Simulation Parameters Research Articles

Diagnosing the factors that contribute to the intermodel spread of climate feedback in CMIP6

Abstract The global surface temperature climate feedback parameter (λ) varies significantly across climate models, and its real-world value remains uncertain. Studies have found that sea surface temperature (SST) response pattern and atmospheric model physics can each affect the climate feedback parameter in historical and idealized warming simulations. In this study, we design and analyze a series of targeted atmospheric global climate model (AGCM) experiments to quantify how much the SST (both the warming pattern and the base climatology) and atmospheric model physics contributes to the intermodel spread of the climate feedback parameter and cloud feedback in in the Coupled Model Intercomparison Project Phase 6 (CMIP6). We use three AGCMs, HiRAM, AM2.5, and AM4, developed in the Geophysical Fluid Dynamics Laboratory (GFDL), which span a wide range of feedbacks in response to uniform surface warming. The three GFDL models indicate that historical patterns of SST change systematically alter the λ, supporting the hypothesized role for a “pattern effect” in historical climate sensitivity. However, we found that forcing one AGCM with the different CO2-induced SST warming patterns or model SST climatology from a suite of climate models from CMIP6 can only reproduce ~10% the intermodel spread of CO2-induced λ, while the atmospheric model used determines the magnitude of λ (~45%). This underscores the role of atmospheric model physics in altering λ, particularly the cloud-related schemes. In addition, we demonstrate that the nonlinear interaction between SST and AGCM has a non-negligible role in affecting λ.

How Accurate Are QM/MM Models?

Despite the success and widespread use of QM/MM methods in modeling (bio)chemically important processes, their accuracy is still not well understood. A key reason is because these methods are ultimately approximations to direct QM calculations of very large systems, which are impractical to perform in most cases. We highlight recent progress toward the development of realistic model systems where it is possible to obtain full QM reference data to directly and systematically evaluate the effectiveness of different QM/MM generation schemes. These model systems are highly flexible and can be tailored to probe the sensitivity of a QM/MM model to different reaction types and simulation parameters such as pairing of QM and MM potentials, QM region size, and composition. It is envisaged that this strategy could be used to directly validate different QM/MM generation schemes and spur the development of more robust models in the future.

Dynamic control, routing, and resource assignment in multi-granular optical node topologies combining wavelength, waveband, and spatial switching for 6G transport networks [Invited

Effective management of end-to-end 6G network services is crucial, with peak capacity requirements for 6G transport connections expected to exceed 1 Tb/s. As demand for high bandwidth rises, there is a growing necessity for high-capacity optical fiber links, including ultra-wideband (UWB) and multiple fiber links within the network. Scaling up to accommodate these demands, designing wavelength-selective switches (WSSs) for such networks significantly increases the port count. To tackle this issue, we propose various multi-granular optical node (MG-ON) architectures utilizing heterogeneous wavelength, waveband, and spatial switching. We evaluate these architectures’ performance against high-capacity wavelength division multiplexed (WDM) networks through various simulation parameters.

Impact of Bt direction on the detachment threshold during H-mode discharge on EAST with SOLPS-ITER modeling

Abstract Upper single-null H-mode detachment experiments on EAST were carried out with an ITER-like tungsten divertor, and it was found that the detachment threshold of the upper outer divertor was significantly lower under an unfavorable toroidal magnetic field (B t) than a favorable B t. In this work, SOLPS-ITER simulations are carried out to study the effects of B t direction on the detachment threshold to explain the experimental observations. The simulation results are compared with the experiments, showing reasonable agreement with the diagnostics, and the simulation setups and parameters are thus determined. Based on the simulation data, two-point model formatting is applied to the flux tube where detachment occurs. A quantitative comparison highlights the critical roles of the upstream parallel heat flux and the downstream momentum and energy loss factors on the ion saturation current density at the target ( j sat,t). Further analysis of the simulation results indicates that due to the influence of E×B drift on the convective heat flux, combined with the presence of heat transport associated with parallel current, the upstream heat flux is reduced under unfavorable B t, while in the downstream region the divergence of the E×B drift flux acts as a particle source under unfavorable B t and as a sink under favorable B t, leading to increased momentum and energy losses under unfavorable B t. These factors collectively cause a lower detachment threshold under unfavorable B t.

A LES-ALM Study for the Turbulence Characteristics of Wind Turbine Wake Under Different Roughness Lengths

To investigate the characteristics of wind turbine wakes under different aerodynamic roughness lengths, a series of LES-ALM simulations were carried out in this study. First, a sensitivity analysis of the time step of the simulation results was performed. Then, the study compared the power and thrust of wind turbines under different roughness conditions. Finally, the mean velocity deficit, added turbulence intensity, and Reynolds shear stresses in the wake were analyzed under different roughness conditions. This study finds that a 0.1 s time step can provide satisfactory results for the LES-ALM compared to a 0.02 s time step. Furthermore, for the same hub-height wind speed, the thrust coefficient varies from 0.75 to 0.8 under the different roughness levels. As the roughness length increases, the time-averaged velocity deficit and added turbulence intensity decreases, and the wake recovers more quickly at the incoming level. However, the effect of roughness length on the Reynolds shear stress is weak within the downstream range of x = 6D to 10D. For the velocity deficit, a single Gaussian function is not able to describe its vertical distribution. Additionally, under higher roughness conditions, the height of the wake center is distinctively higher than the hub height as the wake develops downstream. The findings of this paper are beneficial for selecting the approximate numerical parameters for the wake simulations and provide deeper insights into the turbulence mechanisms of wind turbine wake, which are crucial for establishing analytical models to predict the wake field.

Research and design of transient test platform for DC current transformer

This article studies the design and simulation of a transient test platform for DC current transformers to verify the effectiveness of the design scheme. DC current transformers convert large currents on the primary side into small currents on the secondary side through electromagnetic induction principles and are widely used in power systems. In order to verify the feasibility of the design scheme, simulation research was adopted, and the simulation parameters were the same as the actual scheme, including controllable current source parameters, cable parameters, and MOS switch branches. The simulation results show that the designed DC transient square wave current source can effectively compensate for the impulse current, and the output current waveform is stable, meeting the design requirements.

Design and experiment of facility elevated planting strawberry continuous picking manipulator

Strawberry picking manipulator is one of the key factors affecting the effectiveness of robot picking operations, elevated strawberries cultivated off the ground with hanging fruits provide convenient conditions for automatic strawberry picking. In order to adapt to the small space of elevated strawberry picking operation and to solve the problem of discontinuous picking process, a P-P-R-P type continuous strawberry picking manipulator was proposed. Based on the elevated cultivation mode and the clamping-shear picking method, the structural parameters and control system of the continuous picking manipulator were determined. The kinematics model of the continuous picking manipulator was established, the Jacobian condition number and performance of each joint in the picking operation were solved, and its flexibility and operability were analyzed. According to the growth characteristics of elevated strawberries, the picking simulation routes and operation cycles were set. Using MATLAB software, the simulation tests were carried out to analyze the working space of the continuous picking manipulator, the maximum speed and acceleration changes of each joint. In order to improve the success rate of strawberry picking, the operating parameters such as rotational speed, end picking angle and clamping position of the continuous picking manipulator were optimized using Box-Behnken test. Based on the simulation results and parameter optimization results, a continuous picking manipulator and motion control system were constructed and validation tests were conducted. The results showed that the average picking success rate of the continuous picking manipulator was 74.21 %, and the average picking efficiency was 12.87 s per piece. In conclusion, the P-P-R-P type continuous picking manipulator proposed in this study meets the automated operation requirements of strawberry picking and can be adapted to the elevated operating environment.

FRACTAL-BASED MODELING OF RAIL ROUGHNESS

This paper uses fractal modeling techniques to effectively depict the roughness characteristics of the rails examined in this study by means of both the structure function and the Weierstrass–Mandelbrot function to capture the intricate nature of the rail roughness. Experimental analysis of the roughness of the railway rails from Faurei Railway Testing Center (RTC Faurei) in Romania proves that the roughness height exhibits distinct mathematical fractal characteristics. The study evaluated and compared 41 classical statistical parameters derived from roughness measurements with simulated fractal parameters. The classical roughness parameters, including the Autocorrelation function, Amplitude Density Function, as well as Bearing Area Curves, and rail acoustic roughness, were determined from a profile obtained using the Weierstrass function and compared to the measured ones, revealing a noticeable congruence between the generated charts. The research findings indicate a strong convergence between experimental measurements and simulated profiles, with most parameters falling within a 10% relative error range. This aspect highlights the approach of fractal potent in assessing rail roughness behavior. Consequently, the simulated parameters could potentially analyze rail roughness quality to maintain and track grinding and mitigate the rolling noise.

Numerical prediction of impact damage in thick fabric composite laminates

A simulation methodology for assessing the damage in thick fabric Carbon Fibre Reinforced Polymer (CFRP) composite laminates under low- and high-velocity impacts is presented. It encompasses steps for calibration, verification, and validation of the elastic and fracture material properties as well as determination of model parameters for the numerical simulations. Damage is modelled using a discrete fracture approach with cohesive interface elements that capture individual cracks occurring in and between plies. For computational efficiency, the method is implemented in a two-dimensional (2D) axi-symmetric model. Results from double-cantilever beam, end-notched flexure, and quasi-static indentation experiments align well with numerical simulations and serve to calibrate and verify the implementation of the discrete fracture approach. The methodology is extended to dynamic impact analysis to predict damage mechanisms, force–displacement histories, and is validated using test results. This methodology combines meaningful insight in the failure mechanisms with a manageable computational effort, achieving a factor 50 improvement compared to a benchmark. A parametric analysis summarised in failure maps relates damage mechanisms to impact energy, mass, and laminate thickness. The proposed methodology strikes a balance between computational efficiency and accuracy, making it a valuable tool for optimum design and certification of thick CFRP composite laminates under impact.

Computational Investigation of the High-Temperature Chemistry of Small Hydroxy Ethers: Methoxymethanol, Ethoxymethanol, and 2-Methoxyethanol.

In the search for alternative energy carriers that can replace conventional fossil fuels, sustainably produced oxygenated hydrocarbons represent a promising class of potential candidates. An illustrative member of this class of alternative biofuels are oxymethylene ethers (OMEs). This study makes a contribution to this objective by investigating hydroxy ethers, specifically methoxymethanol, ethoxymethanol, and 2-methoxyethanol. These bifunctional oxygenated molecules are relevant intermediates formed during the combustion of OMEs or via an equilibrium reaction in methanol and formaldehyde mixtures. The high-temperature chemistry of hydroxy ethers is examined, with a particular focus on the unimolecular reactions involving H atom migration and bond fission. Quantum chemical calculations are utilized to gain insights into these processes. The results include bond dissociation energies, one-dimensional representations of the potential energy surfaces along the reaction coordinate for the simulated reactions, thermodynamic properties, and reaction rate parameters, which were derived via established ab initio methods. A detailed account of the unimolecular decomposition reactions of methoxymethanol, ethoxymethanol, and 2-methoxyethanol is provided, including the predominant reaction rates and pressure dependencies. The BDEs are benchmarked against data from different literature sources. For ME, a direct comparison of BDEs and reaction constants is performed with the results of previous studies. The derived reaction parameters for all three hydroxy ethers are compared to the results obtained via theoretical methods for dimethoxymethane and diethoxymethane, which feature similar chemical structures. The high-temperature chemistry of methoxymethanol and ethoxymethanol shares similarities and is dominated by rapid H atom migrations, forming alcohol + formaldehyde. This reaction is geometrically constrained in 2-methoxyethanol and therefore bond fission reactions dominate. A brief systematic comparison of the bifunctional oxygenated structure of hydroxy ether with other fuel species, such as n-alkanes, ethers, and alcohols, was conducted. The findings of this study are supported by the results of published chemical kinetic models of 2-methoxyethanol, formaldehyde + methanol, and OME-2. This includes simulations of chemical kinetic mechanisms, which were updated with the newly derived reaction rates, and compared with associated experimental data from literature. These experiments are concentration measurements obtained from a jet-stirred reactor and laminar burning velocities measured in an atmospheric burner. The results are in reasonable agreement with the reported experiments and thus may be considered an update to the models. The updated reaction rate constants showed no observable impact on the ignition delay times of OME-2. However, they did result in a significant increase in the concentration of methoxymethanol as an intermediate. In conclusion, this study provides crucial insights into the high-temperature combustion properties of hydroxy ethers. The data presented is intended to enhance comprehension of the decomposition behavior of these bifunctional oxygenated species and to support the development of detailed chemical kinetic models for combustion applications.

Probabilistic Cellular Automata Simulation of Microstructure Evolution: The Role of Model Parameters on Precision and Uncertainty

Abstract Probabilistic cellular automata (PCA) is a widely used and cost-efficient method for simulating microstructural evolution. In this method, probabilistic state change rules determine the evolution of cell states at each time step. However, its stochastic nature introduces inherent uncertainty, leading to non-repeatable results. Most microstructural simulation studies assess the accuracy of simulations by comparing predicted results with experimental observations, neglecting uncertainties in mathematical models and algorithms. In this study, the precision and stochastic behavior of microstructure evolution in the PCA simulations were investigated. The probabilities of transformations of cell states at each time step were formulated, and discrete probability distribution functions (dPDF) were introduced to analyze the frequency distribution of simulation outcomes. The performance and consistency of these dPDFs were assessed by comparing statistical analyses of PCA simulation results with dPDF predictions, revealing that the variance of simulation results is less than that of the binomial distribution function. Additionally, the effects of modeling parameters, such as model size, cellular resolution, and probability distribution of state changes in two- and three-dimensional PCA modeling, on the precision and reliability of simulation results were studied. In PCA models, simulation uncertainty inversely relates to the square root of model size. Furthermore, in 2D simulations, uncertainty is inversely proportional to the square root of the cellular resolution, while in 3D simulations, it is inversely proportional to the cellular resolution itself. These findings provide a simple and computationally efficient method for evaluating PCA simulation uncertainty and determining optimal simulation parameters, including model size, cellular resolution, and dimensionality.

CFD investigations of a shape-memory polymer foam-based endovascular embolization device for the treatment of intracranial aneurysms.

The hemodynamic and convective heat transfer effects of a patient-specific endovascular therapeutic agent based on shape-memory polymer foam (SMPf) are evaluated using computational fluid dynamics studies for six patient-specific aneurysm geometries. The SMPf device is modeled as a continuous porous medium with full expansion for the flow studies and with various degrees of expansion for the heat transfer studies. The flow simulation parameters were qualitatively validated based on the existing literature. Further, a mesh independence study was conducted to verify an optimal cell size and reduce the computational costs. For convective heat transfer, a worst-case scenario is evaluated where the minimum volumetric flow rate is applied alongside the zero-flux boundary conditions. In the flow simulations, we found a reduction of the average intra-aneurysmal flow of > 85% and a reduction of the maximum intra-aneurysmal flow of > 45% for all presented geometries. These findings were compared with the literature on numerical simulations of hemodynamic and heat transfer of SMPf devices. The results obtained from this study provide a novel and practical framework for optimizing the design of patient-specific SMPf devices, integrating advanced computational models of hemodynamics and heat transfer. This framework could guide the future development of personalized endovascular embolization solutions for intracranial aneurysms with improved therapeutic outcome.

Simulation of peak properties in thermoluminescence dosimeters with the potential stimulation of all electron traps

In thermoluminescence (TL) models, the competition between energy levels for electron trapping is a direct consequence of conduction band mediation in electron transport. From a theoretical perspective, these competitive effects are considered responsible for many TL phenomena. Furthermore, while they are easily detectable in simulation studies, there are no clear and reliable criteria for their experimental verification. In the present work, it is suggested that in certain materials with narrow band gap energies, existing electron traps can be accessed by thermal stimulation up to 500 °C. It is possible to predict properties through simulation that can then be verified experimentally. The simulation is applied to a complex TL glow curve consisting of four electron trapping levels and one luminescence center. With the appropriate selection of simulation parameter values, the competition is limited solely to electron transport, rather than to significant differences in parameter values. The simulation indicated that the last peak serves as the primary competitor. Furthermore, the shape of the TL glow curve depends on the strength of the competition. When the competitor is strong, the lower temperature peaks behave as first-order peaks, and the shape of the glow curve remains stable as a function of dose. The last peak, which acts as the final competitor, exhibits behavior similar to that of the peak derived from the one trap, one recombination (OTOR) model. This OTOR-like behavior provides an excellent experimental test for the simulation results and, consequently, for the model used. The validity of the simulation was assessed by comparing the input values of kinetic parameters with the output values obtained from computerized glow curve deconvolution (CGCD) analysis for each TL glow curve.

An automated toolbox for microcalcification cluster modeling for mammographic imaging.

Mammographic imaging is essential for breast cancer detection and diagnosis. In addition to masses, calcifications are of concern and the early detection of breast cancer also heavily relies on the correct interpretation of suspicious microcalcification clusters. Even with advances in imaging and the introduction of novel techniques such as digital breast tomosynthesis and contrast-enhanced mammography, a correct interpretation can still be challenging given the subtle nature and large variety of calcifications. Computer simulated lesion models can serve to develop, optimize, or improve imaging techniques. In addition to their use in comparative (virtual clinical trial) detection experiments, these models have potential application in training deep learning models and in the understanding and interpretation of breast lesions. Existing simulation methods, however, often lack the capacity to model the diversity occurring in breast lesions or to generate models relevant for a specific case. This study focuses on clusters of microcalcifications and introduces an automated, flexible toolbox designed to generate microcalcification cluster models customized to specific tasks. The toolbox allows users to control a large number of simulation parameters related to model characteristics such as lesion size, calcification shape, or number of microcalcifications per cluster. This leads to the capability of creating models that range from regular to complex clusters. Based on the input parameters, which are either tuned manually or pre-set for a specific clinical type, different sets of models can be simulated depending on the use case. Two lesion generation methods are described. The first method generates three-dimensional microcalcification clusters models based on geometrical shapes and transformations. The second method creates two-dimensional (2D) microcalcification cluster models for a specific 2D mammographic image. This novel method employs radiomics analysis to account for local textures, ensuring the simulated microcalcification cluster is appropriately integrated within the existing breast tissue. The toolbox is implemented in the Python language and can be conveniently run through a Jupyter Notebook interface, openly accessible at https://gitlab.kuleuven.be/medphysqa/deploy/breast-calcifications. Validation studies performed by radiologists assessed the level of malignancy and realism of clusters tuned with specific parameters and inserted in mammographic images. The flexibility of the toolbox with multiple simulation methods is illustrated, as well as the compatibility with different simulation frameworks and image types. The automation allows for the straightforward and fast generation of diverse microcalcification cluster models. The generated models are most likely applicable for various tasks as they can be configured in a variety of ways and inserted in different types of mammographic images of multiple acquisition systems. Validation studies confirmed the capacity to simulate realistic clusters and capture clinical properties when tuned with appropriate parameter settings. This simulation toolbox offers a flexible means of simulating microcalcification cluster models with potential use in both technical and clinical research in mammography imaging. The 3D generation methods allow for specifying many characteristics regarding the calcification shape and cluster architecture, and the 2D generation method presents a novel manner to create microcalcification clusters tailored to existing breast textures.

Aerodynamics evaluation and flight test of a vertical take-off and landing fixed-wing UAV with joined-wing configuration in transition flight state

The transition states between the cruising flight and the taking-off/landing process of a vertical take-off and landing (VTOL) fixed-wing UAV of joined-wing configuration are subject to significant unsteady aerodynamic interference. In this paper, the aerodynamic characteristics of this VTOL UAV during transition flights are evaluated by CFD with and without crosswind interference, in order to reveal the underlying mechanisms of the transition process. Based on these CFD-derived parameters, a flight test with a specifically-designed joined-wing VTOL UAV is proposed. The obtained results demonstrate that the forward flight speed is a crucial parameter during the takeoff transition phase. In the time interval of 5–15 s, significant disturbances are observed in the forces and moments due to the rotor deceleration and forward propeller acceleration, which result in slipstream and downwash flow effects. When crosswind disturbance is added, significant roll and yaw moments arise due to the vast vertical stabilizer area, which requires coordinated attitude adjustments between the rotor and the fixed-wing rudder surface. The descent transition phase is set a duration of 10 s. Four seconds later, as the rotor downwash flow intensifies, the lift force of the fixed wing is transferred to the rotor. When the combined lift becomes insufficient, the flight altitude decreases. When introducing crosswind disturbance, the entire aircraft undergoes significant additional pitch, yaw, and roll moments, with a maximum wave momentum greater than 200 %. Flight tests are then conducted using simulated parameters. The obtained results show that the take-off and landing transition responses without crosswinds are consistent with the predicted outcomes, which demonstrates the high effectiveness of the CFD simulations in predicting these transitions.

A stochastic population model for the impact of cancer cell dormancy on therapy success

Therapy evasion – and subsequent disease progression – is a major challenge in current oncology. An important role in this context seems to be played by various forms of cancer cell dormancy. For example, therapy-induced dormancy, over short timescales, can create serious obstacles to aggressive treatment approaches such as chemotherapy, and long-term dormancy may lead to relapses and metastases even many years after an initially successful treatment.In this paper, we focus on individual cancer cells switching into and out of a dormant state both spontaneously as well as in response to treatment. We introduce an idealized mathematical model, based on stochastic agent-based interactions, for the dynamics of cancer cell populations involving individual short-term dormancy, and allow for a range of (multi-drug) therapy protocols. Our analysis – based on simulations of the many-particle limit – shows that in our model, depending on the specific underlying dormancy mechanism, even a small initial population (of explicitly quantifiable size) of dormant cells can lead to therapy failure under classical single-drug treatments that would successfully eradicate the tumour in the absence of dormancy. We further investigate and quantify the effectiveness of several multi-drug regimes (manipulating dormant cancer cells in specific ways, including increasing or decreasing resuscitation rates or targeting dormant cells directly). Relying on quantitative results for concrete simulation parameters, we provide some general basic rules for the design of (multi-)drug treatment protocols depending on the types and processes of dormancy mechanisms present in the population.

Multi-objective optimization prediction model for building parameters of photovoltaic windows based on NSGA II-BP

This article simulates the indoor useful daylight illuminance (UDI), energy consumption, and power generation of photovoltaic (PV) window buildings using EnergyPlus simulation software. Extensive data on these simulation parameters are obtained using parametric simulation software and combined with actual meteorological data. The factors significantly influencing PV window building performance are determined based on ANOVA. A model is developed to predict energy consumption, power generation, and UDI of PV window buildings using a back propagation neural network. For better lighting quality, lower energy consumption, and greater power generation, NSGA-II is introduced to optimize the windows' performance with multi-objective parameters. Moreover, the resulting energy saving rate, annual average power generation growth rate, and UDI growth rate are compared with the initial values to evaluate the effectiveness of the optimal solution. The results demonstrate that the energy saving rate of the building is 18.23 %, and the growth rates of the useful daylight illuminance and power generation reach 41.6 % and 5.12 %, respectively, compared to the initial values.

Host-Guest Binding Free Energies à la Carte: An Automated OneOPES Protocol.

Estimating absolute binding free energies from molecular simulations is a key step in computer-aided drug design pipelines, but the agreement between computational results and experiments is still very inconsistent. Both the accuracy of the computational model and the quality of the statistical sampling contribute to this discrepancy, yet disentangling the two remains a challenge. In this study, we present an automated protocol based on OneOPES, an enhanced sampling method that exploits replica exchange and can accelerate several collective variables to address the sampling problem. We apply this protocol to 37 host-guest systems. The simplicity of setting up the simulations and producing well-converged binding free energy estimates without the need to optimize simulation parameters provides a reliable solution to the sampling problem. This, in turn, allows for a systematic force field comparison and ranking according to the correlation between simulations and experiments, which can inform the selection of an appropriate model. The protocol can be readily adapted to test more force field combinations and study more complex protein-ligand systems, where the choice of an appropriate physical model is often based on heuristic considerations rather than systematic optimization.

Simulating Real-Time Molecular Electron Dynamics Efficiently Using the Time-Dependent Density Matrix Renormalization Group.

Compared to ground-state electronic structure optimizations, accurate simulations of molecular real-time electron dynamics are usually much more difficult to perform. To simulate electron dynamics, the time-dependent density matrix renormalization group (TDDMRG) has been shown to offer an attractive compromise between accuracy and cost. However, many simulation parameters significantly affect the quality and efficiency of a TDDMRG simulation. So far, it is unclear whether common wisdom from ground-state DMRG carries over to the TDDMRG, and a guideline on how to choose these parameters is missing. Here, in order to establish such a guideline, we investigate the convergence behavior of the main TDDMRG simulation parameters, such as time integrator, the choice of orbitals, and the choice of matrix-product-state representation for complex-valued nonsinglet states. In addition, we propose a method to select orbitals that are tailored to optimize the dynamics. Lastly, we showcase the TDDMRG by applying it to charge migration ionization dynamics in furfural, where we reveal a rapid conversion from an ionized state with a σ character to one with a π character within less than a femtosecond.

NitroNet – a machine learning model for the prediction of tropospheric NO2 profiles from TROPOMI observations

Abstract. We introduce NitroNet, a deep learning model for the prediction of tropospheric NO2 profiles from satellite column measurements. NitroNet is a neural network trained on synthetic NO2 profiles from the regional chemistry and transport model WRF-Chem, which was operated on a European domain for the month of May 2019. This WRF-Chem simulation was constrained by in situ and satellite measurements, which were used to optimize important simulation parameters (e.g. the boundary layer scheme). The NitroNet model receives NO2 vertical column densities (VCDs) from the TROPOspheric Monitoring Instrument (TROPOMI) and ancillary variables (meteorology, emissions, etc.) as input, from which it reproduces NO2 concentration profiles. Training of the neural network is conducted on a filtered dataset, meaning that NO2 profiles showing strong disagreement (>20 %) with colocated TROPOMI column measurements are discarded. We present a first evaluation of NitroNet over a variety of geographical and temporal domains (Europe, the US West Coast, India, and China) and different seasons. For this purpose, we validate the NO2 profiles predicted by NitroNet against satellite, in situ, and MAX-DOAS (Multi-Axis Differential Optical Absorption Spectroscopy) measurements. The training data were previously validated against the same datasets. During summertime, NitroNet shows small biases and strong correlations with all three datasets: a bias of +6.7 % and R=0.95 for TROPOMI NO2 VCDs, a bias of −10.5 % and R=0.75 for AirBase surface concentrations, and a bias of −34.3 % to +99.6 % with R=0.83–0.99 for MAX-DOAS measurements. In comparison to TROPOMI satellite data, NitroNet even shows significantly lower errors and stronger correlation than a direct comparison with WRF-Chem numerical results. During wintertime considerable low biases arise because the summertime/late-spring training data are not fully representative of all atmospheric wintertime characteristics (e.g. longer NO2 lifetimes). Nonetheless, the wintertime performance of NitroNet is surprisingly good and comparable to that of classic regional chemistry and transport models. NitroNet can demonstrably be used outside the geographic and temporal domain of the training data with only slight performance reductions. What makes NitroNet unique when compared to similar existing deep learning models is the inclusion of synthetic model data, which offers important benefits: due to the lack of NO2 profile measurements, models trained on empirical datasets are limited to the prediction of surface concentrations learned from in situ measurements. NitroNet, however, can predict full tropospheric NO2 profiles. Furthermore, in situ measurements of NO2 are known to suffer from biases, often larger than +20 %, due to cross-sensitivities to photooxidants, which other models trained on empirical data inevitably reproduce.