A 3D SPATIO-TEMPORAL MACHINE LEARNING EMULATION OF ATMOSPHERIC TRANSPORT MODELING
Explore the source record for details and available documents.
SEARCH · Engineering Papers
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.
For the development of heat-pipe cooled microreactors, it is crucial to thoroughly understand the characteristics and functioning of heat pipes across a wide spectrum of operating conditions. Passive heat removal and its long-term performance stability are critical factors in this context. Enhanced experimental data is vital for evaluating the operational lifespan of alkali metal heat pipes. Idaho National Laboratory (INL) has successfully conducted an extended duration test on a high-performance sodium-filled heat pipe, closely monitoring the axial temperature profile, power supplied by the heaters, and heat removed by a gas-gap calorimeter. The results from this testing provide valuable data that are instrumental in supporting heat pipe validation efforts. Specifically, this data aids in the development and validation of Sockeye, the Multiphysics Object-Oriented Simulation Environment (MOOSE) tool under the US-DOE NEAMS program designed for heat pipe modeling. By comparing experimental results with Sockeye’s predictions, the tool's accuracy and reliability can be assessed and improved, thereby enhancing its capability to simulate heat pipe operations under various conditions.
No abstract prepared.
Abstract. Climate model output emulation has long been attempted to support impact research, mainly to fill in gaps in the scenario space. Given the computational cost of running coupled earth system models (ESMs), which are usually the domain of supercomputers and require on the order of days to weeks to complete a century-long simulation, only a handful of different scenarios are usually chosen to externally force ESM simulations. An effective emulator, able to run on standard computers in times of the order of minutes rather than days could therefore be used to derive climate information under scenarios that were not run by ESMs. Lately, the necessity of accounting for internal variability has also made the availability of initial-condition ensembles, under a specific scenario, important, further increasing the computational demand. At least so far, emulators have been limited to simplified ESM-like output, either seasonal, annual, or decadal averages of basic quantities, like temperature and precipitation, often emulated independently of one another. With this work, we propose a more comprehensive solution to ESM output emulation. Our emulator, STITCHES, uses existing archives of earth system models' (ESMs) scenario experiments to construct ESM-like output under new scenarios or enrich existing initial-condition ensembles, which is what other emulators also aim to do. Importantly, however, STITCHES' output has the same characteristics of the ESM output it sets out to emulate: multivariate, spatially resolved, and high frequency, representing both the forced component and the internal variability around it. STITCHES extends the idea of time sampling – according to which climate outcomes are stratified by the global warming level at which they manifest themselves, irrespective of the scenario and time at which they occur – to the construction of a continuous history of ESM-like output over the whole 21st century, consistent with a 21st-century trajectory of global surface air temperature (GSAT) derived from the scenario that has been chosen as the target of the emulation. STITCHES does so by first splitting the target GSAT trajectory into decade-long windows, then matching each window in turn to a decade-long window within an existing model simulation from the available scenario runs according to its proximity to the target in absolute size of the temperature anomaly and its rate of change. A look-up table is therefore created of a sequence of existing experiment–time-window combinations that, when stitched together, create a GSAT trajectory “similar” to the target. Importantly, we can then stitch together much more than GSAT from these windows, i.e., any output that the ESM has saved for these existing experiment–time-window combinations, at any frequency and spatial scale available in its archive. We show that the stitching does not introduce artifacts in the great majority of cases (we look at temperature and precipitation at monthly frequency and on the native grid of the ESM and at an index of ENSO activity, the Southern Oscillation Index). This is true even if the criteria for the identification of the decades to be stitched together are chosen to work for a smoothed time series of annual GSAT, a result we expect given the larger amount of noise affecting most other variables at finer spatial scales and higher frequencies, which therefore are more “forgiving” of the stitching. We successfully test the method's performance over many ESMs and scenarios. Only a few exceptions surface, but these less-than-optimal outcomes are always associated with a scarcity of the archived simulations from which we can gather the decade-long windows that form the building blocks of the emulated time series. In the great majority of cases, STITCHES' performance is satisfactory according to metrics that reward consistency in trends, interannual and inter-ensemble variance, and autocorrelation structure of the time series stitched together. The method therefore can be used to create ESM-like output according to new scenarios, on the basis of a trajectory of GSAT produced according to that scenario, which could be easily obtained by a simple climate model. It can also be used to increase the size of existing initial-condition ensembles. There are aspects of our emulator that will immediately disqualify it for specific applications, like when climate information is needed whose characteristics result from accumulated quantities over windows of times longer than those used as pieces by STITCHES, droughts longer than a decade for example. But for many applications, we argue that a stitched product can satisfy the climate information needs of impact researchers. STITCHES cannot emulate ESM output from scenarios that result in GSAT trajectories outside of the envelope available in the archive, nor can it emulate trajectories with shapes different from existing ones (overshoots with negative derivative, for example). Therefore, the size and characteristics of the available archives of ESM output are the principal limitations for STITCHES' deployment. Thus, we argue for the possibility of designing scenario experiments within, for example, the next phase of the Coupled Model Intercomparison Project according to new principles, relieved of the need to produce a number of similar trajectories that vary only in radiative forcing strength but more strategically covering the space of temperature anomalies and rates of change.
Large-scale simulations of cosmic structure formation play an important role in interpreting cosmological observations at high precision. The simulations must cover a parameter range beyond the standard six cosmological parameters and need to be run at high mass and force resolution. A key simulation-based task is the generation of accurate theoretical predictions for observables using a finite number of simulation runs, via the method of emulation. Using a new sampling technique, we explore an eight-dimensional parameter space including massive neutrinos and a variable equation of state of dark energy. We construct trial emulators using two surrogate models (the linear power spectrum and an approximate halo mass function). The new sampling method allows us to build precision emulators from just 26 cosmological models and to systematically increase the emulator accuracy by adding new sets of simulations in a prescribed way. Emulator fidelity can now be continuously improved as new observational data sets become available and higher accuracy is required. Finally, using one ΛCDM cosmology as an example, we study the demands imposed on a simulation campaign to achieve the required statistics and accuracy when building emulators for investigations of dark energy.
Two-dimensional hydrodynamic flood models are commonly employed for simulating flood extent and inundation depth. However, the influence of urban drainage network (UDN) is frequently overlooked in these models, potentially compromising their accuracy. Furthermore, the expensive computational costs and longer processing times make them challenging for large-scale hydrodynamic simulation. To address these challenges, this paper develops a machine learning (ML)-driven emulator for an open-source flood model, the Two-dimensional Runoff Inundation Toolkit for Operational Needs (TRITON). A TRITON-ML Emulator (TR-Emulator) that utilizes Convolutional Long Short-Term Memory is developed to capture the spatiotemporal features of flood events based on the outputs from TRITON. We further enhance the emulator by integrating UDN parameters (TR-UDN), such as the flow capacity of drainage pipes, pipe size, and pipe length, via an ML stacking technique to improve the water surface elevation (WSE) simulation. Hurricane Harvey 2017 in Houston, TX is used as the case study. We compare WSE results from TRITON, TR-Emulator, TR-UDN, and the United States Geological Survey (USGS) observations to evaluate the performance of these models. The results indicate that the TR-Emulator effectively replicates the WSE simulated by TRITON. Additionally, TR-UDN performs well in capturing WSE patterns and peak flows, aligning more closely with USGS observations, except in areas with milder slopes where conveyance discrepancies are observed. We further test the generalizability of our ML-based models using another smaller event. This paper shows that the TR-Emulator is effective for users and engineers to emulate a 2D hydrodynamic model, and the enhanced version of the TR-Emulator, TR-UDN, can be an efficient tool for predicting WSEs during urban flooding.
NASA Glenn Research Center, in collaboration with the aerospace industry and academia, has begun the development of technology for a future hybrid-wing body electric airplane with a turboelectric distributed propulsion (TeDP) system. It is essential to design a subscale system to emulate the TeDP power grid, which would enable rapid analysis and demonstration of the proof-of-concept of the TeDP electrical system. This paper describes how small electrical machines with their controllers can emulate all the components in a TeDP power train. The whole system model in Matlab/Simulink was first developed and tested in simulation, and the simulation results showed that system dynamic characteristics could be implemented by using the closed-loop control of the electric motor drive systems. Then we designed a subscale experimental system to emulate the entire power system from the turbine engine to the propulsive fans. Firstly, we built a system to emulate a gas turbine engine driving a generator, consisting of two permanent magnet (PM) motors with brushless motor drives, coupled by a shaft. We programmed the first motor and its drive to mimic the speed-torque characteristic of the gas turbine engine, while the second motor and drive act as a generator and produce a torque load on the first motor. Secondly, we built another system of two PM motors and drives to emulate a motor driving a propulsive fan. We programmed the first motor and drive to emulate a wound-rotor synchronous motor. The propulsive fan was emulated by implementing fan maps and flight conditions into the fourth motor and drive, which produce a torque load on the driving motor. The stator of each PM motor is designed to travel axially to change the coupling between rotor and stator. This feature allows the PM motor to more closely emulate a wound-rotor synchronous machine. These techniques can convert the plain motor system into a unique TeDP power grid emulator that enables real-time simulation performance using hardware-in-the-loop (HIL).
Recent developments in software-defined infrastructures promise that scientific workflows utilizing supercomputers, instruments, and storage systems will be dynamically composed and orchestrated using software at unprecedented speed and scale in the near future. Testing of the underlying networking software, particularly during initial exploratory stages, remains a challenge due to potential disruptions, and resource allocation and coordination needed over the multi-domain physical infrastructure. To overcome these challenges, we develop the Virtual Science Network Environment (VSNE) that emulates the multi-site host, storage, and network infrastructure using Virtual Machines (VMs), wherein the production and nascent software can be tested. Within each VM, which represents a site, the hosts and local-area networks are emulated using Mininet, and the Software-Defined Network (SDN) controllers and service daemon codes are natively run to support dynamic provisioning of network connections. Additionally, Lustre filesystem support at the sites and an emulation of the long-haul network using Mininet, are provided using separate VMs. As case studies, we describe Lustre file transfers using XDD, Red5 streaming service demonstration, and an emulated experiment with remote monitoring and steering modules, all supported over dynamically configured connections using SDN controllers.
Dedicated network connections are being increasingly deployed in cloud, centralized and edge computing and data infrastructures, whose throughput profiles are critical indicators of the underlying data transfer performance. Due to the cost and disruptions to physical infrastructures, network emulators, such as Mininet, are often used to generate measurements needed to estimate throughput profiles, typically expressed as a function of the connection round trip time. The profiles estimated using measurements from such emulated networks are usually inaccurate for high bandwidth and high latency connections, since they do not accurately reflect the critical network transport dynamics mainly due to computing and memory constraints of the host. We present a machine learning (ML) method to estimate the throughput profiles using emulation measurements to closely match the testbed and production network profiles. In particular, we propose a micro Kernel Network (mKN) that provides baseline throughput measurements on the host running Mininet emulations, which are used to learn a regression map that converts them to the corresponding testbed measurement estimates. Once initially learned, this map is applied to measurements from subsequent network emulations on the same host. We present experimental measurements to illustrate this approach, and derive generalization equations for the proposed mKN-ML method. Using a four-site scenario emulation, we show the effectiveness of this method in providing accurate concave throughput profiles from inaccurate convex or non-smooth ones indicated by Mininet emulation.
ABSTRACT We present a neural network-based emulator for the galaxy redshift-space power spectrum that enables several orders of magnitude acceleration in the galaxy clustering parameter inference, while preserving 3$\sigma$ accuracy better than 0.5 per cent up to $k_{\mathrm{max}}$ = 0.25 $\, h\text{Mpc}^{-1}$ within Lambda-cold dark matter ($\Lambda$CDM) and around 0.5 per cent $w_0$–$w_a$CDM. Our surrogate model only emulates the galaxy bias-invariant terms of one-loop perturbation theory predictions, these terms are then combined analytically with galaxy bias terms, counter-terms, and stochastic terms in order to obtain the non-linear redshift-space galaxy power spectrum. This allows us to avoid any galaxy bias prescription in the training of the emulator, which makes it more flexible. Moreover, we include the redshift $z \in [0,1.4]$ in the training which further avoids the need for re-training the emulator. We showcase the performance of the emulator in recovering the cosmological parameters of $\Lambda$CDM by analysing the suite of 25 AbacusSummit simulations that mimic the Dark Energy Spectroscopic Instrument luminous red galaxies at $z=0.5$ and 0.8, together as the emission line galaxies at $z=0.8$. We obtain similar performance in all cases, demonstrating the reliability of the emulator for any galaxy sample at any redshift in $0 \lt z \lt 1.4$. We will make our emulator public at github repository.
The code and detailed step-by-step instructions for generating the model output data, processing results, and analysis and plotting are provided at https://github.com/IMMM-SFA/Ferencz_et_al_2026_ER_Water. The PyArtes model is a python adaptation of the Artes model. PyArtes uses many of the same input data and optimization model architecture as Artes. Documentation for the PyArtes model is provided in the Supplement to the paper. The primary data product are simulated monthly water shortages for indoor and outdoor demand under a large ensemble of drought scenarios (>13,000). The droughts are hypothetical and are not based on historical time series data of supply sources - though historical data did help inform ranges explored for supply parameters. Demands are informed by recent 2017-2021 water supply data. Demands used for the model can be accessed at https://github.com/IMMM-SFA/Ferencz_et_al_2026_ER_Water. Simulations resolve demand for over 90 water providers in the study region. The results report 36 months of water shortage data for each indoor and outdoor demand node. The study also developed a multilayer perceptron (MLP) neural network trained on a subset of the simulated shortage ensemble to emulate worst annual water shortage for a given set of parameter multipliers -- provided the parameter values fall within the ranges sampled in the ensemble. Emulated water shortages for synthetic ensembles are in the MLP-generated shortages folder. The MLP model was used to generate larger ensembles to support Sobol analysis that would have been extremely computationally expensive to simulate. Datasets provided in this repository*: Simulated shortages. These results are used for the analysis for Figures 5, 8, and 9 in the paper, and also to train the MLP emulator. .zip file containing outputs for the 13,312 scenario ensemble. Separate .csv files for indoor and outdoor shortage for each scenario. Rows = demand ids (~100), Columns = months (36) Units = acre-feet/month of shortage (shortage = monthly demand - supply). 1 acft = 1233.48 m^3 .csv files of aggregated shortages derived from the 13,312 ensemble Rows = scenarios (13,312), Columns = demand ids (~100) Units = acre-feet/year (either worst annual shortage or total shortage over the 3-year drought) .csv file of the parameter multipliers scenarios for the ensemble .csv file of the parameter ranges and baseline values the multipliers were applied to MLP-generated shortages. These results are used for Figures 4, 6, and 7 in the paper. mwd higher folder: scenario ensembles, emulated worst year total shortages (acft), and Sobol results Emulated shortages. Rows = scenarios, columns = demand ids, units acft Sobol results. Rows = demand ids, columns Sobol (S1, ST, or 95% confidence interval) value for each parameter mwd lower folder: scenario ensembles, emulated worst year total shortages (acft), and Sobol results same organization as mwd higher MLP performance: performance metrics (R^2, RMSE, BIAS, MAPE) for the testing subset (20% or 2,662 scenarios) and simulated vs emulated worst year shortage (acre-feet/year) for every demand node, MWD wholesale regions, and the entire study region (LAC). Supporting data for figures. Figure plotting scripts in the associated GitHub repo. These files support analysis and visualization. Geospatial Data used for plotting simulated water shortages and Sobol results. Dictionary of full names for demand nodes in the model and estimates of water supply by source type informed by Artes input files and California Urban Water Management Planning data: https://water.ca.gov/Programs/Water-Use-And-Efficiency/Urban-Water-Use-Efficiency/Urban-Water-Management-Plans *Readme files provided for each folder.
Special-purpose robot hands (end effectors) now under development are intended to enable robots to traverse cliffs much as human climbers do. Potential applications for robots having this capability include scientific exploration (both on Earth and other rocky bodies in space), military reconnaissance, and outdoor search and rescue operations. Until now, enabling robots to traverse cliffs has been considered too difficult a task because of the perceived need of prohibitively sophisticated planning algorithms as well as end effectors as dexterous as human hands. The present end effectors are being designed to enable robots to attach themselves to typical rock-face features with less planning and simpler end effectors. This advance is based on the emulation of the equipment used by human climbers rather than the emulation of the human hand. Climbing-aid equipment, specifically cams, aid hooks, and cam hooks, are used by sport climbers when a quick ascent of a cliff is desired (see Figure 1). Currently two different end-effector designs have been created. The first, denoted the simple hook emulator, consists of three "fingers" arranged around a central "palm." Each finger emulates the function of a particular type of climbing hook (aid hook, wide cam hook, and a narrow cam hook). These fingers are connected to the palm via a mechanical linkage actuated with a leadscrew/nut. This mechanism allows the fingers to be extended or retracted. The second design, denoted the advanced hook emulator (see Figure 2), shares these features, but it incorporates an aid hook and a cam hook into each finger. The spring-loading of the aid hook allows the passive selection of the type of hook used. The end effectors can be used in several different modes. In the aid-hook mode, the aid hook on one of the fingers locks onto a horizontal ledge while the other two fingers act to stabilize the end effector against the cliff face. In the cam-hook mode, the broad, flat tip of the cam hook is inserted into a non-horizontal crack in the cliff face. A subsequent transfer of weight onto the end effector causes the tip to rotate within the crack, creating a passive, self-locking action of the hook relative to the crack. In the advanced hook emulator, the aid hook is pushed into its retracted position by contact with the cliff face as the cam hook tip is inserted into the crack. When a cliff face contains relatively large pockets or cracks, another type of passive self-locking can be used. Emulating the function of the piece of climbing equipment called a "cam" (note: not the same as a "cam hook"; see Figure 1), the fingers can be fully retracted and the entire end effector inserted into the feature. The fingers are then extended as far as the feature allows. Any weight then transferred to the end effector will tend to extend the fingers further due to frictional force, passively increasing the grip on the feature. In addition to the climbing modes, these end effectors can be used to walk on (either on the palm or the fingertips) and to grasp objects by fully extending the fingers.
Purpose: Current commercial 10 MV Linac waveguides are 1.5 m. The authors’ current 6 MV linear accelerator–magnetic resonance imager (Linac–MR) system fits in typical radiotherapy vaults. To allow 10 MV treatments with the Linac–MR and still fit within typical vaults, the authors design a 10 MV Linac with an accelerator waveguide of the same length (27.5 cm) as current 6 MV Linacs. Methods: The first design stage is to design a cavity such that a specific experimental measurement for breakdown is applicable to the cavity. This is accomplished through the use of finite element method (FEM) simulations to match published shunt impedance, Q factor, and ratio of peak to mean-axial electric field strength from an electric breakdown study. A full waveguide is then designed and tuned in FEM simulations based on this cavity design. Electron trajectories are computed through the resulting radio frequency fields, and the waveguide geometry is modified by shifting the first coupling cavity in order to optimize the electron beam properties until the energy spread and mean energy closely match values published for an emulated 10 MV Linac. Finally, Monte Carlo dose simulations are used to compare the resulting photon beam depth dose profile and penumbra with that produced by the emulated 10 MV Linac. Results: The shunt impedance, Q factor, and ratio of peak to mean-axial electric field strength are all matched to within 0.1%. A first coupling cavity shift of 1.45 mm produces an energy spectrum width of 0.347 MeV, very close to the published value for the emulated 10 MV of 0.315 MeV, and a mean energy of 10.53 MeV, nearly identical to the published 10.5 MeV for the emulated 10 MV Linac. The depth dose profile produced by their new Linac is within 1% of that produced by the emulated 10 MV spectrum for all depths greater than 1.5 cm. The penumbra produced is 11% narrower, as measured from 80% to 20% of the central axis dose. Conclusions: The authors have successfully designed and simulated an S-band waveguide of length of 27.5 cm capable of producing a 10 MV photon beam. This waveguide operates well within the breakdown threshold determined for the cavity geometry used. The designed Linac produces depth dose profiles similar to those of the emulated 10 MV Linac (waveguide-length of 1.5 m) but yields a narrower penumbra.
The nucleation of cloud droplets from the ambient aerosol is a critical physical process that must be resolved for global models to faithfully predict aerosol–cloud interactions and aerosol indirect effects on climate. To better represent droplet nucleation from a complex, multimodal, and multicomponent aerosol population within the context of a global model, a new metamodeling framework is applied to derive an efficient and accurate activation parameterization. The framework applies polynomial chaos expansion to a detailed parcel model in order to derive an emulator that maps thermodynamic and aerosol parameters to the supersaturation maximum achieved in an adiabatically ascending parcel and can be used to diagnose droplet number from a single lognormal aerosol mode. The emulator requires much less computational time to build, store, and evaluate than a high-dimensional lookup table. Compared to large sample sets from the detailed parcel model, the relative error in the predicted supersaturation maximum and activated droplet number computed with the best emulator is and (one standard deviation), respectively. On average, the emulators constructed here are as accurate and between 10 and 17 times faster than a leading physically based activation parameterization. Because the underlying parcel model being emulated resolves size-dependent droplet growth factors, the emulator captures kinetic limitations on activation. Furthermore, the results discussed in this work suggest that this metamodeling framework can be extended to accurately account for the detailed activation of a complex aerosol population in an arbitrary coupled global aerosol–climate model.
Conventional generation sources play a critical role in the stability and reliability of the electrical grid, particularly as we transition towards more renewable energy sources. To understand and accurately emulate their behavior for optimizing grid operations and ensuring seamless integration with renewable technologies, it is essential to better emulate the grid- and plant-level impacts of conventional generation sources, such as natural gas (NG) driven heat recovery steam generators (HRSGs) and combustion turbines (CTs). Therefore, a governor model is developed in a programmable logic controller (PLC) to investigate the performance of the conventional generator under various dynamic operating conditions and to identify the impact on grid stability in a controlled environment. The governor model aims to enable the hardware-in-the-loop (HIL) based emulation of these conventional generation sources using the existing 2 MVA synchronous machine/generator that is driven by a flexible 2.5 MW variable speed drive. This setup will allow us to replicate the dynamic characteristics and response behaviors of NG-driven HRSGs and CTs. The controls for the emulated conventional plants follow the industry standard and are adjustable, ensuring they accurately reflect the operational capabilities and limitations of real-world systems. These controls include load-following capabilities, ramp rates, startup and shutdown sequences, and emissions characteristics. By incorporating these adjustable controls, we aim to capture the nuanced impacts of conventional generation, such as their ability to provide ancillary services like frequency regulation, voltage support, and spinning reserve. In this report, we simulate two types of dynamic operations: grid-connected and islanding. For each dynamic operation, representative starting sequences are tested, including turbine purge, ignition, speed ramping up, generator excitation and synchronizing, and breaker close. The HIL based tests provides insights for field deployment, specifically the high-fidelity governor model provides results to predict the potential stability and reliability risk and suggest possible integration measures (e.g., generation and load balancing, tuning of governor control parameters). Ultimately, this enhanced emulation capability will be integrated into our Advanced Research on Integrated Energy Systems (ARIES), enabling us to conduct comprehensive studies on the interactions between conventional and renewable energy sources. By better understanding these interactions, we can develop strategies to optimize the overall performance and reliability of the grid. This will support the deployment of advanced grid management techniques, such as demand response, grid-forming inverters, and energy storage systems. The main contributions are summarized as follows: (1) This report introduces a PLC-based governor model for gas turbines. This model accurately simulates the dynamic behavior of conventional generation sources under various operational scenarios; (2) The model is integrated with an HIL testbed that includes a 2.5 MW variable speed drive and a 2 MVA synchronous machine. This setup enables realistic, real-time emulation of conventional power plants, particularly NG driven HRSGs and CTs; (3) The developed model is adaptable to various gas turbine configurations and allows for precise control over parameters such as MW ramp rates. This flexibility makes it a valuable tool for future research and industry collaboration; and (4) By incorporating the model into the National Renewable Energy Laboratory's Advanced Research on Integrated Energy Systems, the report lays the groundwork for future studies on interactions between conventional and renewable energy sources, enhancing the ability to develop advanced grid management strategies.
We look at cosmological probes that pose an inverse problem where the measurement result is obtained through observations, and the objective is to infer values of model parameters that characterize the underlying physical system-our universe, from these observations and theoretical forward-modeling. The only way to accurately forward-model physical behavior on small scales is via expensive numerical simulations, which are further “emulated” due to their high cost. Emulators are commonly built with a set of simulations covering the parameter space with Latin hypercube sampling and an interpolation procedure; the aim is to establish an approximately constant prediction error across the hypercube. In this paper, we provide a description of a novel statistical framework for obtaining accurate parameter constraints. The proposed framework uses multi-output Gaussian process emulators that are adaptively constructed using Bayesian optimization methods with the goal of maintaining a low emulation error in the region of the hypercube preferred by the observational data. In this paper, we compare several approaches for constructing multi-output emulators that enable us to take possible inter-output correlations into account while maintaining the efficiency needed for inference. Using a Lyα forest flux power spectrum, we demonstrate that our adaptive approach requires considerably fewer-by a factor of a few in the Lyα P(k) case considered here-simulations compared to the emulation based on Latin hypercube sampling, and that the method is more robust in reconstructing parameters and their Bayesian credible intervals.
We describe an emulator of a detailed cloud parcel model which has been trained to assess droplet nucleation from a complex, multimodal aerosol size distribution simulated by a global aerosol–climate model. The emulator is constructed using a sensitivity analysis approach (polynomial chaos expansion) which reproduces the behavior of the targeted parcel model across the full range of aerosol properties and meteorology simulated by the parent climate model. An iterative technique using aerosol fields sampled from a global model is used to identify the critical aerosol size distribution parameters necessary for accurately predicting activation. Across the large parameter space used to train them, the emulators estimate cloud droplet number concentration (CDNC) with a mean relative error of 9.2% for aerosol populations without giant cloud condensation nuclei (CCN) and 6.9% when including them. Versus a parcel model driven by those same aerosol fields, the best-performing emulator has a mean relative error of 4.6%, which is comparable with two commonly used activation schemes also evaluated here (which have mean relative errors of 2.9 and 6.7%, respectively). We identify the potential for regional biases in modeled CDNC, particularly in oceanic regimes, where our best-performing emulator tends to overpredict by 7%, whereas the reference activation schemes range in mean relative error from -3 to 7%. The emulators which include the effects of giant CCN are more accurate in continental regimes (mean relative error of 0.3%) but strongly overestimate CDNC in oceanic regimes by up to 22%, particularly in the Southern Ocean. Finally, the biases in CDNC resulting from the subjective choice of activation scheme could potentially influence the magnitude of the indirect effect diagnosed from the model incorporating it.