Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Statistical forecasting”

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 163 records · Page 9
All sourcesNASA NTRSDOE OSTI
NASA NTRS2012

Improved Impact of Atmospheric Infrared Sounder (AIRS) Radiance Assimilation in Numerical Weather Prediction

Improvements to global and regional numerical weather prediction (NWP) have been demonstrated through assimilation of data from NASA s Atmospheric Infrared Sounder (AIRS). Current operational data assimilation systems use AIRS radiances, but impact on regional forecasts has been much smaller than for global forecasts. Retrieved profiles from AIRS contain much of the information that is contained in the radiances and may be able to reveal reasons for this reduced impact. Assimilating AIRS retrieved profiles in an identical analysis configuration to the radiances, tracking the quantity and quality of the assimilated data in each technique, and examining analysis increments and forecast impact from each data type can yield clues as to the reasons for the reduced impact. By doing this with regional scale models individual synoptic features (and the impact of AIRS on these features) can be more easily tracked. This project examines the assimilation of hyperspectral sounder data used in operational numerical weather prediction by comparing operational techniques used for AIRS radiances and research techniques used for AIRS retrieved profiles. Parallel versions of a configuration of the Weather Research and Forecasting (WRF) model with Gridpoint Statistical Interpolation (GSI) that mimics the analysis methodology, domain, and observational datasets for the regional North American Mesoscale (NAM) model run at the National Centers for Environmental Prediction (NCEP)/Environmental Modeling Center (EMC) are run to examine the impact of each type of AIRS data set. The first configuration will assimilate the AIRS radiance data along with other conventional and satellite data using techniques implemented within the operational system; the second configuration will assimilate AIRS retrieved profiles instead of AIRS radiances in the same manner. Preliminary results of this study will be presented and focus on the analysis impact of the radiances and profiles for selected cases.

Zavodsky, Bradley↗
NASA NTRS2012

Photospheric Magnetic Evolution in the WHI Active Regions

Sequences of line-of-sight (LOS) magnetograms recorded by the Michelson Doppler Imager are used to quantitatively characterize photospheric magnetic structure and evolution in three active regions that rotated across the Sun s disk during the Whole Heliosphere Interval (WHI), in an attempt to relate the photospheric magnetic properties of these active regions to flares and coronal mass ejections (CMEs). Several approaches are used in our analysis, on scales ranging from whole active regions, to magnetic features, to supergranular scales, and, finally, to individual pixels. We calculated several parameterizations of magnetic structure and evolution that have previously been associated with flare and CME activity, including total unsigned magnetic flux, magnetic flux near polarity-inversion lines, amount of canceled flux, the "proxy Poynting flux," and helicity flux. To catalog flare events, we used flare lists derived from both GOES and RHESSI observations. By most such measures, AR 10988 should have been the most flare- and CME-productive active region, and AR 10989 the least. Observations, however, were not consistent with this expectation: ARs 10988 and 10989 produced similar numbers of flares, and AR 10989 also produced a few CMEs. These results highlight present limitations of statistics-based flare and CME forecasting tools that rely upon line-of-sight photospheric magnetic data alone.

Welsch, B. T.↗
NASA NTRS2019

Assimilation of Microwave Observations in the Rainbands of Tropical Cyclones

We propose a novel Bayesian Monte Carlo Integration (BMCI) technique to retrieve the profiles of temperature, water vapor, and cloud liquid/ice water content from microwave cloudy measurements in the presence of tropical cyclones (TC). These retrievals then can either be directly used by meteorologists to analyze the structure of TCs or be assimilated into numerical models to provide accurate initial conditions for the NWP (Numerical Weather Prediction) models. The BMCI technique is applied to the data from the Advanced Technology Microwave Sounder (ATMS) onboard Suomi National Polar-orbiting Partnership (NPP) and Global Precipitation Measurement (GPM) Microwave Imager (GMI). The retrieved profiles are then assimilated into Hurricane WRF (Weather Research and Forecasting) using the GSI (Gridpoint Statistical Interpolation) data assimilation system.

Moradi, Isaac↗
NASA NTRS2000

A New Eddy Dissipation Rate Formulation for the Terminal Area PBL Prediction System(TAPPS)

The TAPPS employs the MASS model to produce mesoscale atmospheric simulations in support of the Wake Vortex project at Dallas Fort-Worth International Airport (DFW). A post-processing scheme uses the simulated three-dimensional atmospheric characteristics in the planetary boundary layer (PBL) to calculate the turbulence quantities most important to the dissipation of vortices: turbulent kinetic energy and eddy dissipation rate. TAPPS will ultimately be employed to enhance terminal area productivity by providing weather forecasts for the Aircraft Vortex Spacing System (AVOSS). The post-processing scheme utilizes experimental data and similarity theory to determine the turbulence quantities from the simulated horizontal wind field and stability characteristics of the atmosphere. Characteristic PBL quantities important to these calculations are determined based on formulations from the Blackadar PBL parameterization, which is regularly employed in the MASS model to account for PBL processes in mesoscale simulations. The TAPPS forecasts are verified against high-resolution observations of the horizontal winds at DFW. Statistical assessments of the error in the wind forecasts suggest that TAPPS captures the essential features of the horizontal winds with considerable skill. Additionally, the turbulence quantities produced by the post-processor are shown to compare favorably with corresponding tower observations.

Charney, Joseph J.↗
NASA NTRS2014

Impact of Land Surface Initialization Approach on Subseasonal Forecast Skill: a Regional Analysis in the Southern Hemisphere

The authors use a sophisticated coupled land-atmosphere modeling system for a Southern Hemisphere subdomain centered over southeastern Australia to evaluate differences in simulation skill from two different land surface initialization approaches. The first approach uses equilibrated land surface states obtained from offline simulations of the land surface model, and the second uses land surface states obtained from reanalyses. The authors find that land surface initialization using prior offline simulations contribute to relative gains in subseasonal forecast skill. In particular, relative gains in forecast skill for temperature of 10%-20% within the first 30 days of the forecast can be attributed to the land surface initialization method using offline states. For precipitation there is no distinct preference for the land surface initialization method, with limited gains in forecast skill irrespective of the lead time. The authors evaluated the asymmetry between maximum and minimum temperatures and found that maximum temperatures had the largest gains in relative forecast skill, exceeding 20% in some regions. These results were statistically significant at the 98% confidence level at up to 60 days into the forecast period. For minimum temperature, using reanalyses to initialize the land surface contributed to relative gains in forecast skill, reaching 40% in parts of the domain that were statistically significant at the 98% confidence level. The contrasting impact of the land surface initialization method between maximum and minimum temperature was associated with different soil moisture coupling mechanisms. Therefore, land surface initialization from prior offline simulations does improve predictability for temperature, particularly maximum temperature, but with less obvious improvements for precipitation and minimum temperature over southeastern Australia.

Precipitation↗
DOE OSTI2025

Huge ensembles – Part 2: Properties of a huge ensemble of hindcasts generated with spherical Fourier neural operators

Abstract. In Part 1, we created an ensemble based on spherical Fourier neural operators. As initial condition perturbations, we used bred vectors, and as model perturbations, we used multiple checkpoints trained independently from scratch. Based on diagnostics that assess the ensemble's physical fidelity, our ensemble has comparable performance to operational weather forecasting systems. However, it requires orders-of-magnitude fewer computational resources. Here in Part 2, we generate a huge ensemble (HENS), with 7424 members initialized each day of summer 2023. We enumerate the technical requirements for running huge ensembles at this scale. HENS precisely samples the tails of the forecast distribution and presents a detailed sampling of internal variability. HENS has two primary applications: (1) as a large dataset with which to study the statistics and drivers of extreme weather and (2) as a weather forecasting system. For extreme climate statistics, HENS samples events 4σ away from the ensemble mean. At each grid cell, HENS increases the skill of the most accurate ensemble member and enhances coverage of possible future trajectories. As a weather forecasting model, HENS issues extreme weather forecasts with better uncertainty quantification. It also reduces the probability of outlier events, in which the verification value lies outside the ensemble forecast distribution.

Mahesh, Ankur↗
NASA NTRS2013

Assimilation of Quality Controlled AIRS Temperature Profiles using the NCEP GFS

We have previously conducted a number of data assimilation experiments using AIRS Version-5 quality controlled temperature profiles as a step toward finding an optimum balance of spatial coverage and sounding accuracy with regard to improving forecast skill. The data assimilation and forecast system we used was the Goddard Earth Observing System Model , Version-5 (GEOS-5) Data Assimilation System (DAS), which represents a combination of the NASA GEOS-5 forecast model with the National Centers for Environmental Prediction (NCEP) operational Grid Point Statistical Interpolation (GSI) global analysis scheme. All analyses and forecasts were run at a 0.5deg x 0.625deg spatial resolution. Data assimilation experiments were conducted in four different seasons, each in a different year. Three different sets of data assimilation experiments were run during each time period: Control; AIRS T(p); and AIRS Radiance. In the "Control" analysis, all the data used operationally by NCEP was assimilated, but no AIRS data was assimilated. Radiances from the Aqua AMSU-A instrument were also assimilated operationally by NCEP and are included in the "Control". The AIRS Radiance assimilation adds AIRS observed radiance observations for a select set of channels to the data set being assimilated, as done operationally by NCEP. In the AIRS T(p) assimilation, all information used in the Control was assimilated as well as Quality Controlled AIRS Version-5 temperature profiles, i.e., AIRS T(p) information was substituted for AIRS radiance information. The AIRS Version-5 temperature profiles were presented to the GSI analysis as rawinsonde profiles, assimilated down to a case-by-case appropriate pressure level p(sub best) determined using the Quality Control procedure. Version-5 also determines case-by-case, level-by-level error estimates of the temperature profiles, which were used as the uncertainty of each temperature measurement. These experiments using GEOS-5 have shown that forecasts resulting from analyses using the AIRS T(p) assimilation system were superior to those from the Radiance assimilation system, both with regard to global 7 day forecast skill and also the ability to predict storm tracks and intensity.

Susskind, Joel↗
DOE OSTI2022

A Comparison of Infectious Disease Forecasting Methods across Locations, Diseases, and Time

Accurate infectious disease forecasting can inform efforts to prevent outbreaks and mitigate adverse impacts. This study compares the performance of statistical, machine learning (ML), and deep learning (DL) approaches in forecasting infectious disease incidences across different countries and time intervals. We forecasted three diverse diseases: campylobacteriosis, typhoid, and Q-fever, using a wide variety of features (n = 46) from public datasets, e.g., landscape, climate, and socioeconomic factors. We compared autoregressive statistical models to two tree-based ML models (extreme gradient boosted trees [XGB] and random forest [RF]) and two DL models (multi-layer perceptron and encoder–decoder model). The disease models were trained on data from seven different countries at the region-level between 2009–2017. Forecasting performance of all models was assessed using mean absolute error, root mean square error, and Poisson deviance across Australia, Israel, and the United States for the months of January through August of 2018. The overall model results were compared across diseases as well as various data splits, including country, regions with highest and lowest cases, and the forecasted months out (i.e., nowcasting, short-term, and long-term forecasting). Overall, the XGB models performed the best for all diseases and, in general, tree-based ML models performed the best when looking at data splits. There were a few instances where the statistical or DL models had minutely smaller error metrics for specific subsets of typhoid, which is a disease with very low case counts. Feature importance per disease was measured by using four tree-based ML models (i.e., XGB and RF with and without region name as a feature). The most important feature groups included previous case counts, region name, population counts and density, mortality causes of neonatal to under 5 years of age, sanitation factors, and elevation. This study demonstrates the power of ML approaches to incorporate a wide range of factors to forecast various diseases, regardless of location, more accurately than traditional statistical approaches.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2023

ORION: Operational FoRecastIng Of INduced Seismicity

The Operational Forecasting of Induced Seismicity toolkit “ORION” (ORION) is an open-source, observation-based ensemble forecasting toolkit which is geared towards helping operators understand the seismic hazard (i.e., probabilistic assessment of the magnitude and frequency of induced seismic events) at a site. ORION analyzes how the seismic hazard evolves during injection and suggests possible mitigation strategies to employ if an earthquake that exceeds certain threshold is observed. Through its ensemble modeling approach, ORION leverages the benefits of statistical-, physics-, and machine learning-based forecasting methodologies, while reducing the impact of each model’s respective limitations. The ORION toolkit consists of an easy-to-use GUI interface that affords a user as much or as little interaction as desired. Advanced capabilities allow the user to upload local, high-precision earthquake catalogs, projected injection profiles and/or spatiotemporal estimates of pressure/stress, and to tune various model parameters. ORION will then provide a spatial and temporal ensemble forecast of seismicity defined as the probability of exceedance of a given earthquake magnitude over a forecast period. Additionally, ORION will provide probability distribution of the statistically derived maximum possible earthquake magnitude that may be expected. Finally, ORION will provide suggested operational management strategies (e.g. reduce injection volumes at specific wells) based on the level of hazard.

NRAP,NRAP Tools,ORION,Operational Forcasting of In↗
NASA NTRS1975

A preliminary study of the statistical analyses and sampling strategies associated with the integration of remote sensing capabilities into the current agricultural crop forecasting system

Extending the crop survey application of remote sensing from small experimental regions to state and national levels requires that a sample of agricultural fields be chosen for remote sensing of crop acreage, and that a statistical estimate be formulated with measurable characteristics. The critical requirements for the success of the application are reviewed in this report. The problem of sampling in the presence of cloud cover is discussed. Integration of remotely sensed information about crops into current agricultural crop forecasting systems is treated on the basis of the USDA multiple frame survey concepts, with an assumed addition of a new frame derived from remote sensing. Evolution of a crop forecasting system which utilizes LANDSAT and future remote sensing systems is projected for the 1975-1990 time frame.

Sand, F.↗
NASA NTRS2005

Analysis of Rapidly Developing Low Cloud Ceilings in a Stable Environment

This report describes the work done by the Applied Meteorology Unit (AMU) in developing a database of days that experienced rapid (< 90 minutes) low cloud formation in a stable atmosphere, resulting in ceilings at the Shuttle Landing Facility (TTS) that violated Space Shuttle Flight Rules (FR). The meteorological conditions favoring the rapid formation of low ceilings include the presence of any inversion below 8000 ft, high relative humidity beneath the inversion, and a clockwise turning of the winds from the surface to the middle troposphere (approx. 15000 ft). The AMU compared and contrasted the atmospheric and thermodynamic conditions between days with rapid low ceiling formation and days with low ceiling resulting from other mechanism. The AMU found that the vertical wind profile is the probable discerning factor between the rapidly-forming ceiling days and other low ceiling days at TTS. Most rapidly-developing low ceiling days had a clockwise turning of the winds with height, whereas other low ceiling days typically had a counter-clockwise turning of the winds with height or negligible vertical wind shear. Forecasters at the Space Meteorology Group (SMG) issue 30 to 90 minute forecasts for low cloud ceilings at TTS to support Space Shuttle landings. Mission verification statistics have shown ceilings to be the number one forecast challenge. More specifically, forecasters at SMG are concerned with any rapidly developing clouds ceilings below 8000 ft in a stable, capped thermodynamic environment, Therefore, the AMU was tasked to examine archived events of rapid stable cloud formation resulting in ceilings below 8000 ft, and document the atmospheric regimes favoring this type of cloud development. The AMU examined the cool season months of November to March during the years of 1993-2003 for days that had low-level inversions and rapid, stable low cloud formation that resulted in ceilings violating the Space Shuttle FR. The AMU wrote and modified existing code to identify inversions from the morning Cape Canaveral, FL rawinsonde (XMR) during the cool season and output pertinent sounding information. They parsed all days with cloud ceilings below 8000 ft at TTS, forming a database of possible rapidly-developing low ceiling events. Days with precipitation or noticeable fog bum-off situations were excluded from the database. Only the daytime hours were examined for possible ceiling development events since low clouds are easier to diagnose with visible satellite imagery. Follow-on work would expand the database to include nighttime cases, using a special enhancement of the infrared imagery for identifying areas of low clouds. The report presents two sample cases of rapidly-developing low cloud ceilings. These cases depict the representative meteorological and thermodynamic characteristics of such events. The cases also illustrate how quickly the cloud decks can develop, sometimes forming in 30 minutes or less. The report also summarizes the composite meteorological conditions for 20 event days with rapid low cloud ceiling formation and 48 non-events days consisting of advection or widespread low cloud ceilings. The meteorological conditions were quite similar for both the event and non-event days, since both types of days experienced low cloud ceilings. Both types of days had a relatively moist environment beneath the inversion based below 8000 ft. In the 20 events identified, de onset of low ceilings occurred between 1200-1800 UTC in every instance. The distinguishing factor between the event and non-event days appears to be the vertical wind profile in the XMR sounding. Eighty-five percent of the event days had a clockwise turning of the winds with height in the lower to middle troposphere whereas 83% of the non-events had a counter-clockwise turning of the winds with height or negligible vertical wind shear. A clockwise turning of the winds with height indicates a warm advection regime, which supports large-scale rising motn and possible cloud formation. Meanwhile, a counter-clockwise turning of the winds with height indicates cold advection or sinking motion in a post-cold frontal environment.

Wheeler, Mark M.↗
NASA NTRS2007

Stable Low Cloud Phase II: Nocturnal Event Study

This report describes the work done by the Applied Meteorology Unit (AMU) in developing a database of nights that experienced rapid (< 90 minutes) low cloud formation in a stable atmosphere, resulting in ceilings at the Shuttle Landing Facility (TTS) that violated Space Shuttle Flight Rules (FR). This work is the second phase of a similar AMU task that examined the same phenomena during the day. In the first phase of this work, the meteorological conditions favoring the rapid formation of low ceilings include the presence of any inversion below 8000 ft, high relative humidity (RH) beneath the inversion and a clockwise turning of the winds from the surface to the middle troposphere (-15000 ft). The AMU compared and contrasted the atmospheric and thermodynamic conditions between nights with rapid low ceiling formation and nights with low ceilings resulting from other mechanisms. The AMU found that there was little to discern between the rapidly-forming ceiling nights and other low ceiling nights at TTS. When a rapid development occurred, the average RH below the inversions was 87% while non-events had an average RH of 79%. One key parameter appeared to be the vertical wind profile in the Cape Canaveral, FL radiosonde (XMR) sounding. Eighty-three percent of the rapid development events had veering winds with height from the surface to the middle troposphere (-15,000 ft) while 61% of the non-events had veering winds with height. Veering winds indicate a warm-advection regime, which supports large-scale rising motion and ultimately cloud formation in a moist environment. However, only six of the nights (out of 86 events examined) with low cloud ceilings had an occurrence of rapidly developing ceilings. Since only 7% rapid development events were observed in this dataset, it is likely that rapid low cloud development is not a common occurrence during the night, or at least not as common as during the day. In the AMU work on the daytime rapid low cloud development (Case and Wheeler 2005), nearly 30% of the low cloud ceiling cases investigated were identified as rapidly developing events. Forecasters at the Space Meteorology Group (SMG) issue 30 to 90 minute forecasts for low cloud ceilings at TTS to support Space Shuttle landings. Mission verification statistics have shown ceilings to be the number one forecast challenge. More specifically, forecasters at SMG are concerned with any rapidly developing clouds/ceilings below 8000 R in a stable, capped thermodynamic environment. Therefore, the AMU was tasked to examine archived events of rapid stable cloud formation resulting in ceilings below 8000 ft, and document the atmospheric regimes favoring this type of cloud development. The AMU examined the cool season months of November to March during the years of 1994-2005 for nights that had low-level inversions and rapid, stable low cloud formation that resulted in ceilings violating the Space Shuttle FR. The AMU wrote and modified existing code to identify inversions from the evening and morning XMR radiosonde during the cool season and output pertinent sounding information. They parsed all days with cloud ceilings below 8000 ft at TTS, forming a database of possible rapidly-developing low ceiling events. Nights with precipitation or noticeable fog burn-off situations were excluded from the database. Only the nighttime hours were examined for possible ceiling development events since the daytime events were examined in the first phase of this work. The report presents one sample case of rapidly-developing low cloud ceilings. The case depicts the representative meteorological and thermodynamic characteristics of such events. The case also illustrates how quickly the cloud decks can develop, sometimes forming in 30 minutes or less. The report also summarizes the composite meteorological conditions for 6 event nights with rapid low cloud ceiling formation and 80 non-events nights consisting of advection or widespread low cloud ceilings. The teorological conditions were quite similar for both the event and non-event nights, since both types of nights experienced low cloud ceilings. Both types of nights had a relatively moist environment beneath an inversion based below 8000 ft.

Bauman, William H., III↗
DOE OSTI2022

Limited Dynamic Triggering in the Utah Region, USA

The state of Utah, USA, experiences around 3800 catalogued earthquakes per year, highlighting that the region is seismically active and susceptible to earthquakes. Following the 2002 Denali Fault (M7.9) earthquake in Alaska, the region showed an elevated seismicity rate for 3 weeks following the passage of high amplitude surface waves, suggesting that the region may be particularly susceptible to dynamic triggering. With over 23,396 faults and each fault presenting a potential fault for triggering, we systematically search for dynamic triggering throughout the state of Utah caused by large, global earthquakes with M ≥ 7. Specifically, we analyse earthquake catalogues and all available waveform data to determine statistical increases of seismicity rate following the passage of seismic arrivals. While we find instances of dynamic triggering, our results show that these events occur sparsely in the region. In total, less than 20 per cent of the 273 main shocks that occur from 2000 to the end of 2017 show a statistical indication of dynamic triggering throughout the Utah region, highlighting that dynamic triggering is limited for stresses created by transient signals from global M ≥ 7 earthquakes, with the exception being the Denali Fault (M7.9), Alaska earthquake (i.e. an instance of significant triggering).

58 GEOSCIENCES↗
NASA NTRS2015

The Impact of Assimilating Precipitation-affected Radiance on Cloud and Precipitation in Goddard WRF-EDAS Analyses

High-frequency TMI and AMSR-E radiances, which are sensitive to precipitation over land, are assimilated into the Goddard Weather Research and Forecasting Model- Ensemble Data Assimilation System (WRF-EDAS) for a few heavy rain events over the continental US. Independent observations from surface rainfall, satellite IR brightness temperatures, as well as ground-radar reflectivity profiles are used to evaluate the impact of assimilating rain-sensitive radiances on cloud and precipitation within WRF-EDAS. The evaluations go beyond comparisons of forecast skills and domain-mean statistics, and focus on studying the cloud and precipitation features in the jointed rainradiance and rain-cloud space, with particular attentions on vertical distributions of height-dependent cloud types and collective effect of cloud hydrometers. Such a methodology is very helpful to understand limitations and sources of errors in rainaffected radiance assimilations. It is found that the assimilation of rain-sensitive radiances can reduce the mismatch between model analyses and observations by reasonably enhancing/reducing convective intensity over areas where the observation indicates precipitation, and suppressing convection over areas where the model forecast indicates rain but the observation does not. It is also noted that instead of generating sufficient low-level warmrain clouds as in observations, the model analysis tends to produce many spurious upperlevel clouds containing small amount of ice water content. This discrepancy is associated with insufficient information in ice-water-sensitive radiances to address the vertical distribution of clouds with small amount of ice water content. Such a problem will likely be mitigated when multi-channel multi-frequency radiances/reflectivity are assimilated over land along with sufficiently accurate surface emissivity information to better constrain the vertical distribution of cloud hydrometers.

WRF-EDAS Analyses↗
DOE OSTI2022

Do Machine Learning Approaches Offer Skill Improvement for Short-Term Forecasting of Wind Gust Occurrence and Magnitude?

Abstract Wind gusts, and in particular intense gusts, are societally relevant but extremely challenging to forecast. This study systematically assesses the skill enhancement that can be achieved using artificial neural networks (ANNs) for forecasting of wind gust occurrence and magnitude. Geophysical predictors from the ERA5 reanalysis are used in conjunction with an autoregressive term in regression and ANN models with different predictors, and varying model complexity. Models are derived and assessed for the warm (April–September) and cold (October–March) seasons for three high passenger volume airports in the United States. Model uncertainty is assessed by deriving models for 1000 different randomly selected training (70%) and testing (30%) subsets. Gust prediction fidelity in independent test samples is critically dependent on inclusion of an autoregressive term. Gust occurrence probabilities derived using five-layer ANNs exhibit consistently higher fidelity than those from regression models and shallower ANNs. Inclusion of the autoregressive term and increasing the number of hidden layers in ANNs from 1 to 5 also improve the model performance for gust magnitudes (lower RMSE, increased correlation, and model standard deviations that more closely approximate observed values). Deeper ANNs (e.g., 20 hidden layers) exhibit higher skill in forecasting strong (17–25.7 m s −1 ) and damaging (≥25.7 m s −1 ) wind gusts. However, such deep networks exhibit evidence of overfitting and still substantially underestimate (by 50%) the frequency of strong and damaging wind gusts at the three airports considered herein. Significance Statement Improved short-term forecasting of wind gusts will enhance aviation safety and logistics and may offer other societal benefits. Here we present a rigorous investigation of the relative skill of models of wind gust occurrence and magnitude that employ different statistical methods. It is shown that artificial neural networks (ANNs) offer considerable skill enhancement over regression methods, particularly for strong and damaging wind gusts. For wind gust magnitudes in particular, application of deeper learning networks (e.g., five or more hidden layers) offers tangible improvements in forecast accuracy. However, deeper networks are vulnerable to overfitting and exhibit substantial variability with the specific training and testing data subset used. Also, even deep ANNs reproduce only half of strong and damaging wind gusts. These results indicate the need for future work to elucidate the dynamical mechanisms of intense wind gusts and advance solutions to their prediction.

54 ENVIRONMENTAL SCIENCES↗
NASA NTRS2006

Use of High-resolution WRF Simulations to Forecast Lightning Threat

Recent observational studies have confirmed the existence of a robust statistical relationship between lightning flash rates and the amount of large precipitating ice hydrometeors in storms. This relationship is exploited, in conjunction with the capabilities of recent forecast models such as WRF, to forecast the threat of lightning from convective storms using the output fields from the model forecasts. The simulated vertical flux of graupel at -15C is used in this study as a proxy for charge separation processes and their associated lightning risk. Six-h simulations are conducted for a number of case studies for which three-dimensional lightning validation data from the North Alabama Lightning Mapping Array are available. Experiments indicate that initialization of the WRF model on a 2 km grid using Eta boundary conditions, Doppler radar radial velocity and reflectivity fields, and METAR and ACARS data yield the most realistic simulations. An array of subjective and objective statistical metrics are employed to document the utility of the WRF forecasts. The simulation results are also compared to other more traditional means of forecasting convective storms, such as those based on inspection of the convective available potential energy field.

McCaul, William E.↗
NASA NTRS1998

Statistical Aspects of Intense Hurricanes in the Atlantic Basin during the Past 49 Hurricane Seasons (1950-1998)

Statistical aspects of intense hurricanes (those of category 3 or higher) in the Atlantic basin for the interval of 1950-1998 are investigated with respect to the ENSC, cycle and the postulated 'more versus less' activity modes for intense hurricane activity. This is accomplished in order to evaluate the statistical basis for these specific groupings and to presage a forecast for the number of expected intense hurricanes during the forthcoming 1999 hurricane season. Statistically speaking, because the 1998-1999 La Nina (onset presumed to be September 1998) should persist longer than about a year - the average duration for La Nina events based on the 10 previously occurring La Nina, each having a duration in the range of 7-22 months - it seems likely that the 1999 season will be classified as a 'non-El Nino-related' (NENR) season. If true, then, greater than or equal to 2 intense hurricanes are to be expected. Based on Poisson statistics, the probability of greater than or equal to 2 events is about 77% when the season is classified as NENR, and it is about 87% when the season is classified as NENR and the 'more active' phase is in fashion; likewise, the probability of greater than or equal to 4 events is about 31% and 48%, respectively, for the two cases. Therefore, an above average rate (possibly, as many as 4 +/- 1, or higher) of intense hurricanes forming in the Atlantic basin seems a very distinct possibility during the 1999 season.

Wilson, Robert M.↗
DOE OSTI2021

Discrete Graph Structure Learning for Forecasting Multiple Time Series

Time series forecasting is an extensively studied subject in statistics, economics, and computer science. Exploration of the correlation and causation among the variables in a multivariate time series shows promise in enhancing the performance of a time series model. When using deep neural networks as forecasting models, we hypothesize that exploiting the pairwise information among multiple (multivariate) time series also improves their forecast. If an explicit graph structure is known, graph neural networks (GNNs) have been demonstrated as powerful tools to exploit the structure. In this work, we propose learning the structure simultaneously with the GNN if the graph is unknown. We cast the problem as learning a probabilistic graph model through optimizing the mean performance over the graph distribution. The distribution is parameterized by a neural network so that discrete graphs can be sampled differentiably through reparameterization. Empirical evaluations show that our method is simpler, more efficient, and better performing than a recently proposed bilevel learning approach for graph structure learning, as well as a broad array of forecasting models, either deep or non-deep learning based, and graph or non-graph based.

Shang, Chao↗