Search NASA⌕ Search

Engineering topics

Forman, Barton A.

Publications and source records attributed to Forman, Barton A..

8 records
All sourcesNASA NTRSDOE OSTI
NASA NTRS2019

Exploring the Utility of Machine Learning-Based Passive Microwave Brightness Temperature Data Assimilation over Terrestrial Snow in High Mountain Asia

This study explores the use of a support vector machine (SVM) as the observation operator within a passive microwave brightness temperature data assimilation framework (herein SVM-DA) to enhance the characterization of snow water equivalent (SWE) over High Mountain Asia (HMA). A series of synthetic twin experiments were conducted with the NASA Land Information System (LIS) at a number of locations across HMA. Overall, the SVM-DA framework is effective at improving SWE estimates (~70% reduction in RMSE relative to the Open Loop) for SWE depths less than 200 mm during dry snowpack conditions. The SVM-DA framework also improves SWE estimates in deep, wet snow (~45% reduction in RMSE) when snow liquid water is well estimated by the land surface model, but can lead to model degradation when snow liquid water estimates diverge from values used during SVM training. In particular, two key challenges of using the SVM-DA framework were observed over deep, wet snowpacks. First, variations in snow liquid water content dominate the brightness temperature spectral difference (TB) signal associated with emission from a wet snowpack, which can lead to abrupt changes in SWE during the analysis update. Second, the ensemble of SVM-based predictions can collapse (i.e., yield a near-zero standard deviation across the ensemble) when prior estimates of snow are outside the range of snow inputs used during the SVM training procedure. Such a scenario can lead to the presence of spurious error correlations between SWE and TB, and as a consequence, can result in degraded SWE estimates from the analysis update. These degraded analysis updates can be largely mitigated by applying rule-based approaches. For example, restricting the SWE update when the standard deviation of the predicted TB is greater than 0.05 K helps prevent the occurrence of filter divergence. Similarly, adding a thin layer (i.e., 5 mm) of SWE when the synthetic TB is larger than 5 K can improve SVM-DA performance in the presence of a precipitation dry bias. The study demonstrates that a carefully constructed SVM-DA framework cognizant of the inherent limitations of passive microwave-based SWE estimation holds promise for snow mass data assimilation.

Kwon, Yonghwan↗
NASA NTRS2019

Analyzing Machine Learning Predictions of Passive Microwave Brightness Temperature Spectral Difference Over Snow-Covered Terrain in High Mountain Asia

Snow is an important component of the terrestrial freshwater budget in high mountainAsia (HMA) and contributes to the runoff in Himalayan rivers through snowmelt. Despitethe importance of snow in HMA, considerable spatiotemporal uncertainty exists across the different estimates of snow water equivalent for this region. In order to better estimate snow water equivalent, radiative transfer models are often used in conjunction with microwave brightness temperature measurements. In this study, the efficacy of support vector machines (SVMs), a machine learning technique, to predict passive microwave brightness temperature spectral difference (1Tb) as a function of geophysical variables (snow water equivalent, snow depth, snow temperature, and snow density) is explored through a sensitivity analysis. The use of machine learning (as opposed to radiative transfer models) is a relatively new and novel approach for improving snow water equivalent estimates. The Noah-MP land surface model within the NASALand Information System framework is used to simulate the hydrologic cycle over HMA and model geophysical variables that are then used for SVM training. The SVMsserve as a nonlinear map between the geophysical space (modeled in Noah-MP) andthe observation space (1Tb as measured by the radiometer). Advanced MicrowaveScanning Radiometer-Earth Observing System measured passive microwave brightness temperatures over snow-covered locations in the HMA region are used as training data during the SVM training phase. Sensitivity of well-trained SVMs to each Noah-MP modeled state variable is assessed by computing normalized sensitivity coefficients. Sensitivity analysis results generally conform with the known first-order physics. Input states that increase volume scattering of microwave radiation, such as snow density and snow water equivalent, exhibit a plurality of positive normalized sensitivity coefficients. In general, snow temperature was the most sensitive input to the SVM predictions. The sensitivity of each state is location and time dependent. The signs of normalized sensitivity coefficients that indicate physical irrationality are ascribed to significant cross-correlation between Noah-MP simulated states and decreased SVM prediction capability at specific locations due to insufficient training data. SVM prediction pitfalls do exist that serve to highlight the limitations of this particular machine learning algorithm.

high mountain Asia↗
NASA NTRS2018

Assimilation of Multi-Frequency, Multi-Polarization Passive Microwave Brightness Temperature Observations in North America over Snow-Covered Regions Using Support Vector Machines

Accurately estimating the mass of water within a snowpack (a.k.a. snow water equivalent, or SWE) across regional or continental scales is a challenge. In order to overcome some of the limitations in traditional SWE retrieval algorithms or radiative transfer-based snow emission models, this study explores the use of a support vector machine (SVM) to merge an advanced land surface model within a radiance emission (i.e., brightness temperature) assimilation framework. The goal of direct radiance assimilation is preferable as it avoids inconsistencies in the use of ancillary data between the assimilation system and the independently-generated geophysical retrieval. The impact of assimilating multiple observations simultaneously at different frequency and polarization combinations is then evaluated via comparisons to state-of-the-art SWE and snow depth products as well as available ground-based measurements across North America for the years 2002 through 2011. It is found that assimilation-derived estimates (relative to estimates without assimilation) tend to better agree with state-of-the-art snow products. In addition, an overall improvement in goodness-of-fit statistics for snow estimates is achieved via assimilation when compared against ground-based snow measurements. In addition, these improvements in snow are shown to translate into improvements in streamflow predictions. Specifically, 11 out of the 13 major snow-dominated basins investigated have improved cumulative runoff estimates versus ground-based discharge measurements compared to the no-assimilation scenario. It is proven that a SVM can serve as an efficient and effective observation operator for a snow mass analysis within a radiance assimilation system.

Xue, Yuan↗
NASA NTRS2014

Using a Support Vector Machine and a Land Surface Model to Estimate Large-Scale Passive Microwave Temperatures over Snow-Covered Land in North America

A support vector machine (SVM), a machine learning technique developed from statistical learning theory, is employed for the purpose of estimating passive microwave (PMW) brightness temperatures over snow-covered land in North America as observed by the Advanced Microwave Scanning Radiometer (AMSR-E) satellite sensor. The capability of the trained SVM is compared relative to the artificial neural network (ANN) estimates originally presented in [14]. The results suggest the SVM outperforms the ANN at 10.65 GHz, 18.7 GHz, and 36.5 GHz for both vertically and horizontally-polarized PMW radiation. When compared against daily AMSR-E measurements not used during the training procedure and subsequently averaged across the North American domain over the 9-year study period, the root mean squared error in the SVM output is 8 K or less while the anomaly correlation coefficient is 0.7 or greater. When compared relative to the results from the ANN at any of the six frequency and polarization combinations tested, the root mean squared error was reduced by more than 18 percent while the anomaly correlation coefficient was increased by more than 52 percent. Further, the temporal and spatial variability in the modeled brightness temperatures via the SVM more closely agrees with that found in the original AMSR-E measurements. These findings suggest the SVM is a superior alternative to the ANN for eventual use as a measurement operator within a data assimilation framework.

Modeling↗
NASA NTRS2013

Connecting Satellite Observations with Water Cycle Variables Through Land Data Assimilation: Examples Using the NASA GEOS-5 LDAS

A land data assimilation system (LDAS) can merge satellite observations (or retrievals) of land surface hydrological conditions, including soil moisture, snow, and terrestrial water storage (TWS), into a numerical model of land surface processes. In theory, the output from such a system is superior to estimates based on the observations or the model alone, thereby enhancing our ability to understand, monitor, and predict key elements of the terrestrial water cycle. In practice, however, satellite observations do not correspond directly to the water cycle variables of interest. The present paper addresses various aspects of this seeming mismatch using examples drawn from recent research with the ensemble-based NASA GEOS-5 LDAS. These aspects include (1) the assimilation of coarse-scale observations into higher-resolution land surface models, (2) the partitioning of satellite observations (such as TWS retrievals) into their constituent water cycle components, (3) the forward modeling of microwave brightness temperatures over land for radiance-based soil moisture and snow assimilation, and (4) the selection of the most relevant types of observations for the analysis of a specific water cycle variable that is not observed (such as root zone soil moisture). The solution to these challenges involves the careful construction of an observation operator that maps from the land surface model variables of interest to the space of the assimilated observations.

Land data assimilation↗
NASA NTRS2012

Assessment and Enhancement of MERRA Land Surface Hydrology Estimates

The Modern-Era Retrospective analysis for Research and Applications (MERRA) is a state-ofthe-art reanalysis that provides, in addition to atmospheric fields, global estimates of soil moisture, latent heat flux, snow, and runoff for 1979-present. This study introduces a supplemental and improved set of land surface hydrological fields ("MERRA-Land") generated by re-running a revised version of the land component of the MERRA system. Specifically, the MERRA-Land estimates benefit from corrections to the precipitation forcing with the Global Precipitation Climatology Project pentad product (version 2.1) and from revised parameter values in the rainfall interception model, changes that effectively correct for known limitations in the MERRA surface meteorological forcings. The skill (defined as the correlation coefficient of the anomaly time series) in land surface hydrological fields from MERRA and MERRA-Land is assessed here against observations and compared to the skill of the state-of-the-art ERA-Interim (ERA-I) reanalysis. MERRA-Land and ERA-I root zone soil moisture skills (against in situ observations at 85 US stations) are comparable and significantly greater than that of MERRA. Throughout the northern hemisphere, MERRA and MERRA-Land agree reasonably well with in situ snow depth measurements (from 583 stations) and with snow water equivalent from an independent analysis. Runoff skill (against naturalized stream flow observations from 18 US basins) of MERRA and MERRA-Land is typically higher than that of ERA-I. With a few exceptions, the MERRA-Land data appear more accurate than the original MERRA estimates and are thus recommended for those interested in using MERRA output for land surface hydrological studies.

Reichle, Rolf H.↗
NASA NTRS2011

Assimilation of Terrestrial Water Storage from GRACE in a Snow-Dominated Basin

Terrestrial water storage (TWS) information derived from Gravity Recovery and Climate Experiment (GRACE) measurements is assimilated into a land surface model over the Mackenzie River basin located in northwest Canada. Assimilation is conducted using an ensemble Kalman smoother (EnKS). Model estimates with and without assimilation are compared against independent observational data sets of snow water equivalent (SWE) and runoff. For SWE, modest improvements in mean difference (MD) and root mean squared difference (RMSD) are achieved as a result of the assimilation. No significant differences in temporal correlations of SWE resulted. Runoff statistics of MD remain relatively unchanged while RMSD statistics, in general, are improved in most of the sub-basins. Temporal correlations are degraded within the most upstream sub-basin, but are, in general, improved at the downstream locations, which are more representative of an integrated basin response. GRACE assimilation using an EnKS offers improvements in hydrologic state/flux estimation, though comparisons with observed runoff would be enhanced by the use of river routing and lake storage routines within the prognostic land surface model. Further, GRACE hydrology products would benefit from the inclusion of better constrained models of post-glacial rebound, which significantly affects GRACE estimates of interannual hydrologic variability in the Mackenzie River basin.

Forman, Barton A.↗