Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “error indicators”

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 667 records · Page 37

Quantum mechanics capacitance molecular mechanics modeling of core-electron binding energies of methanol and methyl nitrite on Ag(111) surface

We study a newly devised quantum mechanics capacitance molecular mechanics (QMCMM) method for the calculation of core-electron binding energies in the case of molecules adsorbed on metal surfaces. This yet untested methodology is applied to systems with monolayer of methanol/methyl nitrite on an Ag(111) surface at 100 K temperature. It was found out that the studied C, N, and O 1s core-hole energies converge very slowly as a function of the radius of the metallic cluster, which was ascribed to build up of positive charge on the edge of the Ag slab. Further analysis revealed that an extrapolation process can be used to obtain binding energies that deviated less than 0.5 eV against experiments, except in the case of methanol O 1s where the difference was as large as 1.8 eV. Additional QM-cluster calculations suggest that the latter error can be connected to the lack of charge transfer over the QM–CMM boundary. Thus, the results indicate that the QMCMM and QM-cluster methods can complement each other in a holistic picture of molecule-adsorbate core-ionization studies, where all types of intermolecular interactions are considered.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

ALFALFA DISCOVERY OF THE NEARBY GAS-RICH DWARF GALAXY LEO P. I. H I OBSERVATIONS

The discovery of a previously unknown 21 cm H I line source identified as an ultra-compact high velocity cloud in the ALFALFA survey is reported. The H I detection is barely resolved by the Arecibo 305 m telescope {approx}4' beam and has a narrow H I linewidth (half-power full width of 24 km s{sup -1}). Further H I observations at Arecibo and with the Very Large Array corroborate the ALFALFA H I detection, provide an estimate of the H I radius, {approx}1' at the 5 Multiplication-Sign 10{sup 19} cm{sup -2} isophote, and show the cloud to exhibit a velocity field which, if interpreted as disk rotation, has an amplitude of {approx_equal}9.0 {+-} 1.5 km s{sup -1}. In other papers, Rhode et al. show the H I source to have a resolved stellar counterpart and ongoing star forming activity, while Skillman et al. reveal it as having extremely low metallicity: 12 + log (O/H) = 7.16 {+-} 0.04. The H I mass to stellar mass ratio of the object is found to be 2.6. We use the Tully-Fisher template relation in its baryonic form to obtain a distance estimate D{sub Mpc}=1.3{sup +0.9}{sub -0.5}. Additional constraints on the distance are also provided by the optical data of Rhode et al. and McQuinn et al., both indicating a distance in the range of 1.5 to 2.0 Mpc. The three estimates are compatible within their errors. The object appears to be located beyond the dynamical boundaries of, but still in close proximity to the Local Group. Its pristine properties are consistent with the sedate environment of its location. At a nominal distance of 1.75 Mpc, it would have an H I mass of {approx_equal} 1.0 Multiplication-Sign 10{sup 6} M {sub Sun }, a stellar mass of {approx_equal} 3.6 Multiplication-Sign 10{sup 5} M {sub Sun }, and a dynamical mass within the H I radius of {approx_equal} 1.5 Multiplication-Sign 10{sup 7} M {sub Sun }. This discovery supports the idea that optically faint-or altogether dark-low mass halos may be detectable through their non-stellar baryons.

79 ASTRONOMY AND ASTROPHYSICS↗

EXPLOSIVE OUTFLOWS POWERED BY THE DECAY OF NON-HIERARCHICAL MULTIPLE SYSTEMS OF MASSIVE STARS: ORION BN/KL

The explosive Becklin-Neugebauer (BN)/Kleinman-Low (KL) outflow emerging from OMC1 behind the Orion Nebula may have been powered by the dynamical decay of a non-hierarchical multiple system {approx}500 years ago that ejected the massive stars I, BN, and source n, with velocities of about 10-30 km s{sup -1}. New proper-motion measurements of H{sub 2} features show that within the errors of measurement, the outflow originated from the site of stellar ejection. Combined with published data, these measurements indicate an outflow age of {approx}500 years, similar to the time since stellar ejection. The total kinetic energy of the ejected stars and the outflow is about 2 to 6 x 10{sup 47} erg. It is proposed that the gravitational potential energy released by the formation of a short-period binary, most likely source I, resulted in stellar ejection and powered the outflow. A scenario is presented for the formation of a compact, non-hierarchical multiple star system, its decay into an ejected binary and two high-velocity stars, and launch of the outflow. Three mechanisms may have contributed to the explosion in the gas: (1) unbinding of the circumcluster envelope following stellar ejection, (2) disruption of circumstellar disks and high-speed expulsion of the resulting debris during the final stellar encounter, and (3) the release of stored magnetic energy. Plausible protostellar disk end envelope properties can produce the observed outflow mass, velocity, and kinetic energy distributions. The ejected stars may have acquired new disks by fall-back or Bondi-Hoyle accretion with axes roughly orthogonal to their velocities. The expulsion of gas and stars from OMC1 may have been driven by stellar interactions.

79 ASTRONOMY AND ASTROPHYSICS↗

Improving seasonal precipitation forecasts in the Western United States through statistical downscaling

Abstract Seasonal precipitation forecasts in the western United States are critical resources for water resource management, especially during winter. While current seasonal forecasting systems provide monthly precipitation forecasts operationally, their coarse resolution limits their effectiveness in capturing the localized precipitation patterns and snowpack conditions essential for water resource managers in the mountainous regions. Here, analog statistical downscaling is demonstrated as an effective approach to enhance the spatial resolution of operational seasonal forecasts provided by the North American Multi-Model Ensemble. Downscaling was performed by building an analog ‘library’, in which corresponding model forecasts and observed values during the training period were stored. In the testing period, unseen model forecasts referenced the closest historical forecast from the analog library and applied the corresponding observational value for each point. This analysis indicates that downscaled products can capture localized features more accurately than the original coarse resolution forecasts, reducing forecast error across the western United States. Moreover, downscaling individual ensemble members—rather than downscaling the ensemble mean—further reduces forecasting error for their multi-model ensemble mean products. The greatest error reductions in the downscaled product, measured by root mean squared error (RMSE), were observed at low to mid-elevations (500–2000 meters), with 50%–70% improvement relative to the original forecasts. In the higher elevations (2000 meters and above), changes in RMSE relative to the original forecast were limited to 10%–30% improvements. The improvement is more substantial for forecast systems with 10 ensemble members compared to that with 4 members, but this relationship does not hold for the system with 24 ensemble members. These findings show that analog statistical downscaling can effectively address the spatial limitations of seasonal precipitation forecasts with minimal computational cost, providing a valuable framework for enhancing coarse resolution forecasting products while providing insights into the timing of ensemble mean calculations during the downscaling process.

Vernon, B. (ORCID:0009000891670689)↗

Predicting images for the dynamics of stellar clusters ( π-DOC ): a deep learning framework to predict mass, distance, and age of globular clusters

ABSTRACT Dynamical mass estimates of simple systems such as globular clusters (GCs) still suffer from up to a factor of 2 uncertainty. This is primarily due to the oversimplifications of standard dynamical models that often neglect the effects of the long-term evolution of GCs. Here, we introduce a new approach to measure the dynamical properties of GCs, based on the combination of a deep-learning framework and the large amount of data from direct N-body simulations. Our algorithm, π-DOC (Predicting Images for the Dynamics Of stellar Clusters) is composed of two convolutional networks, trained to learn the non-trivial transformation between an observed GC luminosity map and its associated mass distribution, age, and distance. The training set is made of V-band luminosity and mass maps constructed as mock observations from N-body simulations. The tests on π-DOC demonstrate that we can predict the mass distribution with a mean error per pixel of 27 per cent, and the age and distance with an accuracy of 1.5 Gyr and 6 kpc, respectively. In turn, we recover the shape of the mass-to-light profile and its global value with a mean error of 12 per cent, which implies that we efficiently trace mass segregation. A preliminary comparison with observations indicates that our algorithm is able to predict the dynamical properties of GCs within the limits of the training set. These encouraging results demonstrate that our deep-learning framework and its forward modelling approach can offer a rapid and adaptable tool competitive with standard dynamical models.

Chardin, Jonathan↗

Distributed State Estimation Over Time-Varying Graphs: Exploiting the Age-of-Information

Here, we study the problem of designing a distributed observer for an LTI system over a time-varying communication graph. The limited existing work on this topic imposes various restrictions either on the observation model or on the sequence of communication graphs. In contrast, we propose a single-time-scale distributed observer that works under mild assumptions. Specifically, our communication model only requires strong-connectivity to be preserved over non-overlapping, contiguous intervals that are even allowed to grow unbounded over time. We show that under suitable conditions that bound the growth of such intervals, joint observability is sufficient to track the state of any discrete-time LTI system exponentially fast, at any desired rate. We also develop a variant of our algorithm that is provably robust to worst-case adversarial attacks, provided the sequence of graphs is sufficiently connected over time. The key to our approach is the notion of a "freshness-index" that keeps track of the age-of-information being diffused across the network. Such indices enable nodes to reject stale estimates of the state, and, in turn, contribute to stability of the error dynamics.

42 ENGINEERING↗

WE-H-207A-07: Image-Based Versus Atlas-Based Internal Dosimetry

Purpose: Monte Carlo (MC) simulation is known as the gold standard method for internal dosimetry. It requires radionuclide distribution from PET or SPECT and body structure from CT for accurate dose calculation. The manual or semi-automatic segmentation of organs from CT images is a major obstacle. The aim of this study is to compare the dosimetry results based on patient’s own CT and a digital humanoid phantom as an atlas with pre-specified organs. Methods: SPECT-CT images of a 50 year old woman who underwent bone pain palliation with Samarium-153 EDTMP for osseous metastases from breast cancer were used. The anatomical date and attenuation map were extracted from SPECT/CT and three XCAT digital phantoms with different BMIs (i.e. matched (38.8) and unmatched (35.5 and 36.7) with patient’s BMI that was 38.3). Segmentation of patient’s organs in CT image was performed using itk-SNAP software. GATE MC Simulator was used for dose calculation. Specific absorbed fractions (SAFs) and S-values were calculated for the segmented organs. Results: The differences between SAFs and S-values are high using different anatomical data and range from −13% to 39% for SAF values and −109% to 79% for S-values in different organs. In the spine, the clinically important target organ for Samarium Therapy, the differences in the S-values and SAF values are higher between XCAT phantom and CT when the phantom with identical BMI is employed (53.8% relative difference in S-value and 26.8% difference in SAF). However, the whole body dose values were the same between the calculations based on the CT and XCAT with different BMIs. Conclusion: The results indicated that atlas-based dosimetry using XCAT phantom even with matched BMI for patient leads to considerable errors as compared to image-based dosimetry that uses the patient’s own CT Patient-specific dosimetry using CT image is essential for accurate results.

60 APPLIED LIFE SCIENCES↗

Extension of the PINN diffusion model to k-eigenvalue problems

This paper extends our recent work on the Physics-Informed Neural Networks (PINN) approach for the fixed source diffusion models and applies it to the diffusion theory based k-eigenvalue problems. To make the PINN equitable for the eigenvalue problems, we introduce a novel integral regularization term to the loss function in the framework, and allow the direct inference of the principal eigenvalue and the associated eigenfunction. The regularization term enforces a pre-defined value on the integration of the model predictions, and this value can be directly related to a physical property of the system. We also introduce an additional learnable parameter to approximate the principal eigenvalue. As a proof of principle, we solve the one-group two-dimensional k-eigenvalue neutron diffusion equation in this work. We then provide two numerical examples to demonstrate the applicability of the PINN approach. In each example, we solve the k-eigenvalue diffusion equation in a multi-region configuration constrained with a set of Robin boundary conditions for generality. We use a FEM solution based on the power-iteration method to verify the results of the PINN solution. The results showed relative percentage error in the predicted eigenvalue of about 0.77% and about 1.2% for example 1 and example 2, respectively. The mean absolute error in the predicted flux for example 1 is ∼ 0.002 and for example 2 is ∼ 0.0024. These results indicate some preliminary successes of the PINN application to k-eigenvalue problems. (authors)

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Development and Application of Predictive Tools for MHD Stability Limits in Tokamaks

This is a project to develop and apply analytic and computational tools to answer physics questions relevant to the onset of non-ideal magnetohydrodynamic (MHD) instabilities in toroidal magnetic confinement plasmas. The focused goal of the research is to develop predictive tools for these instabilities, including an inner layer solution algorithm, a resistive wall with control coils, and energetic particle effects. The production phase compares studies of instabilities in such systems using analytic techniques, PEST- III and NIMROD. Two important physics puzzles are targeted as guiding thrusts for the analyses. The first is to form an accurate description of the physics determining whether the resistive wall mode or a tearing mode will appear first as β is increased at low rotation and low error fields in DIII-D. The second is to understand the physical mechanism behind recent NIMROD results indicating strong damping and stabilization from energetic particle effects on linear resistive modes. The work seeks to develop a highly relevant predictive tool for ITER, advance the theoretical description of this physics in general, and analyze these instabilities in experiments such as ASDEX Upgrade, DIII-D, JET, JT-60U and NTSX. The awardee on this grant is the University of Tulsa. The research efforts are supervised principally by Dr. Brennan. Support is included for two graduate students, and a strong collaboration with Dr. John M. Finn of LANL. The work includes several ongoing collaborations with General Atomics, PPPL, and the NIMROD team, among others.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Data-Driven Reduced-Order Modeling of Convective Heat Transfer in Porous Media

This work presents a data-driven Reduced-Order Model (ROM) for parametric convective heat transfer problems in porous media. The intrusive Proper Orthogonal Decomposition aided Reduced-Basis (POD-RB) technique is employed to reduce the porous medium formulation of the incompressible Reynolds-Averaged Navier–Stokes (RANS) equations coupled with heat transfer. Instead of resolving the exact flow configuration with high fidelity, the porous medium formulation solves a homogenized flow in which the fluid-structure interactions are captured via volumetric flow resistances with nonlinear, semi-empirical friction correlations. A supremizer approach is implemented for the stabilization of the reduced fluid dynamics equations. The reduced nonlinear flow resistances are treated using the Discrete Empirical Interpolation Method (DEIM), while the turbulent eddy viscosity and diffusivity are approximated by adopting a Radial Basis Function (RBF) interpolation-based approach. The proposed method is tested using a 2D numerical model of the Molten Salt Fast Reactor (MSFR), which involves the simulation of both clean and porous medium regions in the same domain. For the steady-state example, five model parameters are considered to be uncertain: the magnitude of the pumping force, the external coolant temperature, the heat transfer coefficient, the thermal expansion coefficient, and the Prandtl number. For transient scenarios, on the other hand, the coastdown-time of the pump is the only uncertain parameter. The results indicate that the POD-RB-ROMs are suitable for the reduction of similar problems. The relative L2 errors are below 3.34% for every field of interest for all cases analyzed, while the speedup factors vary between 54 (transient) and 40,000 (steady-state).

German, Péter (ORCID:0000000307285283)↗

Precise Time Delays from Strongly Gravitationally Lensed Type Ia Supernovae with Chromatically Microlensed Images

Time delays between the multiple images of strongly gravitationally lensed Type Ia supernovae (glSNe Ia) have the potential to deliver precise cosmological constraints, but the effects of microlensing on time delay extraction have not been studied in detail. Here we quantify the effect of microlensing on the glSN Ia yield of the Large Synoptic Survey Telescope (LSST) and the effect of microlensing on the precision and accuracy of time delays that can be extracted from LSST glSNe Ia. Microlensing has a negligible effect on the LSST glSN Ia yield, but it can be increased by a factor of ~2 over previous predictions to 930 systems using a novel photometric identification technique based on spectral template fitting. Crucially, the microlensing of glSNe Ia is achromatic until three rest-frame weeks after the explosion, making the early-time color curves microlensing-insensitive time delay indicators. By fitting simulated flux and color observations of microlensed glSNe Ia with their underlying, unlensed spectral templates, we forecast the distribution of absolute time delay error due to microlensing for LSST, which is unbiased at the sub-percent level and peaked at 1% for color curve observations in the achromatic phase, while for light-curve observations it is comparable to state-of-the-art mass modeling uncertainties (4%). About 70% of LSST glSN Ia images should be discovered during the achromatic phase, indicating that microlensing time delay uncertainties can be minimized if prompt multicolor follow-up observations are obtained. Lastly, accounting for microlensing, the 1-2 day time delay on the recently discovered glSN Ia iPTF16geu can be measured to 40% precision, limiting its cosmological utility.

79 ASTRONOMY AND ASTROPHYSICS↗

BLIND EXTRACTION OF AN EXOPLANETARY SPECTRUM THROUGH INDEPENDENT COMPONENT ANALYSIS

Blind-source separation techniques are used to extract the transmission spectrum of the hot-Jupiter HD189733b recorded by the Hubble/NICMOS instrument. Such a 'blind' analysis of the data is based on the concept of independent component analysis. The detrending of Hubble/NICMOS data using the sole assumption that nongaussian systematic noise is statistically independent from the desired light-curve signals is presented. By not assuming any prior or auxiliary information but the data themselves, it is shown that spectroscopic errors only about 10%-30% larger than parametric methods can be obtained for 11 spectral bins with bin sizes of {approx}0.09 {mu}m. This represents a reasonable trade-off between a higher degree of objectivity for the non-parametric methods and smaller standard errors for the parametric de-trending. Results are discussed in light of previous analyses published in the literature. The fact that three very different analysis techniques yield comparable spectra is a strong indication of the stability of these results.

79 ASTRONOMY AND ASTROPHYSICS↗

Application health monitoring for extreme-scale resiliency using cooperative fault management

Resiliency is and will be a critical factor in determining scientific productivity on current and exascale supercomputers, and beyond. Applications oblivious to and incapable of handling transient soft and hard errors could waste supercomputing resources or, worse, yield misleading scientific insights. In this work, we introduce a novel application-driven silent error detection and recovery strategy based on application health monitoring. Our methodology uses application output that follows known patterns, as indicators of an application's health and knowledge that violation of these patterns could be indication of faults. Information from system monitors that report hardware and software health status is used to corroborate faults. Collectively, this information is used by a fault coordinator agent to take preventive and corrective measures by applying computational steering to an application between checkpoints. This cooperative fault management system uses the Fault Tolerance Backplane as a communication channel. The benefits of this framework are demonstrated with two real application case studies, molecular dynamics, and quantum chemistry simulations, on scalable clusters with simulated memory and I/O corruptions. Lastly, the developed approach is general and can be easily applied to other applications.

97 MATHEMATICS AND COMPUTING↗

Estimating and Calibrating DER Model Parameters Using Levenberg–Marquardt Algorithm in Renewable Rich Power Grid

The proliferation of inverter-based distributed energy resources (IBDERs) has increased the number of control variables and dynamic interactions, leading to new grid control challenges. For stability analysis and designing appropriate protection controls, it is important that IBDER models are accurate. This paper focuses on the accurate estimation and parameter calibration of DER_A, a recently proposed aggregated IBDER model. In particular, we focus on the parameters of the reactive power–voltage regulation module. We formulate the problem of parameter tuning as a non-linear least square minimization problem and solve it using the Levenberg–Marquardt (LM) method. The LM method is primarily chosen due to its flexibility in adaptively selecting between the steepest descent and Gauss–Newton methods through a damping parameter. The LM approach is used to minimize the error between the actual measurements and the estimated response of the model. Further, the computational challenges posed by the numerical calculation of the Jacobian are tackled using a quasi-Newton root-finding approach. The proposed method is validated on a real feeder model in the northeastern part of the United States. The feeder is modeled in OpenDSS and the measurements thus obtained are fed to the DER_A model for calibration. The simulation results indicate that our approach is able to successfully calibrate the relevant model parameters quickly and with high accuracy, with a total sum of square error of 3.57 × 10 –7 .

29 ENERGY PLANNING, POLICY, AND ECONOMY↗

What Is the Value of Alternative Methods for Estimating Ramping Needs?

Power system operators procure and deploy flexibility reserves or ramping products to address balancing needs caused by uncertainty and variability of load and generation. Existing methods estimate ramping needs using calendar information and historical forecast errors. Novel methods investigate if real-time weather information could inform ramping and other balancing requirements. This paper compares estimation methods for ramping requirements in theory and practice. The theoretical framework indicates when an alternative method could yield improved economic or reliability performance than existing methods by requiring lower or higher levels of ramping products. Preliminary simulations on a 118-bus test system for 4 days in May 2019 illustrate how system performance improves or deteriorates when ramping requirements are weather-informed (alternative) instead of calendar-based (baseline). Preliminary results suggest high variability in change of performance and underline the impact of additional factors, such as system conditions, on the realized performance change.

41 EE - Solar Energy Technologies Office (EE-4S)↗

Characterization of ultrasound elevation beamwidth artifacts for prostate brachytherapy needle insertion

Purpose: Ultrasound elevation beamwidth leads to image artifacts and uncertainties in localizing objects (such as a surgical needle) in ultrasound images. The authors examined the clinical significance of errors caused by elevation beamwidth artifacts and imaging parameters in needle insertion procedures. Methods: Beveled prostate brachytherapy needles were inserted through all holes of a grid template under real-time transrectal ultrasound (TRUS) guidance. The needle tip position as indicated by the TRUS image was compared to their observed physical location. A new device was developed to measure the ultrasound elevation beamwidth. Results: Imaging parameters of the TRUS scanner have direct impact on the localization error ranging from 0.5 up to 4 mm. The smallest localization error was observed laterally close to the center of the grid template and axially within the beam's focal zone. Largest localization error occurs laterally around both sides of the grid template and axially within the beam's far field. The authors also found that the localization errors vary with both lateral and elevation offsets. Conclusions: The authors found properly adjusting the TRUS imaging settings to lower the ultrasound gain and power effectively minimized the appearance of elevation beamwidth artifacts and in turn reduced the localization errors of the needle tip.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

An innovative digital image correlation technique for in-situ process monitoring of composite structures in large scale additive manufacturing

As additive manufacturing (AM) continues to develop and become a standardized manufacturing method, there will be a continued need to provide in-situ monitoring during the manufacturing of polymer composite printed components. Thermal residual stress is a primary cause of failures such as interlayer disbonds or delamination, micro cracking, and dimensional instability, which can occur during or after the build. Here, we report a novel digital image correlation (DIC) adaptation to monitor thermal residual stresses during the entire print process for large-scale AM. In this work, DIC has been investigated (a) by the natural speckle produced by the polymer surface for correlation, (b) to monitor AM build, and (c) to evaluate the effect of thermal residual stress on warpage of the printed component. The natural speckle pattern of the AM material resulted in a respectable 3.57% error compared to the traditional painted speckle pattern of 3.05% error. DIC measured a 190% increase in vertical displacement at the edge of the wall compared to the center, indicating warpage during AM. This work is a step towards a non-intrusive residual stress measuring technique using DIC for large-scale AM.

36 MATERIALS SCIENCE↗

Predicting lake surface water phosphorus dynamics using process-guided machine learning

Phosphorus (P) loading to lakes is degrading the quality and usability of water globally. Accurate predictions of lake P dynamics are needed to understand whole-ecosystem P budgets, as well as the consequences of changing lake P concentrations for water quality. However, complex biophysical processes within lakes, along with limited observational data, challenge our capacity to reproduce short-term lake dynamics needed for water quality predictions, as well as long-term dynamics needed to understand broad scale controls over lake P. Here we use an emerging paradigm in modeling, process-guided machine learning (PGML), to produce a phosphorus budget for Lake Mendota (Wisconsin, USA) and to accurately predict epilimnetic phosphorus over a time range of days to decades. In our implementation of PGML, which we term a Process-Guided Recurrent Neural Network (PGRNN), we combine a process-based model for lake P with a recurrent neural network, and then constrain the predictions with ecological principles. We test independently the process-based model, the recurrent neural network, and the PGRNN to evaluate the overall approach. The process-based model accounted for most of the observed pattern in lake P; however it missed the long-term trend in lake P and had the worst performance in predicting winter and summer P in surface waters. The root mean square error (RMSE) for the process-based model, the recurrent neural network, and the PGRNN was 33.0 μg P L -1 , 22.7 μg P L -1 , and 20.7 μg P L -1 , respectively. All models performed better during summer, with RMSE values for the three models (same order) equal to 14.3 μg P L -1 , 10.9 μg P L -1 , and 10.7 μg P L -1 . Although the PGRNN had only marginally better RMSE during summer, it had lower bias and reproduced long-term decreases in lake P missed by the other two models. For all seasons and all years, the recurrent neural network had better predictions than process alone, with root mean square error (RMSE) of 23.8 μg P L -1 and 28.0 μg P L -1 , respectively. The output of PGRNN indicated that new processes related to water temperature, thermal stratification, and long term changes in external loads are needed to improve the process model. By using ecological knowledge, as well as the information content of complex data, PGML shows promise as a technique for accurate prediction in messy, real-world ecological dynamics, while providing valuable information that can improve our understanding of process.

54 ENVIRONMENTAL SCIENCES↗