scholarly journals An Online-Learned Neural Network Chemical Solver for Stable Long-Term Global Simulations of Atmospheric Chemistry

2021 ◽  
Author(s):  
Makoto Kelp ◽  
Daniel Jacob ◽  
Haipeng Lin ◽  
Melissa Sulprizio
Keyword(s):  
Neural Network ◽  
Atmospheric Chemistry ◽  
Surface Ozone ◽  
Kinetic Equations ◽  
Online Training ◽  
Chemical Mechanism ◽  
Error Growth ◽  
Time Step ◽  
Implicit Solvers ◽  
Order Of Magnitude

A major computational barrier in global modeling of atmospheric chemistry is the numerical integration of the coupled kinetic equations describing the chemical mechanism. Machine-learned (ML) solvers can offer order-of-magnitude speedup relative to conventional implicit solvers, but past implementations have suffered from fast error growth and only run for short simulation times (<1 month). A successful ML solver for global models must avoid error growth over year-long simulations and allow for re-initialization of the chemical trajectory by transport at every time step. Here we explore the capability of a neural network solver equipped with an autoencoder to achieve stable full-year simulations of tropospheric oxidant chemistry in the global 3-D GEOS-Chem model, replacing its standard mechanism (228 species) by the Super-Fast mechanism (12 species) to avoid the curse of dimensionality. We find that online training of the ML solver within GEOS-Chem is essential for accuracy, whereas offline training from archived GEOS-Chem inputs/outputs produces large errors. After online training we achieve stable 1-year simulations with five-fold speedup compared to the standard implicit Rosenbrock solver with global tropospheric normalized mean biases of -0.3% for ozone, 1% for hydrogen oxide radicals, and -5% for nitrogen oxides. The ML solver captures the diurnal and synoptic variability of surface ozone at polluted and clean sites. There are however large regional biases for ozone and NOx under remote conditions where chemical aging leads to error accumulation. These regional biases remain a major limitation for practical application, and ML emulation would be more difficult in a more complex mechanism.

A Modified Implicit Iterative Difference Algorithm for Stiff Chemical Kinetic Equations in Complex Combustion System

2011 ◽  
Vol 295-297 ◽  
pp. 2333-2340 ◽  
Author(s):  
Bo He ◽  
Wan Sheng Nie ◽  
Song Jiang Feng ◽  
Guo Qiang Li
Keyword(s):  
Mass Fraction ◽  
Kinetic Equations ◽  
Operator Method ◽  
Chemical Kinetic ◽  
Chemical Mechanism ◽  
Combustion System ◽  
Time Step ◽  
Nitrogen Tetroxide ◽  
Elementary Reactions ◽  
Computation Efficiency

The splitting-operator method has been widely used in the numerical simulation of complex combustion system with its high computation efficiency. The main difficult encountered in the method is how to integrate the stiff chemical kinetics ordinary differential equation (ODE) efficiently and accurately in each grid node of flow field. Although several ODE software packages such as LSODE and VODE have very predominant performance in integrating stiff and non-stiff case, it couldn’t ensure the positivity of mass fraction in integrating stiff chemical kinetic ODE. A detail liquid rocket hypergolic bipropellant combination of Monomethylhydrazine (MMH)/Red Fuming Nitric Acid (RFNA) chemical mechanism consisting of 550 elementary reactions among 83 species was constituted basing on several relevant literatures firstly. And then a modified implicit iterative difference algorithm with variable time steps was developed to tackle the problem of mass fraction non-positive. The results from the integration of MMH /nitrogen tetroxide (NTO) chemical kinetic ODE indicate that the algorithm is always retain stability and positive value of mass fraction. The time step length would also be automatically adjusted at different chemical reaction time in the algorithm to keep its computational efficiency.

scholarly journals An adaptive method for speeding up the numerical integration of chemical mechanisms in atmospheric chemistry models: application to GEOS-Chem version 12.0.0

2019 ◽  
Author(s):  
Lu Shen ◽  
Daniel J. Jacob ◽  
Mauricio Santillana ◽  
Xuan Wang ◽  
Wei Chen
Keyword(s):  
Numerical Integration ◽  
Atmospheric Chemistry ◽  
Computational Cost ◽  
Adaptive Method ◽  
Chemical Mechanism ◽  
Individual Species ◽  
Time Step ◽  
Chemical Mechanisms ◽  
Chemical Complexity ◽  
Loss Rates

Abstract. The major computational bottleneck in atmospheric chemistry models is the numerical integration of the stiff coupled system of kinetic equations describing the chemical evolution of the system as defined by the model chemical mechanism (typically over 100 coupled species). We present an adaptive method to greatly reduce the computational cost of that numerical integration in global 3-D models while maintaining high accuracy. Most of the atmosphere does not in fact require solving for the full chemical complexity of the mechanism, so considerable simplification is possible if one can recognize the dynamic continuum of chemical complexity required across the atmospheric domain. We do this by constructing a limited set of reduced chemical mechanisms (chemical regimes) to cover the range of atmospheric conditions, and then pick locally and on the fly which mechanism to use for a given gridbox and time step on the basis of computed production and loss rates for individual species. Application to the GEOS-Chem global 3-D model for oxidant-aerosol chemistry in the troposphere and stratosphere (full mechanism of 228 species) is presented. We show that 20 chemical regimes can largely encompass the range of conditions encountered in the model. Results from a 2-year GEOS-Chem simulation shows that our method can reduce the computational cost of chemical integration by 30–40 % while maintaining accuracy better than 1 % and with no error growth. Our method retains the full complexity of the original chemical mechanism where it is needed, provides the same model output diagnostics (species production and loss rates, reaction rates) as the full mechanism, and can accommodate changes in the chemical mechanism or in model resolution without having to reconstruct the chemical regimes.

scholarly journals Technical note: application of α-QSS to the numerical integration of kinetic equations in tropospheric chemistry

2005 ◽  
Vol 5 (4) ◽  
pp. 6215-6262
Author(s):  
F. Liu ◽  
E. Schaller ◽  
D. R. Mott
Keyword(s):  
Numerical Integration ◽  
Atmospheric Chemistry ◽  
Single Point ◽  
Kinetic Equations ◽  
Flow Simulation ◽  
Tropospheric Chemistry ◽  
Box Model ◽  
Reacting Flow ◽  
Time Step ◽  
Transport Modelling

Abstract. A major task in many applications of atmospheric chemistry transport problems is the numerical integration of stiff systems of Ordinary Differential Equations (ODEs) describing the chemical transformations. A faster solver that is easier to couple to the other physics in the problem is still needed. The integration method, α-QSS, corresponding to the solver CHEMEQ2 aims at meeting the demands of a process-split, reacting-flow simulation (Mott 2000; Mott and Oran, 2001). However, this integrator has yet to be applied to the numerical integration of kinetic equations in tropospheric chemistry. A zero-dimensional (box) model is developed to test how well CHEMEQ2 works on the tropospheric chemistry equations. This paper presents the testing results. The reference chemical mechanisms herein used are Regional Atmospheric Chemistry Mechanism (RACM) (Stockwell et al., 1997) and its secondary lumped successor Regional Lumped Atmospheric Chemical Scheme (ReLACS) (Crassier et al., 2000). The box model is forced and initialized by the DRY scenarios of Protocol Ver. 2 developed by EUROTRAC (Poppe et al., 2001). The accuracy of CHEMEQ2 is evaluated by comparing the results to solutions obtained with VODE. This comparison is made with parameters of the error tolerance, relative difference with respect to VODE scheme, trade off between accuracy and efficiency, global time step for integration etc. The study based on the comparison concludes that the single-point α-QSS approach is fast and moderately accurate as well as easy to couple to reacting flow simulation models, which makes CHEMEQ2 one of the best candidates for three-dimensional atmospheric Chemistry Transport Modelling (CTM) studies. In addition the RACM mechanism may be replaced by ReLACS mechanism for tropospheric chemistry transport modelling. The testing results also imply that the accuracy for chemistry numerical simulations is highly different from species to species. Therefore ozone is not the good choice for testing numerical ODE solvers or for evaluation of mechanisms because current tropospheric chemistry mechanisms are mainly designed for troposphere ozone prediction.

scholarly journals Isoprene chemistry in pristine and polluted Amazon environments: Eulerian and Lagrangian model frameworks and the strong bearing they have on our understanding of surface ozone and predictions of rainforest exposure to this priority pollutant

2015 ◽  
Vol 15 (17) ◽  
pp. 24251-24310 ◽  
Author(s):  
J. G. Levine ◽  
A. R. MacKenzie ◽  
O. J. Squire ◽  
A. T. Archibald ◽  
P. T. Griffiths ◽  
...  
Keyword(s):  
Atmospheric Chemistry ◽  
Amazon Basin ◽  
Surface Ozone ◽  
Chemical Mechanism ◽  
Atmospheric Composition ◽  
Lagrangian Model ◽  
Model Framework ◽  
Eulerian Model ◽  
Airborne Measurements ◽  
The Impact

Abstract. This study explores our ability to simulate the atmospheric chemistry stemming from isoprene emissions in pristine and polluted regions of the Amazon basin. We confront two atmospheric chemistry models – a global, Eulerian chemistry-climate model (UM-UKCA) and a trajectory-based Lagrangian model (CiTTyCAT) – with recent airborne measurements of atmospheric composition above the Amazon made during the SAMBBA campaign of 2012. The simulations with the two models prove relatively insensitive to the chemical mechanism employed; we explore one based on the Mainz Isoprene Mechanism, and an updated one that includes changes to the chemistry of first generation isoprene nitrates (ISON) and the regeneration of hydroxyl radicals via the formation of hydroperoxy-aldehydes (HPALDS) from hydroperoxy radicals (ISO2). In the Lagrangian model, the impact of increasing the spatial resolution of trace gas emissions employed from 3.75° × 2.5° to 0.1° × 0.1° varies from one flight to another, and from one chemical species to another. What consistently proves highly influential on our simulations, however, is the model framework itself – how the treatment of transport, and consequently mixing, differs between the two models. The lack of explicit mixing in the Lagrangian model yields variability in atmospheric composition more reminiscent of that exhibited by the measurements. In contrast, the combination of explicit (and implicit) mixing in the Eulerian model removes much of this variability but yields better agreement with the measurements overall. We therefore explore a simple treatment of mixing in the Lagrangian model that, drawing on output from the Eulerian model, offers a compromise between the two models. We use this Lagrangian/Eulerian combination, in addition to the separate Eulerian and Lagrangian models, to simulate ozone at a site in the boundary layer downwind of Manaus, Brazil. The Lagrangian/Eulerian combination predicts a value for an AOT40-like accumulated exposure metric of around 1000 ppbv h, compared to just 20 ppbv h with the Eulerian model. The model framework therefore has considerable bearing on our understanding of the frequency at which, and the duration for which, the rainforest is exposed to damaging ground-level ozone concentrations.

scholarly journals An adaptive method for speeding up the numerical integration of chemical mechanisms in atmospheric chemistry models: application to GEOS-Chem version 12.0.0

2020 ◽  
Vol 13 (5) ◽  
pp. 2475-2486 ◽  
Author(s):  
Lu Shen ◽  
Daniel J. Jacob ◽  
Mauricio Santillana ◽  
Xuan Wang ◽  
Wei Chen
Keyword(s):  
Numerical Integration ◽  
Atmospheric Chemistry ◽  
Computational Cost ◽  
Adaptive Method ◽  
Chemical Mechanism ◽  
Individual Species ◽  
Time Step ◽  
Chemical Mechanisms ◽  
Chemical Complexity ◽  
Loss Rates

Abstract. The major computational bottleneck in atmospheric chemistry models is the numerical integration of the stiff coupled system of kinetic equations describing the chemical evolution of the system as defined by the model chemical mechanism (typically over 100 coupled species). We present an adaptive method to greatly reduce the computational cost of that numerical integration in global 3-D models while maintaining high accuracy. Most of the atmosphere does not in fact require solving for the full chemical complexity of the mechanism, so considerable simplification is possible if one can recognize the dynamic continuum of chemical complexity required across the atmospheric domain. We do this by constructing a limited set of reduced chemical mechanisms (chemical regimes) to cover the range of atmospheric conditions and then pick locally and on the fly which mechanism to use for a given grid box and time step on the basis of computed production and loss rates for individual species. Application to the GEOS-Chem global 3-D model for oxidant–aerosol chemistry in the troposphere and stratosphere (full mechanism of 228 species) is presented. We show that 20 chemical regimes can largely encompass the range of conditions encountered in the model. Results from a 2-year GEOS-Chem simulation shows that our method can reduce the computational cost of chemical integration by 30 %–40 % while maintaining accuracy better than 1 % and with no error growth. Our method retains the full complexity of the original chemical mechanism where it is needed, provides the same model output diagnostics (species production and loss rates, reaction rates) as the full mechanism, and can accommodate changes in the chemical mechanism or in model resolution without having to reconstruct the chemical regimes.

scholarly journals Statistical Uncertainty of DNS in Geometries without Homogeneous Directions

2021 ◽  
Vol 11 (4) ◽  
pp. 1399
Author(s):  
Jure Oder ◽  
Cédric Flageul ◽  
Iztok Tiselj
Keyword(s):  
Channel Flow ◽  
A Priori ◽  
Time Integration ◽  
Navier Stokes ◽  
Statistical Uncertainty ◽  
Time Step ◽  
Order Of Magnitude ◽  
Averaging Time ◽  
The Individual ◽  
Time Required

In this paper, we present uncertainties of statistical quantities of direct numerical simulations (DNS) with small numerical errors. The uncertainties are analysed for channel flow and a flow separation case in a confined backward facing step (BFS) geometry. The infinite channel flow case has two homogeneous directions and this is usually exploited to speed-up the convergence of the results. As we show, such a procedure reduces statistical uncertainties of the results by up to an order of magnitude. This effect is strongest in the near wall regions. In the case of flow over a confined BFS, there are no such directions and thus very long integration times are required. The individual statistical quantities converge with the square root of time integration so, in order to improve the uncertainty by a factor of two, the simulation has to be prolonged by a factor of four. We provide an estimator that can be used to evaluate a priori the DNS relative statistical uncertainties from results obtained with a Reynolds Averaged Navier Stokes simulation. In the DNS, the estimator can be used to predict the averaging time and with it the simulation time required to achieve a certain relative statistical uncertainty of results. For accurate evaluation of averages and their uncertainties, it is not required to use every time step of the DNS. We observe that statistical uncertainty of the results is uninfluenced by reducing the number of samples to the point where the period between two consecutive samples measured in Courant–Friedrichss–Levy (CFL) condition units is below one. Nevertheless, crossing this limit, the estimates of uncertainties start to exhibit significant growth.

scholarly journals Deep Learning Forecasts of Soil Moisture: Convolutional Neural Network and Gated Recurrent Unit Models Coupled with Satellite-Derived MODIS, Observations and Synoptic-Scale Climate Index Data

2021 ◽  
Vol 13 (4) ◽  
pp. 554
Author(s):  
A. A. Masrur Ahmed ◽  
Ravinesh C Deo ◽  
Nawin Raj ◽  
Afshin Ghahramani ◽  
Qi Feng ◽  
...  
Keyword(s):  
Neural Network ◽  
Soil Moisture ◽  
Deep Learning ◽  
Convolutional Neural Network ◽  
Climate Index ◽  
Climate Indices ◽  
High Temporal Resolution ◽  
Synoptic Scale ◽  
Time Step ◽  
Gated Recurrent Unit

Remotely sensed soil moisture forecasting through satellite-based sensors to estimate the future state of the underlying soils plays a critical role in planning and managing water resources and sustainable agricultural practices. In this paper, Deep Learning (DL) hybrid models (i.e., CEEMDAN-CNN-GRU) are designed for daily time-step surface soil moisture (SSM) forecasts, employing the gated recurrent unit (GRU), complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), and convolutional neural network (CNN). To establish the objective model’s viability for SSM forecasting at multi-step daily horizons, the hybrid CEEMDAN-CNN-GRU model is tested at 1st, 5th, 7th, 14th, 21st, and 30th day ahead period by assimilating a comprehensive pool of 52 predictor dataset obtained from three distinct data sources. Data comprise satellite-derived Global Land Data Assimilation System (GLDAS) repository a global, high-temporal resolution, unique terrestrial modelling system, and ground-based variables from Scientific Information Landowners (SILO) and synoptic-scale climate indices. The results demonstrate the forecasting capability of the hybrid CEEMDAN-CNN-GRU model with respect to the counterpart comparative models. This is supported by a relatively lower value of the mean absolute percentage and root mean square error. In terms of the statistical score metrics and infographics employed to test the final model’s utility, the proposed CEEMDAN-CNN-GRU models are considerably superior compared to a standalone and other hybrid method tested on independent SSM data developed through feature selection approaches. Thus, the proposed approach can be successfully implemented in hydrology and agriculture management.

scholarly journals A two-dimensional glacier–fjord coupled model applied to estimate submarine melt rates and front position changes of Hansbreen, Svalbard

2018 ◽  
Vol 64 (247) ◽  
pp. 745-758 ◽  
Author(s):  
E. DE ANDRÉS ◽  
J. OTERO ◽  
F. NAVARRO ◽  
A. PROMIŃSKA ◽  
J. LAPAZARAN ◽  
...  
Keyword(s):  
Coupled Model ◽  
Small Scale ◽  
Two Dimensional ◽  
Time Step ◽  
Software Packages ◽  
Front Position ◽  
Circulation Patterns ◽  
Glacier Front ◽  
Order Of Magnitude ◽  
Frontal Ablation

ABSTRACTWe have developed a two-dimensional coupled glacier–fjord model, which runs automatically using Elmer/Ice and MITgcm software packages, to investigate the magnitude of submarine melting along a vertical glacier front and its potential influence on glacier calving and front position changes. We apply this model to simulate the Hansbreen glacier–Hansbukta proglacial–fjord system, Southwestern Svalbard, during the summer of 2010. The limited size of this system allows us to resolve some of the small-scale processes occurring at the ice–ocean interface in the fjord model, using a 0.5 s time step and a 1 m grid resolution near the glacier front. We use a rich set of field data spanning the period April–August 2010 to constrain, calibrate and validate the model. We adjust circulation patterns in the fjord by tuning subglacial discharge inputs that best match observed temperature while maintaining a compromise with observed salinity, suggesting a convectively driven circulation in Hansbukta. The results of our model simulations suggest that both submarine melting and crevasse hydrofracturing exert important controls on seasonal frontal ablation, with submarine melting alone not being sufficient for reproducing the observed patterns of seasonal retreat. Both submarine melt and calving rates accumulated along the entire simulation period are of the same order of magnitude, ~100 m. The model results also indicate that changes in submarine melting lag meltwater production by 4–5 weeks, which suggests that it may take up to a month for meltwater to traverse the englacial and subglacial drainage network.

scholarly journals Comparisons of observed and modeled OH and HO2concentrations during the ambient measurement period of the HOxComp field campaign

2012 ◽  
Vol 12 (5) ◽  
pp. 2567-2585 ◽  
Author(s):  
Y. Kanaya ◽  
A. Hofzumahaus ◽  
H.-P. Dorn ◽  
T. Brauers ◽  
H. Fuchs ◽  
...  
Keyword(s):  
Atmospheric Chemistry ◽  
Chemical Mechanism ◽  
Ozone Production ◽  
High Yield ◽  
Base Case ◽  
Peroxy Radicals ◽  
Field Campaign ◽  
Mixing Ratios ◽  
Measurement Mode ◽  
Single Instrument

Abstract. A photochemical box model constrained by ancillary observations was used to simulate OH and HO2 concentrations for three days of ambient observations during the HOxComp field campaign held in Jülich, Germany in July 2005. Daytime OH levels observed by four instruments were fairly well reproduced to within 33% by a base model run (Regional Atmospheric Chemistry Mechanism with updated isoprene chemistry adapted from Master Chemical Mechanism ver. 3.1) with high R2 values (0.72–0.97) over a range of isoprene (0.3–2 ppb) and NO (0.1–10 ppb) mixing ratios. Daytime HO2(*) levels, reconstructed from the base model results taking into account the sensitivity toward speciated RO2 (organic peroxy) radicals, as recently reported from one of the participating instruments in the HO2 measurement mode, were 93% higher than the observations made by the single instrument. This also indicates an overprediction of the HO2 to OH recycling. Together with the good model-measurement agreement for OH, it implies a missing OH source in the model. Modeled OH and HO2(*) could only be matched to the observations by addition of a strong unknown loss process for HO2(*) that recycles OH at a high yield. Adding to the base model, instead, the recently proposed isomerization mechanism of isoprene peroxy radicals (Peeters and Müller, 2010) increased OH and HO2(*) by 28% and 13% on average. Although these were still only 4% higher than the OH observations made by one of the instruments, larger overestimations (42–70%) occurred with respect to the OH observations made by the other three instruments. The overestimation in OH could be diminished only when reactive alkanes (HC8) were solely introduced to the model to explain the missing fraction of observed OH reactivity. Moreover, the overprediction of HO2(*) became even larger than in the base case. These analyses imply that the rates of the isomerization are not readily supported by the ensemble of radical observations. One of the measurement days was characterized by low isoprene concentrations (∼0.5 ppb) and OH reactivity that was well explained by the observed species, especially before noon. For this selected period, as opposed to the general behavior, the model tended to underestimate HO2(*). We found that this tendency is associated with high NOx concentrations, suggesting that some HO2 production or regeneration processes under high NOx conditions were being overlooked; this might require revision of ozone production regimes.

scholarly journals Accelerating simulations using REDCHEM_v0.0 for atmospheric chemistry mechanism reduction

2018 ◽  
Vol 11 (8) ◽  
pp. 3391-3407 ◽  
Author(s):  
Zacharias Marinou Nikolaou ◽  
Jyh-Yuan Chen ◽  
Yiannis Proestos ◽  
Jos Lelieveld ◽  
Rolf Sander
Keyword(s):  
Atmospheric Chemistry ◽  
Transport Model ◽  
A Priori ◽  
Chemical Mechanism ◽  
Climate Modelling ◽  
Mechanism Reduction ◽  
Computational Speed ◽  
Speed Up ◽  
Spatio Temporal ◽  
Skeletal Mechanism

Abstract. Chemical mechanism reduction is common practice in combustion research for accelerating numerical simulations; however, there have been limited applications of this practice in atmospheric chemistry. In this study, we employ a powerful reduction method in order to produce a skeletal mechanism of an atmospheric chemistry code that is commonly used in air quality and climate modelling. The skeletal mechanism is developed using input data from a model scenario. Its performance is then evaluated both a priori against the model scenario results and a posteriori by implementing the skeletal mechanism in a chemistry transport model, namely the Weather Research and Forecasting code with Chemistry. Preliminary results, indicate a substantial increase in computational speed-up for both cases, with a minimal loss of accuracy with regards to the simulated spatio-temporal mixing ratio of the target species, which was selected to be ozone.

Sign in / Sign up

Export Citation Format

Share Document