H31H-0754
A Global Soil Moisture Data Assimilation System for the USDA-FAS Crop Assessment Data Retrieval and Evaluation (CADRE) Decision Support System
An Ensemble Kalman Filter (EnKF) system has been designed to integrate soil moisture retrievals from the EOS Advanced Microwave Scanning Radiometer (AMSR-E) into the U.S. Department of Agriculture (USDA) Crop Assessment Data Retrieval and Evaluation (CADRE) Decision Support System (DSS). The operational soil moisture model currently used by CADRE is forced by daily meteorological observations (precipitation and temperature) provided by the Air Force Weather Agency (AFWA) and World Meteorological Organization (WMO). The improved coverage and temporal resolution of AMSR-E soil moisture retrievals upon the AFWA and WMO data is envisaged to provide a better characterization of surface wetness conditions particularly where the AFWA and WMO data are sparse. This study evaluates the added value of the AMSR-E soil moisture data assimilation over the conterminous United States. The experimental methodology is based on designating a single model realization forced with reliable precipitation as truth. The EnKF is then applied to assimilate AMSR-E soil moisture estimates into the model runs forced by an error-prone precipitation dataset. Effectiveness of the soil moisture data assimilation system is evaluated by comparing the EnKF output soil moisture with the truth soil moisture. System design and model uncertainly is evaluated by comparisons with in-situ observations, and analyzing the filter divergence, and innovation statistics. Applying global AMSR-E soil moisture retrievals to the soil moisture data assimilation system for the CADRE decision support system of USDA-FAS will be discussed.
H31H-0755
An Integrated Hierarchical Approach to Hydrologic Data Assimilation and Uncertainty Analysis
The issue of hydrologic uncertainty needs to be adequately addressed if hydrologic modeling is to realize its maximum practical potential in environmental decision-making. The key to properly addressing hydrologic uncertainty is to understand, quantify, and reduce the uncertainties involved in hydrologic modeling in a cohesive and systematic manner. Although general principles and techniques for addressing hydrologic uncertainty are being discussed in the literature, there exist no well-accepted guidelines about how to actually implement such principles and methods in various hydrologic settings in an integrated manner. This presentation discusses the various types of hydrologic uncertainty from a systems perspective, and presents an integrated hierarchical Bayesian framework for pursuing hydrologic data assimilation - using a progression of several steps to maximally reduce the uncertainty in hydrologic predictions (including system identification, parameter estimation, state estimation, and final output prediction).
H31H-0756
Correcting Unintended Perturbation Biases in Hydrologic Data Assimilation Using the Ensemble Kalman Filter
Recent advances in hydrologic data assimilation have demonstrated the value of remotely sensed surface soil moisture in improving forecasts of key hydrologic variables such as root-zone soil moisture and surface runoff. In hydrologic data assimilation, the ensemble Kalman filter (EnKF) provides a robust framework to optimally update model predictions using observations based on the uncertainties of the model and observations. In the EnKF, model uncertainty is obtained using a Monte Carlo approach where the states, parameters, or forcing data for the model are perturbed to create an ensemble of states, and the covariance of the ensemble members represents the model uncertainty. However, although adding mean-zero perturbations to the model should not affect the mean performance of the underlying model, it often causes systematic model state biases due to nonlinear land surface model physics and the bounded nature of key land surface states (e.g. soil moisture). Here we show an example of systematic biases caused by land surface model perturbations, present a diagnosis for these biases, and propose a simple method to reduce them. The Noah land surface model within the Land Information System (LIS) is applied to a medium-scale basin in the United States Southern Great Plains for the demonstration. Our proposed method removes virtually all the bias caused by the mean-zero ensemble perturbations and significantly improves the performance of the EnKF. Compared to non-corrected EnKF results, combining the bias-correction scheme with EnKF updating reduces forecast errors in evapotranspiration by 51%, and surface runoff by 74%. Additional effects of the nonlinear model physics and bounded states on model performance will be discussed and alternative methods to optimally summarize the nonlinear diagnostic variables will also be presented.
H31H-0757
The development of adaptive filtering techniques for land surface models
Due to the complexity of potential error sources in land surface models, the accurate specification of model error parameters has emerged as a major challenge in the development of effective land data assimilation systems for hydrologic and hydro-climatic applications. While several on-line procedures for estimating model error parameters - based on the statistical analysis of filtering innovations - have been introduced for geophysical models, such procedures have not been widely applied to land surface models. Consequently, little is currently known about their relative merits with regards to land surface data assimilation applications. Using a series of synthetic twin experiments and an Ensemble Kalman filter, this paper will inter-compare the performance of a number of existing adaptive filtering approaches when applied to an observation and modeling error estimation problem within a land data assimilation system. These comparisons will highlight the suitability of classical adaptive filtering approaches (designed for purely linear systems) for nonlinear land surface models. Special emphasis will also be placed on identifying the suitability of various approaches with regards to the unique characteristics of the land surface data assimilation problem (relative to similar problems in ocean and atmosphere modeling). These attributes include the high degree of land surface spatial heterogeneity which precludes the use of ergodic techniques for sampling innovation statistics and the nonlinear, yet fundamentally dissipative, structure of land surface processes. Preliminary real data results based on the assimilation of remotely-sensed surface soil moisture retrievals into a land surface model forced by satellite-based precipitation will be used to demonstrate the potential value of these approaches in an operational setting.
H31H-0758
Uncertainty Analysis in Environmental Modeling: Towards Improved Evaluation and Diagnostics of Hydrological and Environmental Models (2nd International IAHS-PUB Workshop 15th-18th July 2007 in Bertinoro, Italy)
This paper reports back to the community some of the major findings from a recent discussion-based workshop to explore state-of-the-art uncertainty analysis techniques and their value for diagnostic model analysis. The participants of the previous 'uncertainty in environmental modelling' workshop in Italy in 2004 concluded that there is an urgent need for new and better diagnostic approaches to evaluate environmental models and to contribute to the shift from model calibration to understanding - the main focus of Prediction in Ungauged Basins (PUB) and of its working group on 'Uncertainty analyses and model diagnostics' who organized this workshop. Current, largely performance-based approaches tend to lack the power to extract information from model applications that helps to diagnose where models fail and how models could be improved. The focus of the recent second workshop was, therefore, 'diagnostic evaluation of hydrological and environmental models' and dealt with four specific science questions in the context of diagnostic model evaluation: [1] How should we measure model performance? [2] How can we separate and identify observation error versus model error? [3] What is the information content of hydrologic data? [4] How can we analyze temporal parameter variability to guide hydrological and environmental model diagnostics and improvement? We report here the major conclusions of the workshop and of the subsequent review papers prepared by the participants, with the main objective to further engage with the research community and to identify opportunities for future collaborations and developments within the PUB uncertainty working group.
H31H-0759
Variational Estimation of Energy Balance Partitioning and Soil Heat Diffusion Using Remotely Sensed Land Surface Temperature Data
Estimates of two major parameters of surface energy balance that control the partitioning of available energy into sensible, latent, and ground heat fluxes are made based on a variational data assimilation (VDA) approach. This method minimizes the estimated soil surface temperature against observations. Sequences of radiometric surface temperature measurements are the only input data source. The one dimensional parabolic heat diffusion equation is used as a physical constraint (the adjoint method). The land data assimilation scheme is formulated such that it does not need ancillary data such as soil texture and vegetation. The two key unknown parameters for estimating fluxes are: near-surface air turbulent conductivity (bulk heat transfer coefficient) and evaporative fraction (ratio of latent heat flux to the sum of the sensible and latent heat fluxes), which is almost constant for near-peak radiation hours. The inclusion of the multi-layer heat diffusion model reduces the phase errors associated with ground heat flux. As a result the new model performs better than the other VDA methods which use parsimonious force-restore equation in place of heat diffusion equation.
H31H-0760
Estimating Point Concentrations of Total Organic Carbon in Water Sediments by Geostatistical Downscaling
Attribute data such as contaminant concentrations in water sediments are typically obtained in core sections of varying lengths, and only the average concentration of each section is measured. Estimating the attribute distribution at a uniform support (i.e. spatial resolution) is often needed to characterize the site and for the design of appropriate risk-based remediation alternatives. Because attributes exhibit spatial autocorrelation, geostatistical methods have become an essential tool for estimating the spatial distribution of attributes based on limited sampling. The purpose of this work is to infer fine resolution concentrations from average concentrations using downscaling, formulated as a geostatistical inverse problem. Taking sediment total organic carbon (TOC) concentration observations as an example, we compare inverse modeling to the more traditional ordinary kriging. Traditional kriging methods are not able to estimate the point concentration using the average concentrations accurately, because these approaches are designed for data with uniform support. Geostatistical inverse modeling, on the other hand, can resolve this problem by accounting for the relationship between the known average concentrations and the unknown point concentrations to be estimated. The Restricted Maximum Likelihood (RML) approach is used to estimate the spatial covariance of the concentration distribution at finer resolutions. Results from both pseudodata and field data show that, in general, inverse modeling is better able to estimate the concentration distribution for data with variable support. Pseudodata examples confirm that the estimates of both covariance parameters and point concentrations from inverse modeling are closer to the true situation, relative to estimates obtained from ordinary kriging. Field data from a ten-kilometer stretch of the Passaic River were also used to validate the proposed approach. Consistent with our initial hypothesis, inverse modeling is better able to represent small scale variability, while honoring the average concentrations measured at larger scales.
H31H-0761
An Uncertainty Framework For Downward Modeling Of Watershed Hydrology
We present an uncertainty-guided downward approach to model the hydrologic behavior of watersheds. A series of watershed models with increasing complexity is applied to interpret signatures of inter-annual, intra-annual and daily streamflow response behavior for 12 US watersheds across a range of hydro-climatic conditions. Model performance for each of these signatures is evaluated using a Monte Carlo framework. Probabilistic model performance measures are combined with fuzzy rules to provide guidance on the appropriate levels of model complexity that are necessary to represent watershed behavior at a certain scale. The objective of this work is to provide better a priori understanding of model selection for a range of hydro-climate conditions and timescales.
H31H-0762
Richards' Equation and its Constitutive Relations as a System of Differential-Algebraic Equations
Richards' Equation is commonly used to understand how water flows in unsaturated soils. We present a new formulation of Richards' Equation which will allow us to incorporate model and observation errors. In addition, we can address spatial and temporal inconsistencies existing between the model and observations. There are two basic formulations for Richards' Equation: the pressure head form and the mixed form, the latter of which explicitly incorporates soil moisture content. The mixed form is typically solved using HYDRUS, a freely available program that uses finite elements with Picard iteration to handle the nonlinearities. However, recent results suggest considering Richards' Equation as a differential-algebraic equation (DAE), where the algebraic models for soil moisture content (van Genuchten's equation) is solved simultaneously with Richards' Equation (Kees, et. al., 2002). This formulation can give more accurate forward model solutions, however, we note that it also allows us to consider the uncertainties in the pressure head ψ and the soil moister content θ during the inversion process. We extend the DAE formulation to include the algebraic constraint for hydraulic conductivity K, so that its uncertainty can also be considered in an inversion. This poster focuses on the efficiency and accuracy of the forward numerical solution of this particular DAE formulation of Richards' Equation and how it compares to other forward solutions, such as HYDRUS.
H31H-0763
A Bayesian Uncertainty Framework for Conceptual Snowmelt and Hydrologic Models Applied to the Tenderfoot Creek Experimental Forest
In many mountainous regions, the single most important parameter in forecasting the controls on regional water resources is snowpack (Williams et al., 1999). In an effort to bridge the gap between theoretical understanding and functional modeling of snow-driven watersheds, a flexible hydrologic modeling framework is being developed. The aim is to create a suite of models that move from parsimonious structures, concentrated on aggregated watershed response, to those focused on representing finer scale processes and distributed response. This framework will operate as a tool to investigate the link between hydrologic model predictive performance, uncertainty, model complexity, and observable hydrologic processes. Bayesian methods, and particularly Markov chain Monte Carlo (MCMC) techniques, are extremely useful in uncertainty assessment and parameter estimation of hydrologic models. However, these methods have some difficulties in implementation. In a traditional Bayesian setting, it can be difficult to reconcile multiple data types, particularly those offering different spatial and temporal coverage, depending on the model type. These difficulties are also exacerbated by sensitivity of MCMC algorithms to model initialization and complex parameter interdependencies. As a way of circumnavigating some of the computational complications, adaptive MCMC algorithms have been developed to take advantage of the information gained from each successive iteration. Two adaptive algorithms are compared is this study, the Adaptive Metropolis (AM) algorithm, developed by Haario et al (2001), and the Delayed Rejection Adaptive Metropolis (DRAM) algorithm, developed by Haario et al (2006). While neither algorithm is truly Markovian, it has been proven that each satisfies the desired ergodicity and stationarity properties of Markov chains. Both algorithms were implemented as the uncertainty and parameter estimation framework for a conceptual rainfall-runoff model based on the Probability Distributed Model (PDM), developed by Moore (1985). We implement the modeling framework in Stringer Creek watershed in the Tenderfoot Creek Experimental Forest (TCEF), Montana. The snowmelt-driven watershed offers that additional challenge of modeling snow accumulation and melt and current efforts are aimed at developing a temperature- and radiation-index snowmelt model. Auxiliary data available from within TCEF's watersheds are used to support in the understanding of information value as it relates to predictive performance. Because the model is based on lumped parameters, auxiliary data are hard to incorporate directly. However, these additional data offer benefits through the ability to inform prior distributions of the lumped, model parameters. By incorporating data offering different information into the uncertainty assessment process, a cross-validation technique is engaged to better ensure that modeled results reflect real process complexity.
H31H-0764
Estimation of Evapotranspiration Supported by MODIS-MM5 Four Dimensional Data Assimilation system
Shortwave and net radiations are one of the key terms of the surface energy budget and is vitally important for climate studies and many applications such as agricultural meteorology and air–sea–ice interaction studies. The accurate monitoring of net radiation is a fundamental process in various meteorological and ecological studies including the estimation of evapotranspiration (ET) that is a critical process for the energy and hydrologic partitioning at the land surface. Many methods have been developed, and recently the Moderate Resolution Imaging Spectroradiometer (MODIS) also offers an opportunity to improve regional monitoring of shortwave radiation. However, MODIS has a difficulty in obtaining data in cloudy-sky conditions. In this study, we developed and tested applicability of MODIS-MM5 Four Dimensional Data Assimilation (FDDA) system for continuous estimation of ET. On a clear-sky condition, MODIS was utilized to calculate ET, while MODIS-MM5 FDDA system provides input variables for calculation of ET on a cloudy-sky condition. We evaluated the performance of MODIS- MM5 FDDA system by using meteorological data from 72 national weather stations. The reliability of ET estimates was tested with ET measured at two flux tower sites in Korea. Our preliminary results indicate that the MODIS- MM5 FDDA successfully provide input meteorological data for ET estimation on cloudy-sky days and hence, continuous ET monitoring is enhanced when compared with using MODIS data only.
H31H-0765
Comparison of two perturbation methods to estimate the land surface modeling uncertainty
In land surface modeling, it is almost impossible to simulate the land surface processes without any error because the earth system is highly complex and the physics of the land processes has not yet been understood sufficiently. In most cases, people want to know not only the model output but also the uncertainty in the modeling, to estimate how reliable the modeling is. Ensemble perturbation is an effective way to estimate the uncertainty in land surface modeling, since land surface models are highly nonlinear which makes the analytical approach not applicable in this estimation. The ideal perturbation noise is zero mean Gaussian distribution, however, this requirement can't be satisfied if the perturbed variables in land surface model have physical boundaries because part of the perturbation noises has to be removed to feed the land surface models properly. Two different perturbation methods are employed in our study to investigate their impact on quantifying land surface modeling uncertainty base on the Land Information System (LIS) framework developed by NASA/GSFC land team. One perturbation method is the built-in algorithm named "STATIC" in LIS version 5; the other is a new perturbation algorithm which was recently developed to minimize the overall bias in the perturbation by incorporating additional information from the whole time series for the perturbed variable. The statistical properties of the perturbation noise generated by the two different algorithms are investigated thoroughly by using a large ensemble size on a NASA supercomputer and then the corresponding uncertainty estimates based on the two perturbation methods are compared. Their further impacts on data assimilation are also discussed. Finally, an optimal perturbation method is suggested. http://crew.iges.org/research/LISW
H31H-0766
Improvements to modeled latent heat fluxes through the assimilation of soil moisture variability
Soil moisture is highly variable in space and time due to variations in climatic, topographic, vegetative, and soil properties. While it has been demonstrated in numerous studies that assimilation of soil moisture improves modeled estimates of evapotranspiration (ET), few studies have examined the incorporation of sub-grid-scale soil moisture variability into data assimilation schemes. In this study we assimilate soil moisture data collected at the field scale over a 70x70m grid into the CLASS model (Canadian Land Surface Scheme) using three versions of an ensemble Kalman filter. The results demonstrate that assimilating field soil moisture variability into CLASS improved model latent heat flux estimates by up to 14 percent when compared to estimates from a flux tower. However, the amount of improvement depends on the method and timing of assimilation. The effect was maximum at the beginning of the growing season, while it was minimum at peak growth. This study also demonstrates that assimilation of soil moisture variability into CLASS can result in greater improvement in modeled ET than assimilation of the mean of the sampling area.
H31H-0767
Evaluation and Assimilation of Cloud Cleared Radiances for AIRS in GEOS-5
The use of clear (cloud-free) channels for AIRS in GEOS-5 had shown positive impact on forecast skills in both hemispheres. However, improvements in forecast skills due to the assimilation of AIRS data are less impressive since the number of assimilated channels from AIRS is much larger than that from other Infrared sounders such as HIRS-3 onboard NOAA 15-17 satellites. This limitation of AIRS radiance data to improve the forecast skill is mainly due to the fact that channels capable of peaking below clouds are not used in the assimilation and yet those have highest vertical resolving capability of AIRS instrument are concentrated in the lower troposphere. On average, the percentage of AIRS footprints completely clear for all channels is less than 10%. The percentage of assimilated AIRS channel radiances however ranges from 100% for channels peaking in the upper stratosphere, above the cloud, to no more that 5% in the lower atmosphere due to cloud contamination. Our current ability to model and predict clouds accurately in global model, and to fully characterize and parameterize optical properties of cloud particles in radiative transfer model are the two major obstacles prohibiting us to use cloudy radiance directly in the assimilation. To further improve forecast skill using AIRS data, we ought to use the channels peaking below the clouds in the troposphere, which can be accomplished by assimilating cloud-cleared radiance. The cloud-cleared radiance data for AIRS used in this study were obtained from optimal cloud clearing procedures developed by researchers at CIMSS of University of Wisconsin at Madison to retrieve clear column radiances for all AIRS channels by collocating multi-band MODIS IR clear radiance observations with the AIRS cloudy radiances on a single footprint basis. Two adjacent AIRS cloudy footprints are used to retrieve one AIRS cloud-cleared radiance spectrum and no background information (first guess) is needed. To assimilate the cloud-cleared radiance data, the errors of the cloud-cleared radiances need to be addressed. The details of convolving AIRS radiances with MODIS spectral response function and comparison with MODIS-measured cloud- free radiance will be presented. The range of errors of cloud-cleared radiances for AIRS using collocated MODIS clear and near-by AIRS clear data will be shown. The NASA global data assimilation model, GEOS-5, is used to evaluate and assimilate the cloud-cleared radiance for AIRS. The residues between the cloud-cleared brightness temperature and the simulated brightness temperature from background (i.e., OMFs) will be investigated. The quality control procedures will be documented based on error estimation and the OMFs. Finally, the impacts between assimilation of clear channel radiances and cloud-cleared radiances will be addressed.