NASA SCIENCE MISSION DIRECTORATE APPLIED SCIENCES PROGRAM ___________________________________________________________________________ Earth Science for Water Management Enhancement of USDA SCAN using NASA LIS and AMSR-E Rapid Prototyping Capability - Evaluation Report 31 October, 2008 Prepared by: Mississippi State University Geosystems Research Institute Box 9652, Mississippi State, MS 39762 Contributors: Valentine Anantharaj, Georgy Mostovoy and Robert Moorhead (Mississippi State University) Yan Luo and Paul Houser (Center for Research on Environment and Water) Bailing Li, Christa Peters-Lidard and Sujay Kumar (NASA Goddard Space Flight Center) Earth Science Exploration Serving Society Accelerating the realization of economic and societal benefits from Earth science, information, and technology … i This page is left blank intentionally. ii Table of Contents Executive Summary...................................................................................................................................... iv 1 Introduction .......................................................................................................................................... 1 1.1 Objectives...................................................................................................................................... 1 1.2 Decision Support using the USDA-NRCS SCAN ............................................................................. 1 1.3 Land Data Assimilation activities at NASA .................................................................................... 2 1.3.1 NASA Land Information System ............................................................................................ 2 2 Evaluation Methodology and Approach ............................................................................................... 4 3 Evaluation Experiments ........................................................................................................................ 4 3.1 Evaluation of Precipitation Forcing Data ...................................................................................... 4 3.2 Evaluation of Noah Model Physics................................................................................................ 5 3.3 Data Assimilation Experiment I - AMSR-E and LIS 5.0 at 1x1 km2) ............................................... 6 3.3.1 Model Spin-up Experiment ................................................................................................... 6 3.3.2 Control Run I (CR1)................................................................................................................ 7 3.3.3 AMSR-E Assimilation Module in LIS ...................................................................................... 7 3.3.4 AMSR-E Data Assimilation..................................................................................................... 8 3.4 Data Assimilation Experiment II - AMSR-E (1/8 deg) and SCAN (1x1 km2) ................................... 9 3.4.1 Assimilated Data ................................................................................................................... 9 3.4.2 Land surface modeling and assimilation ............................................................................. 10 3.4.3 Experiment results .............................................................................................................. 11 4 Summary and Recommendations ....................................................................................................... 12 5 Socioeconomic Benefits ...................................................................................................................... 13 6 LIS Evaluation Team ............................................................................................................................ 13 7 References .......................................................................................................................................... 15 8 Tables and Figures............................................................................................................................... 17 iii Executive Summary The NASA Land Information System (LIS) is designed using advanced software engineering principles, allowing the interoperability of land surface models with advanced data assimilation capabilities. The AMSR-E data assimilation experiments carried out in this project demonstrate the utility of the flexible, extensible LIS data assimilation framework to apply hydrological observations and modeling tools. A new module to support AMSR-E data was implemented in LIS to facilitate data assimilation. Several deficiencies in the system have been identified and partially rectified. The approach and results of the Noah model physics evaluation demonstrate that an in-depth examination of the modeled soil moisture fields against observations at all levels can reveal deficiencies in model physics and result in more accurate soil moisture profile predictions. The SCAN measurements have been crucial in identifying the problem with the free drainage and finding an alternative. The CDF matching technique, used in the data assimilation process, does not correct the mean of the modeled soil moisture fields. When observations are transformed through the CDF matching process, they assume the mean of the modeled fields. If the model has a systematic bias, the assimilation with CDF matching will not correct it. Statistically and meteorologically, the mean behavior of the soil moisture fields is more important than others. Without a correct mean, the increased correlation from any data assimilation may not improve the soil moisture prediction. Therefore, data assimilation should be conducted in conjunction with examining model physics such as the one presented in this report to achieve optimum soil moisture prediction. Preliminary results show that soil moisture assimilation of SCAN observations is reasonably better, compared to the model simulation alone or the AMSR-E assimilation. At sites where the correlation between AMSR-E and SCAN is comparable to the correlation between the model baseline and SCAN, the assimilation did improve the modeled performance. In addition, given the large systematic discrepancies in soil moisture estimation, it is difficult to reach a definite conclusion on whether or not soil moisture estimation from AMSR-E assimilation is superior to both the model simulation and satellite estimates individually. Nevertheless, these results suggest that there is still room for further developmental efforts in application of AMSRE remotely sensed soil moisture. Finally, we wish to comment on future use of the data assimilation system built into LIS and its possible developments. In order to improve the performance of data assimilation, we urgently need to improve our knowledge of the scaling issues, and thus narrow down the uncertainties in soil moisture estimation. We are also interested in extending 1D-EnKF to 3D-EnKF, which can spread information from observed to unobserved locations. This approach is especially attractive for assimilation of sparse ground observations. These studies are important for soil moisture estimation, and eventually improving our understanding of weather prediction and climate change. In summary, the Noah model in the LIS implemented in the Mississippi Delta (even without the need for data assimilation) has demonstrated adequate skill in dynamically extrapolating the soil moisture estimates across a range of spatial and temporal scales that would be of interest for water management applications, such as conservation, irrigation planning etc. Hence, further transitions toward routine applications would be justified in this domain. iv 1 Introduction 1.1 Objectives The goal of this RPC experiment is to evaluate the cross-cutting capabilities and utility of NASA land surface research results and resources toward enabling water management applications of national priority. The award-winning NASA Land Information System (LIS) has been evaluated for its potential to extend the utility of the Soil Climate Analysis Network (SCAN) Decision Support Tool (DST), deployed and operated by the USDA Natural Resource Conservation Service (NRCS), for conservation planning and management. The NASA LIS has been evaluated for: its capabilities to enhance and extend the SCAN DST to derive physically consistent soil moisture maps at a wide range of spatial resolutions, scaling from 25x25 km2 to 1x1 km2; its usefulness in assimilating the NASA Advanced Microwave Sensor Radiometer – EOS (AMSR-E). 1.2 Decision Support using the USDA-NRCS SCAN The stated vision of USDA NRCS is “Harmony between people in the land.” Toward fulfilling this vision, NRCS has adopted the future goal to be "A globally recognized source for a top quality spatial snow, water, climate, and hydrologic network of information and technology." The USDA NRCS operates a network of nationwide cooperative soil moisture and climate information system in order to support the assessment and conservation of natural resources. The NRCS utilizes the SCAN data in conjunction with other sources information in order to provide decision support for the management of irrigation, nutrients, animal waste, pests, salinity, and water quality. Currently, the NRCS manages a network of 115 SCAN station in the United States (see Fig. 1). The USDA NRCS has an immediate identified need to provide high-resolution analysis of soil moisture across the continental United States. The National Research Council has also identified the national soil moisture monitoring network as an essential scientific and decision support tool to enhance our understanding of surface layer processes. The National Integrated Drought Information System (NIDIS) is envisioned to be a “drought early warning system”. The U.S.Group on Earth Observation (USGEO) has identified the SCAN DST as an essential and significant component of NIDIS. Hence, the evaluation of LIS to enhance the SCAN DST will serve to meet and an immediate societal need. The current network of stations focuses primarily on the agricultural regions of the nation. The expansion of this network requires careful planning to intelligently distribute the future locations of the stations so that the network can be optimized to support the requirements of NIDIS. A set of Observing System Simulation Experiments (OSSE) involving LIS can help identify, optimize, and prioritize the extension and continued deployment of SCAN. 1 1.3 Land Data Assimilation activities at NASA Effective real-world decisions making in the areas of weather and climate prediction, crop productivity estimation, water resource management, air quality monitoring and prediction, disaster management, and human health depends upon a proper understanding and knowledge of the water, energy, and carbon budgets in the land surface and their changes over time across a range of spatial scales. NASA continues to advance the scientific research in the study of the energy and water cycles via current and future missions (see Table 1) and investigations involving systematic observations and modeling of the earth’s atmosphere and land surface on global, continental and regional scales. The Land Data Assimilation System (LDAS) project at NASA is a joint effort involving a number of partner agencies and academia, including NOAA (NCEP & NESDIS), Princeton, Rutgers, University of Washington, University of Maryland, and the GEWEX international program. Accurate initialization of land surface water and energy stores is critical in environmental prediction because of their regulation of land-atmosphere fluxes over a variety of spatial and temporal scales. Errors in land surface forcing and parameterization accumulate in these integrated land stores leading to incorrect surface water and energy partitioning. However, many relatively new land surface observations from current (or future) remote sensing and other sources, based on AMSR-E, ASTER, GOES, GOES-R, GPM, MODIS, NPOESS, and TRMM, are becoming (or will become) available. These observations can be used to constrain the dynamics of land surface states. These constraints can be imposed by (a) forcing the land surface primarily by observations, thereby avoiding the often severe model biases, and (b) using data assimilation techniques to constrain unrealistic storage dynamics. The LDAS conceptual framework aims to develop the best estimation of the current state of land surfaces through a best possible integration of these land surface and atmospheric observations. Several LDAS systems have been implemented in near real time and at high spatial resolution for North American, European, and global domains. These systems are forced with real time output from numerical prediction models, satellite and other in-situ data, and radar precipitation measurements. Various land state observations can be incorporated as a constraint to the model dynamics using hydrologic data assimilation methods. The National Land Data Assimilation System (NLDAS) has developed and incorporated land data assimilation schemes to provide continually updated, 1/8 degree fields of land-surface states over central North America (Cosgrove et al., 2003; Mitchell et al., 2004). The Global Land Data Assimilation System (GLDAS) extends the NLDAS concepts to the global scale (Rodell et al., 2004). They optimally integrate multiple observation based data products, and use them to parameterize, force, and constrain (via data assimilation) the model, as well as validate results. The output fields provide more accurate land surface states than is currently available. Hence, the incorporation of these NASA research results and validated data products should increase the accuracy of weather forecast models (such as WRF), augment water management decision capabilities (using decision support systems and tools such as RiverWare and BASINS), and enhance and extend the continental scale soil moisture analysis of the USDA SCAN DST. 1.3.1 NASA Land Information System The Land Information System has its lineage in NLDAS and GLDAS. LIS is a high performance and terrascale extension of LDAS, overcoming the limitations and enhancing the capabilities of GLDAS to perform 1x1 km2 global land data assimilation (Kumar et al., 2004). 2 Figure 1: A sensor web-enabled architecture of the NASA Land Information (Courtesy: Paul Houser, CREW) LIS incorporates a suite of land surface models (LSM) of various level of sophistication encapsulating various approaches for physical solution. The default set of LSMs include the Noah, CLM, and Vic models. It has a user-friendly web-based user interface for the configuration of models and visualization of the output results. LIS also incorporates community standards and conventions such as the Earth System Modeling Framework (ESMF) to enable coupling with other ESMF-enabled models (that include WRF and COAMPS), and the Assistance for Land Modeling (ALMA), an internal data exchange structure, to facilitate the generic coupling of the LSMs via the specialized ESMF super-structure (Hill et al., 2004). The high-resolution capabilities of LIS facilitate the evaluation and implementation of decision support solutions at the same fine spatial scales of physical processes that are important in the application domain (such as the atmospheric boundary layer and cumulus cloud development); and thereby improving the surface layer parameter and flux estimates. It is also developed and implemented using advanced software and systems engineering concepts and interoperable design principles. Hence, the components of LIS can be readily integrated into the relevant systems components of the Rapid Prototyping Node(s). 3 2 Evaluation Methodology and Approach The RPC LIS evaluations are structured to clearly establish the suitability and capabilities of NASA research results and data to adequately meet the identified operational requirements and needs of the partner agencies. The specified needs and operational goals of the USDA SCAN DST are being analyzed to define the RPC experiment. Further, a coupled WRF-LIS evaluation experiment, in conjunction with a MRC funded ISS project, was also conducted to demonstrate the cross-cutting nature of potential LIS applications to serve national priorities for societal benefits. The experiments utilized the high-performance computing facilities available at the Mississippi State University (MSU) – High Performance Computing Collaboratory (HPC2). The general approach for the evaluation of LIS involved the identification and evaluation of (1) SCAN DST requirements, (2) relevant NASA observations, (3) relevant AMSR-E soil moisture data (4) future soil moisture observation capabilities, (5) relevant LIS modeling components, (6) NASA generated land surface forcing, and (7) NASA data assimilation resources. The following categories of evaluation activities are envisioned: LIS Performance Analysis: The suite of LSMs in LIS were exercised to study the performance requirements of the partner applications, specifically the USDA SCAN DST and the NASA SPoRT research to operations activities to support NOAA via the couple WRF-LIS experiment evaluations. A small region of interest that encompasses the states of Mississippi and Alabama was selected due to the relatively dense network of SCAN stations. Further, independent ground observations, acquired by agriculture producers, are also available seasonally in this region. The experiments were performed for a range of spatial resolutions, and the uncertainties involved have been quantified. Data Assimilation and Observation Sensitivity Experiments (DA-OSE): LIS has the capabilities to assimilate or otherwise incorporate observations and products from a suite of NASA assets, models as well as from other sources of partner agencies. Normally, the OSEs are conducted as part of the verification and validation process. Since the assimilation of AMSR-E data is an essential component of the evaluations, it is necessary to understand its impacts using OSEs. Hence, a candidate set of OSEs will be identified and defined to evaluate the value added by AMSR-E observations. During the process, the uncertainties associated with the individual observations have also been been characterized. For this task, we will evaluate the use of the Ensemble Kalman Filter (EnKF) to assimilate the available remote sensing and in-situ (SCAN) observations (Reichle et al., 2002). The same EnKF data assimilation method can also be used to assimilate surface temperature observations from MODIS, GOES, NPP, or NPOESS sensors, if needed. 3 Evaluation Experiments 3.1 Evaluation of Precipitation Forcing Data The most important input data for LIS and its land surface models (LSM) for soil moisture simulation are the precipitation data. With the release of LIS 5.0, a new precipitation forcing source, Stage IV, became available to LIS users. Stage IV precipitation data are based on the regional multi-sensor analyses performed by the 12 CONUS River Forecast Centers (RFC) and transmitted to NCEP. The images generated by the RFCs are quality checked through 4 automated and manual processes before transmission and therefore are more reliable than others. On the other hand, NLDAS precipitation forcing uses hourly WSR-88D radar images to disaggregate daily gauge measurements to generate precipitation information. In order to evaluate which precipitation forcing should be used for the RPC project, the two forcing data were statistically analyzed against measured rainfall amounts at five SCAN sites. The location of the SCAN sites are shown in Figure 1 and the yearly statistical results are summarized in Table 1. It can be seen that NLDAS, in general, has a lower bias and root mean squared error (RMSE) than Stage IV forcing. While NLDAS shows slightly higher false alarm rates (FAR) than Stage IV, it has better probability of detection (POD) rates. The annual rainfall amounts estimated from the two analyses are in reasonable agreement with the in situ measured rainfall amounts at each site. In short, Stage IV shows no advantage over NLDAS, so either set of forcing data can be used as the precipitation forcing for LIS runs in Mississippi areas. 3.2 Evaluation of Noah Model Physics For this project, the Noah land surface model is chosen to carry out the numerical simulation of soil moisture in the land surface. Noah is a physically based distributed model which was developed by the NOAA/NCEP for coupled weather/land surface modeling. The underlining mathematic principle of the Noah model implies that the simulated soil moisture fields from Noah are point values and so can be readily compared to in situ soil moisture measurements. This evaluation focuses on using point values of the measured soil moisture field to examine the simulated soil moisture profiles by the Noah LSM. The soil moisture measurements provided by the SCAN sites are used as the bench mark for the evaluation. Hourly volumetric soil moisture contents are measured at SCAN sites at five different depths: 5 cm, 10 cm, 20 cm, 51 cm and 102 cm. Noah is configured with twenty soil layers (as opposed to the four layers used by NOAA/NCEP) to obtain a more accurate solution to the Richards equation which is the governing equation used by Noah for soil moisture simulation. When compared to the in situ soil moisture measurements, the modeled moisture profiles are interpolated to the exact depth of each measurement for a point to point comparison. The simulation experiments are conducted for a two-year period (from March, 2004 to February, 2006) to eliminate any possible spin-up effect. The study area for this particular evaluation is marked by the black square (with latitude and longitude ranging from 32.455 to 33.725 and from -91.465 to -90.195, respectively) shown in Figure 1. Figures 2 and 3 show the simulated versus measured soil moisture contents averaged over the five SCAN site, at the 5 cm and 102 cm depths, respectively. The simulation results show that, in general, there are better agreements between the modeled and observed soil moisture contents in the top soil layer than in the lower layers. Noah is found to systematically underestimate soil moisture in the lower soil layers at all the SCAN sites within the study area. This is attributed to the free drainage condition applied at the lower boundary of the 2-meter simulation domain. This underestimation of soil moisture becomes even more severe in the late summer and the early fall seasons when less precipitation is measured. To correct this problem, a constant head (i.e. constant soil moisture) boundary condition, obtained by extrapolating measured moisture content at 102 cm from the SCAN sites, is implemented at the 2 meter depth. The results show that the modified boundary condition 5 yields much wetter soil moisture profiles in the lower part of the land surface, leading to a better agreement with the observations at all sites. Figures 4 and 5 show the simulated soil moisture fields with free drainage, constant water content, and measured soil moisture contents averaged over all five sites, at the 5 cm and 102 cm depths respectively. Both boundary conditions produce good predictions of soil moisture at the 5 cm depth. The free drainage actually does slightly better than the constant water content in mimicking the drying process in the summer months. This is primarily due to the fact that many parameters related to the surface processes such as infiltration and ET are selected or ‘calibrated’ based on the free drainage condition (Ek et al., 2003). As a result, these parameters may not work well with the constant water content condition. On the other hand, at the deeper soil profile (102 cm) shown in Figure 5, the constant water content boundary condition clearly out performs the free drainage by reducing the underestimation of soil moisture significantly. To summarize the performance of Noah with these two boundary conditions, the averaged (over five SCAN sites and five measuring depths) bias is calculated and displayed in Figure 6. The bias here is defined as the simulation minus observation. It can be seen that the constant water content boundary condition produces near unbiased soil moisture estimation while the free drainage clearly underestimates the soil moisture fields. The underestimation is most severe in the late summer and the fall season when less precipitation is measured in the area. The improvements can be seen through the root zone soil moisture. Figure 7 shows the time series of the daily root zone soil moisture for the simulated and observed soil moisture fields averaged at the five SCAN sites. The constant water content condition again performs exceedingly better than the free drainage condition for the water storage estimation. In conclusion, the free drainage condition used by many land surface models including Noah is not applicable in this region and possibly many other areas as well; instead, a constant head boundary condition may be more appropriate based on the observations and the hydrogeological conditions in the region. The constant water content condition improves Noah’s soil moisture simulation in Mississippi with unbiased estimation of soil moisture. Though this conclusion may not be applied to other areas without further investigations, this approach and results of this evaluation demonstrate that an in-depth examination of the modeled soil moisture fields against observations at all levels can reveal deficiencies in model physics and result in more accurate soil moisture profile predictions. The SCAN measurements (in depth and continuous observations) are crucial in identifying the problem with the free drainage and finding an alternative. 3.3 Data Assimilation Experiment I - AMSR-E and LIS 5.0 at 1x1 km2) 3.3.1 Model Spin-up Experiment Spin-up is a common approach used in the land surface modeling community to generate initial conditions needed for numerical simulations. The need for generating initial conditions is due to the lack of observations for the state variable for every grid point at any given model initial time. A spin-up process usually begins with an arbitrary initial value for the state variable and then the land surface model is run for a certain period of time with external atmospheric forcing data. At the end of the run, it is expected that the state variable has been brought up to a 6 state similar to its true state since realistic atmospheric forcing data have been used to ‘coerce’ the soil moisture into the final state. A spin-up experiment is said to reach a converging solution if the simulated profiles remain unchanged at the end of the model run as the run period increases. Three Noah runs were conducted using a six months, eight months and one year spin up period, respectively, with all runs ending at 00Z, May 1, 2004. In each spin-up run, initial water content was set at 0.3 and the entire model was forced with NLDAS precipitation data. As illustrated in Figure 8, the three runs yield very similar soil moisture profiles at the end of each run, with the profiles from 10 months and one year spin-ups overlapping each other. It is clear that 10 months are long enough to spin up the soil moisture in this Mississippi region. While much of the attention has been given to the convergence and speed of the spin-up process, it is worthwhile to point out that the converged solution from any spin-up process is not warranted to match the observations. As shown in Figure 8, the differences between the simulations and field measurements at the SCAN site can be significant. The discrepancy can be caused by many sources. But based on the previous analyses, the model physics, i.e., the free drainage boundary condition is mostly responsible for the drier soil moisture profile, especially at the lower profile. This experiment further demonstrates that, without correct model physics, the generated soil moisture state from any spin-up experiment is not warranted to be consistent with in situ measurements, no matter how long the model is spun up. 3.3.2 Control Run I (CR1) The control run is designed to establish baseline simulated soil moisture fields so that any improvements made by the data assimilation can be illustrated. Even though the free drainage boundary condition is found to be inappropriate in this region, we still choose to use it for the data assimilation. The primary reason is that the free drainage condition is used in the official version of the Noah land surface model which is used by the majority of the community. In addition, it is not clear based on existing research if the boundary condition will affect the results of the data assimilation. The run domain for the data assimilation, shown in Figure 9, is larger than the domain we used for evaluating model physics so that more SCAN sites can be included for evaluating data assimilation. The area is about 300 km by 300 km with latitude and longitude ranging from 32.885 to 35.405, and -92.155 to -89.605, respectively. The horizontal grid resolution is again set to be 0.01 degree. The official standard four layers of soil are employed. Noah is forced with NLDAS forcing data. The UMD 1 km land cover data set is used to provide vegetation type and the STATSGO soil texture data set is used for deriving the soil hydraulic parameters. The baseline run period is from 2002 to 2006, which provides enough variations in precipitation to make the final statistical analyses meaningful, in addition to eliminating any spin-up effects. The initial soil moisture is set to 0.3. Soil moisture content is output at threehour intervals. 3.3.3 AMSR-E Assimilation Module in LIS A general data assimilation module, which is based on the ensemble Kalman filter developed by Reichle el al. (2007) for assimilating soil moisture into land surface models, has been implemented in LIS prior to this project. The filter has been tested with the Catchment 7 model and AMSR-E soil moisture retrievals and a synthetic soil moisture assimilation using Noah. For this RPC project, a new module was implemented in LIS to facilitate data assimilation using AMSR-E and the Noah LSM. AMSR-E soil moisture is retrieved based on measured brightness temperature from the NASA polar-orbiting Aqua satellite. The level 3 AMSR-E soil moisture data set, which contains both ascending and descending retrievals, is used. Even though the level 3 are interpolated to the 25 km cell spacing, the actual foot print each retrieval represents is about 56 km. The daily AMSR-E data files are stored in HDF-EOS format and in the Equal-Area Scalable Earth Grid (EASE-Grid) projection. Since LIS uses equal distance cylindrical projection, a re-projection of the gridded AMSR-E (based on a nearest neighbor searching algorithm) is implemented in LIS to convert the equal area based EASE grid projection to the equal latitude/longitude projection. One of the key issues in using AMSR-E soil moisture data is the apparent bias between AMSR-E retrieved soil moisture values and the modeled values. The difference can be attributed to retrieval errors, scale issues and the model bias. To reduce the bias, Reichle et al. (2007) used the cumulative distribution function (CDF) matching technique which maps the CDF of observed soil moisture contents to that of the modeled ones and therefore forces the two sets of soil moisture to share the same mean value. At each LIS grid point, CDFs are derived for the soil moisture obtained from the 5 year baseline run and the 5 year AMSR-E retrievals, respectively. At each point, 500 bins are used for deriving these functions. When AMSR-E is assimilated into Noah, the retrieved soil moisture is converted to model compatible values based on the CDFs at the given location. 3.3.4 AMSR-E Data Assimilation To evaluate the assimilation results, the correlation coefficients of the daily mean anomalies of modeled soil moisture fields with that of the SCAN in situ measurements are calculated and presented in Table 2. The same correlation coefficients for the baseline run and the unconverted AMSR-E soil moisture are also presented in Table 2 for comparison. The five year soil moisture data for both the modeled and observed are treated as a complete continuous time series. Table 2 shows that AMSR-E data in general have a lower correlation with SCAN than the modeled soil moisture fields produced by the Noah baseline. This can be attributed to the NLDAS forcing data used in this study which, as shown earlier, compares very well with the gauged rainfall measurements. The lack of strong correlation for the AMSR-E retrievals with the SCAN data are likely related to the fact that AMSR-E retrievals are not sensitive to the daily changes of soil moisture either due to the retrieval algorithm or due to the larger scale they represent. In addition, there are only about two retrievals daily in the Mississippi region, which may lower dynamic ranges of the soil moisture at any pixel. On average the assimilation results did not improve over the Noah baseline simulation. However, at sites (for instance, Scott, Lonoke Farm, and Earle) where the correlation between AMSR-E and SCAN is comparable to the correlation between the model baseline and SCAN, the assimilation did improve the modeled performance. It can be drawn from this study that the satellite observations need to have compatible quality (i.e., correlation in this study) with the model in order to see improvement through data assimilation. Note that the SCAN measurement time series are not always complete, as indicated in the last column of Table 2. Some sites have only two to three years of observations which may affect the statistics listed in Table 2 as well. 8 The underlining filter used for the data assimilation can be evaluated by examining the mean and variance of the normalized innovation which is defined as the difference between actual observations and the predicted observations divided by the sum of the model and observations errors (standard deviation) at any given location. Here the mean and variance are calculated based on the five year innovation time series. Figure 10 (a) shows that the means are very close to zero because the CDF matching essentially made the observations have the same mean as the modeled. Figure 10 (b) shows that the variance is generally around 1 with a few spots near the border having higher variability. For both the man and variance, there are some spots having higher variability near the border where elevations are higher than the areas near the river banks. The larger heterogeneity may also be due to the high resolution rainfall data and soil texture data used for the simulations. The spatial average of the innovation variance in the study area is around 1.2 which shows the filter is reasonably configured. 3.4 Data Assimilation Experiment II - AMSR-E (1/8 deg) and SCAN (1x1 km2) An Ensemble Kalman Filtering (EnKF) algorithm was implanted in LIS for the purpose of effectively combining the satellite land surface observations with land surface simulations to improve land modeling. Our work was part of an effort to enhance USDA SCAN by using NASA LIS and AMSR-E, dedicated to investigating assimilation of SCAN data into a land surface model within LIS. Thus the main effort of this work was to perform and evaluate AMSR_E and SCAN assimilation using LIS. The schematic diagram in Fig.11 demonstrates the roadmap for soil moisture data assimilation and the evaluation approach used for this project. This report describes ground-based USDA SCAN soil moisture data, satellite-based AMSR-E soil moisture products, and the NCEP Noah Land Surface Model (LSM) applied into the data assimilation framework built in LIS. In-situ SCAN soil moisture observations for 12 sites in the Mississippi Delta were gathered and processed for model assimilation, comparison and validation. Most importantly, this report describes the assimilation experiments, results, summary and future directions of the LIS data assimilation capability. 3.4.1 Assimilated Data 3.4.1.1 USDA SCAN in-situ soil moisture measurements Soil moisture data from SCAN are available at five depths; 5cm, 10cm, 20cm, 51cm and 102cm, and collected every hour. Note these are point samples, with very little spatial information and limited regional coverage, but it was the best verification data available. In order to make the SCAN datasets at the required spatial and temporal resolution for 1km-Noah-LIS assimilation runs and also use the available observed SCAN data as much as possible, some unique procedures and techniques were used. The conversion from local time to UTC time was also taken into account. We focused on a domain covering the lower part of the Mississippi Delta, located mainly in the state of the Mississippi. Over the given Noah-LIS domain of the Delta region, twelve SCAN sites that had a more complete soil moisture data record for the period of June-August 2005 were selected, as indicated in Fig. 9. Observed surface soil moisture at the 5cm top layer from 12 SCAN sites were selected twice daily at 06Z and 18Z, with same availability as the AMSR-E data for use of data assimilation. Observed data at 00Z, 03Z, 09Z, 12Z, 15Z and 21Z (data at assimilated time 06Z and 18Z are not included) from these sites were used to cross validate the model outputs from SCAN and AMSR-E data assimilation experiments. 9 3.4.1.2 Satellite-based AMSR-E soil moisture products NASA’s AMSR-E soil moisture products (Njoku, 2004) taken from Level 3 soil moisture retrievals were employed in this project. They provided approximately the upper 1cm soil moisture field available globally and twice daily at 06Z and 18Z from 18 June, 2002. When the AMSR_E soil moisture data files were collected from NSIDC, we processed them from the original EASE grid in HDF format into 1/8 degree latitude-longitude grid in binary format, with the file format and grid spacing matching the LIS based Noah-EnKF framework. Before assimilating satellite observations into the model, the quality of the data must be checked. We processed the daily data for four years from June 18, 2002 to June 18, 2006. Fig. 12 shows the retrieved North American surface soil moisture for the four-year climatology. In order to understand the overall quality of the soil moisture retrieval dataset, we ran LIS using the Noah land surface model forced by NLDAS forcing data at 1/8 degree resolution for North America for four years from June 18, 2002 to June 18, 2006. Averaging the whole period of simulation, the results were used as surface soil moisture climatology for the four years. An intercomparison between the AMSR_E soil moisture retrievals and the Noah-LIS soil moisture climatology and their difference are demonstrated in Fig. 12. Compared with the simulated climatology, the AMSR-E retrievals showed quite a large departure of soil moisture climatology from the Noah simulations with a significant underestimation over most of the U.S. continent. To correct the systematic difference between the retrieved and modeled soil moisture, a CDF matching approach (Reichle and Koster, 2004) was utilized, by scaling AMSR-E soil moisture so that it had a similar climatology to the Noah model. A four year common training period (2002-2006) is used for the CDF matching. Fig. 13 shows one typical example of the CDF matching results. In general, with CDF matching, the climatology of scaled AMSR-E soil moisture estimates is obviously close to that of model estimates. 3.4.2 Land surface modeling and assimilation The Noah LSM v.2.7.1 (Ek et. al, 2003) was used for model simulation and assimilation runs in this study. In the current configuration, the four model soil layers had been specified at the depths of 10, 40, 60 and 100cm, respectively. It operated on a 30-minute time step and required the following parameters and gridded data inputs: NLDAS forcing data (Cosgrove et al., 2003) and UMD land cover and vegetation classification, NOAA FAO soil map, NCEP quarterly albedo climatology, monthly greenness fraction climatology, maximum snow albedo and bottom temperature. Generally the parameters of the Noah LSM and fixed fields are similar to that of the NLDAS project. The North American Land Data Assimilation System (NLDAS) dataset (Cosgrove et al., 2003), provided near surface atmospheric forcing data to force the Noah model. The forcing dataset was specified at approximately 15-km grid spacing and covered the CONUS region and some adjacent regions of Canada and Mexico. It had available data from late 1996. The Noah model was configured at 0.01 x 0.01 latitude-longitude resolution (approximately 1x1 km2) over a domain covering the lower part of the Mississippi Delta region. The 1-km domain size covers 2.55x2.55 latitude-longitude bounded by 32.855-35.405 N, 92.15589.605W, including a total of 256x256 points. Additional AMSR-E assimilation runs were configured on the North America LDAS grid, which is a coarser 21x 21 1/8th-degree grid over the same Mississippi Delta domain. To delineate the effects of CDF matching, another run was conducted using the scaled AMSR_E soil moisture. 10 The model started with a spin-up run from January 1, 2000 until June, 1, 2005 and then started running designed experiments until August 31, 2005 which produced 3-hourly model outputs. Noah model simulations and soil moisture data assimilation were performed at either 1/8 deg for AMSR-E assimilation or 1km grid spacing for SCAN assimilation. The only assimilated variables were soil moisture at the four layers, which were updated by the observed surface soil moisture. The EnKF experiments were performed for a period of 1 June – 31 August 2005. In this case, assimilation by the EnKF was repeated every 12-hour cycle at 06Z and 18Z. The filters used 20 ensemble members. It is a 1D-EnKF scheme, which means the analysis was computed independently at each grid point. If there was no observation, there would be no update (no data assimilation). Random perturbations of Gaussian distribution were added to the state variable and forcing terms prior to assimilation. The parameters related to EnKF scheme were tuned to get the optimal performance. The model outputs for each of the 12 SCAN sites for the experiment period were analyzed. The modeled 10-cm top layer soil moisture content estimates were examined comprehensively. Note the spatial interpolation to site locations for SCAN data was not performed, as we simply used the nearest model grid point data. 3.4.3 Experiment results In this project, both in-situ measurements from the USDA Soil Climate Analysis Network (SCAN) and remotely sensed data from the AMSR-E instrument onboard the Aqua satellite have been assimilated into the Noah model using the data assimilation capabilities of LIS. Therefore, four experiments were conducted: 1. AMSR-E DA Run (termed as AMSR-E EnKF) with observations derived from AMSR-E retrievals; 2. Scaled AMSR_E DA Run (termed as Scaled AMSR-E EnKF) with observations derived from Scaled AMSR-E retrievals; 3. SCAN DA Run (termed as SCAN EnKF) with observations derived from SCAN measurements; 4. Open Loop Run; Noah simulation perturbed with forcing and states and without assimilation. Fig. 14 shows spatial maps of the soil moisture assimilation results by assimilating AMSR_E or scaled AMSR-E data, compared with Noah simulations and their original satellite retrievals. The assimilation performance was evaluated against soil moisture observations at 12 sites. While the analysis was conducted using results from all of the sites, here we provide an example for one representative site − GOODWIN CK TIMBER (ms_2025). The effect of each assimilation on soil moisture is shown in Figs.15-17, respectively, along with the no assimilation case (Open Loop) and corresponding observations. In the AMSR-E assimilation case shown in Fig.15, the bias between simulated and observed values is apparent. This is partly because this bias can be attributed to lower values and lower variability of AMSR-E retrievals. There is another drastic difference due to the scaled AMSR_E assimilation also observed in Fig. 16. In Fig. 17, model simulation overestimates soil moisture, but SCAN assimilation lowers the modeled soil moisture, bringing it closer to the observations. Fig.18 illustrates a comparison plot of RMSEs against SCAN point-scale measurements between AMSR-E and SCAN assimilation results at each of the 12 SCAN sites. Likewise, Fig.19 gives a correlation comparison. It was observed that the performance varied among assimilations of various real observations with the best one resulting from the SCAN EnKF. It has smallest RMSE for 11 out of 12 sites and highest correlations for 10 out of 12 sites. Compared to the Open Loop, EnKF improved Soil moisture estimation in the SCAN assimilation, but performance was degraded in AMSR-E assimilation. Moreover, whenever soil moisture model 11 simulation (Noah) performs better than satellite retrieval (AMSR-E) compared with field measurements (SCAN), using CDF-matching to calibrate the retrievals may make the retrieval (Scaled AMSR-E) accuracy better and consequently causes their assimilation (Scaled AMSR-E EnKF) to be improved relative to non-calibrated case (AMSR-E EnKF). 4 Summary and Recommendations The LIS architecture is designed using advanced software engineering principles, allowing the interoperability of land surface models, meteorological inputs, land surface parameters and observational data, and data assimilation capabilities. The AMSR-E data assimilation experiments carried out in this project demonstrate the utility of the flexible, extensible LIS data assimilation framework to apply hydrological observations and modeling tools. A new module to re-project the gridded AMSR-E data was implemented in LIS to facilitate data assimilation using AMSR-E data and the Noah LSM. The approach and results of the Noah model physics evaluation demonstrate that an indepth examination of the modeled soil moisture fields against observations at all levels can reveal deficiencies in model physics and result in more accurate soil moisture profile predictions. The SCAN measurements (in depth and continuous observations) are crucial in identifying the problem with the free drainage and finding an alternative. A fundamental issue with the CDF matching technique is that it does not correct the mean of the modeled soil moisture fields. When observations are transformed through the CDF matching process, they assume the mean of the modeled fields. If the model has a systematic bias, the assimilation with CDF matching will not correct it. Statistically and meteorologically, the mean behavior of the soil moisture fields is more important than others. Without a correct mean, the increased correlation from any data assimilation may not improve the soil moisture prediction. Therefore, data assimilation should be conducted in conjunction with examining model physics such as the one presented in this report to achieve optimum soil moisture prediction. Preliminary results show that soil moisture assimilation of SCAN observations is reasonably better, compared to the model simulation alone or the AMSR-E assimilation. At sites where the correlation between AMSR-E and SCAN is comparable to the correlation between the model baseline and SCAN, the assimilation did improve the modeled performance. In addition, we also found that, given the large systematic discrepancies in soil moisture estimation, it is difficult to reach a conclusion whether or not soil moisture estimation from AMSR-E assimilation is superior to both the model simulation and satellite estimates individually. Nevertheless, these results suggest that there is still room for further efforts in application of AMSR-E remotely sensed soil moisture data in the land data assimilation applications. Accounting for the errors from those independent soil moisture sources, these disparities are manifestations of a wide range of uncertainties in the soil moisture estimation that need to be reduced, and thus the challenges imposed on land data assimilation approaches. Note that point measurements may have discrepancy when compared to the spatially averaged remote sensing and model simulations. In-situ observations may be affected by whether or not such in-situ data correctly represents true soil moisture at the scale described in the land data assimilation system. 12 The inconsistent effective top layer soil depth in model (10cm), satellite observation (~1cm) and ground measurement (~5cm) is another important issue that needs to be resolved. Finally, we wish to comment on future use of the data assimilation system built into LIS and its possible developments. Bias correction schemes for correcting model errors through EnKF were implemented in LIS, during a previous JCSDA project. Bias correction to observation errors by means of EnKF is currently underway in a new Aquarius OSSE project. In order to improve the performance of data assimilation, we urgently need to improve our knowledge of the scaling issues, and thus narrow down the uncertainties in soil moisture estimation. We are also interested in extending 1D-EnKF to 3D-EnKF, which can spread information from observed to unobserved locations. This approach is especially attractive for assimilation of sparse ground observations. These studies are important for soil moisture estimation, and eventually improving our understanding of weather prediction and climate change. 5 Socioeconomic Benefits There are a wide range of expected societal beneficial results from the real time integration of USDA SCAN information and assimilation of NASA remote sensing data using LIS in order to derive a high resolution analysis of soil moisture information across the continental United States, as follows: (1) provides critical information to support drought monitoring and mitigation, (2) provides critical information for predicting droughts based on weather and climate predictions, (3) supports irrigation water management, (4) supports fire risk assessment, (5) supports water supply forecasting and NWS flood forecasting, (6) supplies a critical missing component to assist with snow, climate and associated hydrometeorological data analysis, (7) supports climate change assessment, (8) enables water quality monitoring, (9) supports soil survey interruption and mapping, and (10) supports a wide variety of natural resource management & research activities such as NASA remote sensing activities of soil moisture and ARS watershed studies. The primary contribution of NASA data, research results and models will be validated high-quality satellite data products, such as capabilities to be provided by current and/or future sensors and missions including AMSR, TRMM, and GPM, and water availability and surface energy parameters from LIS. 6 LIS Evaluation Team The LIS evaluation team for this RPC experiment consists of research scientists and students from Mississippi State University, George Mason University (GMU), and NASA GSFC. The interface with this science team was via Valentine Anantharaj, the single point of contact for the RPC LIS activities. The RPC LIS evaluation efforts were closely coordinated with the RPC node development activities. External partners included: (a) Dr. Christa PetersLidard of NASA GSFC for the technical and scientific guidance and application of LIS, the design and evaluation of experiments, the coupling of models, liaison with external partners as necessary and provision of necessary data, and for the development of PMW; and (b) Prof. Paul Houser at the GMU Center for Research on Environment and Water (CREW) for scientific leadership and guidance, data assimilation activities, and liaison with USDA NRCS, and for the development of PMW. Other consultations included: Dr. Lars Peter Riishogaard of the NASA 13 Global Modeling and Assimilation Office (GMAO) regarding the design and analysis of OSSEs; and Dr. Gary Jedlovec of NASA SPoRT regarding the coupled WRF-LIS case study; and the National Center for Atmospheric Research for the implementation of the WRF model. Support for GSFC, GMU, and NCAR are programmed via the rapid-prototyping proposal. 14 7 References Burgan, R. E., P. L. Andrews, L. S. Bradshaw, C. H. Chase, R. A. Hartford, and D. J. Latham. 1997. WFAS: Wildland Fire Assessment System. Fire Management Notes 57:14-17. Cooke, W., V. Anantharaj, C. Wax, J. Choi, K. Grala, M. Jolly, G.P. Dixon, J. Dyer, D.L. Evans, G.B. Goodrich, 2007: Integrating climatic and fuels information into national fire risk decision support tools. Pages 555-569 in B.W. Butler and W. Cook, editors, The fire environment--innovations, management, and policyLannom, Keith B., Evans D.L., and Cooke III, W.H., Forest Mapping of Central America and Mexico with AVHRR Data, March 2001. Geocarto International, Vol. 16, No. 1. pp. 45-53. Cosgrove, B.A. and Coauthors, 2003: Real-time and retrospective forcing in the North American Land Data Assimilation System (NLDAS) project. J.Geophys. Res., 108, D22, 12pp. Covington, W. W., 2000: Helping western forests heal. Nature, 408, 135-136. Ek, M.B. and Coauthors, 2003: Implementation of NOAH land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta model. J. Geophys. Res., 108, D22, 16 pp. Hill, C., DeLuca, C., Balaji, V., Suarez, M. and da Silva, A. (2004). The architecture of the Earth System Modeling Framework. Computing in Science and Engineering, 6(1). Houser, P. R., W. J. Shuttleworth, H. V. Gupta, J. S. Famiglietti, K. H. Syed, and D. C. Goodrich, 1998: Integration of Soil Moisture Remote Sensing and Hydrologic Modeling using Data Assimilation. Water Resources Research, 34(12):3405-3420 Koster, R.D. and M.J. Suarez, 1992: Modeling the land surface boundary in climate models as a composite of independent vegetation stands. J. Geophys. Res., 97, D3, 2697-2715. Kumar, S. V., C. D. Peters-Lidard, Y. Tian, P. R. Houser, J. Geiger, S. Olden, L. Lighty, J. L. Eastman, B. Doty, P. Dirmeyer, J. Adams, K. Mitchell, E. F. Wood and J. Sheffield, 2006: Land Information System - An Interoperable Framework for High Resolution Land Surface Modeling. Environmental Modelling & Software, v21, p1402-1415. Mitchell, K.E., and 21 co-authors (2004). The Multi-institution North American Land Data Assimilation System (NLDAS): Utilization of multiple GCIP products and partners in a continental distributed hydrological modeling system. Journal of Geophysical Research, 109: doi:10.1029/2003JD003823. Mölders, N. , Jankov, M., and G. Kramm, 2005: Application of Gaussian error propagation principles for theoretical assessment of model uncertainty in simulated soil processes caused by thermal and hydraulic parameters. J. Hydrometeor., 6, 1045-1062. NASA, The Science Plan for NASA’s Science Mission Directorate (2007-2016). NRC "Confronting the Nation's Water Problems: The Role of Research" (2004); Committee on the Assessment of Water Resources, Water Science and Technology Board, National Academies Press, Washington, DC; prepublication version available at: http://www.nap.edu Peters-Lidard, C. D., P. R. Houser, Y. Tian, S. V. Kumar, J. Geiger, S. Olden, L. Lighty, B. Doty, P. Dirmeyer, J. Adams, K. Mitchell, E. F. Wood and J. Sheffield, 2007: High-performance Earth system modeling with NASA/GSFC's Land Information System. Innovations in Systems and Software Engineering . Vol. 3(3), 157-165. Reichle, R.H., McLaughlin D.H, and Entekhabi D., 2002, Hydrologic data assimilation using the Ensemble Kalman filter, Mon. Wea. Rev., 130: 103-115. Reichle, R. H., R. D. Koster, P. Liu, S. P. P. Mahanama, E. G. Njoku, and M. Owe, Comparison and assimilation of global soil moisture retrievals from the Advanced Microwave Scanning Radiometer 15 for the Earth Observing System (AMSR-E) and the Scanning Multichannel Microwave Radiometer (SMMR), J. Geophys. Res., 112, D09108, doi:10.1029/2006JD008033, 2007. Rodell, M., P. R. Houser, U. Jambor, J. Gottschalck, K. Mitchell, C.-J. Meng, K. Arsenault, B. Cosgrove, J. Radakovich, M. Bosilovich, J. K. Entin, J. P. Walker, D. Lohmann, and D. Toll, The Global Land Data Assimilation System, Bull. Amer. Meteor. Soc., 85 (3), 381–394, 2004. Santanello, J.A. Jr. and T. N. Carlson, 2001: Mesoscale simulation of rapid soil drying and its implications for predicting daytime temperature. J. Hydrometeor., 2, 71-88. Tian, Y., C. D. Peters-Lidard, S. Kumar, J. Geiger, P. R. Houser, J. L. Eastman, P. Dirmeyer, B. Doty and J. Adams, 2004. High Performance Land Surface Modeling with a Beowulf Cluster. Submitted to Computing in Science & Engineering. USDA Forest Service. 2003. Weather Information Management System User's Guide. Department of Agriculture, Forest Service, Fire and Aviation Management, Washington, D.C. US Deparment of the Interior (USDOI) and US Department of Agriculture (USDA). 2005. Interagency Standards for Fire and Fire Aviation Operations. NFES 2724. 16 8 Tables and Figures 17 TABLE 1. LIST OF NASA DATA PRODUCTS THAT HAVE POTENTIAL APPLICATION USING LIS Class Observation Technique Example Platform Temporal Spatial Land Parameters Leaf area and greenness optical/IR AVHRR, MODIS, NPOESS weekly 1km Albedo optical/IR MODIS, NPOESS weekly 1km Emissivity optical/IR MODIS, NPOESS weekly 1km Vegetation structure Lidar ICESAT, ESSP lidar mission weekly-monthly 100m Topography in-situ survey, radar GTOPO30, SRTM episodic 30m–1km Wind profile radar Air Humidity and temperature Near- surface radiation IR, MW TOVS, GOES, AVHRR, MODIS, AMSR hourly-weekly 5 km optical/IR GOES, MODIS, CERES, ERBS, etc. hourly-weekly 1km Precipitation microwave/IR TRMM, GPM, SSMI, GEO-IR, etc. hourly-monthly 10km Temperature IR, in-situ IR-GEO, MODIS, AVHRR, TOVS hourly-monthly 10m-4km Thermal anomalies IR, NIR, optical AVHRR, MODIS, TRMM daily-weekly Snow cover and water optical, microwave weekly-monthly Freeze/thaw radar SSMI, TM, MODIS, AMSR, AVHRR, etc. Quickscat, HYDROS, IceSAT, CryoSAT 250m– 1km 1km weekly 3km Total water storage gravity GRACE monthly 1000km Soil moisture active/passive microwave SSMI, AMSR, HYDROS, SMOS, etc. 3-30 day 10-100 km Evapotranspiration optical/IR, in-situ MODIS, GOES hourly-weekly 10m-4km Solar radiation optical, IR MODIS, GOES, CERES, ERBS hourly-monthly Longwave radiation optical, IR MODIS, GOES hourly-monthly 10m-4km Sensible heat flux IR MODIS, ASTER, GOES hourly-monthly 10m-4km Land Forcings Land States Land Fluxes 18 TABLE 2. COMPARISON OF NLDAS (RED) AND STAGE IV (GREEN) YEARLY STATISTICS RELATIVE TO MEASURED RAINFALL AT FIVE SCAN SITES. Table 1: Yearly Statistics and Estimated Rainfall SCAN site ID Bias RMSE FAR POD (NLD/STG4) (mm/sec) (NLD/STG4) (mm/sec) (NLD/STG4) (NLD/STG4) TotalRain TotalRain (NLD/STG4) (observation) (mm) (mm) Beasley Lake -6.96 e-06 -8.27 e-06 1.53 e-07 1.53 e-07 38 29 65 58 1,441 1,403 Perthshire 1.55 e-06 2.50 e-06 1.06 e-07 1.56 e-07 45 42 65 60 1,523 1,553 1,475 Scott 5.81 e-06 7.02 e-06 9.55 e-08 44 36 70 62 1,597 1,635 1,415 1,640 1.32 e-07 Silver City 5.54 e-06 -4.14 e-06 8.56 e-08 7.94 e-08 47 30 77 70 1,296 1,043 1,151 North Issaquena 4.59 e-06 -7.94 e-06 9.76 e-08 3.08 e-08 46 22 81 81 1,331 1,134 1,342 19 TABLE 3. CORRELATION COEFFICIENTS OF DAILY MEAN ANOMALIES WITH SCAN SCAN Sites Model Assimilation AMSR-E Years of SCAN Perthshire 0.59 0.33 0.27 4.75 Silver City 0.65 0.65 0.57 3 Scott 0.43 0.49 0.43 3.6 Beasely Lake 0.35 0.31 0.18 5 NIssaquena 0.63 0.50 0.37 3 Tunica 0.68 0.44 0.38 5 Vance 0.62 0.41 0.35 5 Lonoke Farm 0.61 0.65 0.58 5 Campus PB 0.71 0.57 0.41 3 Marianna 0.65 0.35 0.35 2.5 Earle 0.55 0.63 0.54 2.75 DeWitt 0.22 0.30 0.28 2.6 Average 0.56 0.47 0.39 20 Figure 1: The Soil Climate Analysis Network (SCAN) stations in the United States. Mississippi and Alabama have a dense network of SCAN sites. 21 Figure 2: At 5 cm depth and averaged over all five sites, SCAN observed soil moisture content (red diamonds) versus Noah simulated soil moisture content (blue line) for 2004 (left) and 2005 (right). 22 Figure 3: At 102 cm depth and averaged over all five sites, SCAN observed soil moisture content (red diamonds) versus Noah simulated soil moisture content (blue line) for 2004 (left) and 2005 (right). 23 Figure 4: At 5 cm depth and averaged over all five sites, SCAN observed soil moisture content (red diamonds), Noah free drainage (blue line), and Noah constant water content (brown starred line) for 2004 (left) and 2005 (right). 24 Figure 5: At 102 cm depth and averaged over all five sites, SCAN observed soil moisture content (red diamonds), Noah free drainage (blue line), and Noah constant water content (brown starred line) for 2004 (left) and 2005 (right). 25 Figure 6: Bias of simulated soil moisture field averaged on five depths and at five sites, Noah free drainage (blue line), and Noah constant water content (brown starred line) for 2004 (left) and 2005 (right). 26 Figure 7: Daily root zone soil moisture of simulated soil moisture field averaged at five sites, with Noah free drainage (blue line), Noah constant water content (brown starred line) and SCAN (red cross line) for 2004 (left) and 2005 (right). 27 Figure 8: Vertical water content profiles of three spin-up runs and the observations at Silver City, Mississippi at 23Z April 30, 2004. 28 Figure 9: The simulation domain and location of the SCAN sites used for data assimilation 29 Figure 10: The normalized innovation: (a) mean (b) variance 30 Soil Moisture Observations (SCAN or AMSR-E) Soil Moisture Data EnKF DA Noah Land Surface Model of NASA Land Information System Soil Climate Analysis Network No DA Soil Moisture Data Soil Moisture Data Soil Moisture Data Evaluation Study Figure 11: Schematic diagram roadmap of the soil moisture data assimilation and evaluation approach. 31 (a) (b) Figure 12: Spatial distribution of soil moisture climatology estimated from Noah simulations (a) and AMSR_E retrievals (b) averaged for the four years (2002-2006), showing their differences in (c), while (d) is scaled AMSR_E soil moisture. Units are v/v(%). 32 Without CDF Matching With CDF Matching NOAH NOAH AMSR-E Scaled AMSR-E SCAN SCAN Ames, IA Figure 13: Monthly soil moisture time series estimated from Noah simulations (blue curve), AMSR_E retrievals (green curve, left panel) and scaled AMSR_E retrievals (green curve, right panel), SCAN measurements (black curve) in Ames, Iowa for the four years (2002-2006). Units are v/v%. 33 EnKF Assimilation of AMSR-E SM Retrievals EnKF Assimilation of Scaled AMSR-E SM Retrievals Figure 14: Results of the Ensemble Kalman Filter soil moisture estimates compared with the results either from Noah model integration or from AMSR-E soil moisture retrievals. 34 Impact of AMSR-E assimilation Open Loop AMSR-E EnKF AMSR-E SCAN Mean(-0.36618) Variance (1.03958) Figure15: NOAH modeled (with and without AMSR-E assimilation) and observed (SCAN) soil moisture in the Goodwin CK Timber (ms_2025, 34.23˚N, 89.9˚W). 35 Impact of Scaled AMSR-E assimilation Open Loop Scaled AMSR-E EnKF Scaled AMSR-E SCAN Mean(-0.020981) Variance (1.14694) Figure 16: NOAH modeled (with and without Scaled AMSR-E assimilation) and observed (SCAN) soil moisture in the Goodwin CK Timber (ms_2025, 34.23˚N, 89.9˚W). 36 Impact of SCAN assimilation Open Loop SCAN EnKF SCAN Mean(-0.01867) Variance (1.06464) Figure17: NOAH modeled (with and without SCAN assimilation) and observed (SCAN) soil moisture in the Goodwin CK Timber (ms_2025, 34.23˚N, 89.9˚W). 37 RMSE (v/v%) 25 20 15 10 5 0 5 2 4 5 6 6 7 0 3 4 5 1 02 03 03 03 04 08 08 03 08 08 08 09 2 2 2 2 2 2 2 2 2 2 2 2 _ _ _ _ _ s_ s_ s_ s_ s_ s_ s_ ar ar ar ar ar m m m m m m m SCAN Sites Open Loop AMSR-E EnKF Scaled AMSR-E EnKF SCAN EnKF Figure 18: Mean RMS errors (RMSE) of Noah top layer soil moisture estimates computed against the SCAN observations for each of the 12 SCAN sites in the Open-Loop experiment (no DA), and three assimilation experiments, respectively. Units are v/v%. 38 Correlation 1 0.8 0.6 0.4 0.2 0 m 2 s_ 5 02 m 2 s_ 2 03 m 2 s_ 4 03 m 2 s_ 5 03 m 2 s_ 6 04 m 2 s_ 6 08 m 2 s_ 7 30 83 84 85 91 08 20 _20 _20 20 _20 _ _ ar ar ar ar ar SCAN Sites Open Loop AMSR-E EnKF Scaled AMSR-E EnKF SCAN EnKF Figure 19: Correlations of Noah top layer soil moisture estimates computed against the SCAN observations for each of the 12 SCAN sites in the Open-Loop experiment (no DA), and three assimilation experiments, respectively. Note negative correlation is not shown. (Some correlation bars with negative values are missing in some sites, e.g. ms_2035, ms_2087) 39