Speaker
Description
Subsurface flow models are often used to predict states and fluxes in the subsurface. Soil moisture predictions are important for irrigation planning, weather prediction or flood forecasting, while groundwater-level and recharge predictions are needed for water resources management. Integrated models that represent the groundwater system and the unsaturated zone as one system are becoming increasingly popular for these purposes. Due to the lack of knowledge of the hydraulic parameters and subsurface structure, predictions are highly uncertain. Observations can be used to reduce the uncertainty and to improve predictions. When observations are available as continuous time series, sequential data assimilation can be used for this purpose.
Typical observations are point measurements of soil moisture in the unsaturated zone and groundwater-table heights in aquifers. Using integrated models, all available observations can be assimilated with the aim of enhancing predictions for both compartments—a multivariate data‑assimilation approach. For example, point observations of soil moisture often improve predictions of spatially averaged soil moisture at the soil surface, yet near the groundwater table, observations are often not available and predictions are poor. Incorporating groundwater‑table height observations could therefore improve soil‑moisture forecasts at greater depths.
Model errors may cause data assimilation to degrade predictions relative to forecasts that ignore observations. As model errors in the compartments differ, multivariate data assimilation can often lead to deterioration of predictions. The transition zone between the unsaturated zone and aquifer is a domain prone to artefacts, such as unrealistically high fluxes generated by soil moisture updates. Univariate data assimilation has often been found to outperform multivariate data assimilation (for example Zhang et al., 2016).
We examine the potential drawbacks and benefits of cross-compartmental and multivariate data assimilation for a subsurface system comprising unsaturated zone and unconfined aquifer, focusing on predictions of local and spatially averaged variables, such as averaged soil moisture in the root zone, as well as groundwater recharge. We use an integrated unsaturated‑zone–aquifer model and the Ensemble Kalman Filter for data assimilation to address this question. The impact of model errors due to non-resolved structure and the use of bias correction and localization for compensation as well as weakly or strongly coupled data assimilation strategies are discussed. A general finding is that soil moisture predictions benefit from groundwater-table-head observations, whereas groundwater-table predictions can hardly be improved by soil moisture observations. Nevertheless, deteriorations can be mitigated with bias corrections. Updating not only the groundwater states but also the states in the layer immediately above the water table improves groundwater predictions. Also, it is beneficial to acknowledge the layering of soil structure.
Zhang, D., Madsen, H., Ridler, M.E., Kidmose, J., Jensen, K.H. and Refsgaard, J.C. (2016). Multivariate hydrological data assimilation of soil moisture and groundwater head. Hydrology and Earth System Sciences 20(10), 4341-4357.
| Country | Germany |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
