H32A-01 INVITED
Interannual variations in terrestrial water storage from basin to continental scales
In this study we assess spatial and temporal variations in terrestrial water storage using the latest releases of the GRACE data. In particular, we focus on short-term variations for each of the continents, and in selected major river basins within them. We interpret our results in the context of observed hydroclimatology and known anthropogenic influences. Results have implications for using GRACE to monitor hydrologic change in response to changing climate.
H32A-02 INVITED
An adaptive Ensemble Kalman filter for soil moisture data assimilation
Accurate estimates of model and observation error parameters are key ingredients of a data assimilation system. To date, two main approaches for obtaining model error parameters have been used in soil moisture data assimilation. The first approach derives model error parameters by comparing the open loop trajectory to validating measurements (from field or synthetic data) outside of the cycling data assimilation system. The second approach is based on repeating the entire data assimilation experiment many times over with differents sets of model error parameters, that is, the model error parameters that produce the best validation of assimilation estimates with the cycling assimilation system are selected by enumeration. We demonstrate in a fraternal twin experiment for the Red-Arkansas river basin that the first approach yields poor assimilation estimates because the error parameters are not determined within the cycling assimilation system. While theoretically correct, the second approach of enumeration is computationally not feasible for large systems. We demonstrate that a computationally affordable, adaptive assimilation system provides improved assimilation estimates. In the adaptive assimilation system, the model error parameters are continually adjusted in cycling assimilation mode in response to the innovation information provided by the observation minus forecast misfits.
H32A-03 INVITED
Enhancement of Satellite Observations of the Land Surface Via Data Assimilation
Satellite based observation of land surface conditions has transformed the field of hydrology, enabling water cycle studies which span local to global scales. However, such studies are limited by errors in the retrieval algorithms, data gaps, and sometimes low resolutions. Land surface models (LSMs) simulate the redistribution of water and energy incident on the land surface based on our understanding of physical processes, but with limited accuracy. The advantages of each can be harnessed by data assimilation. Here we present examples of how data assimilation can be used for spatial, temporal, and vertical interpolation and downscaling of observations from GRACE, MODIS, and other sources, thus enhancing their value for water resources applications and hydroclimatological investigations.
H32A-04
An Integrated System for Sequential Hydrologic Data Assimilation using the Land Information System
The Land Information System (LIS; http:lis.gsfc.nasa.gov) is a hydrologic modeling system that integrates various community land surface models, ground and satellite-based observations, and high performance computing and data management tools to enable assessment and prediction of hydrologic conditions at various spatial and temporal scales. Recently, the LIS framework has been enhanced by developing an interoperable extension for sequential data assimilation, thereby providing a comprehensive framework that can integrate data assimilation techniques, hydrologic models, observations, and the required computing infrastructure. The extensible LIS data assimilation framework allows the incorporation and interplay of multiple observational sources, multiple data assimilation algorithms, and multiple land surface models. These capabilities are demonstrated using a suite of observing system simulation experiments (OSSEs) that assimilate different sources of observational data into different land surface models to propagate observational information in space and time using assimilation algorithms with varying complexity ranging from rule-based approaches to ensemble Kalman Filtering (EnKF). The assimilation of soil moisture, snow cover, and snow water equivalent data is demonstrated using the Noah and Catchment land surface models using a number of sequential assimilation algorithms. These experiments illustrate the sensitivity of model parameterizations and physical representations on the efficiency of the assimilation process and the relative merits of the assimilation approaches. Further, the system also provides an infrastructure to diagnose the consequences of assumptions on model and observation error properties on the accuracy of assimilated products. These experiments demonstrate the use of LIS data assimilation framework as an ideal testbed for development and evaluation of techniques in hydrologic data assimilation.
H32A-05
A Bayesian approach to snow water equivalent reconstruction
For nearly three decades, remotely sensed observations of snow cover depletion have been used to forecast seasonal snowmelt runoff and (indirectly) seasonal snow water equivalent (SWE) accumulation. The use of these snow covered area (SCA) data to reconstruct snow accumulation is based on the simple concept that deeper snow takes more time (or energy) to melt than shallower snow. Traditional reconstruction methodologies do not extract all of the available information from the remote sensing measurements, however. Indeed, snow cover states show considerable temporal autocorrelation due to the fact that accumulation and ablation take place over several months during each winter. Furthermore, there is not a convenient method for taking advantage of the spatial correlations in the remote sensing measurements, which stymies exploitation of mutual information in neighboring pixels. Finally, the uncertainty in the remote sensing observations and snowmelt models cannot be treated in a rigorous way within the current schemes. These uncertainties are time-dependent, and vary with geophysical factors such as elevation and forest cover. To address the drawbacks of traditional reconstruction methods, we demonstrate a Bayesian approach to estimating the spatial distribution of snow accumulation by using the Ensemble Kalman Smoother (EnKS) to combine a land surface model (LSM) with remote sensing SCA observations in a synthetic test. Specifically, the weights used to interpolate gage-based precipitation measurements to the model pixel resolution are estimated using the EnKS. All synthetic SCA measurements are applied in a single end-of-winter single analysis step, which capitalizes on the significant SWE autocorrelation time, allowing for the full exploitation of the time series of SCA measurements. The SWE at each pixel is updated based on the SCA estimates at all pixels within the assumed correlation length, which allows for the exploitation of the mutual information of neighboring pixels. The uncertainty of the various mass and energy products and model parameterizations are modeled explicitly. The potential of this Bayesian approach to SWE reconstruction is explored by probing the sensitivity of the estimates to the uncertainty of the various inputs.
H32A-06
A One-Dimensional Data Assimilation Experiment Using 3D Eddy Covariance Heat Flux Observations to Improve Land Surface Modelling
Land surface models such as the CSIRO Biosphere Model are often coupled with weather and climate forecast models to provide a continuous feedback of latent and sensible heat flux values as the lower boundary condition for weather and climate forecasting. However, these flux estimates are typically poor due to approximations of complex physical processes and errors in model forcing data and parameters. Hence, the technique of data assimilation is commonly applied to improve latent and sensible heat flux prediction, with research focussing on the assimilation of soil moisture measurements. However, variables such as soil moisture typically share a weak or uncertain relationship with the latent and sensible heat fluxes in land surface models. This is often exacerbated by a lack of soil and vegetation property data required to accurately parameterise the models. The assimilation of latent and sensible heat flux observations to improve land surface model predictions of latent and sensible heat fluxes and associated soil moisture and temperature states has received very little attention in the scientific community thus far. In this study, data assimilation was performed using 3D eddy correlation measurements of latent and sensible heat flux, together with meteorological forcing data from south eastern Australia. An Ensemble Kalman Filter assimilation algorithm was applied and results validated against 3D eddy flux data and measured soil moisture and temperature profiles to determine the impact on model estimates of fluxes and states and whether assimilating latent and sensible heat fluxes is an approvement over soil moisture assimilation.
H32A-07
Incorporating remotely sensed cloud and atmospheric thermodynamic data into a microphysically based precipitation model
In this work we formulate a precipitation model driven by remotely sensed cloud microphysical parameters and atmospheric thermodynamic structure. The primary objective in developing the model is to retain a simple structure capable of easily generating ensemble fields of precipitation at relatively high spatial and temporal resolution. The motivation for doing so is to ultimately use the model in a data assimilation framework. The precipitation model is based on a one-dimensional conceptualization of an atmospheric column that derives the liquid mass balance of a cloud layer and surface rainfall rate from thermodynamic principles and state-of-the-art and readily available satellite information. Specifically, cloud microphysical parameters obtained from the VISST/SIST algorithm include cloud top and base pressure, liquid and ice water content, and characteristic hydrometeor size, and are used to estimate precipitation leaving the cloud base. Profiles of atmospheric temperature and humidity obtained from the AIRS sensor aboard the AQUA satellite are used in estimating rainfall at ground level after accounting for evaporation and updraft in the subcloud layer. Combination of the aforementioned data within the proposed model yields high-resolution (4 x 4 km, half hourly) precipitation fields. Uncertainty in the precipitation fields are simulated using postulated a priori probability density functions for the relatively few model input parameters. The computed ensemble of precipitation fields can subsequently serve as a prior estimate for a data assimilation procedure that incorporates information from the variety of products used while reflecting the appropriate uncertainty in the estimates. This approach can be coupled with a recent application with insolation fields, yielding an ensemble of physically consistent forcing fields for hydrologic models conditioned on the wealth of available multi-scale data products.
H32A-08
An algorithm for estimation of evapotranspiration for all sky conditions
An algorithm is developed for estimation of actual evapotranspiration (ET) for all sky conditions using primarily remote sensing data. Our approach is based on an extension of the Priestley-Taylor equation, a contextual interpretation of remotely sensed surface temperature and vegetation index, and estimation of net radiation from remotely sensed data. For cloud free days, net radiation is computed using MODIS-Terra data and a simple sinusoidal model to obtain the diurnal variation of radiation. If a particular day is cloudy for a MODIS-Terra pixel, we use data from MODIS-Aqua which has a different orbital pass time. If a pixel is cloudy for Terra and Aqua overpass times, then a set of regression relationships are used to estimate net radiation and surface temperature. For cloudy days, the algorithm first computes the shortwave radiation using a regression model between clear sky shortwave radiation and cloud fraction. Then, net radiation is computed using a linear regression model between estimated shortwave radiation for cloudy days. Surface temperature is estimated using regression model between calculated net radiation and cloud cover. Once the surface temperature is estimated, a contextual interpretation of surface temperature and vegetation index is used to obtain ET. Results from calibration and validation of the proposed ET estimation algorithm for all sky conditions over the Southern Great Plains will be presented.