Data assimilation of soil water flow by considering multiple uncertainty sources and spatial–temporal features: a field-scale real case study

Accurate estimates of soil moisture and soil hydraulic parameters via data assimilation largely depend on the quality of the model structure, input data and observations. Often, however, all of this information is subject to uncertainty under real circumstances. This real-case study seeks to understand the effects of different uncertainty sources and observation scales on data assimilation performance. Ensemble Kalman filter method based on the soil water-flow model, a sub-module of soil–water–atmosphere–plant model, is established to simultaneously estimate the model states and parameters. The soil hydraulic parameters are extensively measured or calibrated to examine the parameter estimation accuracy. Furthermore, considering the high spatial and temporal variability of soil moisture observation in the field-scale problem, an analysis of spatiotemporal characteristics is combined with data assimilation. Results indicated that simultaneously considering parameter and initial conditions uncertainty leads to a better soil moisture and parameter estimation than that ignoring initial uncertainty in realistic practice. Unlike the other error sources, an inadequate description to the meteorological forcing has a negative influence on surface soil moisture estimation, which might be attributed to the persistent disturbances of evaporation uncertainty and the lack of observations at shallow soil depth. Moreover, a prior knowledge of spatiotemporal features of soil moisture observation is beneficial to efficiently improve data assimilation performance. It is possible to implement field-scale data assimilation with a few representative points, instead of using the spatial average of all observations at a high cost. The assimilation results highlight the possibly positive outcomes of accounting for the multi-source of uncertainties and emphasize the significant importance of characterizing the spatial–temporal feature of soil moisture for a field-scale application.