H41I-01 INVITED
New methods for assimilating remotely sensed observations of snow covered area into land surface models
Snow cover is the fastest varying land surface feature on Earth, and it has a dramatic impact on atmospheric processes and hydrology at the local, regional, and global scales. Snow is also an important memory component of the climate system, as seasonal establishment of the snow pack has a strong influence on further snow accumulation, while seasonal melt can impact soil moisture for months into the warm season. For these reasons it is essential that land surface models (LSMs) provide an accurate representation of snow cover for forecast initialization and retrospective analysis. Data assimilation provides a tool for updating snow within an LSM, but assimilation algorithms that utilize observations of snow covered area (SCA) are complicated by (1) an information deficit between the retrieved variable and those simulated by the LSM, and (2) by the potential to inadvertently disturb other water storage variables during the assimilation update. The results of two assimilation algorithms are presented. In the first, remotely sensed snow cover observations are introduced to the Noah LSM using a rule-based function that converts SCA to snow water equivalent (SWE). In the second, observations are introduced to the model one day in advance of the observation time. These advance observations are used to adjust the atmospheric forcing fields within the LSM, pulling the model into agreement with satellite data. In global simulations, both algorithms improved the LSM's simulation of total snow covered area, as evaluated against independent datasets, and both provided some improvement in the simulation of SWE. The algorithms differed in their impact on surface energy fluxes and the local hydrologic budget, and the implications of these differences will be discussed.
H41I-02
The contribution of soil moisture retrievals to land data assimilation products
Satellite retrievals of surface soil moisture are subject to errors and cannot provide complete space-time coverage. Data assimilation systems merge the available satellite retrievals with information from land surface models that is spatio-temporally complete but likewise uncertain. For the design of new satellite missions it is critical to understand just how uncertain satellite retrievals can be and still be useful. Here, we present a synthetic data assimilation experiment that determines the contribution of satellite retrievals to the skill of land assimilation products as a function of retrieval and land model skill. We quantified how the skill of the assimilation products increases with the skill of the model and that of the retrievals. The skill of the surface and root zone soil moisture assimilation products always exceeds that of the model, that is, even retrievals of low quality contribute some information to the assimilation product, particularly if model skill is modest.
H41I-03
Estimating rootzone soil moisture by assimilating both microwave based surface soil moisture and thermal based soil moisture proxy observations
Remote sensing can be used to retrieve soil moisture through either microwave- or thermal-based satellite observations. Microwave satellite data can be used to retrieve top 5-cm soil moisture (surface soil moisture), whereas thermal observations are indirectly related to root-zone soil moisture through the thermal response of the vegetation canopy to soil water stress. Microwave-derived soil moisture data does not depend on weather conditions and can provide almost daily coverage. However, the resolution is relatively low (>10 km). In contrast, thermal satellite data has high resolution (100 m to 1 km), but low temporal frequency, since retrieval is not possible in the presence of clouds. The assimilation of microwave-derived surface soil moisture into land surface models has been an active area of research for nearly a decade. Likewise, a number of past studies have inputted thermal-based land surface temperature retrievals into a data assimilation system. However, relatively little work has focused on the simultaneous assimilation of both observations into a land surface model. In this paper, a number of synthetic data assimilation experiments are carried out at the USDA Economic and Environmental Enhancement (OPE3) site in Beltsville, Maryland. As a first case, only surface soil moisture retrievals are assimilated into a land surface model using the Ensemble Kalman filter (EnKF). This case is compared to dual EnKF assimilation results where both surface and root-zone soil moisture observations (derived from thermal remote sensing observations) are simultaneously assimilated. The ability of the dual assimilation case to enhance results (relative to the single surface soil moisture assimilation case) is examined for a range of root-zone soil moisture accuracies and frequencies to determine how valuable root-zone moisture retrievals - degraded to realistic levels of frequency and accuracy - are for the accurate constraint of land surface model predictions. Problems associated with the low temporal frequency of thermal remote sensing data, the accurate specification of modeling uncertainties within the EnKF, and the nonlinearity of the land surface model with respect to soil moisture will be discussed. Preliminary results for the assimilation of real observations at the OPE3 site will also be presented.
H41I-04 INVITED
Dealing with systematic errors in land surface modeling, soil moisture observations and assimilation
A fundamental assumption of the Kalman filter is that both the observations and model predictions are unbiased. Bias in observations typically reflects instrumental inaccuracies, representativeness errors, or, in the case of remote sensing observations, errors in the retrieval algorithm. This bias is typically (at best) removed prior to assimilation. Land surface models are usually biased in at least a subset of the simulated variables even after calibration. On-line forecast bias estimation may therefore be needed for data assimilation. Here, in situ soil moisture observations in a small agricultural field (OPE3) were merged with Community Land Model (CLM2.0) simulations using different algorithms for state and bias estimation with and without bias correction feedback. The different bias correction schemes were tested to study the impact of the state correction on depending model fluxes. The best variant for state and bias estimation depends on the nature of the model bias: an improper bias correction scheme could distort the water balance. The lack of knowledge of the bias `dynamics' in time and space and the approximation of the bias uncertainty structure limit successful bias estimation and correction to directly observed state variables. However, all assimilation schemes including bias correction algorithms yield far improved state analysis results compared to standard state filter analyses.
H41I-05
Modeling uncertainty and correlation in soil properties using Restricted Pairing and implications for ensemble-based hillslope-scale soil moisture and temperature estimation
Soil hydraulic and thermal properties (SHTPs) affect both the rate of moisture redistribution in the soil column and the volumetric soil water capacity. Adequately constraining these properties through field and lab analysis to parameterize spatially-distributed hydrology models is often prohibitively expensive. Because SHTPs vary significantly at small spatial scales individual soil samples are also only reliably indicative of local conditions, and these properties remain a significant source of uncertainty in soil moisture and temperature estimation. In ensemble-based soil moisture data assimilation, uncertainty in the model-produced prior estimate due to associated uncertainty in SHTPs must be taken into account to avoid under-dispersive ensembles. To treat SHTP uncertainty for purposes of supplying inputs to a distributed watershed model we use the restricted pairing (RP) algorithm, an extension of Latin Hypercube (LH) sampling. The RP algorithm generates an arbitrary number of SHTP combinations by sampling the appropriate marginal distributions of the individual soil properties using the LH approach, while imposing a target rank correlation among the properties. A previously-published meta- database of 1309 soils representing 12 textural classes is used to fit appropriate marginal distributions to the properties and compute the target rank correlation structure, conditioned on soil texture. Given categorical soil textures, our implementation of the RP algorithm generates an arbitrarily-sized ensemble of realizations of the SHTPs required as input to the TIN-based Realtime Integrated Basin Simulator with vegetation dynamics (tRIBS+VEGGIE) distributed parameter ecohydrology model. Soil moisture ensembles simulated with RP- generated SHTPs exhibit less variance than ensembles simulated with SHTPs generated by a scheme that neglects correlation among properties. Neglecting correlation among SHTPs can lead to physically unrealistic combinations of parameters that exhibit implausible hydrologic behavior when input to the tRIBS+VEGGIE model.
H41I-06
Fractal Interpolation and Monte Carlo Simulations to Address Aliasing in Well Hydrographs
Calibrating transient models to time-series observations is a challenge in many aspects of hydrology due to incomplete or inadequate time-series data. For example, when calibrating groundwater models to well hydrographs researchers must account for a time scales ranging from minutes to years. In such cases, aliasing is an inherent problem because it is often impractical to measure water levels in wells for extended periods of time at a rate greater than the highest-frequency stress on the aquifer. (Aliasing refers to spurious low-frequency components of a signal introduced by an inadequate sampling rate.) To address the aliasing problem for well hydrographs it is necessary to interpolate water level data so that sufficiently high frequency components are present in the calibration signal. This can be achieved using Monte Carlo simulations in which stochastic hydrographs are generated that (1) match the data points on the observed (low frequency) hydrograph, and (2) have a fractal dimension consistent with that of the observed time series. We tested the inversion method by generating synthetic recharge signals to generate synthetic water-table hydrographs. Because the method works well with the synthetic data, we have applied these methods to water- table hydrographs measured on Hatteras Island, North Carolina.
H41I-07
Using Ensemble Kalman Filter to Simulate Groundwater Flow and Solute Transport in Heterogeneous Media with Unknown Contamination Sources
Hydraulic conductivity distribution and plume initial source are two important factors to affect solute transport in a naturally heterogeneous medium. Due to current economic and technologic limitations, hydraulic conductivity can only be measured at limited locations in a field. Therefore, its spatial distribution in a complex heterogeneous medium is generally uncertain. In many groundwater contamination sites, solute initial conditions are generally unknown. The plume distributions are available only at sometimes after the contaminations occurred. The uncertain spatial distribution of the hydraulic conductivity field and plume of the initial condition will lead to uncertain predictions to groundwater flow and solute transport in subsurface. In this study, a data assimilation method is developed for calibrating a hydraulic conductivity field and improving solute transport prediction with unknown initial condition. Ensemble Kalman filter (EnKF) is used to update the model parameter, hydraulic conductivity, and model variables, hydraulic head and solute concentration, when data are available. Two- dimensional numerical experiments are designed to assess the performance of the EnKF method on data assimilation for solute transport prediction. The study results indicate that the EnKF method will significantly improve the estimation of the hydraulic conductivity distribution and solute transport prediction by assimilating hydraulic head measurements with a known solute initial condition. When solute source is unknown, solute prediction by assimilating continuous measurements of solute concentration at a few points in the plume will well capture the plume evolution process in downstream.
H41I-08
Combined assimilation of soil moisture and streamflow data by an ensemble Kalman filter in a coupled model of surface–subsurface flow.
Hydrologic models can largely benefit from the use of data assimilation algorithms, which allow to update the modeled system state incorporating in the solution of the model itself information coming from experimental measurements of various quantities, as soon as the data become available. In this context, data assimilation seems to be well fit for coupled surface--subsurface models, which, considering the watershed as the ensemble of surface and subsurface domains, allow a more accurate description of the hydrological processes at the catchment scale, where soil moisture largely influences the partitioning of rain between runoff and infiltration and thus controls the flow at the outlet. The need for a better determination of the variables of interest (streamflow at the outlet section, water table, soil water content, etc.) has led to a many efforts focused on the development of coupled numerical models, together with field and laboratory observations. Nevertheless, uncertainty in the schematic description of physical processes and inaccuracies on source data collection induce errors in the model predictions. The ensemble Kalman filter (EnKF) represents an extension to nonlinear problems of the classic Kalman filter by means of a Monte Carlo approach. A sequential assimilation procedure based on EnKF is developed and integrated in a process-based numerical model, which couples a three-dimensional finite element Richards equation solver for variably saturated porous media and a finite difference diffusion wave approximation based on a digital elevation data for surface water dynamics. A detailed analysis of the data assimilation algorithm behavior within the coupled model has been carried out on a synthetic 1D test case in order to verify the correct implementation and derive a series of fundamental parameters, such as the minimum ensemble size that can ensure a sufficient accuracy in the statistical estimates. The assimilation frequency, as well as the effects induced by assimilation on the surface and/or subsurface system state, was tested on a 3D synthetic test case represented by a 1.62 km2 tilted v-catchment, for which observations of pressure head and streamflow data are assimilated in order to retrieve the true watershed state in 2 scenarios: i) starting from a drier initial condition and ii) intentionally imposing a biased atmospheric forcing. In general, streamflow prediction is improved by assimilation of both pressure head and streamflow individually and by coupled assimilation. However, assimilation of streamflow data only does not improve the subsurface system state, leading to a deficit in soil moisture compared to both the true and the open loop simulations. Combined assimilation is therefore more adequate for the description of the entire surface—subsurface system state. The sensitivity analysis to the assimilation frequency yields contradictory results: as expected, a higher assimilation frequency improves the true state retrieval in the drier initial condition scenario, while for the biased atmospheric forcing scenario an analogous improvement is not manifest.