Robust Bayesian Optimal Experimental Design for Data Fusion with Model Discrepancy
Explore the source record for details and available documents.
SEARCH · Search NASA
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.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Data fusion is the operation of integrating data from different modalities to construct a single consistent representation. Here, this paper proposes variations of coupled dictionary learning through an additional factorization. One variation of this model is applicable to the pansharpening data fusion problem. Real world pansharpening data was applied to train and test our proposed formulation. The results demonstrate that the data fusion model can successfully be applied to the pan-sharpening problem.
This report is a brief overview and evaluation of Physics Informed Neural Networks (PINNs). Karniadakis and co-workers, e.g., Karniadakis et al. (2021) assert that the PINNs approach integrates seamlessly both data and mathematical physics models, even in partially understood, uncertain and high-dimensional contexts. They further claim that PINNs are effective and efficient for ill-posed and inverse problems, and when combined with domain decomposition, are scalable to large problems and a tool to discover hidden physics. While they demonstrate the capabilities in specific academic instances, their overarching claims about PINNs seem to be an overstatement, at least at the current time. We have briefly considered a few of the limitations of PINNs in this investigation. It is not clear to us if the PINNs approach can ever be competitive with approaches that use specialized algorithms to achieve high-accuracy solutions of governing equations and other techniques that can combine observational data with such solutions. As an example of the latter, consistent with the principles of Bayesian inference, data assimilation, or more generally data-model fusion, is a process that fuses observational data typically with a computational model that respects certain constraints such as conservation laws. For example, improvements in observational network combined with data assimilation have been key in improving weather predictions over the past four decades Kalnay (2003).
Predicting how forest carbon cycling will change in response to climate change and management depends on the collective knowledge from measurements across environmental gradients, ecosystem manipulations of global change factors, and mathematical models. Formally integrating these sources of knowledge through data assimilation, or model–data fusion, allows the use of past observations to constrain model parameters and estimate prediction uncertainty. Data assimilation (DA) focused on the regional scale has the opportunity to integrate data from both environmental gradients and experimental studies to constrain model parameters. Here, we introduce a hierarchical Bayesian DA approach (Data Assimilation to Predict Productivity for Ecosystems and Regions, DAPPER) that uses observations of carbon stocks, carbon fluxes, water fluxes, and vegetation dynamics from loblolly pine plantation ecosystems across the southeastern US to constrain parameters in a modified version of the Physiological Principles Predicting Growth (3-PG) forest growth model. The observations included major experiments that manipulated atmospheric carbon dioxide (CO 2 ) concentration, water, and nutrients, along with nonexperimental surveys that spanned environmental gradients across an 8.6 × 10 5 km 2 region. We optimized regionally representative posterior distributions for model parameters, which dependably predicted data from plots withheld from the data assimilation. While the mean bias in predictions of nutrient fertilization experiments, irrigation experiments, and CO 2 enrichment experiments was low, future work needs to focus modifications to model structures that decrease the bias in predictions of drought experiments. Predictions of how growth responded to elevated CO 2 strongly depended on whether ecosystem experiments were assimilated and whether the assimilated field plots in the CO 2 study were allowed to have different mortality parameters than the other field plots in the region. We present predictions of stem biomass productivity under elevated CO 2 , decreased precipitation, and increased nutrient availability that include estimates of uncertainty for the southeastern US. Overall, we (1) demonstrated how three decades of research in southeastern US planted pine forests can be used to develop DA techniques that use multiple locations, multiple data streams, and multiple ecosystem experiment types to optimize parameters and (2) developed a tool for the development of future predictions of forest productivity for natural resource managers that leverage a rich dataset of integrated ecosystem observations across a region.
Probabilistic aerodynamic databases are a crucial component of the development lifecycle for aerospace vehicles. A key challenge when building aerodynamic databases is that most data used to construct them represent various simplifications of the real flight vehicle. For example, wind tunnel models often simplify the vehicle geometry and surface roughness characteristics, while CFD computations often make simplifications to the physics being modeled, such as fully laminar or turbulent calculations. Multifidelity data fusion models rely on a user being able to define a hierarchy of fidelity levels anchored to some "truth" data. This approach is unsatisfactory when no data can be considered to accurately reflect real flight conditions. In this work, we provide an alternative approach by presenting a consistent mathematical framework for building probabilistic aerodynamic databases in the form of a conditional probability distribution described by an ensemble of multifidelity Gaussian Processes. Instead of relying on a single hierarchy of data fidelity levels, the presented framework identifies a "nominal" configuration and potential corrections to the nominal which represent specific physical phenomena not represented in the nominal data. The nominal and correction functions themselves are constructed as multifidelity Gaussian Processes and linearly combined to form an ensemble model which fuses the uncertainties associated nominal and correction models. Results obtained using the proposed framework on a simplified Orion Crew Module wind tunnel dataset demonstrate the predictive capability of the multihierarchy framework. We further demonstrate the benefits of such a probabilistic aerodynamic database approach through function sampling and computing the conditional distributions of derived quantities, such as the trim angle of attack and aerodynamic coefficients at trim.
Abstract Wetlands are the largest natural source of methane (CH 4 ) to the atmosphere. The eddy covariance method provides robust measurements of net ecosystem exchange of CH 4 , but interpreting its spatiotemporal variations is challenging due to the co‐occurrence of CH 4 production, oxidation, and transport dynamics. Here, we estimate these three processes using a data‐model fusion approach across 25 wetlands in temperate, boreal, and Arctic regions. Our data‐constrained model—iPEACE—reasonably reproduced CH 4 emissions at 19 of the 25 sites with normalized root mean square error of 0.59, correlation coefficient of 0.82, and normalized standard deviation of 0.87. Among the three processes, CH 4 production appeared to be the most important process, followed by oxidation in explaining inter‐site variations in CH 4 emissions. Based on a sensitivity analysis, CH 4 emissions were generally more sensitive to decreased water table than to increased gross primary productivity or soil temperature. For periods with leaf area index (LAI) of ≥20% of its annual peak, plant‐mediated transport appeared to be the major pathway for CH 4 transport. Contributions from ebullition and diffusion were relatively high during low LAI (<20%) periods. The lag time between CH 4 production and CH 4 emissions tended to be short in fen sites (3 ± 2 days) and long in bog sites (13 ± 10 days). Based on a principal component analysis, we found that parameters for CH 4 production, plant‐mediated transport, and diffusion through water explained 77% of the variance in the parameters across the 19 sites, highlighting the importance of these parameters for predicting wetland CH 4 emissions across biomes. These processes and associated parameters for CH 4 emissions among and within the wetlands provide useful insights for interpreting observed net CH 4 fluxes, estimating sensitivities to biophysical variables, and modeling global CH 4 fluxes.
Air quality managers, researchers, and concerned community scientists around the world have a variety of sources for air quality information, ranging from traditional regulatory monitoring networks and atmospheric chemistry models to remote sensing data products and low-cost sensor networks. However, the ability to incorporate data from these disparate sources and synthesize a comprehensive overview of the local air quality situation remains a considerable barrier for many end-users. This presentation will outline a tool, currently in development, which will address this need using a flexible data fusion approach. The tool will make use of air quality forecast model outputs (primarily from the NASA GEOS-CF composition forecast modeling system), satellite remote sensing data (from instruments including MODIS, VIIRS, TROPOMI, plus TEMPO for the US when available), and in-situ data from official regulatory and/or low-cost networks where these are available. The ability to incorporate data from low-cost sensor networks will be a key feature of the tool; it will make use of other available data sources to calibrate the low-cost sensor data on a regional scale, then use these calibrated low-cost sensor data for localized updating to resolve finer-scale air quality patterns. Development of this tool is taking place with the help of national and international partners and end-user groups, coordinated through the US EPA and the United Nations Environment Programme (UNEP). The tool is being developed on the Google Earth Engine cloud computing platform to facilitate integration of diverse data sources and free access by a broad community of end-users. Stewardship of the tool will be passed to US EPA and UNEP to support future activities with end-users in the US and around the world, and the tool itself will remain freely accessible. We hope that this tool will lower the barrier to entry for various user groups worldwide, including community scientists, who struggle to integrate disparate data sources to gain insight into their local air quality situations. This presentation will cover the early stages of the development of the tool, including the underlying methods and some pilot case studies in integrating low-cost sensor data.
The rapid increase in the volume and variety of terrestrial biosphere observations (i.e., remote sensing data and in situ measurements) offers a unique opportunity to derive ecological insights, refine process‐based models, and improve forecasting for decision support. However, despite their potential, ecological observations have primarily been used to benchmark process‐based models, as many past and current models lack the capability to directly integrate observations and their associated uncertainties for parameterization. In contrast, data assimilation frameworks such as the CARbon DAta MOdel fraMework (CARDAMOM) and its suite of process‐based models, known as the Data Assimilation Linked Ecosystem Carbon Model (DALEC), are specifically designed for model‐data fusion. This review, motivated by a recent CARDAMOM community workshop, examines the development and applications of CARDAMOM, with an emphasis on its role in advancing ecosystem process understanding. CARDAMOM employs a Bayesian approach, using a Markov Chain Monte Carlo algorithm to enable data‐driven calibration of DALEC parameters and initial states (i.e., carbon pool sizes) through observation operators. CARDAMOM's unique ability to retrieve localized model process parameters from diverse datasets—ranging from in situ measurements to global satellite observations—makes it a highly flexible tool for analyzing spatially variable ecosystem responses to environmental change. However, assimilating these data also presents challenges, including data quality issues that propagate into model skill, as well as trade‐offs between model complexity, parameter equifinality, and predictive performance. We discuss potential solutions to these challenges, such as reducing parameter equifinality by incorporating new observations. This review also offers community recommendations for incorporating emerging datasets, integrating machine learning techniques, strengthening collaboration with remote sensing, field, and modeling communities, and expanding CARDAMOM's relevance for localized ecosystem monitoring and decision‐making. CARDAMOM enables a deep, mechanistic understanding of terrestrial ecosystem dynamics that cannot be achieved through empirical analyses of observational datasets or weakly constrained models alone.
The research proposed here focuses on improving the predictive power of the land component of earth system models (ESMs) using (1) model-data fusion enabled by machine learning (ML) and artificial intelligence (AI), (2) predictive modeling through the combination of ML, AI, and big-data (comprising both model output and observations), and (3) insight of ESM structure and process mechanisms gleaned from complex data using ML and AI.
Abstract We partitioned the soil carbon dioxide flux ( R s ) into its respective autotrophic and heterotrophic components in a mature temperate‐boreal forest (Howland Forest in Maine, USA). We combined automated chamber measurements of R s with two different partitioning methods: (1) a classic root trenching experiment and (2) a radiocarbon ( 14 C) mass balance approach. With a model‐data fusion approach, we used these data to constrain a parsimonious ecosystem model (FöBAAR), and we investigated differences in modeled C fluxes and pools under both current and future climate scenarios. The trenching experiment indicated that heterotrophic respiration accounted for 53 ± 11% of total R s . In comparison, using the 14 C method, the heterotrophic contribution was 42 ± 9%. For both current and future model runs, incorporating the partitioning data as constraints substantially reduced the uncertainties of autotrophic and heterotrophic respiration fluxes. Moreover, with best fit model parameters, the two partitioning methods yielded fundamentally different estimates of the relative contributions of autotrophic and heterotrophic respiration to total R s , especially at the annual time scale. Surprisingly, however, modeled soil C and biomass C pool size trajectories did not differ significantly between model runs based on the different methods. Instead, model differences in partitioning were compensated for by changes in C allocation, resulting in similar, but still highly uncertain, soil C pool trajectories. Our findings show that incorporating constraints on the partitioning of R s can reduce model uncertainties of fluxes but not pools, and the results are sensitive to the partitioning method used.
Agriculture contributes nearly a quarter of global greenhouse gas (GHG) emissions, which is motivating interest in adopting certain farming practices that have the potential to reduce GHG emissions or sequester carbon in soil. The related GHG emission (including N 2 O and CH 4 ) and changes in soil carbon stock are defined here as “agricultural carbon outcomes”. Accurate quantification of agricultural carbon outcomes is the basis for achieving emission reductions for agriculture, but existing approaches for measuring carbon outcomes (including direct measurements, emission factors, and process-based modeling) fall short of achieving the required accuracy and scalability necessary to support credible, verifiable, and cost-effective measurement and improvement of these carbon outcomes. Here we propose a foundational and scalable framework to quantify field-level carbon outcomes for farmland, which is based on the holistic carbon balance of the agroecosystem: Agroecosystem Carbon Outcomes = Environment (E) × Management (M) × Crop (C). Following a comprehensive review of the scientific challenges associated with existing approaches, as well as their tradeoffs between cost and accuracy, we propose that the most viable path for the quantification of field-level carbon outcomes in agricultural land is through an effective integration of various approaches (e.g. diverse observations, sensor/in-situ data, and modeling), defined as the “System-of-Systems” solution. Such a “System-of-Systems” solution should simultaneously comprise the following components: (1) scalable collection of ground truth data and cross-scale sensing of environment variables (E), management practices (M), and crop conditions (C) at the local field level; (2) advanced modeling with necessary processes to support the quantification of carbon outcomes; (3) systematic Model-Data Fusion (MDF), i.e. robust and efficient methods to integrate sensing data and models at each local farmland level; (4) high computation efficiency and artificial intelligence (AI) to scale to millions of individual fields with low cost; and (5) robust and multi-tier validation systems and infrastructures to ensure solution fidelity and true scalability, i.e. the ability of a solution to perform robustly with accepted accuracy on all targeted fields. In this regard, we provide here the detailed scientific rationale, current progress, and future research and development (R&D) priorities to achieve different components of the “System-of-Systems” solution, thus accomplishing the Environment×Management×Crop framework to quantify field-level agricultural carbon outcomes.
The Feature Analyst is a computer program for assisted (partially automated) recognition of targets in images. This program was developed to accelerate the processing of high-resolution satellite image data for incorporation into geographic information systems (GIS). This program creates an advanced user interface that embeds proprietary machine-learning algorithms in commercial image-processing and GIS software. A human analyst provides samples of target features from multiple sets of data, then the software develops a data-fusion model that automatically extracts the remaining features from selected sets of data. The program thus leverages the natural ability of humans to recognize objects in complex scenes, without requiring the user to explain the human visual recognition process by means of lengthy software. Two major subprograms are the reactive agent and the thinking agent. The reactive agent strives to quickly learn the user's tendencies while the user is selecting targets and to increase the user's productivity by immediately suggesting the next set of pixels that the user may wish to select. The thinking agent utilizes all available resources, taking as much time as needed, to produce the most accurate autonomous feature-extraction model possible.
The Feature Analyst is a computer program for assisted (partially automated) recognition of targets in images. This program was developed to accelerate the processing of high-resolution satellite image data for incorporation into geographic information systems (GIS). This program creates an advanced user interface that embeds proprietary machine-learning algorithms in commercial image-processing and GIS software. A human analyst provides samples of target features from multiple sets of data, then the software develops a data-fusion model that automatically extracts the remaining features from selected sets of data. The program thus leverages the natural ability of humans to recognize objects in complex scenes, without requiring the user to explain the human visual recognition process by means of lengthy software. Two major subprograms are the reactive agent and the thinking agent. The reactive agent strives to quickly learn the user s tendencies while the user is selecting targets and to increase the user s productivity by immediately suggesting the next set of pixels that the user may wish to select. The thinking agent utilizes all available resources, taking as much time as needed, to produce the most accurate autonomous feature-extraction model possible.
We describe a model-data fusion (MDF) inter-comparison project (REFLEX), which compared various algorithms for estimating carbon (C) model parameters consistent with both measured carbon fluxes and states and a simple C model. Participants were provided with the model and with both synthetic net ecosystem exchange (NEE) ofCO2 and leaf area index (LAI) data, generated from the model with added noise, and observed NEE and LAI data from two eddy covariance sites. Participants endeavoured to estimate model parameters and states consistent with the model for all cases over the two years for which data were provided, and generate predictions for one additional year without observations. Nine participants contributed results using Metropolis algorithms, Kalman filters and a genetic algorithm. For the synthetic data case, parameter estimates compared well with the true values. The results of the analyses indicated that parameters linked directly to gross primary production (GPP) and ecosystem respiration, such as those related to foliage allocation and turnover, or temperature sensitivity of heterotrophic respiration,were best constrained and characterised. Poorly estimated parameters were those related to the allocation to and turnover of fine root/wood pools. Estimates of confidence intervals varied among algorithms, but several algorithms successfully located the true values of annual fluxes from synthetic experiments within relatively narrow 90% confidence intervals, achieving>80% success rate and mean NEE confidence intervals <110 gCm-2 year-1 for the synthetic case. Annual C flux estimates generated by participants generally agreed with gap-filling approaches using half-hourly data. The estimation of ecosystem respiration and GPP through MDF agreed well with outputs from partitioning studies using half-hourly data. Confidence limits on annual NEE increased by an average of 88% in the prediction year compared to the previous year, when data were available. Confidence intervals on annual NEE increased by 30% when observed data were used instead of synthetic data, reflecting and quantifying the addition of model error. Finally, our analyses indicated that incorporating additional constraints, using data on C pools (wood, soil and fine roots) would help to reduce uncertainties for model parameters poorly served by eddy covariance data.