The Biosphere 2 Landscape Evolution Observatory (LEO) has been developed to investigate hydrological, chemical, biological, and geological processes in a large-scale, controlled infrastructure. The experimental hillslopes at LEO are instrumented with a large number of different sensors that allow detailed monitoring of local and global dynamics and changes in the hydrological state and structure of the landscapes. Sensor failure, i.e., a progressive reduction in the number of active or working sensors, in such an evolving system can have a dramatic impact on observability of flow dynamics and estimation of the model parameters that characterize the soil properties. In this study we assess the retrieval of the spatial distributions of soil water content and saturated hydraulic conductivity under different scenarios of heterogeneity (different values of correlation length of the random field describing the hydraulic conductivity) and a variable number of active sensors. To avoid the influence of model structural errors and measurement bias, the analysis is based on a synthetic representation of the first hydrological experiment at LEO simulated with the physically-based hydrological model CATHY. We assume that the true hydraulic conductivity is a particular random realization of a stochastic field with lognormal distribution and exponential correlation length. During the true run, we collect volumetric water content measurements at an hourly interval. Perturbed observations are then used to estimate the total water storage via linear interpolation and to retrieve the conductivity field via the ensemble Kalman filter technique. The results show that when less than 100 out of 496 total sensors are active, the reconstruction of volumetric water content may introduce large errors in the estimation of total water storage. In contrast, retrieval of the saturated hydraulic conductivity distribution allows the CATHY model to reproduce the integrated hydrological response of LEO for all sensor configurations investigated.