation degree were found to be positively correlated with the VGM parameters σθ and n, and to be negatively correlated with the VGM parameters α and KS. In the third paper, a new closed-form expression of soil water variability was developed to explore the relationship between standard deviation (σθ) and mean of soil water content (). The novel closed-form expression is based on the VGM model and uses stochastic theory of 1D unsaturated gravitational water flow in soils. A sensitivity study of the closed-form expression revealed that the n parameter has the strongest effect on the σθ() relationship, followed by the parameters KS, θS, and α. The closed-form expression was used to estimate σθ() using information on percentages of sand, silt, and clay content, and bulk density from datasets of eight test sites with varying soil properties, vegetation, climate conditions and topographies. Six out of eight datasets showed good agreement between observed and predicted σθ() with R2-values ranging between 0.55 and 0.84. Furthermore, The closed-form expression was successfully used to estimate the variability of hydraulic properties from observed σθ() data, with R2-values ranging between 0.69 and 0.88. It is anticipated that an improved understanding of the σθ() pattern provides better insight for an improved upscaling of point-scale information to scales required for climate or hydrological modeling