Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Inference”

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 1,513 records · Page 84

Numerical study of a micro flow reactor at engine pressures: Flames with repetitive extinction and ignition and simulations with a reduced chemical model

Combustion in a narrow channel with an imposed temperature gradient is studied numerically at elevated pressure with engine-relevant fuels. Here, the focus is placed on unsteady flames with repetitive extinction and ignition (FREI) to determine the potential of this regime for fuel testing and calibration of reduced chemical mechanisms. First, it is shown that the FREI regime does occur at elevated pressures for sufficiently small tube diameters. The sensitivity of the extinction and ignition temperatures to low-temperature chemistry is found to be significantly enhanced at 25 bar compared to atmospheric conditions. The ignition and extinction temperatures differ by up to 100 K between PRF mixtures with varying octane numbers. Ternary mixtures of iso-octane/n-heptane/toluene and iso-octane/n-heptane/ethanol at similar research and motor octane numbers are also studied. Second, the potential of using data from the micro flow reactor to infer reaction rates is assessed. A reduced chemical mechanism combining a small fuel-dependent submechanism with a detailed fuel-independent submechanism for the core species chemistry is used for that purpose. Only the most sensitive fuel-dependent reactions are inferred. the calibrated reduced model is compared to a detailed model and good agreement in ignition delay times and laminar flame speeds is observed. Finally, this illustrates the potential of a micro flow reactor with a controlled temperature profile to test fuels and infer kinetic data.

33 ADVANCED PROPULSION SYSTEMS↗

Detecting regular sound changes in linguistics as events of concerted evolution

Background: Concerted evolution is normally used to describe parallel changes at different sites in a genome, but it is also observed in languages where a specific phoneme changes to the same other phoneme in many words in the lexicon—a phenomenon known as regular sound change. We develop a general statistical model that can detect concerted changes in aligned sequence data and apply it to study regular sound changes in the Turkic language family. Results: Linguistic evolution, unlike the genetic substitutional process, is dominated by events of concerted evolutionary change. Our model identified more than 70 historical events of regular sound change that occurred throughout the evolution of the Turkic language family, while simultaneously inferring a dated phylogenetic tree. Including regular sound changes yielded an approximately 4-fold improvement in the characterization of linguistic change over a simpler model of sporadic change, improved phylogenetic inference, and returned more reliable and plausible dates for events on the phylogenies. The historical timings of the concerted changes closely follow a Poisson process model, and the sound transition networks derived from our model mirror linguistic expectations. Conclusions: We demonstrate that a model with no prior knowledge of complex concerted or regular changes can nevertheless infer the historical timings and genealogical placements of events of concerted change from the signals left in contemporary data. Our model can be applied wherever discrete elements—such as genes, words, cultural trends, technologies, or morphological traits—can change in parallel within an organism or other evolving group.

59 BASIC BIOLOGICAL SCIENCES↗

Variational Bayesian Blind Image Deconvolution: A review

In this study, we provide a review of the recent literature on Bayesian Blind Image Deconvolution (BID) methods. We believe that two events have marked the recent history of BID: the predominance of Variational Bayes (VB) inference as a tool to solve BID problems and the increasing interest of the computer vision community in solving BID problems. VB inference in combination with recent image models like the ones based on Super Gaussian (SG) and Scale Mixture of Gaussians (SMG) representations have led to the use of very general and powerful tools to provide clear images from blurry observations. In the provided review emphasis is paid on VB inference and the use of SG and SMG models with coverage of recent advances in sampling methods. We also provide examples of current state of the art BID methods and discuss problems that very likely will mark the near future of BID.

97 MATHEMATICS AND COMPUTING↗

Protein-based forensic identification using genetically variant peptides in human bone

Bone tissue contains organic material that is useful for forensic investigations and may contain preserved endogenous protein that can persist in the environment for extended periods of time over a range of conditions. Single amino acid polymorphisms in these proteins reflect genetic information since they result from non-synonymous single nucleotide polymorphisms (SNPs) in DNA. Detection of genetically variant peptides (GVPs) — those peptides that contain amino acid polymorphisms — in digests of bone proteins allows for the corresponding SNP alleles to be inferred. Resulting genetic profiles can be used to calculate statistical measures of association between a bone sample and an individual. In this study proteomic analysis on rib cortical bone samples from 10 recently deceased individuals demonstrates this concept. A straight-forward acidic demineralization protocol yielded proteins that were digested with trypsin. Tryptic digests were analyzed by liquid chromatography mass spectrometry. A total of 1736 different proteins were identified across all resulting datasets. On average, individual samples contained 454 ± 121 ($\bar{x}$ ± σ) proteins. Thirty-five genetically variant peptides were identified from 15 observed proteins. Overall, 134 SNP inferences were made based on proteomically detected GVPs, which were confirmed by sequencing of subject DNA. Inferred individual SNP genetic profiles ranged in random match probability (RMP) from 1/6 to 1/42,472 when calculated with European population frequencies in the 1000 Genomes Project, Phase 3. Similarly, RMPs based on African population frequencies were calculated for each SNP genetic profile and likelihood ratios (LR) were obtained by dividing each European RMP by the corresponding African RMP. In conclusion, resulting LR values ranged from 1.4 to 825 with a median value of 16. GVP markers offer a basis for the identification of compromised skeletal remains independent of the presence of DNA template.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

An automated design process for short pulse laser driven opacity experiments

Stellar-relevant conditions can be reached by heating a buried layer target with a short pulse laser. Previous design studies of iron buried layer targets found that plasma conditions are dominantly controlled by the laser energy while the accuracy of the inferred opacity is limited by tamper emission and optical depth effects. In this paper, we developed a process to simultaneously optimize laser and target parameters to meet a variety of design goals. We explored two sets of design cases: a set focused on conditions relevant to the upper radiative zone of the sun (electron temperatures of 200 to 400 eV and densities greater than 1/10 of solid density) and a set focused on reaching temperatures consistent with deep within the radiative zone of the sun (500 to 1000 eV) at a fixed density. We found optimized designs for iron targets and determined that the appropriate dopant, for inferring plasma conditions, depends on the goal temperature: magnesium for up to 300 eV, aluminum for 300 to 500 eV, and sulfur for 500 to 1000 eV. The optimal laser energy and buried layer thickness increase with goal temperature. The accuracy of the inferred opacity is limited to between 11% and 31%, depending on the design. Finally, overall, short pulse laser heated iron experiments reaching stellar-relevant conditions have been designed with consideration of minimizing tamper emission and optical depth effects while meeting plasma condition and x-ray emission goals.

36 MATERIALS SCIENCE↗

Overview of oxygen opacity experiments at the National Ignition Facility and investigation of potential systematic errors

Experiments to measure oxygen opacity at stellar interior conditions have been performed at the National Ignition Facility in a Discovery Science campaign. These experiments utilize the Opacity-on-NIF platform with a sample comprised of O, Mg, and Si. The spectral data from the Opacity Spectrometer cover the 1000–2000 eV photon energy range showing bound-free continuum absorption from O and line absorption from Mg and Si. DANTE and the Gated X-ray Detector are employed to measure the sample plasma’s temperature and density, respectively. Initial data show lower transmission than expected by theoretical models, raising questions of whether potential background or data uniformity concerns could produce systematic errors in the inferred transmission. Here, we investigate three concerns thought to be important for the oxygen opacity data, including instrumental scattered background, sample self-emission non-uniformity, and backlight continuum non-uniformity. Additionally, we show the effect of a recently developed method to account for 2nd order crystal reflection. The total effect of these concerns on one experiment is found to be small compared to the observed difference between the inferred transmission and a model calculation at the inferred temperature and density. Thus, we conclude that these potential sources of systematic error cannot account for the observed difference, increasing the likelihood of a real effect due to the high temperature and density conditions. However, because this is only a single experiment, we cannot make a firm conclusion. More experiments measuring the opacity and necessary calibrations are needed to assess the reproducibility and uncertainty of this result.

79 ASTRONOMY AND ASTROPHYSICS↗

Simulation-based optimal Bayesian experimental design for nonlinear systems

The optimal selection of experimental conditions is essential to maximizing the value of data for inference and prediction, particularly in situations where experiments are time-consuming and expensive to conduct. We propose a general mathematical framework and an algorithmic approach for optimal experimental design with nonlinear simulation-based models; in particular, we focus on finding sets of experiments that provide the most information about targeted sets of parameters. Our framework employs a Bayesian statistical setting, which provides a foundation for inference from noisy, indirect, and incomplete data, and a natural mechanism for incorporating heterogeneous sources of information. An objective function is constructed from information theoretic measures, reflecting expected information gain from proposed combinations of experiments. Polynomial chaos approximations and a two-stage Monte Carlo sampling method are used to evaluate the expected information gain. Stochastic approximation algorithms are then used to make optimization feasible in computationally intensive and high-dimensional settings. These algorithms are demonstrated on model problems and on nonlinear parameter inference problems arising in detailed combustion kinetics.

97 MATHEMATICS AND COMPUTING↗

Multi-fidelity Bayesian neural networks: Algorithms and applications

Here we propose a new class of Bayesian neural networks (BNNs) that can be trained using noisy data of variable fidelity, and we apply them to learn function approximations as well as to solve inverse problems based on partial differential equations (PDEs). These multi-fidelity BNNs consist of three neural networks: The first is a fully connected neural network, which is trained following the maximum a posteriori probability (MAP) method to fit the low-fidelity data; the second is a Bayesian neural network employed to capture the cross-correlation with uncertainty quantification between the low- and high-fidelity data; and the last one is the physics-informed neural network, which encodes the physical laws described by PDEs. For the training of the last two neural networks, we first employ the mean-field variational inference (VI) to maximize the evidence lower bound (ELBO) to obtain informative prior distributions for the hyperparameters in the BNNs, and subsequently we use the Hamiltonian Monte Carlo (HMC) method to estimate accurately the posterior distributions for the corresponding hyperparameters. We demonstrate the accuracy of the present method using synthetic data as well as real measurements. Specifically, we first approximate a one- and four-dimensional function, and then infer the reaction rates in one- and two-dimensional diffusion-reaction systems. Moreover, we infer the sea surface temperature (SST) in the Massachusetts and Cape Cod Bays using satellite images and in-situ measurements. Taken together, our results demonstrate that the present method can capture both linear and nonlinear correlation between the low- and high-fidelity data adaptively, identify unknown parameters in PDEs, and quantify uncertainties in predictions, given a few scattered noisy high-fidelity data. Finally, we demonstrate that we can effectively and efficiently reduce the uncertainties and hence enhance the prediction accuracy with an active learning approach, using as examples a specific one-dimensional function approximation and an inverse PDE problem.

97 MATHEMATICS AND COMPUTING↗

Active operator learning with predictive uncertainty quantification for partial differential equations

With the increased prevalence of neural operators being used to provide rapid solutions to partial differential equations (PDEs), understanding the accuracy of model predictions and the associated error levels is necessary for deploying reliable surrogate models in scientific applications. Existing uncertainty quantification (UQ) frameworks employ ensembles or Bayesian methods, which can incur substantial computational costs during both training and inference. Here, we propose a lightweight predictive UQ method tailored for Deep operator networks (DeepONets) that also generalizes to other operator networks. Numerical experiments on linear and nonlinear PDEs demonstrate that the framework’s uncertainty estimates are unbiased and provide accurate out-of-distribution uncertainty predictions with a sufficiently large training dataset. Our framework provides fast inference and uncertainty estimates that can efficiently drive outer-loop analyses that would be prohibitively expensive with conventional solvers. We demonstrate how predictive uncertainties can be used in the context of Bayesian optimization and active learning problems to yield improvements in accuracy and data-efficiency for outer-loop optimization procedures. In the active learning setup, we extend the framework to Fourier Neural Operators (FNO) and describe a generalized method for other operator networks. To enable real-time deployment, we introduce an inference strategy based on precomputed trunk outputs and a sparse placement matrix, reducing evaluation time by more than a factor of five. Our method provides a practical route to uncertainty-aware operator learning in time-sensitive settings.

97 MATHEMATICS AND COMPUTING↗

Prediction of electric and magnetic fields from spectral data using machine learning algorithms for Doppler-free saturation spectroscopy diagnostics

The prediction of electric and magnetic field amplitudes from atomic spectral data is critical for plasma control in fusion devices such as tokamaks. Conventional approaches that rely on physics-based models are computationally expensive and unsuitable for real-time applications. In this work, we develop and benchmark three machine learning algorithms—simulation-based inference (SBI), fully connected neural networks (FCNN), and histogram-based gradient boosting regression (GBR-Hist)—to infer field intensities directly from Doppler-free saturation spectroscopy (DFSS) spectra. Synthetic datasets of spectra were generated using the EZSSS code and evaluated both with and without added Poisson noise to mimic experimental conditions. We find that SBI achieves the highest accuracy and robustness, FCNN provides a strong balance of accuracy and computational efficiency for real-time applications, and GBR-Hist offers the fastest inference but is more sensitive to noise. Furthermore, these results demonstrate the potential of machine learning to accelerate DFSS analysis and enhance its utility for plasma diagnostics and control.

Doppler-free saturation spectroscopy↗

25 April 2015 Gorkha Earthquake in Nepal Himalaya (Part 2)

The response from the geosciences community working on Himalaya in general, and the 2015 Nepal earthquakes in specific, was overwhelming, and after a rigorous review process, thirteen papers were selected and published in Part-1. We are still left with a few good papers which are being brought out as Part–2 of the special issue. Here, we describe how in the opening article Jean-Louis Mugnier and colleagues attempt to provide a structural geological perspective of the 25 April 2015 Gorkha earthquake and highlight the role of segmentation in generating the Himalayan mega-thrusts. They could infer segmentation by stable barriers in the HT that define barrier-type earthquake families. In yet another interesting piece of work, Pandey and colleagues map the crustal structure across the earthquake volume using Receiver function approach and infer a 5-km thick low velocity layer that connects to the MHT ramp. They are also able to correlate the rupture termination with the highest point of coseismic uplift. The last paper by Shen et al. highlights the usefulness of INSAR technique in mapping the coseismic slip distribution applied to the 25 April 2015 Gorkha earthquake. Finally, they infer low stress drop and corner frequency which coupled with hybrid modeling explain the low level of slip heterogeneity and frequency of ground motion.

Rao, N. Purnachandra↗

Shared micromobility as a first- and last-mile transit solution? Spatiotemporal insights from a novel dataset

The first- and last-mile (FM/LM) problem is a major deterrent to public transit use. With the rise of shared micromobility options such as shared e-scooters in recent years, there is a growing interest in understanding their potential to serve as a last-mile transit solution. However, empirical data regarding the integrated use of shared micromobility and public transit have been limited so far. As a result, much is unknown regarding the spatiotemporal patterns and characteristics of shared micromobility trips serving as an FM/LM connection to transit. Here, this paper addresses these knowledge gaps by leveraging a novel dataset (i.e., the Spin post-ride survey dataset) that records thousands of transit-connecting shared e-scooter trips in Washington DC. Specifically, we used the dataset to reveal the spatiotemporal patterns of transit-connecting shared e-scooter trips in Washington DC, resulting in some major policy insights regarding the integral use of shared e-scooters and public transit. We further leveraged the dataset to validate if and to what extent a commonly applied buffer-zone approach can infer FM/LM micromobility trips accurately. Statistical tests showed that the actual FM/LM Spin e-scooter trips differ from inferred FM/LM Spin e-scooter trips in both spatial and temporal dimensions. This indicates that the common practice of inferring FM/LM micromobility trips with a buffer-zone approach can lead to inaccurate estimates of transit-connecting micromobility trips.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗

Injection data analysis using material balance time for CO 2 storage capacity estimation in deep closed saline aquifers

Estimating the ultimate storage capacity of deep saline aquifers is important to address the formation potential to store the envisioned large volumes of CO 2 . Injection data (i.e. injection rate, bottomhole pressure, and cumulative injected volume of CO 2 ) are routinely recorded during storage operations. These data contain valuable information on the subsurface (e.g. the reservoir pore volume and the formation storage capacity) that can be extracted. In this paper, we present a two-step graphical technique to infer the pore volume and the ultimate storage capacity of closed saline aquifers by analyzing the available injection data. First, the pore volume is inferred through adapting the concept of the material balance time. Material balance time is an approximate superposition time function developed to interpret production data from oil and gas wells operating at variable pressure/rate conditions during the boundary-dominated flow period. Using material balance techniques, the ultimate storage capacity is then estimated through linear extrapolation of the average pressure trend to the maximum allowable pressure the formation can withstand. The average pressure is not available in practice, but is can be obtained from the injection data. Two approaches are presented in this study to calculate the average pressure; namely the rigorous and the approximate approaches. Unlike the rigorous approach, the approximate approach does not require a prior knowledge of some reservoir properties (e.g. relative permeability, absolute permeability, formation porosity and thickness) to calculate the average pressure. To investigate its potential and reliability in analyzing CO 2 injection data, the proposed technique is applied to four synthetic cases representing different well operating conditions. Results indicate that the approximate approach consistently overestimates the actual (simulated) storage capacity as compared to the rigorous approach. The agreement - between the inferred and the simulated reservoir pore volume, and between the analytical and numerical estimates of storage capacity - validates the potential application of the technique to CO 2 storage in closed saline aquifers. The technique is further substantiated through application to a field data set utilized from a commercial-scale geological storage (CGS) project. Finally, field data interpretation shows that the proposed technique can be utilized to identify the degree of hydraulic continuity and reservoir compartmentalization within a target formation by interpreting the corresponding pressure and rate responses.

02 PETROLEUM↗

Efficient bifurcation and tabulation of multi-dimensional combustion manifolds using deep mixture of experts: An a priori study

This work describes and validates an approach for autonomously bifurcating turbulent combustion man-ifolds to divide regression tasks amongst specialized artificial neural networks (ANNs). This approach relies on the mixture of experts (MoE) framework, where each neural network is trained to be specialized in a given portion of the input space. The assignment of different input regions to the experts is determined by a gating network, which is a neural network classifier. In some previous studies, it has been demonstrated that bifurcation of a complex combustion manifold and fitting different ANNs for each part leads to better fits or faster inference speeds. However, the manner of bifurcation in these studies was based on heuristic approaches or clustering techniques. In contrast, the proposed technique enables automatic bifurcation using non-linear planes in high-dimensional turbulent combustion manifolds that are often associated with complex behavior due to different dominating physics in various zones. Furthermore, the proposed concept is validated using 4-dimensional (4D) and 5D flamelet tables, showing that the errors obtained with a given network size, or conversely the network size required to achieve a given accuracy, is considerably reduced. The effect of the number of experts on inference speed is also investigated, showing that by increasing the number of experts from 1 to 8, the inference time can be approximately reduced by a factor of two. Moreover, it is shown that the MoE approach divides the input manifold in a physically intuitive manner, suggesting that the MoE framework can elucidate high-dimensional datasets in a physically meaningful way.

42 ENGINEERING↗

Modeling pair distribution functions of rare-earth phosphate glasses using principal component analysis

The use of principal component analysis (PCA) to statistically infer features of local structure from experimental pair distribution function (PDF) data is assessed on a case study of rare-earth phosphate glasses (REPGs). Such glasses, co-doped with two rare-earth ions (R and R’) of different sizes and optical properties, are of interest to the laser industry. The determination of structure-property relationships in these materials is an important aspect of their technological development. Yet, realizing the local structure of co-doped REPGs presents significant challenges relative to their singly-doped counterparts; specifically, R and R’ are difficult to distinguish in terms of establishing relative material compositions, identifying atomic pairwise correlation profiles in a PDF that are associated with each ion, and resolving peak overlap of such profiles in PDFs. This study demonstrates that PCA can be employed to help overcome these structural complications, by statistically inferring trends in PDFs that exist for a restricted set of experimental data on REPGs, and using these as training data to predict material compositions and PDF profiles in unknown co-doped REPGs. The application of these PCA methods to resolve individual atomic pairwise correlations in t(r) signatures is also presented. The training methods developed for these structural predictions are pre-validated by testing their ability to reproduce known physical phenomena, such as the lanthanide contraction, on PDF signatures of the structurally simpler singly-doped REPGs. The intrinsic limitations of applying PCA to analyze PDFs relative to the quality control of source data, data processing, and sample definition, are also considered. Furthermore, while this case study is limited to lanthanide-doped REPGs, this type of statistical inference may easily be extended to other inorganic solid-state materials, and be exploited in large-scale data-mining efforts that probe many t(r) functions.

36 MATERIALS SCIENCE↗

PySIDT: Subgraph Isomorphic Decision Trees for Molecular Property Prediction

Accurate molecular property prediction is important across all fields of chemistry. Deep neural networks (DNNs) have become increasingly popular due to their ability to train automatically, avoiding the incredibly tedious process of constructing and extending traditional property estimation schemes. However, DNNs require large amounts of training data, are challenging to interpret, require large amounts of memory to load even during inference, and have severe difficulties incorporating qualitative chemical knowledge, which are often desired for molecular property prediction tasks. Here, in this study, we present PySIDT (https://github.com/zadorlab/PySIDT), a software for training and running inference on Subgraph Isomorphic Decision Trees (SIDTs). SIDTs are graph-based decision trees made of nodes associated with molecular substructures. Inference is done by descending target molecular structures down the decision tree to nodes with matching subgraph isomorphic substructures and making predictions based on the final (most specific) nodes matched. SIDTs scale down well to dataset sizes much smaller than is feasible for DNNs. As trees of molecular substructures, SIDTs are inherently readable and easy to visualize, making them easy to analyze. They are also straightforward to extend and retrain, facilitate uncertainty estimation, and enable easy integration of expert knowledge. We demonstrate the SIDT approach discussing its application to a diverse range of molecular prediction tasks: rate coefficient estimation, diffusion coefficient estimation, thermochemistry estimation, transition state bond stretch prediction, p K a prediction, stability of molecular structures, stability of surface structures, and prediction of surface lateral interaction energetics. Additionally, we demonstrate the power of the SIDT algorithms in two direct learning curve vanilla comparisons with the popular DNN-based software Chemprop and the popular gradient boosted trees-based software XGBoost on enthalpy of formation and rate coefficient prediction tasks. In particular, in the enthalpy of formation case, vanilla PySIDT is able to outperform vanilla Chemprop and XGBoost across the full range of training/validation set sizes out to 11,560 data points.

Johnson, Matthew Sean [Sandia National Laboratorie↗

A Novel Modeling Framework for Computationally Efficient and Accurate Real-Time Ensemble Flood Forecasting With Uncertainty Quantification

A novel modeling framework that simultaneously improves accuracy, predictability, and computational efficiency is presented. It embraces the benefits of three modeling techniques integrated together for the first time: surrogate modeling, parameter inference, and data assimilation. The use of polynomial chaos expansion (PCE) surrogates significantly decreases computational time. Parameter inference allows for model faster convergence, reduced uncertainty, and superior accuracy of simulated results. Ensemble Kalman filters assimilate errors that occur during forecasting. To examine the applicability and effectiveness of the integrated framework, we developed 18 approaches according to how surrogate models are constructed, what type of parameter distributions are used as model inputs, and whether model parameters are updated during the data assimilation procedure. We conclude that (1) PCE must be built over various forcing and flow conditions, and in contrast to previous studies, it does not need to be rebuilt at each time step; (2) model parameter specification that relies on constrained, posterior information of parameters (so-called Selected specification) can significantly improve forecasting performance and reduce uncertainty bounds compared to Random specification using prior information of parameters; and (3) no substantial differences in results exist between single and dual ensemble Kalman filters, but the latter better simulates flood peaks. The use of PCE effectively compensates for the computational load added by the parameter inference and data assimilation (up to ~80 times faster). Therefore, the presented approach contributes to a shift in modeling paradigm arguing that complex, high-fidelity hydrologic and hydraulic models should be increasingly adopted for real-time and ensemble flood forecasting.

54 ENVIRONMENTAL SCIENCES↗

A Differentiable Hydrology Approach for Modeling With Time-Varying Parameters

In the analysis of observational data with mathematical models, it is often desirable to let one or more model parameters vary with time to account for changing environmental conditions or to offer more flexibility. A standard approach for dynamic models is to allow for the number of parameters to grow as a function of time, leading to a high-dimensional inference problem for long data records. We propose using Hamiltonian Monte Carlo, a gradient-based Markov chain Monte Carlo (MCMC) method, to perform Bayesian inference for time-varying parameters (TVP) in hydrology models. As derivatives of model error functions with regard to parameters are not available in closed form, we implemented the GR4J rainfall-runoff models in Theano/PyMC3, and Jax/NumPyro, allowing for the application of automatic differentiation techniques. A simulation experiment assessing the viability of these methods for recovering underlying temporal variation in parameters indicates that a discrete Gaussian random walk-based prior appears to be best suited across a range of scenarios while the continuous-time Gaussian process compares relatively poorly. In an analysis of 20 years of daily real-world streamflow records from the Model Parameter Estimation Experiment (MOPEX) hydrology data set with increasing data sparsity and at multiple temporal resolutions, we find that TVP inference with gradient-based MCMC is a flexible and reliable approach for analyzing hydrology models with dynamic parameter sets.

54 ENVIRONMENTAL SCIENCES↗