Search NASA⌕ Search

SEARCH · Search NASA

Results for “distributed estimation”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 631 records · Page 35

Solving inverse problems in stochastic models using deep neural networks and adversarial training

Inverse problems associated with stochastic models constitute a significant portion of scientific and engineering applications. In such cases the unknown quantities are distributions. The applicability of traditional methods is limited because of their demanding assumptions or prohibitive computational consumption; for example, maximum likelihood methods require closed-form density functions, and Markov Chain Monte Carlo needs a large number of simulations. We propose a new method that estimates the unknown distribution by matching the statistical properties between observed and simulated random processes. We leverage the expressive power of neural networks to approximate the unknown distribution and use a discriminative neural network for computing the statistical discrepancies between the observed and simulated random processes. Here we demonstrated numerically that the proposed methods can estimate both the model parameters and learn complicated unknown distributions.

42 ENGINEERING↗

Investigation of liquid cloud microphysical properties of deep convective systems: 1. Parameterization raindrop size distribution and its application for stratiform rain estimation

Abstract To investigate liquid‐phase ( T > 3°C) cloud and precipitation microphysical properties within Deep Convective Systems (DCSs), eight DCS cases sampled by the University of North Dakota Citation II research aircraft during Midlatitude Continental Convective Clouds Experiment were selected. A full spectrum of raindrop size distribution (DSD) was constructed from 120 µm to 4000 µm through a combination of two‐dimensional cloud probe (120 to 900 µm) and High Volume Precipitation Spectrometer (900 to 4000 µm) data sets. A total of 1126 five second DSDs have been used to fit to Gamma and Exponential functions within the stratiform rain (SR) regions of DCSs. The Gamma shape μ Γ and slope λ Γ parameters are then compared with those derived from surface disdrometer measurements. The similar μ Γ ‐ λ Γ relationships but different μ Γ and λ Γ value ranges from two independent platforms at different elevations may represent the real nature of DSD shape information in clouds and at the surface. To apply the exponentially fitted DSD parameters to precipitation estimation using Next Generation Weather Radar (NEXRAD) radar reflectivity factor Z e , the terms N 0 E and λ E have been parameterized as a function of Z e using an empirical N 0 E ‐ λ E relationship. The averaged SR rain rate retrieved from this study is almost identical to the surface measurements, while the NEXRAD Q2 precipitation is twice as large. The comparisons indicate that the new DSD parameterization scheme is robust, while the Q2 SR precipitation estimation based on Marshall‐Palmer Z‐R relationship, where a constant DSD intercept parameter ( N 0 E ) was assumed, needs to be improved for heavy precipitation cases.

Wang, Jingyu↗

The DECADE cosmic shear project II: photometric redshift calibration of the source galaxy sample

We present the photometric redshift characterization and calibration for the Dark Energy Camera All Data Everywhere (DECADE) weak lensing dataset: a catalog of 107 million galaxies observed by the Dark Energy Camera (DECam) in the northern Galactic cap. The redshifts are estimated from a combination of wide-field photometry, deep-field photometry with associated redshift estimates, and a transfer function between the wide field and deep field that is estimated using a source injection catalog. We construct four tomographic bins for the galaxy catalog, and estimate the redshift distribution, $n(z)$, within each one using the Self-organizing Map Photo-Z (SOMPZ) methodology. Our estimates include the contributions from sample variance, zeropoint calibration uncertainties, and redshift biases, as quantified for the deep-field dataset. The total uncertainties on the mean redshifts are $σ_{\langle}$$_z$$_{\rangle} ≈ 0.01$. The SOMPZ estimates are then compared to those from the clustering redshift method, obtained by cross-correlating our source galaxies with galaxies in spectroscopic surveys, and are shown to be consistent with each other.

Anbajagane, D. [Univ. of Chicago, IL (United State↗

Hydroacoustic Evaluation of Juvenile Salmonid Passage and Distribution at Detroit Dam, 2011

Pacific Northwest National Laboratory evaluated juvenile salmonid passage and distribution at Detroit Dam (DET) on the North Santiam River, Oregon for the U.S. Army Corps of Engineers (USACE) to provide data to support decisions on long-term measures to enhance downstream passage at DET and others dams in USACE’s Willamette Valley Project. This study was conducted in response to regulatory requirements necessitated by the listing of Upper Willamette River Spring Chinook salmon (Oncorhynchus tshawytscha) and Upper Willamette River steelhead (O. mykiss) as threatened under the Endangered Species Act. The goal of the study was to provide information of juvenile salmonid passage and distribution at DET from February 2011 through February 2012. The results of the hydroacoustic study provide new and, in some cases, first-ever data on passage estimates, run timing, distributions, and relationships between fish passage and environmental variables at the dam. This information will inform management decisions on the design and development of surface passage and collection devices to help restore Chinook salmon populations in the North Santiam River watershed above DET. During the entire study period, an estimated total of 182,526 smolt-size fish (±4,660 fish, 95% CI) passed through turbine penstock intakes. Run timing peaked in winter and early spring months. Passage rates were highest during late fall, winter and early spring months and low during summer. Horizontal distribution for hours when both turbine units were operated simultaneously indicated Unit 2 passed almost twice as much fish as Unit 1. Diel distribution for smolt-size fish during the study period was fairly uniform, indicating fish were passing the turbines at all times of the day. A total of 5,083 smolt-size fish (± 312 fish, 95% CI) were estimated passed via the spillway when it was open between June 23 and September 27, 2011. Daily passage was low at the spillway during the June-August period, and increased somewhat in September 2011. When the spillway was operated simultaneously with the turbines, spillway efficiency (efficiency is estimated as spillway passage divided by total project passage) was 0.72 and effectiveness (fish:flow ratio—proportion fish passage at a route (e.g., spillway) divided by proportion water through that route out of the total project) was 2.69. That is, when the spillway was open, 72% of the fish passing the dam used the spillway and 28% passed into the turbine penstocks. Diel distribution for smolt-size fish at the spillway shows a distinct peak in passage between mid-morning and mid-afternoon and low passage at night. We estimated that 23,339 smolt-size fish (± 572 fish, 95% CI) passed via the Regulating Outlet (RO) when it was open from October 29 through November 12, 2011, January 2-6, and January 20 through February 3, 2012. During the October–November period, RO passage peaked at 1,086 fish on November 5, with a second peak on November 7 (1,075 fish). When the RO was operated simultaneously with the turbines, RO efficiency was 0.33 and effectiveness was 0.89. In multiple regression analyses, a relatively parsimonious model was selected that predicted the observed fish passage data well. The best model included forebay temperature at depth, forebay elevation, total discharge, hours of daylight, and the operation period. The vertical distribution of fish in the forebay near the face of the dam where the transducers sampled showed fish were generally distributed throughout the water column during all four operational periods. During the refill and full pool periods, vertical distribution was bi-modal with surface-layer and mid-water modes. Patterns for day and night distributions were variable. Fish were distributed above and below the thermocline when it was present (full pool and drawdown periods).

Hydroacoustics↗

TU-H-CAMPUS-IeP1-05: A Framework for the Analytic Calculation of Patient-Specific Dose Distribution Due to CBCT Scan for IGRT

Purpose: To investigate the feasibility of an analytic framework to estimate patients’ absorbed dose distribution owing to daily cone-beam CT scan for image-guided radiation treatment. Methods: To compute total absorbed dose distribution, we separated the framework into primary and scattered dose calculations. Using the source parameters such as voltage, current, and bowtie filtration, for the primary dose calculation, we simulated the forward projection from the source to each voxel of an imaging object including some inhomogeneous inserts. Then we calculated the primary absorbed dose at each voxel based on the absorption probability deduced from the HU values and Beer’s law. In sequence, all voxels constructing the phantom were regarded as secondary sources to radiate scattered photons for scattered dose calculation. Details of forward projection were identical to that of the previous step. The secondary source intensities were given by using scatter-to- primary ratios provided by NIST. In addition, we compared the analytically calculated dose distribution with their Monte Carlo simulation results. Results: The suggested framework for absorbed dose estimation successfully provided the primary and secondary dose distributions of the phantom. Moreover, our analytic dose calculations and Monte Carlo calculations were well agreed each other even near the inhomogeneous inserts. Conclusion: This work indicated that our framework can be an effective monitor to estimate a patient’s exposure owing to cone-beam CT scan for image-guided radiation treatment. Therefore, we expected that the patient’s over-exposure during IGRT might be prevented by our framework.

60 APPLIED LIFE SCIENCES↗

THE IMPACT OF EVOLVING INFRARED SPECTRAL ENERGY DISTRIBUTIONS OF GALAXIES ON STAR FORMATION RATE ESTIMATES

We combine Herschel-Photodetector Array Camera and Spectrometer (PACS) data from the PACS Evolutionary Probe (PEP) program with Spitzer 24 {mu}m and 16 {mu}m photometry and ultra deep Infrared Spectrograph (IRS) mid-infrared spectra to measure the mid- to far-infrared spectral energy distribution (SED) of 0.7 < z < 2.5 normal star-forming galaxies (SFGs) around the main sequence (the redshift-dependent relation of star formation rate (SFR) and stellar mass). Our very deep data confirm from individual far-infrared detections that z {approx} 2 SFRs are overestimated if based on 24 {mu}m fluxes and SED templates that are calibrated via local trends with luminosity. Galaxies with similar ratios of rest-frame {nu}L{sub {nu}}(8) to 8-1000 {mu}m infrared luminosity (LIR) tend to lie along lines of constant offset from the main sequence. We explore the relation between SED shape and offset in specific star formation rate (SSFR) from the redshift-dependent main sequence. Main-sequence galaxies tend to have a similar {nu}L{sub {nu}}(8)/LIR regardless of LIR and redshift, up to z {approx} 2.5, and {nu}L{sub {nu}}(8)/LIR decreases with increasing offset above the main sequence in a consistent way at the studied redshifts. We provide a redshift-independent calibration of SED templates in the range of 8-60 {mu}m as a function of {Delta}log(SSFR) offset from the main sequence. Redshift dependency enters only through the evolution of the main sequence with time. Ultra deep IRS spectra match these SED trends well and verify that they are mostly due to a change in ratio of polycyclic aromatic hydrocarbon (PAH) to LIR rather than continua of hidden active galactic nuclei (AGNs). Alternatively, we discuss the dependence of {nu}L{sub {nu}}(8)/LIR on LIR. The same {nu}L{sub {nu}}(8)/LIR is reached at increasingly higher LIR at higher redshift, with shifts relative to local by 0.5 and 0.8 dex in log(LIR) at redshifts z {approx} 1 and z {approx} 2. Corresponding SED template calibrations are provided for use if no stellar masses are on hand. For most of those z {approx} 2 SFGs that also host an AGN, the mid-infrared is dominated by the star-forming component.

79 ASTRONOMY AND ASTROPHYSICS↗

A direct correlation of x-ray diffraction orientation distributions to the in-plane stiffness of semi-crystalline organic semiconducting films

Large charge mobilities of semi-crystalline organic semiconducting films could be obtained by mechanically aligning the material phases of the film with the loading axis. A key element is to utilize the inherent stiffness of the material for optimal or desired alignment. However, experimentally determining the moduli of semi-crystalline organic thin films for different loading directions is difficult, if not impossible, due to film thickness and material anisotropy. In this paper, we address these challenges by presenting an approach based on combining a composite mechanics stiffness orientation formulation with a Gaussian statistical distribution to directly estimate the in-plane stiffness (transverse isotropy) of aligned semi-crystalline polymer films based on crystalline orientation distributions obtained by X-ray diffraction experimentally at different applied strains. Our predicted results indicate that the in-plane stiffness of an annealing film was initially isotropic, and then it evolved to transverse isotropy with increasing mechanical strains. This study underscores the significance of accounting for the crystalline orientation distributions of the film to obtain an accurate understanding and prediction of the elastic anisotropy of semi-crystalline polymer films.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Bayesian network ensemble as a multivariate strategy to predict radiation pneumonitis risk

Purpose: Prediction of radiation pneumonitis (RP) has been shown to be challenging due to the involvement of a variety of factors including dose–volume metrics and radiosensitivity biomarkers. Some of these factors are highly correlated and might affect prediction results when combined. Bayesian network (BN) provides a probabilistic framework to represent variable dependencies in a directed acyclic graph. The aim of this study is to integrate the BN framework and a systems’ biology approach to detect possible interactions among RP risk factors and exploit these relationships to enhance both the understanding and prediction of RP. Methods: The authors studied 54 nonsmall-cell lung cancer patients who received curative 3D-conformal radiotherapy. Nineteen RP events were observed (common toxicity criteria for adverse events grade 2 or higher). Serum concentration of the following four candidate biomarkers were measured at baseline and midtreatment: alpha-2-macroglobulin, angiotensin converting enzyme (ACE), transforming growth factor, interleukin-6. Dose-volumetric and clinical parameters were also included as covariates. Feature selection was performed using a Markov blanket approach based on the Koller–Sahami filter. The Markov chain Monte Carlo technique estimated the posterior distribution of BN graphs built from the observed data of the selected variables and causality constraints. RP probability was estimated using a limited number of high posterior graphs (ensemble) and was averaged for the final RP estimate using Bayes’ rule. A resampling method based on bootstrapping was applied to model training and validation in order to control under- and overfit pitfalls. Results: RP prediction power of the BN ensemble approach reached its optimum at a size of 200. The optimized performance of the BN model recorded an area under the receiver operating characteristic curve (AUC) of 0.83, which was significantly higher than multivariate logistic regression (0.77), mean heart dose (0.69), and a pre-to-midtreatment change in ACE (0.66). When RP prediction was made only with pretreatment information, the AUC ranged from 0.76 to 0.81 depending on the ensemble size. Bootstrap validation of graph features in the ensemble quantified confidence of association between variables in the graphs where ten interactions were statistically significant. Conclusions: The presented BN methodology provides the flexibility to model hierarchical interactions between RP covariates, which is applied to probabilistic inference on RP. The authors’ preliminary results demonstrate that such framework combined with an ensemble method can possibly improve prediction of RP under real-life clinical circumstances such as missing data or treatment plan adaptation.

60 APPLIED LIFE SCIENCES↗

Testing Nesting DXTRAN Spheres for Neutron Transport in Air

The deterministic transport sphere (DXTRAN sphere) is a technology that can be utilized in MCNP for improving sampling in the vicinity of a particular geometry, when the solid angle, between the source and particular geometry, is small. At a collision event, a DXTRAN particle is created and deterministically transported to the DXTRAN sphere, while the colliding particle (NONDXTRAN particle) continues transporting as if no DXTRAN event had occurred. Weight is balanced by killing the NONDXTRAN particle when traversing the DXTRAN sphere. If set up correctly, the DXTRAN particle placed on the DXTRAN sphere has a high probability of interacting with the geometry within the DXTRAN sphere. The Central Limit Theorem (CLT) states that the estimated mean approaches a normal distribution with known variance as the number of random samples grows sufficiently large. In reality the variance is seldom known and therefore must be estimated. In a Monte Carlo calculation, one must determine when the number of random samples is sufficiently large for the CLT to be satisfied. For the CLT to be satisfied, the first two moments (mean and variance) must exist. The second moment, the variance of a normal distribution is estimated. The slope test, which uses the largest 201 history scores of each tally fluctuation chart (TFC) bin fit to a Pareto function, is used to determine if the largest history scores decrease faster than 1/x{sup 3}. If the slope is less than 3, then enough histories have not yet been sampled such to meet the second moment existence requirement of the CLT. Usually, a slope of less than 3 indicates that the effects of high weight scores have not been satisfactorily sampled; such a consequence can also be manifested in examining trends in the 4. moment, the variance of the variance (VOV). For example, when examining the TFC, if the VOV is monotonically decreasing for a few sets of histories and then drastically increases for the next set of histories, chances are a rare high weight score, significantly greater than the average score, was encountered. As a result of the 1/R{sup 2} term, a point detector can achieve an extremely high weight, much greater than the average, as the distance from collision to point detector approaches zero. To remedy this difficulty, a 'radius of exclusion' is applied, where within the radius of exclusion 'the point detector estimate is assumed to be the average flux uniformly distributed'. A DXTRAN sphere can suffer from the same weight fluctuation as collisions in the vicinity of the DXTRAN sphere produce much larger weights than collisions happening far from the DXTRAN sphere. If the rare large weight put on the DXTRAN sphere then scores in a tally, there will be a significant increase in the VOV and the slope may drop below 3. To compliment that analysis, here we utilize an air composition and density that varies as a function of altitude, detailing the trials and tribulations of attempting to pass the 10 statistical checks for a variety of single and nested DXTRAN sphere approaches. The objective here is to illuminate important considerations for radiation transport problems with small solid angles in a scatter medium of few mean free paths (MFPs)

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Modeling the dark current histogram induced by gold contamination in complementary-metal-oxide-semiconductor image sensors

We propose a model for dark current induced by metallic contamination in a CMOS image sensor. Based on Shockley-Read-Hall kinetics, the expression of dark current proposed accounts for the electric field enhanced emission factor due to the Poole-Frenkel barrier lowering and phonon-assisted tunneling mechanisms. To that aim, we considered the distribution of the electric field magnitude and metal atoms in the depth of the pixel. Poisson statistics were used to estimate the random distribution of metal atoms in each pixel for a given contamination dose. Then, we performed a Monte-Carlo-based simulation for each pixel to set the number of metal atoms the pixel contained and the enhancement factor each atom underwent, and obtained a histogram of the number of pixels versus dark current for the full sensor. Excellent agreement with the dark current histogram measured on an ion-implanted gold-contaminated imager has been achieved, in particular, for the description of the distribution tails due to the pixel regions in which the contaminant atoms undergo a large electric field. The agreement remains very good when increasing the temperature by 15 °C. We demonstrated that the amplification of the dark current generated for the typical electric fields encountered in the CMOS image sensors, which depends on the nature of the metal contaminant, may become very large at high electric field. The electron and hole emissions and the resulting enhancement factor are described as a function of the trap characteristics, electric field, and temperature.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Reinforcement Learning for Load-balanced Parallel Particle Tracing

We explore an online reinforcement learning (RL) paradigm to dynamically optimize parallel particle tracing performance in distributed-memory systems. Our method combines three novel components: (1) a work donation algorithm, (2) a high-order workload estimation model, and (3) a communication cost model. First, we design an RL-based work donation algorithm. Our algorithm monitors workloads of processes and creates RL agents to donate data blocks and particles from high-workload processes to low-workload processes to minimize program execution time. The agents learn the donation strategy on the fly based on reward and cost functions designed to consider processes' workload changes and data transfer costs of donation actions. Second, we propose a workload estimation model, helping RL agents estimate the workload distribution of processes in future computations. Third, we design a communication cost model that considers both block and particle data exchange costs, helping RL agents make effective decisions with minimized communication costs. We demonstrate that our algorithm adapts to different flow behaviors in large-scale fluid dynamics, ocean, and weather simulation data. Our algorithm improves parallel particle tracing performance in terms of parallel efficiency, load balance, and costs of I/O and communication for evaluations with up to 16,384 processors.

Distributed and parallel particle tracing↗

Time domain-induced polarization geophysical data collected in the Rifle Floodplain, Colorado, USA

The IP data set was collected using a time domain-induced polarization (TDIP) method along 65 profiles of various lengths over the floodplain. Surface time domain?induced polarization (TDIP) data was collected to create the 3?D images of the complex electrical resistivity, in terms of magnitude and phase, which are associated with mineral precipitation and other lithological properties. The TDIP data was then used to estimate the spatial distribution of naturally reduced zones, which are known to be biogeochemical hotspots in floodplains having increased uranium and other redox-sensitive metal concentrations as well as high carbon contents. In the TDIP method, the transient decay of voltage is measured after current shut-off, typically in the form of an integral of decay curves over a predefined time window (so-called integral chargeability). TDIP measurements at the site were collected using the Syscal Iris Pro Switch equipment with a square-wave current injection, 50% duty cycle, and a pulse length of 2 s. The integral chargeability measurements were carried out using 20 windows during voltage decay between 240 and 1840 ms after current shut-off. Tomographic measurements were collected by deploying stainless steel electrodes with an electrode separation of 1.8 m and using a dipole-dipole ‘‘skip-2’’ and ‘‘skip-3’’ measuring protocol (i.e., for a dipoles length of 5.4 and 7.2 m, respectively). The sequence of dipole-dipole measurements was carefully arranged to (1) minimize unwanted electromagnetic coupling effects in the data, avoiding potential measurements with electrodes located inside the current dipole, (2) prevent voltage measurements using electrodes, which might be polarized due to previous current injection, and (3) increase the signal-to-noise ratio for an intended exploration depth of 8 m, i.e., the bottom of the aquifer. All measurements were collected as normal and reciprocal pairs for estimation of the data error. The IP measurements were collected with symmetric arrays (i.e., the measuring equipment placed at the center of the electrode array) with a maximum of 36 electrodes, considering that longer profiles revealed a significant increase in the normal-reciprocal misfit for the measurements of the decay curve, probably due to greater impact of electromagnetic coupling on the data. The TDIP data sets were inverted in a twodimensional domain along each transect using CRTomo, which is a smoothness-constraint inversion code based on a finite element algorithm. The resistivity and phase shift values at each pixel were then assigned at the corresponding point within the 3-D floodplain domain (the black rectangle Figure 1b) and used in the 3-D estimation. The TDIP inversion results provided the distribution of the complex resistivity, expressed in terms of its magnitude and phase-shift. The analysis of the normal-reciprocal misfit was used to estimate the relative error. We removed the measurement with smallest voltage difference (<2 mV), representing about 2% variations in the data. For the inversion of TDIP measurements, chargeability values were linearly converted to frequency domain phase values (at the fundamental frequency of 0.125 Hz), by assuming a constant-phase response. This approach has been demonstrated to provide consistent results in previous studies

54 ENVIRONMENTAL SCIENCES↗

Clutter identification based on kernel density estimation and sparse recovery

A cognitive radar framework is being developed to dynamically detect changes in the clutter characteristics, and to adapt to these changes by identifying the new clutter distribution. In our previous work, we have presented a sparse-recovery based clutter identification technique. In this technique, each column of the dictionary represents a specific distribution. More specifically, calibration radar clutter data corresponding to a specific distribution is transformed into a distribution through kernel density estimation. When the new batch of radar data arrives, the new data is transformed to a distribution through the same kernel density estimation method and its distribution characteristics is identified through sparse-recovery. In this paper, we extend our previous work to consider different kernels and kernel parameters for sparse-recovery-based clutter identification and the numerical results are presented as well. The impact of different kernels and kernel parameters are analyzed by comparing the identification accuracy of each scenario.

Wang, Haokun↗

Out-of-distribution detection with non-parametric density estimation for models predicting processing history of uranium ore concentrates

The rapid advancement in machine learning (ML) and computer vision (CV) coincides with the growth of interest in deploying these ML/CV models in numerous fields from medicine to social science. Similar to those areas, we have witnessed a great number of works in materials science employing ML/CV models – neural networks in particular – in their studies in recent years. These models have proven to obtain accurate performance in various tasks. However, these models struggle to attain a similar performance when encountering test samples coming from a distribution that is different from the training set. More importantly, they fail without providing any warning to the users. Therefore, we propose a framework for detecting out-of-distribution (OOD) samples to alert users when a human intervention might be necessary in this work. Specifically, we explore the use of a non-parametric density estimation method to detect OOD samples. Here, we assess OOD detection capability of the proposed framework on ML models developed for categorizing precipitation routes of U 3 O 8 when encountering OOD datasets that contain samples (1) undergone different imaging acquisition process, (2) undergone different material synthesis process, and (3) different materials than ID set. Through those experiments, we achieve an average area under the receiver operating characteristic (AUROC) of at least 91% on average in detecting OOD samples. With minimal overhead cost and superior performance, the proposed framework enables a reliable and safe system when deploying in real-world scenarios.

Convolutional neural networks↗

Estimating Watershed Subsurface Permeability From Stream Discharge Data Using Deep Neural Networks

Subsurface permeability is a key parameter in watershed models that controls the contribution from the subsurface flow to stream flows. Since the permeability is difficult and expensive to measure directly at the spatial extent and resolution required by fully distributed watershed models, estimation through inverse modeling has had a long history in subsurface hydrology. The wide availability of stream surface flow data, compared to groundwater monitoring data, provides a new data source to infer soil and geologic properties using integrated surface and subsurface hydrologic models. As most of the existing methods have shown difficulty in dealing with highly nonlinear inverse problems, we explore the use of deep neural networks for inversion owing to their successes in mapping complex, highly nonlinear relationships. We train various deep neural network (DNN) models with different architectures to predict subsurface permeability from stream discharge hydrograph at the watershed outlet. The training data are obtained from ensemble simulations of hydrographs corresponding to an permeability ensemble using a fully-distributed, integrated surface-subsurface hydrologic model. The trained model is then applied to estimate the permeability of the real watershed using its observed hydrograph at the outlet. Our study demonstrates that the permeabilities of the soil and geologic facies that make significant contributions to the outlet discharge can be more accurately estimated from the discharge data. Their estimations are also more robust with observation errors. Compared to the traditional ensemble smoother method, DNNs show stronger performance in capturing the nonlinear relationship between permeability and stream hydrograph to accurately estimate permeability. Our study sheds new light on the value of the emerging deep learning methods in assisting integrated watershed modeling by improving parameter estimation, which will eventually reduce the uncertainty in predictive watershed models.

54 ENVIRONMENTAL SCIENCES↗

Cropland carbon fluxes in the United States: increasing Geospatial Resolution of Inventory-Based Carbon Accounting

Net annual soil carbon change, fossil fuel emissions from cropland production, and cropland net primary productivity were estimated and spatially distributed using land cover defined by the Moderate Resolution Imaging Spectroradiometer (MODIS) and by the Cropland Data Layer (CDL). Spatially resolved estimates of net ecosystem exchange (NEE) and net ecosystem carbon balance (NECB) were developed. NEE represents net on-site vertical fluxes of carbon. NECB represents all on-site and off-site carbon fluxes associated with crop production. Estimates of cropland NEE using moderate resolution (~1km2) land cover data were generated for the conterminous US and compared with higher resolution (30m) estimates of NEE and with direct measurements of CO2 flux from croplands in Illinois and Nebraska. Estimates of NEE using the CDL (30m resolution) had a higher correlation with eddy covariance flux tower estimates compared with estimates of NEE using MODIS. Estimates of NECB are primarily driven by net soil carbon change, fossil-fuel emissions associated with crop production, and CO2 emissions from the application of agricultural lime. NEE and NECB for US croplands were -274 and 7 Tg C yr-1 for 2004, respectively. Use of moderate to high resolution satellite-based land cover data enables improved estimates of cropland carbon dynamics.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗

A posteriori error estimators for the discrete ordinates approximation of the one-speed neutron transport equation

When calculating numerical solutions of the neutron transport equation it is important to have a measure of the accuracy of the solution. As the true solution is generally not known, a suitable estimation of the error must be made. The steady state transport equation possesses discretization errors in all its independent variables: angle, energy and space. In this work only spatial discretization errors are considered. An exact transport solution, in which the degree of regularity of the exact flux across the singular characteristic is controlled, is manufactured to determine the numerical solutions true discretization error. This solution is then projected onto a Legendre polynomial space in order to form an exact solution on the same basis space as the numerical solution, Discontinuous Galerkin Finite Element Method (DGFEM), to enable computation of the true error. Over a series of test problems the true error is compared to the error estimated by: Ragusa and Wang (RW), residual source (LER) and cell discontinuity estimators (JD). The validity and accuracy of the considered estimators are primarily assessed by considering the effectivity index and global L2 norm of the error. In general RW excels at approximating the true error distribution but usually under-estimates its magnitude; the LER estimator emulates the true error distribution but frequently over-estimates the magnitude of the true error; the JD estimator poorly captures the true error distribution and generally under-estimates the error about singular characteristics but over-estimates it elsewhere. (authors)

97 MATHEMATICS AND COMPUTING↗

Two-level Clustering-based Target Detection Through Sensor Deployment and Data Fusion

Target detection is one fundamental problem in many sensor network-based applications, which is typically tackled in two separate stages for sensor deployment and data fusion. We propose an integrated solution, referred to as 2C-SSEM,which combines 2-level clustering-based sensor deployment and source strength estimate map-based data fusion for the detection of a single static or moving target. 2C-SSEM conducts the first level of clustering to determine a sensor deployment scheme and the second level of clustering to divide the deployed sensors into several subsets. For each sensor, the source strength is estimated at each grid point of the entire region based on a signal attenuation model, and for each subset of sensors, the target location is estimated using a distribution map-based statistical analysis method. A final detection decision is made by thresholding the clustering degree of the target location estimates computed by all subsets of sensors. Compared with traditional grid-based target detection methods, 2C-SSEM significantly reduces the computation complexity and improves the detection performance through an integrated optimization strategy. Extensive simulation results show the performance superiority of the proposed solution over several well-known methods for target detection.

Rao, Nageswara↗