Search NASA⌕ Search

SEARCH · Search NASA

Results for “flow 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 127 records · Page 7
All sourcesNASA NTRSDOE OSTI
NASA NTRSDate not supplied

Understanding and Utilizing PBL Height Data from Multiple Observing Systems in the GEOS System

The accuracy of PBL height simulation is a key issue in many applications including forecasting near surface meteorology and air quality, however, it is a very challenging problem due to the lack of not only comprehensive, global Planetary Boundary Layer (PBL) observations but also a strategy and infrastructure to utilize PBL height data from a variety of sensors. Following the designation of PBL as an incubation class observable in the 2017 Decadal Survey, the PBL Incubation Study Team Report [14] made clear that “a future global PBL observing system requires modeling and data assimilation as essential components.” There is an urgent need for global modeling development in order to utilize Program of Record (POR) observations, assess their impacts, and identify gaps to be filled by future PBL missions. Our overall objective is to develop PBL data assimilation capabilities in the NASA Global Earth Observing System (GEOS), focusing on PBL height from multiple observing systems, to support the assessment and use of future PBL observations. The NASA GEOS system is composed of the GEOS global atmospheric general circulation model (AGCM) and the atmospheric data assimilation system (ADAS). The PBL parameterizations include the “Lock” K-profile scheme driven by surface and cloud-top buoyancy fluxes ([4]), and the “Louis” local scheme for stable conditions based on the Richardson number ([5]). Above the mixed layer defined by the Lock surface plume, shallow cumulus convection is represented by the mass flux scheme of [9]. Additional parameterizations are summarized in [1]. The ADAS employs the hybrid 4D Ensemble- Variational (EnVar) configuration ([15]), with the ensemble providing flow-dependent background error covariance information. The resultant analysis increments are fed back to the forecast model through the 4D incremental analysis update (IAU) approach ([11]). In this study, PBL height data are being or have been generated from radiosondes, GNSS RO, satellite (CATS, CALIPSO and ICESat-2) and ground-based (MPLNET) lidars, and wind profiler. Investigations have been conducted to specify quality marks for PBL height retrievals for the data assimilation purpose. These PBL height data have different strengths and weaknesses ([2], [3], [6], [7], [8], [10]), and the satellite PBL height data provide better global coverage and complement in-situ PBL height data. Radiosondes offer high accuracy and in situ measurement of temperature and humidity profiles, but with poor spatio-temporal sampling. The in-situ observing systems like MPLNET and wind profiler provide long history of PBL height records at each station. The GNSS RO based PBL height is retrieved based on the sharp gradients in refractivity profile that represent the fine vertical structure of temperature and moisture changes above the PBL. However, not all RO refractivity profiles reach the surface depending on location and regime, and RO refractivity retrievals can be negatively biased below 2km. The PBL height data from satellite lidars provide high resolution along track PBL height retrievals, but over land they are affected by previous day convective PBL aerosol and strongly associated with mixing layer and retrievals cannot be made below thick, attenuating clouds. A successful assimilation of PBL height data requires a thorough understanding of the observing method and the retrieval algorithm for each observing system in order to use the PBL height data from multiple observing systems properly. Due to the sensitivity of PBL height data to the observing method and choice of algorithm, it is important to use a model definition appropriate for each observation type to compute differences between PBL height data and model PBL height (OmFs). The GEOS model currently includes two PBL height definitions suitable for direct comparison with observed PBL height, and additional definitions are being added in this study. Evaluation of different model PBL height definitions is underway. Meanwhile, efforts have been made in the GEOS data assimilation system to develop PBL height data assimilation capability. PBL height data can be assimilated using two different approaches. The traditional approach is to construct an observation operator and its tangent linear and adjoint, which link control variables to PBL height data from each observing system. This observation operator can be very complicated, e.g., the lidar-based PBL height observation operator includes the backscatter lidar forward observation operator, the algorithm to derive PBL height from attenuated total backscatter, interpolation, and calculations handling the mismatch between observed and model scales. The other approach is to augment PBL height to the control variable vector, and it is adopted in this study. The latter approach was also used in previous studies, e.g., the assimilation of PBL height data from radiosonde and aircraft in the Real Time Mesoscale Analysis (RTMA) system for a dispersion modelling study ([13]); the PBL height assimilation study using lidar PBL height data at Greensburg, Kansas for a field campaign ([12]). The PBL height assimilation from multiple observing systems in this study allows us to take advantage of the diverse PBL height data that provide much better global coverage collectively under different meteorological conditions and with different temporal and spatial scales. As all the PBL heights are tightly coupled with the PBL thermodynamic variables, the strong correlations, which are provided by the 4D ensemble forecast, enable PBL height data from various sources to interact and combine coherently and provide additional information for PBL temperature and moisture fields. The results of comparisons among PBL height data from different sources and the evaluation of the model PBL height definitions with the PBL height data will be presented, and the PBL height data synergy strategies and preliminary results will also be discussed at the conference.

Y. Zhu↗
NASA NTRS2016

Unstructured Grid Adaptation: Status, Potential Impacts, and Recommended Investments Toward CFD Vision 2030

Unstructured grid adaptation is a powerful tool to control discretization error for Computational Fluid Dynamics (CFD). It has enabled key increases in the accuracy, automation, and capacity of some fluid simulation applications. Slotnick et al. provides a number of case studies in the CFD Vision 2030 Study: A Path to Revolutionary Computational Aerosciences to illustrate the current state of CFD capability and capacity. The authors forecast the potential impact of emerging High Performance Computing (HPC) environments forecast in the year 2030 and identify that mesh generation and adaptivity continue to be significant bottlenecks in the CFD work flow. These bottlenecks may persist because very little government investment has been targeted in these areas. To motivate investment, the impacts of improved grid adaptation technologies are identified. The CFD Vision 2030 Study roadmap and anticipated capabilities in complementary disciplines are quoted to provide context for the progress made in grid adaptation in the past fifteen years, current status, and a forecast for the next fifteen years with recommended investments. These investments are specific to mesh adaptation and impact other aspects of the CFD process. Finally, a strategy is identified to diffuse grid adaptation technology into production CFD work flows.

Park, Michael A.↗
DOE OSTI2025

Sensitivity-based voltage constraints for optimal power flow in low-voltage distribution feeders

The optimal power flow (OPF) problem for distribution systems can include network details down to the low-voltage (LV) points of interconnection of individual customers. This paper addresses the implementation of voltage magnitude constraints, and sets forth a practicable approach for capturing the effects on voltage from the switching behavior of loads (e.g., heat pumps, air conditioners, water heaters, or pool pumps) and from the variability of renewable generation (e.g., rooftop solar). The proposed method adjusts the OPF voltage constraints based on forecasts of load and generation upper and lower bounds, in conjunction with sensitivity factors derived from the power flow equations. An illustrative OPF formulation is also provided, which incorporates transformer models that include core loss. We demonstrate that accurate modeling of these LV network components is critical to avoid voltage violations at customer points of interconnection. Furthermore, the ideas are validated through numerical case studies on a realistic distribution feeder.

24 POWER TRANSMISSION AND DISTRIBUTION↗
NASA NTRS2006

Forecast, Measurement, and Modeling of an Unprecedented Polar Ozone Filament Event over Mauna Loa Observatory, Hawaii

In mid-March 2005 the northern lower stratospheric polar vortex experienced a severe stretching episode, bringing a large polar filament far south of Alaska toward Hawaii. This meridional intrusion of rare extent, coinciding with the polar vortex final warming and breakdown, was followed by a zonal stretching in the wake of the easterly propagating subtropical main flow. This caused polar air to remain over Hawaii for several days before diluting into the subtropics. After being successfully forecasted to pass over Hawaii by the high-resolution potential vorticity advection model Modele Isentrope du transport Meso-echelle de l'Ozone Stratospherique par Advection (MIMOSA), the filament was observed on isentropic surfaces between 415 K and 455 K (17-20 km) by the Jet Propulsion Laboratory stratospheric ozone lidar measurements at Mauna Loa Observatory, Hawaii, between 16 and 19 March 2005. It was materialized as a thin layer of enhanced ozone peaking at 1.6 ppmv in a region where the climatological values usually average 1.0 ppmv. These values were compared to those obtained by the three dimensional Chemistry-Transport Model MIMOSA-CHIM. Agreement between lidar and model was excellent, particularly in the similar appearance of the ozone peak near 435 K (18.5 km) on 16 March, and the persistence of this layer at higher isentropic levels for the following three days. Passive ozone, also modeled by MIMOSA-CHIM, was at about 3-4 ppmv inside the filament while above Hawaii. A detailed history of the modeled chemistry inside the filament suggests that the air mass was still polar ozone- depleted when passing over Hawaii. The filament quickly separated from the main vortex after its Hawaiian overpass. It never reconnected and, in less than 10 days, dispersed entirely in the subtropics.

atmospheric processes↗
DOE OSTI2025

Preserving nonlinear constraints in variational flow filtering data assimilation

Data assimilation aims to estimate the states of a dynamical system by optimally combining sparse and noisy observations of the physical system with uncertain forecasts produced by a computational model. The states of many dynamical systems of interest obey nonlinear physical constraints, and the corresponding dynamics is confined to a certain sub-manifold of the state space. Standard data assimilation techniques applied to such systems yield posterior states lying outside the manifold, violating the physical constraints. This work focuses on particle flow filters which use stochastic differential equations to evolve state samples from a prior distribution to samples from an observation-informed posterior distribution. The variational Fokker-Planck (VFP)—a generic particle flow filtering framework—is extended to incorporate non-linear, equality state constraints in the analysis. To this end, two algorithmic approaches that modify the VFP stochastic differential equation are discussed: (i) VFPSTAB, to inexactly preserve constraints with the addition of a stabilizing drift term, and (ii) VFPDAE, to exactly preserve constraints by treating the VFP dynamics as a stochastic differential-algebraic equation (SDAE). Additionally, an implicit-explicit time integrator is developed to evolve the VFPDAE dynamics. The strength of the proposed approach for constraint preservation in data assimilation is demonstrated on three test problems: the double pendulum, Korteweg-de-Vries, and the incompressible Navier-Stokes equations.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS2005

Forecasting Lightning at Kennedy Space Center/Cape Canaveral Air Force Station, Florida

The Applied Meteorology Unit (AMU) developed a set of statistical forecast equations that provide a probability of lightning occurrence on Kennedy Space Center (KSC) I Cape Canaveral Air Force Station (CCAFS) for the day during the warm season (May September). The 45th Weather Squadron (45 WS) forecasters at CCAFS in Florida include a probability of lightning occurrence in their daily 24-hour and weekly planning forecasts, which are briefed at 1100 UTC (0700 EDT). This information is used for general scheduling of operations at CCAFS and KSC. Forecasters at the Spaceflight Meteorology Group also make thunderstorm forecasts for the KSC/CCAFS area during Shuttle flight operations. Much of the current lightning probability forecast at both groups is based on a subjective analysis of model and observational data. The objective tool currently available is the Neumann-Pfeffer Thunderstorm Index (NPTI, Neumann 1971), developed specifically for the KSCICCAFS area over 30 years ago. However, recent studies have shown that 1-day persistence provides a better forecast than the NPTI, indicating that the NPTI needed to be upgraded or replaced. Because they require a tool that provides a reliable estimate of the daily thunderstorm probability forecast, the 45 WS forecasters requested that the AMU develop a new lightning probability forecast tool using recent data and more sophisticated techniques now possible through more computing power than that available over 30 years ago. The equation development incorporated results from two research projects that investigated causes of lightning occurrence near KSCICCAFS and over the Florida peninsula. One proved that logistic regression outperformed the linear regression method used in NPTI, even when the same predictors were used. The other study found relationships between large scale flow regimes and spatial lightning distributions over Florida. Lightning, probabilities based on these flow regimes were used as candidate predictors in the equation development. Fifteen years (1 989-2003) of warm season data were used to develop the forecast equations. The data sources included a local network of cloud-to-ground lightning sensors called the Cloud-to-Ground Lightning Surveillance System (CGLSS), 1200 UTC Florida synoptic soundings, and the 1000 UTC CCAFS sounding. Data from CGLSS were used to determine lightning occurrence for each day. The 1200 UTC soundings were used to calculate the synoptic-scale flow regimes and the 1000 UTC soundings were used to calculate local stability parameters, which were used as candidate predictors of lightning occurrence. Five logistic regression forecast equations were created through careful selection and elimination of the candidate predictors. The resulting equations contain five to six predictors each. Results from four performance tests indicated that the equations showed an increase in skill over several standard forecasting methods, good reliability, an ability to distinguish between non-lightning and lightning days, and good accuracy measures and skill scores. Given the overall good performance the 45 WS requested that the equations be transitioned to operations and added to the current set of tools used to determine the daily lightning probability of occurrence.

Lambert, Winfred↗
NASA NTRS2018

Insight and Evidence Motivating the Simplification of Dual-Analysis Hybrid Systems into Single-Analysis Hybrid Systems

Many hybrid data assimilation systems currently used for NWP employ some form of dual-analysis system approach. Typically a hybrid variational analysis is responsible for creating initial conditions for high-resolution forecasts, and an ensemble analysis system is responsible for creating sample perturbations used to form the flow-dependent part of the background error covariance required in the hybrid analysis component. In many of these, the two analysis components employ different methodologies, e.g., variational and ensemble Kalman filter. In such cases, it is not uncommon to have observations treated rather differently between the two analyses components; recentering of the ensemble analysis around the hybrid analysis is used to compensated for such differences. Furthermore, in many cases, the hybrid variational high-resolution system implements some type of four-dimensional approach, whereas the underlying ensemble system relies on a three-dimensional approach, which again introduces discrepancies in the overall system. Connected to these is the expectation that one can reliably estimate observation impact on forecasts issued from hybrid analyses by using an ensemble approach based on the underlying ensemble strategy of dual-analysis systems. Just the realization that the ensemble analysis makes substantially different use of observations as compared to their hybrid counterpart should serve as enough evidence of the implausibility of such expectation. This presentation assembles numerous anecdotal evidence to illustrate the fact that hybrid dual-analysis systems must, at the very minimum, strive for consistent use of the observations in both analysis sub-components. Simpler than that, this work suggests that hybrid systems can reliably be constructed without the need to employ a dual-analysis approach. In practice, the idea of relying on a single analysis system is appealing from a cost-maintenance perspective. More generally, single-analysis systems avoid contradictions such as having to choose one sub-component to generate performance diagnostics to another, possibly not fully consistent, component.

Todling, Ricardo↗
DOE OSTI2025

pnnl/battery-degradation-prognosis

This is a tool for long term prognosis of redox flow battery parameter degradation which uses Transformer-based deep learning models for time series forecasting which was developed as part of the ROVI project.

Saldanha, Emily [Pacific Northwest National Labora↗
NASA NTRS1985

Evaluation of the satellite derived snow cover area - Runoff forecasting models for the inaccessible basins of western Himalayas

In this study, the existing seasonal snow cover area runoff forecasting models of the Indus, Kabul, Sutlej and Chenab basins were evaluated with the concurrent flow correlation model for the period 1975-79. In all the basins under study, correlation of concurrent flow model explained the variability in flow better than by the snow cover area runoff models. Actually, the concurrent flow correlation model explained more than 90 percent of the variability in the flow of these rivers. Compared to this model, the snow cover area runoff models explained less of the variability in flow. In the Himalayan river basins under study and at least for the period under observation, the concurrent flow correlation model provided a set of results with which to compare the estimates from the snow cover area runoff models.

Dey, B.↗
NASA NTRS1998

Coordinated Parameterization Development and Large-Eddy Simulation for Marine and Arctic Cloud-Topped Boundary Layers

The goal of this project was to compare observations of marine and arctic boundary layers with (i) parameterization systems used in climate and weather forecast models, and (ii) two and three dimensional eddy resolving (LES) models for turbulent fluid flow. Based on this comparison, we hoped to better understand, predict, and parameterize the boundary layer structure and cloud amount, type and thickness as functions of large scale conditions that are predicted by global climate models.

Bretherton, Christopher S.↗
NASA NTRS2004

Objective Lightning Forecasting at Kennedy Space Center/Cape Canaveral Air Force Station using Cloud-to-Ground Lightning Surveillance System Data

The 45th Weather Squadron (45 WS) forecasters at Cape Canaveral Air Force Station (CCAFS) in Florida include a probability of thunderstorm occurrence in their daily morning briefings. This information is used by personnel involved in determining the possibility of violating Launch Commit Criteria, evaluating Flight Rules for the Space Shuttle, and daily planning for ground operation activities on Kennedy Space Center (KSC)/CCAFS. Much of the current lightning probability forecast is based on a subjective analysis of model and observational data. The forecasters requested that a lightning probability forecast tool based on statistical analysis of historical warm-season (May - September) data be developed in order to increase the objectivity of the daily thunderstorm probability forecast. The tool is a set of statistical lightning forecast equations that provide a lightning occurrence probability for the day by 1100 UTC (0700 EDT) during the warm season. This study used 15 years (1989-2003) of warm season data to develop the objective forecast equations. The local CCAFS 1000 UTC sounding was used to calculate stability parameters for equation predictors. The Cloud-to-Ground Lightning Surveillance System (CGLSS) data were used to determine lightning occurrence for each day. The CGLSS data have been found to be more reliable indicators of lightning in the area than surface observations through local informal analyses. This work was based on the results from two earlier research projects. Everitt (1999) used surface observations and rawinsonde data to develop logistic regression equations that forecast the daily thunderstorm probability at CCAFS. The Everitt (1999) equations showed an improvement in skill over the Neumann-Pfeffer thunderstorm index (Neumann 1971), which uses multiple linear regression, and also persistence and climatology forecasts. Lericos et al. (2002) developed lightning distributions over the Florida peninsula based on specific flow regimes. The flow regimes were inferred from the average wind direction in the 1000-700 mb layer at Miami (MIA), Tampa (TBW), and Jacksonville (JAX), Florida, and the lightning data were from the National Lightning Detection Network. The results suggested that the daily flow regime may be an important predictor of lightning occurrence on KSC/CCAFS.

Lambert, Winifred↗
NASA NTRS1985

Response of the GLA fourth order model to changes in horizontal resolution and terrain heights

The effects the selection of the orographic data for the boundary conditions has on results of simulations with the NASA-Goddard Laboratory for Atmospherics fourth order GCM were examined. Two different lower boundary conditions were compared in generating preliminary weather forecasts: mean areally averaged heights, and an enhanced significant height orography for rugged terrain. The latter condition was developed to emphasize the effects the tallest peaks in a given region have on the atmospheric flow by considering only the highest 1/3 of the 1 deg x 1 deg values in the averaging process. Simulations were carried out for five 10 day forecasts with a 4 deg lat x 5 deg long resolution and six with a horizontal grid resolution of 2 deg lat by 2.5 deg long, three with each boundary condition. The rms errors of the sea level pressure and the 500 mb geopotential heights were calculated. The mean errors decreased after the second or third day with the enhanced significant height orography.

Pfaendtner, J.↗
NASA NTRS1990

Diabatic forcing and initialization with assimilation of cloud and rain water in a forecast model: Methodology

The focus of this part of the investigation is to find one or more general modeling techniques that will help reduce the time taken by numerical forecast models to initiate or spin-up precipitation processes and enhance storm intensity. If the conventional data base could explain the atmospheric mesoscale flow in detail, then much of our problem would be eliminated. But the data base is primarily synoptic scale, requiring that a solution must be sought either in nonconventional data, in methods to initialize mesoscale circulations, or in ways of retaining between forecasts the model generated mesoscale dynamics and precipitation fields. All three methods are investigated. The initialization and assimilation of explicit cloud and rainwater quantities computed from conservation equations in a mesoscale regional model are examined. The physical processes include condensation, evaporation, autoconversion, accretion, and the removal of rainwater by fallout. The question of how to initialize the explicit liquid water calculations in numerical models and how to retain information about precipitation processes during the 4-D assimilation cycle are important issues that are addressed. The explicit cloud calculations were purposely kept simple so that different initialization techniques can be easily and economically tested. Precipitation spin-up processes associated with three different types of weather phenomena are examined. Our findings show that diabatic initialization, or diabatic initialization in combination with a new diabatic forcing procedure, work effectively to enhance the spin-up of precipitation in a mesoscale numerical weather prediction forecast. Also, the retention of cloud and rain water during the analysis phase of the 4-D data assimilation procedure is shown to be valuable. Without detailed observations, the vertical placement of the diabatic heating remains a critical problem.

Raymond, William H.↗
NASA NTRS1988

Meteorological aspects of cloud-to-ground lightning in the Kennedy Space Center region

This paper examines the meteorological environment in which cloud-to-ground lightning occurs at the Kennedy Space Center (KSC) region during summer. Using the lightning patterns of central Florida, flashes from individual thunderstorms were analyzed for five low-level wind regimes during three summers. The results showed large differences between the number, timing, and location of flashes for varying flow regimes, with southwest flow being the most prolific producer of flashes in the KSC area. In the second part of the study, short-term forecasts of lightning were investigated using the divergence measured by the KSC surface wind network. An interval of about 1 hour was found from the beginning of a convergence event to first lightning. Monitoring 5-min surface convergence, in combination with understanding the climatology of lightning related to the daily flow regime, made it possible to isolate the lightning. Results were used successfully in operations at KSC.

Holle, Ronald L.↗
NASA NTRS2013

Ensemble Data Assimilation Without Ensembles: Methodology and Application to Ocean Data Assimilation

Two methods to estimate background error covariances for data assimilation are introduced. While both share properties with the ensemble Kalman filter (EnKF), they differ from it in that they do not require the integration of multiple model trajectories. Instead, all the necessary covariance information is obtained from a single model integration. The first method is referred-to as SAFE (Space Adaptive Forecast error Estimation) because it estimates error covariances from the spatial distribution of model variables within a single state vector. It can thus be thought of as sampling an ensemble in space. The second method, named FAST (Flow Adaptive error Statistics from a Time series), constructs an ensemble sampled from a moving window along a model trajectory. The underlying assumption in these methods is that forecast errors in data assimilation are primarily phase errors in space and/or time.

Data Assimilation↗
NASA NTRS2014

Background Error Covariance Estimation Using Information from a Single Model Trajectory with Application to Ocean Data Assimilation

An attractive property of ensemble data assimilation methods is that they provide flow dependent background error covariance estimates which can be used to update fields of observed variables as well as fields of unobserved model variables. Two methods to estimate background error covariances are introduced which share the above property with ensemble data assimilation methods but do not involve the integration of multiple model trajectories. Instead, all the necessary covariance information is obtained from a single model integration. The Space Adaptive Forecast error Estimation (SAFE) algorithm estimates error covariances from the spatial distribution of model variables within a single state vector. The Flow Adaptive error Statistics from a Time series (FAST) method constructs an ensemble sampled from a moving window along a model trajectory.SAFE and FAST are applied to the assimilation of Argo temperature profiles into version 4.1 of the Modular Ocean Model (MOM4.1) coupled to the GEOS-5 atmospheric model and to the CICE sea ice model. The results are validated against unassimilated Argo salinity data. They show that SAFE and FAST are competitive with the ensemble optimal interpolation (EnOI) used by the Global Modeling and Assimilation Office (GMAO) to produce its ocean analysis. Because of their reduced cost, SAFE and FAST hold promise for high-resolution data assimilation applications.

Error Covariance↗
NASA NTRS2014

Background Error Covariance Estimation using Information from a Single Model Trajectory with Application to Ocean Data Assimilation into the GEOS-5 Coupled Model

An attractive property of ensemble data assimilation methods is that they provide flow dependent background error covariance estimates which can be used to update fields of observed variables as well as fields of unobserved model variables. Two methods to estimate background error covariances are introduced which share the above property with ensemble data assimilation methods but do not involve the integration of multiple model trajectories. Instead, all the necessary covariance information is obtained from a single model integration. The Space Adaptive Forecast error Estimation (SAFE) algorithm estimates error covariances from the spatial distribution of model variables within a single state vector. The Flow Adaptive error Statistics from a Time series (FAST) method constructs an ensemble sampled from a moving window along a model trajectory. SAFE and FAST are applied to the assimilation of Argo temperature profiles into version 4.1 of the Modular Ocean Model (MOM4.1) coupled to the GEOS-5 atmospheric model and to the CICE sea ice model. The results are validated against unassimilated Argo salinity data. They show that SAFE and FAST are competitive with the ensemble optimal interpolation (EnOI) used by the Global Modeling and Assimilation Office (GMAO) to produce its ocean analysis. Because of their reduced cost, SAFE and FAST hold promise for high-resolution data assimilation applications.

Data Assimilation↗
DOE OSTI2024

Forecasting high-dimensional spatio-temporal systems from sparse measurements

This paper introduces a new neural network architecture designed to forecast high-dimensional spatio-temporal data using only sparse measurements. The architecture uses a two-stage end-to-end framework that combines neural ordinary differential equations (NODEs) with vision transformers. Initially, our approach models the underlying dynamics of complex systems within a low-dimensional space; and then it reconstructs the corresponding high-dimensional spatial fields. Many traditional methods involve decoding high-dimensional spatial fields before modeling the dynamics, while some other methods use an encoder to transition from high-dimensional observations to a latent space for dynamic modeling. In contrast, our approach directly uses sparse measurements to model the dynamics, bypassing the need for an encoder. This direct approach simplifies the modeling process, reduces computational complexity, and enhances the efficiency and scalability of the method for large datasets. We demonstrate the effectiveness of our framework through applications to various spatio-temporal systems, including fluid flows and global weather patterns. Although sparse measurements have limitations, our experiments reveal that they are sufficient to forecast system dynamics accurately over long time horizons. Our results also indicate that the performance of our proposed method remains robust across different sensor placement strategies, with further improvements as the number of sensors increases. This robustness underscores the flexibility of our architecture, particularly in real-world scenarios where sensor data is often sparse and unevenly distributed.

97 MATHEMATICS AND COMPUTING↗