Results for “ensemble prediction” — Page 7

NASA NTRS2009

Dynamics and Predictability of Hurricane Humberto (2007) Revealed from Ensemble Analysis and Forecasting

This study uses short-range ensemble forecasts initialized with an Ensemble-Kalman filter to study the dynamics and predictability of Hurricane Humberto, which made landfall along the Texas coast in 2007. Statistical correlation is used to determine why some ensemble members strengthen the incipient low into a hurricane and others do not. It is found that deep moisture and high convective available potential energy (CAPE) are two of the most important factors for the genesis of Humberto. Variations in CAPE result in as much difference (ensemble spread) in the final hurricane intensity as do variations in deep moisture. CAPE differences here are related to the interaction between the cyclone and a nearby front, which tends to stabilize the lower troposphere in the vicinity of the circulation center. This subsequently weakens convection and slows genesis. Eventually the wind-induced surface heat exchange mechanism and differences in landfall time result in even larger ensemble spread. 1

Sippel, Jason A.↗

DOE OSTI2021

Quantifying sensitivity in numerical weather prediction-modeled offshore wind speeds through an ensemble modeling approach

A decade of research has shown that numerical weather prediction (NWP)-modeled wind speeds can be highly sensitive to the inputs and setups within the NWP model. For wind resource characterization applications, this sensitivity is often addressed by constructing a range of setups and selecting the one that best validates against observations. However, this approach is not possible in areas that lack high-quality hub height observations, especially offshore wind areas. In such cases, techniques to quantify and disseminate confidence in NWP-modeled wind speeds in the absence of observations are needed. We address this need in the present study and propose best practices for quantifying the spread in NWP-modeled wind speeds. We implement an ensemble approach in which we consider 24 different setups to the Weather Research and Forecasting (WRF) model. We construct the ensemble by considering variations in WRF version, WRF namelist, atmospheric forcing, and sea surface temperature (SST) forcing. Our analysis finds that the standard deviation produces more consistent estimates compared to the interquartile range and tends to be the more conservative estimator for ensemble variability. We further find that model spread increases closer to the surface and on shorter time scales. In conclusion, we explore methods to attribute total ensemble variability to the different ensemble components (e.g., atmospheric forcing and SST product) and find that contributions by components also vary depending on time scale. We anticipate that the methods and results presented in this paper will provide a reasonable foundation for further research into ensemble-based wind resource data sets.

17 WIND ENERGY↗

NASA NTRS1977

Model predictions of wind and turbulence profiles associated with an ensemble of aircraft accidents

The feasibility of predicting conditions under which wind/turbulence environments hazardous to aviation operations exist is studied by examining a number of different accidents in detail. A model of turbulent flow in the atmospheric boundary layer is used to reconstruct wind and turbulence profiles which may have existed at low altitudes at the time of the accidents. The predictions are consistent with available flight recorder data, but neither the input boundary conditions nor the flight recorder observations are sufficiently precise for these studies to be interpreted as verification tests of the model predictions.

Williamson, G. G.↗

NASA NTRS2014

Statistical Mechanics of Turbulent Dynamos

Incompressible magnetohydrodynamic (MHD) turbulence and magnetic dynamos, which occur in magnetofluids with large fluid and magnetic Reynolds numbers, will be discussed. When Reynolds numbers are large and energy decays slowly, the distribution of energy with respect to length scale becomes quasi-stationary and MHD turbulence can be described statistically. In the limit of infinite Reynolds numbers, viscosity and resistivity become zero and if these values are used in the MHD equations ab initio, a model system called ideal MHD turbulence results. This model system is typically confined in simple geometries with some form of homogeneous boundary conditions, allowing for velocity and magnetic field to be represented by orthogonal function expansions. One advantage to this is that the coefficients of the expansions form a set of nonlinearly interacting variables whose behavior can be described by equilibrium statistical mechanics, i.e., by a canonical ensemble theory based on the global invariants (energy, cross helicity and magnetic helicity) of ideal MHD turbulence. Another advantage is that truncated expansions provide a finite dynamical system whose time evolution can be numerically simulated to test the predictions of the associated statistical mechanics. If ensemble predictions are the same as time averages, then the system is said to be ergodic; if not, the system is nonergodic. Although it had been implicitly assumed in the early days of ideal MHD statistical theory development that these finite dynamical systems were ergodic, numerical simulations provided sufficient evidence that they were, in fact, nonergodic. Specifically, while canonical ensemble theory predicted that expansion coefficients would be (i) zero-mean random variables with (ii) energy that decreased with length scale, it was found that although (ii) was correct, (i) was not and the expected ergodicity was broken. The exact cause of this broken ergodicity was explained, after much investigation, by greatly extending the statistical theory of ideal MHD turbulence. The mathematical details of broken ergodicity, in fact, give a quantitative explanation of how coherent structure, dynamic alignment and force-free states appear in turbulent magnetofluids. The relevance of these ideal results to real MHD turbulence occurs because broken ergodicity is most manifest in the ideal case at the largest length scales and it is in these largest scales that a real magnetofluid has the least dissipation, i.e., most closely approaches the behavior of an ideal magnetofluid. Furthermore, the effects grow stronger when cross and magnetic helicities grow large with respect to energy, and this is exactly what occurs with time in a real magnetofluid, where it is called selective decay. The relevance of these results found in ideal MHD turbulence theory to the real world is that they provide at least a qualitative explanation of why confined turbulent magnetofluids, such as the liquid iron that fills the Earth's outer core, produce stationary, large-scale magnetic fields, i.e., the geomagnetic field. These results should also apply to other planets as well as to plasma confinement devices on Earth and in space, and the effects should be manifest if Reynolds numbers are high enough and there is enough time for stationarity to occur, at least approximately. In the presentation, details will be given for both theoretical and numerical results, and references will be provided.

Shebalin, John V.↗

NASA NTRS2014

MHD Turbulence and Magnetic Dynamos

Incompressible magnetohydrodynamic (MHD) turbulence and magnetic dynamos, which occur in magnetofluids with large fluid and magnetic Reynolds numbers, will be discussed. When Reynolds numbers are large and energy decays slowly, the distribution of energy with respect to length scale becomes quasi-stationary and MHD turbulence can be described statistically. In the limit of infinite Reynolds numbers, viscosity and resistivity become zero and if these values are used in the MHD equations ab initio, a model system called ideal MHD turbulence results. This model system is typically confined in simple geometries with some form of homogeneous boundary conditions, allowing for velocity and magnetic field to be represented by orthogonal function expansions. One advantage to this is that the coefficients of the expansions form a set of nonlinearly interacting variables whose behavior can be described by equilibrium statistical mechanics, i.e., by a canonical ensemble theory based on the global invariants (energy, cross helicity and magnetic helicity) of ideal MHD turbulence. Another advantage is that truncated expansions provide a finite dynamical system whose time evolution can be numerically simulated to test the predictions of the associated statistical mechanics. If ensemble predictions are the same as time averages, then the system is said to be ergodic; if not, the system is nonergodic. Although it had been implicitly assumed in the early days of ideal MHD statistical theory development that these finite dynamical systems were ergodic, numerical simulations provided sufficient evidence that they were, in fact, nonergodic. Specifically, while canonical ensemble theory predicted that expansion coefficients would be (i) zero-mean random variables with (ii) energy that decreased with length scale, it was found that although (ii) was correct, (i) was not and the expected ergodicity was broken. The exact cause of this broken ergodicity was explained, after much investigation, by greatly extending the statistical theory of ideal MHD turbulence. The mathematical details of broken ergodicity, in fact, give a quantitative explanation of how coherent structure, dynamic alignment and force-free states appear in turbulent magnetofluids. The relevance of these ideal results to real MHD turbulence occurs because broken ergodicity is most manifest in the ideal case at the largest length scales and it is in these largest scales that a real magnetofluid has the least dissipation, i.e., most closely approaches the behavior of an ideal magnetofluid. Furthermore, the effects grow stronger when cross and magnetic helicities grow large with respect to energy, and this is exactly what occurs with time in a real magnetofluid, where it is called selective decay. The relevance of these results found in ideal MHD turbulence theory to the real world is that they provide at least a qualitative explanation of why confined turbulent magnetofluids, such as the liquid iron that fills the Earth's outer core, produce stationary, large-scale magnetic fields, i.e., the geomagnetic field. These results should also apply to other planets as well as to plasma confinement devices on Earth and in space, and the effects should be manifest if Reynolds numbers are high enough and there is enough time for stationarity to occur, at least approximately. In the presentation, details will be given for both theoretical and numerical results, and references will be provided.

Shebalin, John V↗

DOE OSTI2020

A Multi-Instance learning Framework for Seismic Detectors

In this report, we construct and test a framework for fusing the predictions of a ensemble of seismic wave detectors. The framework is drawn from multi-instance learning and is meant to improve the predictive skill of the ensemble beyond that of the individual detectors. We show how the framework allows the use of multiple features derived from the seismogram to detect seismic wave arrivals, as well as how it allows only the most informative features to be retained in the ensemble. The computational cost of the "ensembling" method is linear in the size of the ensemble, allowing a scalable method for monitoring multiple features/transformations of a seismogram. The framework is tested on teleseismic and regional p-wave arrivals at the IMS (International Monitoring System) station in Warramunga, NT, Australia and the PNSU station in University of Utah's monitoring network.

58 GEOSCIENCES↗

DOE OSTI2022

A Bayesian Deep Learning Approach to Near-Term Climate Prediction

Since model bias and associated initialization shock are serious shortcomings that reduce prediction skills in state-of-the-art decadal climate prediction efforts, we pursue a complementary machine-learning-based approach to climate prediction. The example problem setting we consider consists of predicting natural variability of the North Atlantic sea surface temperature on the interannual timescale in the pre-industrial control simulation of the Community Earth System Model. While previous works have considered the use of recurrent networks such as convolutional LSTMs and reservoir computing networks in this and other similar problem settings, we currently focus on the use of feedforward convolutional networks. In particular, we find that a feedforward convolutional network with a Densenet architecture is able to outperform a convolutional LSTM in terms of predictive skill. Next, we go on to consider a probabilistic formulation of the same network based on Stein variational gradient descent and find that in addition to providing useful measures of predictive uncertainty, the probabilistic (Bayesian) version improves on its deterministic counterpart in terms of predictive skill. Finally, we characterize the reliability of the ensemble of machine learning models obtained in the probabilistic setting by using analysis tools developed in the context of ensemble numerical weather prediction.

54 ENVIRONMENTAL SCIENCES↗

DOE OSTI2023

Optimizing Seasonal‐To‐Decadal Analog Forecasts With a Learned Spatially‐Weighted Mask

Abstract Seasonal‐to‐decadal climate prediction is crucial for decision‐making in a number of industries, but forecasts on these timescales have limited skill. Here, we develop a data‐driven method for selecting optimal analogs for seasonal‐to‐decadal analog forecasting. Using an interpretable neural network, we learn a spatially‐weighted mask that quantifies how important each grid point is for determining whether two climate states will evolve similarly. We show that analogs selected using this weighted mask provide more skillful forecasts than analogs that are selected using traditional spatially‐uniform methods. This method is tested on two prediction problems using the Max Planck Institute for Meteorology Grand Ensemble: multi‐year prediction of North Atlantic sea surface temperatures, and seasonal prediction of El Niño Southern Oscillation. This work demonstrates a methodical approach to selecting analogs that may be useful for improving seasonal‐to‐decadal forecasts and understanding their sources of skill.

Geology↗

DOE OSTI2021

Ensemble‐Based Assimilation of Satellite All‐Sky Microwave Radiances Improves Intensity and Rainfall Predictions for Hurricane Harvey (2017)

Abstract Ensemble‐based data assimilation of radar observations across inner‐core regions of tropical cyclones (TCs) in tandem with satellite all‐sky infrared (IR) radiances across the TC domain improves TC track and intensity forecasts. This study further investigates potential enhancements in TC track, intensity, and rainfall forecasts via assimilation of all‐sky microwave (MW) radiances using Hurricane Harvey (2017) as an example. Assimilating Global Precipitation Measurement constellation all‐sky MW radiances in addition to GOES‐16 all‐sky IR radiances reduces the forecast errors in the TC track, rapid intensification (RI), and peak intensity compared to assimilating all‐sky IR radiances alone, including a 24‐hr increase in forecast lead‐time for RI. Assimilating all‐sky MW radiances also improves Harvey's hydrometeor fields, which leads to improved forecasts of rainfall after Harvey's landfall. This study indicates that avenues exist for producing more accurate forecasts for TCs using available yet underutilized data, leading to better warnings of and preparedness for TC‐associated hazards in the future.

54 ENVIRONMENTAL SCIENCES↗

DOE OSTI2023

Non-asymptotic analysis of ensemble Kalman updates: effective dimension and localization

Many modern algorithms for inverse problems and data assimilation rely on ensemble Kalman updates to blend prior predictions with observed data. Ensemble Kalman methods often perform well with a small ensemble size, which is essential in applications where generating each particle is costly. This paper develops a non-asymptotic analysis of ensemble Kalman updates, which rigorously explains why a small ensemble size suffices if the prior covariance has moderate effective dimension due to fast spectrum decay or approximate sparsity. Here, we present our theory in a unified framework, comparing everal implementations of ensemble Kalman updates that use perturbed observations, square root filtering and localization. As part of our analysis, we develop new dimension-free covariance estimation bounds for approximately sparse matrices that may be of independent interest.

Mathematics↗

DOE OSTI2023

Tree-Based Ensemble Learning Models for Wall Temperature Predictions in Post-Critical Heat Flux Flow Regimes at Subcooled and Low-Quality Conditions

Accurately predicting post-critical heat flux (CHF) heat transfer is an important but challenging task in water-cooled reactor design and safety analysis. Although numerous heat transfer correlations have been developed to predict post-CHF heat transfer, these correlations are only applicable to relatively narrow ranges of flow conditions due to the complex physical nature of the post-CHF heat transfer regimes. In this paper, a large quantity of experimental data is collected and summarized from the literature for steady-state subcooled and low-quality film boiling regimes with water as the working fluid in vertical tubular test sections. In addition, a low-quality water film boiling (LWFB) database is consolidated with a total of 22,813 experimental data points, which cover a wide flow range of the system pressure from 0.1 to 9.0 MPa, mass flux from 25 to 2750 kg/m 2 s, and inlet subcooling from 1 to 70 °C. Two machine learning (ML) models, based on random forest (RF) and gradient boosted decision tree (GBDT), are trained and validated to predict wall temperatures in post-CHF flow regimes. The trained ML models demonstrate significantly improved accuracies compared to conventional empirical correlations. To further evaluate the performance of these two ML models from a statistical perspective, three criteria are investigated and three metrics are calculated to quantitatively assess the accuracy of these two ML models. For the full LWFB database, the root-mean-square errors between the measured and predicted wall temperatures by the GBDT and RF models are 5.7% and 6.2%, respectively, confirming the accuracy of the two ML models.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

DOE OSTI2026

A 1-km Multi-ICs/BCs and Mixed-Physics WRF Ensemble for Sea-Breeze Convection Prediction during ESCAPE: Precipitation Evaluation

Explore the source record for details and available documents.

Hu, Yishi↗

DOE OSTI2023

CMIP6 skill at predicting interannual to multi-decadal summer monsoon precipitation variability

Monsoons affect the economy, agriculture, and human health of two thirds of the world’s population. Therefore, predicting variations in monsoon precipitation is societally important. We explore the ability of climate models from the sixth phase of the Climate Model Intercomparison Project to predict summer monsoon precipitation variability by using hindcasts from the Decadal Climate Prediction Project (Component A). The multi-model ensemble-mean shows significant skill at predicting summer monsoon precipitation from one year to 6–9 years ahead. However, this skill is dependent on the model, monsoon domain, and lead-time. In general, the skill of the multi-model ensemble-mean prediction is low in year 1 but increases for longer-lead times and is largely consistent with externally forced changes. The best captured region is northern Africa for the 2–5 and 6–9 year forecast lead times. In contrast, there is no significant skill using the ensemble-mean over East and South Asia and, furthermore, there is significant spread in skill among models for these domains. By sub-sampling the ensemble we show that the difference in skill between models is tied to the simulation of the externally forced response over East and South Asia, with models with a more skilful forced response capable of better predictions. A further contribution is from skilful prediction of Pacific Ocean temperatures for the South Asian summer monsoon at longer lead-times. Therefore, these results indicate that predictions of the East and South Asian monsoons could be significantly improved.

54 ENVIRONMENTAL SCIENCES↗

DOE OSTI2023

Fast uncertainty estimates in deep learning interatomic potentials

Deep learning has emerged as a promising paradigm to give access to highly accurate predictions of molecular and material properties. A common short-coming shared by current approaches, however, is that neural networks only give point estimates of their predictions and do not come with predictive uncertainties associated with these estimates. Existing uncertainty quantification efforts have primarily leveraged the standard deviation of predictions across an ensemble of independently trained neural networks. This incurs a large computational overhead in both training and prediction, resulting in order-of-magnitude more expensive predictions. Here, we propose a method to estimate the predictive uncertainty based on a single neural network without the need for an ensemble. This allows us to obtain uncertainty estimates with virtually no additional computational overhead over standard training and inference. We demonstrate that the quality of the uncertainty estimates matches those obtained from deep ensembles. We further examine the uncertainty estimates of our methods and deep ensembles across the configuration space of our test system and compare the uncertainties to the potential energy surface. Finally, we study the efficacy of the method in an active learning setting and find the results to match an ensemble-based strategy at order-of-magnitude reduced computational cost.

Chemistry↗

DOE OSTI2020

Sensitivity Study for Forecasting Variables of WRF-Solar Using a Tangent Linear Approach

Integrating solar generation in recent years has highlighted the need for improved accuracy in predicting solar power. Confidence in solar power forecasting can be achieved by designing an ensemble that provides reliable probabilistic information for solar radiation with reduced uncertainty and error. Ideally, ensemble members are created through the optimized perturbation of the initial conditions in numerical weather prediction (NWP) models. Tangent linear models are capable of efficiently investigating the sensitivity of solar radiation to model input parameters because they do not require individual perturbation of each variable. This sensitivity study using tangent linear models provide us the capability to identify the right variables to perturb in an ensemble prediction system. In this study, we developed tangent linear models for WRF-Solar modules that directly impact the computation of solar radiation and the simulation of cloud formation and dissipation including the Fast All-sky Model for Solar Applications (FARMS), the Noah land surface model (LSM), the Thompson microphysics, the Mello-Yamada-Nakanishi-Niino (MYNN) boundary layer parameterization, and the Deng scheme for a shallow-convection parameterization. A sensitivity analysis was conducted under various scenarios based on satellite observations and model simulations from the National Solar Radiation Data Base (NSRDB) and WRF-Solar, respectively. Critical forecasting variables that are highly sensitive to the forecasting of global horizontal irradiance (GHI), direct normal irradiance (DNI), cloud mixing ratio, cloud tendency, cloud fraction, and sensible and latent heat fluxes were determined using the relevant WRF-Solar module. This study will be used as a guidance on future research leading to high-quality probabilistic solar forecasting. In this presentation, we discuss the validation of tangent linear approach for WRF-Solar modules and illustrate how the sensitivity results are valuable in the improvement of probabilistic solar prediction.

24 POWER TRANSMISSION AND DISTRIBUTION↗

DOE OSTI2020

Fouling modeling and prediction approach for heat exchangers using deep learning

In this article, we develop a generalized and scalable statistical model for accurate prediction of fouling resistance using commonly measured parameters of industrial heat exchangers. This prediction model is based on deep learning where a scalable algorithmic architecture learns non-linear functional relationships between a set of target and predictor variables from large number of training samples. Here, the efficacy of this modeling approach is demonstrated for predicting fouling in an analytically modeled cross-flow heat exchanger, designed for waste heat recovery from flue-gas using room temperature water. The performance results of the trained models demonstrate that the mean absolute prediction errors are under 10 –4 KW –1 for flue-gas side, water side and overall fouling resistances. The coefficients of determination (R 2 ), which characterize the goodness of fit between the predictions and observed data, are over 99%. Even under varying levels of measurement noise in the inputs, we demonstrate that predictions over an ensemble of multiple neural networks achieves better accuracy and robustness to noise. We find that the proposed deep-learning fouling prediction framework learns to follow heat exchanger flow and heat transfer physics, which we confirm using locally interpretable model agnostic explanations around randomly selected operating points. Overall, we provide a robust algorithmic framework for fouling prediction that can be generalized and scaled to various types of industrial heat exchangers.

42 ENGINEERING↗

DOE OSTI2021

Formation of a Ti–Cu(111) single atom alloy: Structure and CO binding

A single atom Ti–Cu(111) surface alloy can be generated by depositing small amounts of Ti onto Cu(111) at slightly elevated surface temperatures (~500 to 600 K). Here, scanning tunneling microscopy shows that small Ti-rich islands covered by a Cu single layer form preferentially on ascending step edges of Cu(111) during Ti deposition below about 400 K but that a Ti–Cu(111) alloy replaces these small islands during deposition between 500 and 600 K, producing an alloy in the brims of the steps. Larger partially Cu-covered Ti-containing islands also form on the Cu(111) terraces at temperatures between 300 and 700 K. After surface exposure to CO at low temperatures, reflection absorption infrared spectroscopy (RAIRS) reveals distinct C–O stretch bands at 2102 and 2050 cm –1 attributed to CO adsorbed on Cu-covered Ti-containing domains vs sites in the Ti–Cu(111) surface alloy. Calculations using density functional theory (DFT) suggest that the lower frequency C–O stretch band originates specifically from CO adsorbed on isolated Ti atoms in the Ti–Cu(111) surface alloy and predicts a higher C–O stretch frequency for CO adsorbed on Cu above subsurface Ti ensembles. DFT further predicts that CO preferentially adsorbs in flat-lying configurations on contiguous Ti surface structures with more than one Ti atom and thus that CO adsorbed on such structures should not be observed with RAIRS. The ability to generate a single atom Ti–Cu(111) alloy will provide future opportunities to investigate the surface chemistry promoted by a representative early transition metal dopant on a Cu(111) host surface.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

DOE OSTI2021

A Bayesian Neural Network Ensemble Approach for Improving Large-Scale Streamflow Predictability

Predictive modeling through the use of AI techniques and AI-derived model components; the use of AI and other tools to design a prediction system comprising of a hierarchy of models (e.g.,AI driven model/component/parameterization selection). We propose a Bayesian neural network ensemble approach for improving large-scale streamflow predictability and understanding in a changing environment that combines multiple Earth system land model predictions by calculating spatiotemporally varying model weights and biases while accounting for various types of observations at multiple scales with uncertainty.

58 GEOSCIENCES↗