Articles | Volume 15, issue 8
Development and technical paper
02 May 2022
Development and technical paper | 02 May 2022
Efficient high-dimensional variational data assimilation with machine-learned reduced-order models
Romit Maulik et al.
No articles found.
Caleb Phillips, Lindsay M. Sheridan, Patrick Conry, Dimitrios K. Fytanidis, Dmitry Duplyakin, Sagi Zisman, Nicolas Duboc, Matt Nelson, Rao Kotamarthi, Rod Linn, Marc Broersma, Timo Spijkerboer, and Heidi Tinnesand
Wind Energ. Sci., 7, 1153–1169,Short summary
Adoption of distributed wind turbines for energy generation is hindered by challenges associated with siting and accurate estimation of the wind resource. This study evaluates classic and commonly used methods alongside new state-of-the-art models derived from simulations and machine learning approaches using a large dataset from the Netherlands. We find that data-driven methods are most effective at predicting production at real sites and new models reliably outperform classic methods.
William Shaw, Larry Berg, Mithu Debnath, Georgios Deskos, Caroline Draxl, Virendra Ghate, Charlotte Hasager, Rao Kotamarthi, Jeffrey Mirocha, Paytsar Muradyan, William Pringle, David Turner, and James Wilczak
Wind Energ. Sci. Discuss.,
Preprint under review for WESShort summary
This paper provides a review of prominent scientific challenges to characterizing the offshore wind resource using as examples phenomena that occur in the rapidly developing wind energy areas off the United States. The paper also describes the current state of modelling and observations in the marine atmospheric boundary layer and provides specific recommendations for filling key current knowledge gaps.
Chuxuan Li, Alexander L. Handwerger, Jiali Wang, Wei Yu, Xiang Li, Noah J. Finnegan, Yingying Xie, Giuseppe Buscarnera, and Daniel E. Horton
Nat. Hazards Earth Syst. Sci. Discuss.,
Revised manuscript accepted for NHESSShort summary
In January 2021 a storm triggered numerous debris flows in a wildfire burn scar in California. We use a hydrologic model to assess debris flow hazards in pre-fire and postfire scenarios. Compared to pre-fire conditions, the postfire simulation yields dramatic increases in total and peak discharge, substantially increasing debris flow hazards. Our work demonstrates the utility of 3-D hydrologic models for investigating and potentially forecasting postfire debris flow hazards at regional scales.
Jiali Wang, Zhengchun Liu, Ian Foster, Won Chang, Rajkumar Kettimuthu, and V. Rao Kotamarthi
Geosci. Model Dev., 14, 6355–6372,Short summary
Downscaling, the process of generating a higher spatial or time dataset from a coarser observational or model dataset, is a widely used technique. Two common methodologies for performing downscaling are to use either dynamic (physics-based) or statistical (empirical). Here we develop a novel methodology, using a conditional generative adversarial network (CGAN), to perform the downscaling of a model's precipitation forecasts and describe the advantages of this method compared to the others.
Jaydeep Singh, Narendra Singh, Narendra Ojha, Amit Sharma, Andrea Pozzer, Nadimpally Kiran Kumar, Kunjukrishnapillai Rajeev, Sachin S. Gunthe, and V. Rao Kotamarthi
Geosci. Model Dev., 14, 1427–1443,Short summary
Atmospheric models often have limitations in simulating the geographically complex and climatically important central Himalayan region. In this direction, we have performed regional modeling at high resolutions to improve the simulation of meteorology and dynamics through a better representation of the topography. The study has implications for further model applications to investigate the effects of anthropogenic pressure over the Himalaya.
Jiali Wang, Prasanna Balaprakash, and Rao Kotamarthi
Geosci. Model Dev., 12, 4261–4274,Short summary
Parameterizations are frequently used in models representing physical phenomena and are often the computationally expensive portions of the code. Using model output from simulations performed using a weather model, we train deep neural networks to provide an accurate alternative to a physics-based parameterization. We demonstrate that a domain-aware deep neural network can successfully simulate the entire diurnal cycle of the boundary layer physics and the results are transferable.
Jiali Wang, Cheng Wang, Vishwas Rao, Andrew Orr, Eugene Yan, and Rao Kotamarthi
Geosci. Model Dev., 12, 3523–3539,Short summary
WRF-Hydro needs to be calibrated to optimize its output with respect to observations. However, when applied to a relatively large domain, both WRF-Hydro simulations and calibrations require intensive computing resources and are best performed in parallel. This study ported an independent calibration tool (parameter estimation tool – PEST) to high-performance computing clusters and adapted it to work with WRF-Hydro. The results show significant speedup for model calibration.
Jeffrey D. Mirocha, Matthew J. Churchfield, Domingo Muñoz-Esparza, Raj K. Rai, Yan Feng, Branko Kosović, Sue Ellen Haupt, Barbara Brown, Brandon L. Ennis, Caroline Draxl, Javier Sanz Rodrigo, William J. Shaw, Larry K. Berg, Patrick J. Moriarty, Rodman R. Linn, Veerabhadra R. Kotamarthi, Ramesh Balakrishnan, Joel W. Cline, Michael C. Robinson, and Shreyas Ananthan
Wind Energ. Sci., 3, 589–613,Short summary
This paper validates the use of idealized large-eddy simulations with periodic lateral boundary conditions to provide boundary-layer flow quantities of interest for wind energy applications. Sensitivities to model formulation, forcing parameter values, and grid configurations were also examined, both to ascertain the robustness of the technique and to characterize inherent uncertainties, as required for the evaluation of more general wind plant flow simulation approaches under development.
K. K. Shukla, K. Niranjan Kumar, D. V. Phanikumar, R. K. Newsom, V. R. Kotamarthi, T. B. M. J. Ouarda, and M. V. Ratnam
Atmos. Meas. Tech. Discuss.,
Revised manuscript not acceptedShort summary
Estimation of Cloud base height was carried out by using various ground based instruments (Doppler Lidar and Ceilometer) and satellite datasets (MODIS) over central Himalayan region for the first time. The present study demonstrates the potential of Doppler Lidar in precise estimation of cloud base height and updraft velocities. More such deployments will be invaluable inputs for regional weather prediction models over complex Himalayan terrains.
Y. Feng, V. R. Kotamarthi, R. Coulter, C. Zhao, and M. Cadeddu
Atmos. Chem. Phys., 16, 247–264,Short summary
Aerosol radiative effects are of great importance for climate studies over South Asia, such as the weakening of the South Asian monsoon in the 20th century. This study reveals the altitude dependence of commonly underestimated aerosol radiative properties over this region. It further demonstrates the importance of constraining aerosol vertical distributions and partitioning of scattering vs absorbing aerosols in simulating the subsequent regional dynamical and hydrological responses to aerosols.
B. A. Drewniak, U. Mishra, J. Song, J. Prell, and V. R. Kotamarthi
Biogeosciences, 12, 2119–2129,
J. Elliott, C. Müller, D. Deryng, J. Chryssanthacopoulos, K. J. Boote, M. Büchner, I. Foster, M. Glotter, J. Heinke, T. Iizumi, R. C. Izaurralde, N. D. Mueller, D. K. Ray, C. Rosenzweig, A. C. Ruane, and J. Sheffield
Geosci. Model Dev., 8, 261–277,Short summary
We present and describe the Global Gridded Crop Model Intercomparison (GGCMI) project, an ongoing international effort to 1) validate global models of crop productivity, 2) improve models through detailed analysis of processes, and 3) assess the impacts of climate change on agriculture and food security. We present analysis of data inputs for the project, detailed protocols for conducting and evaluating simulation outputs, and example results.
V. S. Manoharan, R. Kotamarthi, Y. Feng, and M. P. Cadeddu
Atmos. Chem. Phys., 14, 1159–1165,
Y. Feng, V. Ramanathan, and V. R. Kotamarthi
Atmos. Chem. Phys., 13, 8607–8621,
B. Drewniak, J. Song, J. Prell, V. R. Kotamarthi, and R. Jacob
Geosci. Model Dev., 6, 495–515,
Related subject area
Numerical methodsUniversity of Warsaw Lagrangian Cloud Model (UWLCM) 2.0: adaptation of a mixed Eulerian–Lagrangian numerical model for heterogeneous computing clustersPrediction error growth in a more realistic atmospheric toy model with three spatiotemporal scalesOn numerical broadening of particle-size spectra: a condensational growth study using PyMPDATA 1.0Lossy checkpoint compression in full waveform inversion: a case study with ZFPv0.5.5 and the overthrust modelBlockworlds 0.1.0: a demonstration of anti-aliased geophysics for probabilistic inversions of implicit and kinematic geological modelsImproved double Fourier series on a sphere and its application to a semi-implicit semi-Lagrangian shallow-water modelSciKit-GStat 1.0: a SciPy-flavored geostatistical variogram estimation toolbox written in PythonFlow-Py v1.0: a customizable, open-source simulation tool to estimate runout and intensity of gravitational mass flowsMulti-dimensional hydrological-hydraulic model with variational data assimilation for river networks and floodplainsEmulation of high-resolution land surface models using sparse Gaussian processes with application to JULESA three-dimensional variational data assimilation system for aerosol optical properties based on WRF-Chem v4.0: design, development, and application of assimilating Himawari-8 aerosol observationsImplementation of a Gaussian Markov random field sampler for forward uncertainty quantification in the Ice-sheet and Sea-level System Model v4.19A method for assessment of the general circulation model quality using the K-means clustering algorithm: a case study with GETM v2.5An explicit GPU-based material point method solver for elastoplastic problems (ep2-3De v1.0)MagIC v5.10: a two-dimensional message-passing interface (MPI) distribution for pseudo-spectral magnetohydrodynamics simulations in spherical geometryMachine-learning models to replicate large-eddy simulations of air pollutant concentrations along boulevard-type streetsRecalculation of error growth models' parameters for the ECMWF forecast systemHow biased are our models? – a case study of the alpine regionB-flood 1.0: an open-source Saint-Venant model for flash-flood simulation using adaptive refinementA micro-genetic algorithm (GA v1.7.1a) for combinatorial optimization of physics parameterizations in the Weather Research and Forecasting model (v4.0.3) for quantitative precipitation forecast in KoreaSymPKF (v1.0): a symbolic and computational toolbox for the design of parametric Kalman filter dynamicsNDCmitiQ v1.0.0: a tool to quantify and analyse greenhouse gas mitigation targetsCombining ensemble Kalman filter and reservoir computing to predict spatiotemporal chaotic systems from imperfect observations and modelsThe Coastline Evolution Model 2D (CEM2D) V1.1An iterative process for efficient optimisation of parameters in geoscientific models: a demonstration using the Parallel Ice Sheet Model (PISM) version 0.7.3Ocean Plastic Assimilator v0.2: assimilation of plastic concentration data into Lagrangian dispersion modelsDevelopment of a moving point source model for shipping emission dispersion modeling in EPISODE–CityChem v1.3Efficient Bayesian inference for large chaotic dynamical systemsConstraining stochastic 3-D structural geological models with topology information using approximate Bayesian computation in GemPy 2.1Retrieval of process rate parameters in the general dynamic equation for aerosols using Bayesian state estimation: BAYROSOL1.0A discontinuous Galerkin finite-element model for fast channelized lava flows v1.0A nested multi-scale system implemented in the large-eddy simulation model PALM model system 6.0Extending legacy climate models by adaptive mesh refinement for single-component tracer transport: a case study with ECHAM6-HAMMOZ (ECHAM6.3-HAM2.3-MOZ1.0)Using the Després and Lagoutière (1999) antidiffusive transport scheme: a promising and novel method against excessive vertical diffusion in chemistry-transport modelsPorosity and permeability prediction through forward stratigraphic simulations using GPM™ and Petrel™: application in shallow marine depositional settingsEffects of transient processes for thermal simulations of the Central European BasinAssessment of stochastic weather forecast of precipitation near European cities, based on analogs of circulationA note on precision-preserving compression of scientific dataAn N-dimensional Fortran interpolation programme (NterGeo.v2020a) for geophysics sciences – application to a back-trajectory programme (Backplumes.v2020r1) using CHIMERE or WRF outputsA framework to evaluate IMEX schemes for atmospheric modelsInequality-constrained free-surface evolution in a full Stokes ice flow model (evolve_glacier v1.1)A fast and efficient MATLAB-based MPM solver: fMPMM-solver v1.1Necessary conditions for algorithmic tuning of weather prediction models using OpenIFS as an exampleDevelopment of a submerged aquatic vegetation growth model in the Coupled Ocean–Atmosphere–Wave–Sediment Transport (COAWST v3.4) modelRetrieving monthly and interannual total-scale pH (pHT) on the East China Sea shelf using an artificial neural network: ANN-pHT-v1Development of a semi-Lagrangian advection scheme for the NEMO ocean model (3.1)Efficient multi-scale Gaussian process regression for massive remote sensing data with satGP v0.1.2PDE-NetGen 1.0: from symbolic partial differential equation (PDE) representations of physical processes to trainable neural network representationsSimple algorithms to compute meridional overturning and barotropic streamfunctions on unstructured meshesDevelopment of a two-way-coupled ocean–wave model: assessment on a global NEMO(v3.6)–WW3(v6.02) coupled configuration
Piotr Dziekan and Piotr Zmijewski
Geosci. Model Dev., 15, 4489–4501,Short summary
Detailed computer simulations of clouds are important for understanding Earth's atmosphere and climate. The paper describes how the UWLCM has been adapted to work on supercomputers. A distinctive feature of UWLCM is that air flow is calculated by processors at the same time as cloud droplets are modeled by graphics cards. Thanks to this, use of computing resources is maximized and the time to complete simulations of large domains is not affected by communications between supercomputer nodes.
Hynek Bednář and Holger Kantz
Geosci. Model Dev., 15, 4147–4161,Short summary
A scale-dependent error growth described by a power law or by a quadratic hypothesis is studied in Lorenz’s system with three spatiotemporal levels. The validity of power law is extended by including a saturation effect. The quadratic hypothesis can only serve as a first guess. In addition, we study the initial error growth for the ECMWF forecast system. Fitting the parameters, we conclude that there is an intrinsic limit of predictability after 22 days.
Michael A. Olesik, Jakub Banaśkiewicz, Piotr Bartman, Manuel Baumgartner, Simon Unterstrasser, and Sylwester Arabas
Geosci. Model Dev., 15, 3879–3899,Short summary
In systems such as atmospheric clouds, droplets undergo growth through condensation of vapor. The broadness of the resultant size spectrum of droplets influences precipitation likelihood and the radiative properties of clouds. One of the inherent limitations of simulations of the problem is the so-called numerical diffusion causing overestimation of the spectrum width, hence the term numerical broadening. In the paper, we take a closer look at one of the algorithms used in this context: MPDATA.
Navjot Kukreja, Jan Hückelheim, Mathias Louboutin, John Washbourne, Paul H. J. Kelly, and Gerard J. Gorman
Geosci. Model Dev., 15, 3815–3829,Short summary
Full waveform inversion (FWI) is a partial-differential equation (PDE)-constrained optimization problem that is notorious for its high computational load and memory footprint. In this paper we present a method that combines recomputation with lossy compression to accelerate the computation with minimal loss of precision in the results. We show this using experiments running FWI with a variety of compression settings on a popular academic dataset.
Richard Scalzo, Mark Lindsay, Mark Jessell, Guillaume Pirot, Jeremie Giraud, Edward Cripps, and Sally Cripps
Geosci. Model Dev., 15, 3641–3662,Short summary
This paper addresses numerical challenges in reasoning about geological models constrained by sensor data, especially models that describe the history of an area in terms of a sequence of events. Our method ensures that small changes in simulated geological features, such as the position of a boundary between two rock layers, do not result in unrealistically large changes to resulting sensor measurements, as occur presently using several popular modeling packages.
Geosci. Model Dev., 15, 2561–2597,Short summary
This paper proposes a new double Fourier series (DFS) method on a sphere that improves the numerical stability of a model compared with conventional DFS methods. The shallow-water model and the advection model using the new DFS method give stable results without the appearance of high-wavenumber noise near the poles. The model using the new DFS method is faster than the model using spherical harmonics (especially at high resolutions) and gives almost the same results.
Geosci. Model Dev., 15, 2505–2532,Short summary
I preset SciKit-GStat, a well-documented and tested Python package for variogram estimation. The variogram is the core means of geostatistics, which almost all other methods rely on. Geostatistical interpolation and field generation are widely spread in geoscience, i.e., for data assimilation or modeling. While SciKit-GStat focuses on effective and intuitive variogram estimation, it can interface with other prominent packages and make its variograms available for a multitude of methods.
Christopher J. L. D'Amboise, Michael Neuhauser, Michaela Teich, Andreas Huber, Andreas Kofler, Frank Perzl, Reinhard Fromm, Karl Kleemayr, and Jan-Thomas Fischer
Geosci. Model Dev., 15, 2423–2439,Short summary
The term gravitational mass flow (GMF) covers various natural hazard processes such as snow avalanches, rockfall, landslides, and debris flows. Here we present the open-source GMF simulation tool Flow-Py. The model equations are based on simple geometrical relations in three-dimensional terrain. We show that Flow-Py is an educational, innovative GMF simulation tool with three computational experiments: 1. validation of implementation, 2. performance, and 3. expandability.
Léo Pujol, Pierre-André Garambois, and Jérôme Monnier
This contribution presents a new numerical model for representing hydraulic-hydrological quantities at the basin scale. It allows modeling large areas at a low computational cost, with fine zooms where needed. It allows the integration of local and satellite measurements, via data assimilation methods, to improve the model's match to observations. Using this capability, good matches to in situ observations are obtained on a model of the complex Adour river network with fine zooms on floodplains.
Evan Baker, Anna B. Harper, Daniel Williamson, and Peter Challenor
Geosci. Model Dev., 15, 1913–1929,Short summary
We have adapted machine learning techniques to build a model of the land surface in Great Britain. The model was trained using data from a very complex land surface model called JULES. Our model is faster at producing simulations and predictions and can investigate many different scenarios, which can be used to improve our understanding of the climate and could also be used to help make local decisions.
Daichun Wang, Wei You, Zengliang Zang, Xiaobin Pan, Yiwen Hu, and Yanfei Liang
Geosci. Model Dev., 15, 1821–1840,Short summary
This paper presents a 3D variational data assimilation system for aerosol optical properties, including aerosol optical thickness (AOT) retrievals and lidar-based aerosol profiles, which was developed for a size-resolved sectional model in WRF-Chem. To directly assimilate aerosol optical properties, an observation operator based on the Mie scattering theory was designed. The results show that Himawari-8 AOT assimilation can significantly improve model aerosol analyses and forecasts.
Kevin Bulthuis and Eric Larour
Geosci. Model Dev., 15, 1195–1217,Short summary
We present and implement a stochastic solver to sample spatially and temporal varying uncertain input parameters in the Ice-sheet and Sea-level System Model, such as ice thickness or surface mass balance. We represent these sources of uncertainty using Gaussian random fields with Matérn covariance function. We generate random samples of this random field using an efficient computational approach based on solving a stochastic partial differential equation.
Urmas Raudsepp and Ilja Maljutenko
Geosci. Model Dev., 15, 535–551,Short summary
A model's ability to reproduce the state of a simulated object is always a subject of discussion. A new method for the multivariate assessment of numerical model skills uses the K-means algorithm for clustering model errors. All available data that fall into the model domain and simulation period are incorporated into the skill assessment. The clustered errors are used for spatial and temporal analysis of the model accuracy. The method can be applied to different types of geoscientific models.
Emmanuel Wyser, Yury Alkhimenkov, Michel Jaboyedoff, and Yury Y. Podladchikov
Geosci. Model Dev., 14, 7749–7774,Short summary
We propose an implementation of the material point method using graphical processing units (GPUs) to solve elastoplastic problems in three-dimensional configurations, such as the granular collapse or the slumping mechanics, i.e., landslide. The computational power of GPUs promotes fast code executions, compared to a traditional implementation using central processing units (CPUs). This allows us to study complex three-dimensional problems tackling high spatial resolution.
Rafael Lago, Thomas Gastine, Tilman Dannert, Markus Rampp, and Johannes Wicht
Geosci. Model Dev., 14, 7477–7495,Short summary
In this work we discuss a two-dimensional distributed parallelization of MagIC, an open-source code for the numerical solution of the magnetohydrodynamics equations. Such a parallelization involves several challenges concerning the distribution of work and data. We detail our algorithm and compare it with the established, optimized, one-dimensional distribution in the context of the dynamo benchmark and discuss the merits of both implementations.
Moritz Lange, Henri Suominen, Mona Kurppa, Leena Järvi, Emilia Oikarinen, Rafael Savvides, and Kai Puolamäki
Geosci. Model Dev., 14, 7411–7424,Short summary
This study aims to replicate computationally expensive high-resolution large-eddy simulations (LESs) with regression models to simulate urban air quality and pollutant dispersion. The model development, including feature selection, model training and cross-validation, and detection of concept drift, has been described in detail. Of the models applied, log-linear regression shows the best performance. A regression model can replace LES unless high accuracy is needed.
Hynek Bednář, Aleš Raidl, and Jiří Mikšovský
Geosci. Model Dev., 14, 7377–7389,Short summary
Forecast errors in numerical weather prediction systems grow in time. To quantify the impacts of this growth, parametric error growth models may be employed. This study recalculates and newly defines parameters for several statistic models approximating error growth in the ECMWF forecasting system. Accurate values of parameters are important because they are used to evaluate improvements of the forecasting systems or to estimate predictability.
Denise Degen, Cameron Spooner, Magdalena Scheck-Wenderoth, and Mauro Cacace
Geosci. Model Dev., 14, 7133–7153,Short summary
In times of worldwide energy transitions, an understanding of the subsurface is increasingly important to provide renewable energy sources such as geothermal energy. To validate our understanding of the subsurface we require data. However, the data are usually not distributed equally and introduce a potential misinterpretation of the subsurface. Therefore, in this study we investigate the influence of measurements on temperature distribution in the European Alps.
Geoffroy Kirstetter, Olivier Delestre, Pierre-Yves Lagrée, Stéphane Popinet, and Christophe Josserand
Geosci. Model Dev., 14, 7117–7132,Short summary
The development of forecasting tools may help to limit the impacts of flash floods. Our purpose here is to demonstrate the possibility of using b-flood, which is a 2D tool based on shallow-water equations and adaptive mesh refinement.
Sojung Park and Seon K. Park
Geosci. Model Dev., 14, 6241–6255,Short summary
One of the biggest uncertainties in numerical weather predictions (NWPs) comes from treating subgrid-scale physical processes. Physical processes, such as cumulus, microphysics, and planetary boundary layer processes, are parameterized in NWP models by empirical and theoretical backgrounds. We developed an interface between a micro-genetic algorithm and the WRF model for a combinatorial optimization of physics for heavy rainfall events in Korea. The system improved precipitation forecasts.
Olivier Pannekoucke and Philippe Arbogast
Geosci. Model Dev., 14, 5957–5976,Short summary
This contributes to research on uncertainty prediction, which is important either for determining the weather today or estimating the risk in prediction. The problem is that uncertainty prediction is numerically very expensive. An alternative has been proposed wherein uncertainty is presented in a simplified form with only the dynamics of certain parameters required. This tool allows for the determination of the symbolic equations of these parameter dynamics and their numerical computation.
Annika Günther, Johannes Gütschow, and Mairi Louise Jeffery
Geosci. Model Dev., 14, 5695–5730,Short summary
The mitigation components of the nationally determined contributions (NDCs) under the Paris Agreement are essential in our fight against climate change. Regular updates with increased ambition are requested to limit global warming to 1.5–2 °C. The new and easy-to-update open-source tool NDCmitiQ can be used to quantify the NDCs' mitigation targets and construct resulting emissions pathways. In use cases, we show target uncertainties from missing clarity, data, and methodological challenges.
Futo Tomizawa and Yohei Sawada
Geosci. Model Dev., 14, 5623–5635,Short summary
A new method to predict chaotic systems from observation and process-based models is proposed by combining machine learning with data assimilation. Our method is robust to the sparsity of observation networks and can predict more accurately than a process-based model when it is biased. Our method effectively works when both observations and models are imperfect, which is often the case in geoscience. Therefore, our method is useful to solve a wide variety of prediction problems in this field.
Chloe Leach, Tom Coulthard, Andrew Barkwith, Daniel R. Parsons, and Susan Manson
Geosci. Model Dev., 14, 5507–5523,Short summary
Numerical models can be used to understand how coastal systems evolve over time, including likely responses to climate change. However, many existing models are aimed at simulating 10- to 100-year time periods do not represent a vertical dimension and are thus unable to include the effect of sea-level rise. The Coastline Evolution Model 2D (CEM2D) presented in this paper is an advance in this field, with the inclusion of the vertical coastal profile against which the water level can be altered.
Steven J. Phipps, Jason L. Roberts, and Matt A. King
Geosci. Model Dev., 14, 5107–5124,Short summary
Simplified schemes, known as parameterisations, are sometimes used to describe physical processes within numerical models. However, the values of the parameters are uncertain. This introduces uncertainty into the model outputs. We develop a simple approach to identify plausible ranges for model parameters. Using a model of the Antarctic Ice Sheet, we find that the value of one parameter can depend on the values of others. We conclude that a single optimal set of parameter values does not exist.
Axel Peytavin, Bruno Sainte-Rose, Gael Forget, and Jean-Michel Campin
Geosci. Model Dev., 14, 4769–4780,Short summary
We present a new algorithm developed at The Ocean Cleanup to update ocean plastic models based on measurements from the field to improve future cleaning operations. Prepared in collaboration with MIT researchers, this initial study presents its use in several analytical and real test cases in which two observers in a flow field record regular observations to update a plastic forecast. We demonstrate this improves the prediction, even with inaccurate knowledge of the water flows driving plastic.
Kang Pan, Mei Qi Lim, Markus Kraft, and Epaminondas Mastorakos
Geosci. Model Dev., 14, 4509–4534,Short summary
A new moving point source (MPS) model was developed to simulate the dispersion of emissions generated by the moving ships. Compared to the commonly used line source (LS) or fixed point source (FPS) model, the MPS model provides more emission distribution details generated by the moving ships and matches reasonably with the measurements. Therefore, the MPS model should be a valuable alternative for the environmental society to evaluate the pollutant dispersion contributed from the moving ships.
Sebastian Springer, Heikki Haario, Jouni Susiluoto, Aleksandr Bibov, Andrew Davis, and Youssef Marzouk
Geosci. Model Dev., 14, 4319–4333,Short summary
Model predictions always contain uncertainty. But in some cases, such as weather forecasting or climate modeling, chaotic unpredictability increases the difficulty to say exactly how much uncertainty there is. We combine two recently proposed mathematical methods to show how the uncertainty can be analyzed in models that are simplifications of true weather models. The results can be extended in the future to show how forecasts from large-scale models can be improved.
Alexander Schaaf, Miguel de la Varga, Florian Wellmann, and Clare E. Bond
Geosci. Model Dev., 14, 3899–3913,Short summary
Uncertainty is an inherent property of any model of the subsurface. We show how geological topology information – how different regions of rocks in the subsurface are connected – can be used to train uncertain geological models to reduce uncertainty. More widely, the method demonstrates the use of probabilistic machine learning (Bayesian inference) to train structural geological models on auxiliary geological knowledge that can be encoded in graph structures.
Matthew Ozon, Aku Seppänen, Jari P. Kaipio, and Kari E. J. Lehtinen
Geosci. Model Dev., 14, 3715–3739,Short summary
Experimental research has provided large amounts of high-quality data on aerosol over the last 2 decades. However, inference of the process rates (e.g., the rates at which particles are generated) is still typically done by simple curve-fitting methods and does not assess the credibility of the estimation. The devised method takes advantage of the Bayesian framework to not only retrieve the state of the observed aerosol system but also to estimate the process rates (e.g., growth rate).
Colton J. Conroy and Einat Lev
Geosci. Model Dev., 14, 3553–3575,Short summary
Lava flows present a natural hazard to communities around volcanoes and are usually slow-moving (< 1-5 cm/s). Lava flows during the 2018 eruption of Kilauea volcano, Hawai’i, however, reached speeds as high as 11 m/s. To investigate these dynamics we develop a new lava flow computer model that incorporates a nonlinear expression for the fluid viscosity. Model results indicate that the lava flows at Site 8 of the eruption displayed shear thickening behavior due to the flow's high bubble content.
Antti Hellsten, Klaus Ketelsen, Matthias Sühring, Mikko Auvinen, Björn Maronga, Christoph Knigge, Fotios Barmpas, Georgios Tsegas, Nicolas Moussiopoulos, and Siegfried Raasch
Geosci. Model Dev., 14, 3185–3214,Short summary
Large-eddy simulation (LES) of the urban atmospheric boundary layer involves a large separation of turbulent scales, leading to prohibitive computational costs. An online LES–LES nesting scheme is implemented into the PALM model system 6.0 to overcome this problem. Test results show that the accuracy within the high-resolution nest domains approach the non-nested high-resolution reference results. The nesting can reduce the CPU by time up to 80 % compared to the fine-resolution reference runs.
Yumeng Chen, Konrad Simon, and Jörn Behrens
Geosci. Model Dev., 14, 2289–2316,Short summary
Mesh adaptivity can reduce overall model error by only refining meshes in specific areas where it us necessary in the runtime. Here we suggest a way to integrate mesh adaptivity into an existing Earth system model, ECHAM6, without having to redesign the implementation from scratch. We show that while the additional computational effort is manageable, the error can be reduced compared to a low-resolution standard model using an idealized test and relatively realistic dust transport tests.
Sylvain Mailler, Romain Pennel, Laurent Menut, and Mathieu Lachâtre
Geosci. Model Dev., 14, 2221–2233,Short summary
Representing the advection of thin polluted plumes in numerical models is a challenging task since these models usually tend to excessively diffuse these plumes in the vertical direction. This numerical diffusion process is the cause of major difficulties in representing such dense and thin polluted plumes in numerical models. We propose here, and test in an academic framework, a novel method to solve this problem through the use of an antidiffusive advection scheme in the vertical direction.
Daniel Otoo and David Hodgetts
Geosci. Model Dev., 14, 2075–2095,Short summary
The forward stratigraphic simulation method is used to predict lithofacies, porosity, and permeability in a reservoir model. The objective of using this approach is to enhance subsurface property modelling through geologic realistic 3-D stratigraphic patterns. Results show realistic stratigraphic sequences. Given this, we can derive spatial and geometric data as secondary data to constrain property simulation in a reservoir model. The approach can reduce the uncertainty of property modelling.
Denise Degen and Mauro Cacace
Geosci. Model Dev., 14, 1699–1719,Short summary
In this work, we focus on improving the understanding of subsurface processes with respect to interactions with climate dynamics. We present advanced, open-source mathematical methods that enable us to investigate the influence of various model properties on the final outcomes. By relying on our approach, we have been able to showcase their importance in improving our understanding of the subsurface and highlighting the current shortcomings of currently adopted models.
Meriem Krouma, Pascal Yiou, Céline Déandreis, and Soulivanh Thao
Geosci. Model Dev. Discuss.,
Revised manuscript accepted for GMDShort summary
We aim to evaluate the skill of a stochastic weather generator (SWG) to forecast precipitation at different time scales and in different areas of Western Europe from analogs of Z500 hPa. SWG has skill to simulate precipitation for 5 and 10 days. We found that forecast weaknesses can be associated with specific weather patterns. Comparing with ECMWF forecasts confirms the skill of our model. This work is important because it provides information about weather forecasts over specific areas.
Geosci. Model Dev., 14, 377–389,Short summary
Resetting of non-significant figures (precision trimming) enables efficient data compression and helps to avoid excessive use of storage space and network bandwidth while having well-constrained distortion to the data. The paper analyses accuracy losses and artifacts caused by trimming methods and by the widely used linear packing method. The paper presents several methods with implementation, evaluation, and illustrations and includes subroutines directly usable in geoscientific models.
Bertrand Bessagnet, Laurent Menut, and Maxime Beauchamp
Geosci. Model Dev., 14, 91–106,Short summary
This paper presents a new interpolator useful for geophysics applications. It can explore N-dimensional meshes, grids or look-up tables. The code accepts irregular but structured grids. Written in Fortran, it is easy to implement in existing codes and very fast and portable. We have compared it with a Python library. Python is convenient but suffers from portability and is sometimes not optimized enough. As an application case, this method is applied to atmospheric sciences.
Oksana Guba, Mark A. Taylor, Andrew M. Bradley, Peter A. Bosler, and Andrew Steyer
Geosci. Model Dev., 13, 6467–6480,
Anna Wirbel and Alexander Helmut Jarosch
Geosci. Model Dev., 13, 6425–6445,Short summary
We present an open-source numerical tool to simulate the free-surface evolution of gravity-driven flows (e.g. glaciers) constrained by bed topography. No ad hoc post-processing is required to enforce positive ice thickness and mass conservation. We utilise finite elements, define benchmark tests, and showcase glaciological examples. In addition, we provide a thorough analysis of the applicability and robustness of different spatial stabilisation and time discretisation methods.
Emmanuel Wyser, Yury Alkhimenkov, Michel Jaboyedoff, and Yury Y. Podladchikov
Geosci. Model Dev., 13, 6265–6284,Short summary
In this work, we present an efficient and fast material point method (MPM) implementation in MATLAB. We first discuss the vectorization strategies to adapt this numerical method to a MATLAB implementation. We report excellent agreement of the solver compared with classical analysis among the MPM community, such as the cantilever beam problem. The solver achieves a performance gain of 28 compared with a classical iterative implementation.
Lauri Tuppi, Pirkka Ollinaho, Madeleine Ekblom, Vladimir Shemyakin, and Heikki Järvinen
Geosci. Model Dev., 13, 5799–5812,Short summary
This paper presents general guidelines on how to utilise computer algorithms efficiently in order to tune weather models so that they would produce better forecasts. The main conclusions are that the computer algorithms work most efficiently with a suitable cost function, certain forecast length and ensemble size. We expect that our results will facilitate the use of algorithmic methods in the tuning of weather models.
Tarandeep S. Kalra, Neil K. Ganju, and Jeremy M. Testa
Geosci. Model Dev., 13, 5211–5228,Short summary
The paper covers the description of a 3-D open-source model that dynamically couples the biophysical interactions between submerged aquatic vegetation (SAV), hydrodynamics (currents, waves), sediment dynamics, and nutrient loading. Based on SAV growth model, SAV can use growth or dieback while contributing and sequestering nutrients from the water column (modifying the biological environment) and subsequently affect the hydrodynamics and sediment transport (modifying the physical environment).
Xiaoshuang Li, Richard Garth James Bellerby, Jianzhong Ge, Philip Wallhead, Jing Liu, and Anqiang Yang
Geosci. Model Dev., 13, 5103–5117,Short summary
We have developed an ANN model to predict pH using 11 cruise datasets from 2013 to 2017, demonstrated its reliability using three cruise datasets during 2018 and applied it to retrieve monthly pH for the period 2000 to 2016 on the East China Sea shelf using the ANN model in combination with input variables from the Changjiang biology Finite-Volume Coastal Ocean Model. This approach may be a valuable tool for understanding the seasonal variation of pH in poorly observed regions.
Christopher Subich, Pierre Pellerin, Gregory Smith, and Frederic Dupont
Geosci. Model Dev., 13, 4379–4398,Short summary
This work presents a semi-Lagrangian advection module for the NEMO (OPA) ocean model. Semi-Lagrangian advection transports fluid properties (temperature, salinity, velocity) between time steps by following fluid motion and interpolating from upstream locations of fluid parcels. This method is commonly used in atmospheric models to extend time step size, but it has not previously been applied to operational ocean models. Overcoming this required a new approach for solid boundaries (coastlines).
Jouni Susiluoto, Alessio Spantini, Heikki Haario, Teemu Härkönen, and Youssef Marzouk
Geosci. Model Dev., 13, 3439–3463,Short summary
We describe a new computer program that is able produce maps of carbon dioxide or other quantities based on data collected by satellites that orbit the Earth. When working with such data there is often too much data in one area and none in another. The program is able to describe the fields even when data is not available. To be able to do so, new computational methods were developed. The program is also able to describe how uncertain the estimated carbon dioxide or other fields are.
Olivier Pannekoucke and Ronan Fablet
Geosci. Model Dev., 13, 3373–3382,Short summary
Learning physics from data using a deep neural network is a challenge that requires an appropriate but unknown network architecture. The package introduced here helps to design an architecture by translating known physical equations into a network, which the experimenter completes to capture unknown physical processes. A test bed is introduced to illustrate how this learning allows us to focus on truly unknown physical processes in the hope of making better use of data and digital resources.
Dmitry Sidorenko, Sergey Danilov, Nikolay Koldunov, Patrick Scholz, and Qiang Wang
Geosci. Model Dev., 13, 3337–3345,Short summary
Computation of barotropic and meridional overturning streamfunctions for models formulated on unstructured meshes is commonly preceded by interpolation to a regular mesh. This operation destroys the original conservation, which can be then be artificially imposed to make the computation possible. An elementary method is proposed that avoids interpolation and preserves conservation in a strict model sense.
Xavier Couvelard, Florian Lemarié, Guillaume Samson, Jean-Luc Redelsperger, Fabrice Ardhuin, Rachid Benshila, and Gurvan Madec
Geosci. Model Dev., 13, 3067–3090,Short summary
Within the framework of the Copernicus Marine Environment Monitoring Service (CMEMS), an objective is to demonstrate the contribution of coupling the high-resolution analysis and forecasting system with a wave model. This study describes the necessary steps and discusses the various choices made for coupling a wave model and an oceanic model for global-scale applications.
Akella, S. and Navon, I.: Different approaches to model error formulation in 4D-Var: A study with high-resolution advection schemes, Tellus A, 61, 112–128, 2009. a
Bauer, H.-S., Schwitalla, T., Wulfmeyer, V., Bakhshaii, A., Ehret, U., Neuper, M., and Caumont, O.: Quantitative precipitation estimation based on high-resolution numerical weather prediction and data assimilation with WRF – a performance test, Tellus A, 67, 25047, https://doi.org/10.3402/tellusa.v67.25047, 2015. a
Buehner, M.: Ensemble-derived stationary and flow-dependent background-error covariances: Evaluation in a quasi-operational NWP setting, Q. J. Roy. Meteor. Soc., 131, 1013–1043, 2005. a
Cardinali, C., Žagar, N., Radnoti, G., and Buizza, R.: Representing model error in ensemble data assimilation, Nonlinear Proc. Geophys., 21, 971–985, 2014. a
Carmichael, G. R., Sandu, A., Chai, T., Daescu, D. N., Constantinescu, E. M., and Tang, Y.: Predicting air quality: Improvements through advanced methods to integrate models and measurements, J. Comput. Phys., 227, 3540–3571, 2008. a
Chatterjee, A.: An introduction to the proper orthogonal decomposition, Current Science, 78, 808–817, 2000. a
Errico, R. M.: What is an adjoint model?, B. Am. Meteorol. Soc., 78, 2577–2592, 1997. a
Errico, R. M. and Raeder, K. D.: An examination of the accuracy of the linearization of a mesoscale model with moist physics, Q. J. R. Meteor. Soc., 125, 169–195, 1999. a
Errico, R. M., Vukicevic, T., and Raeder, K.: Examination of the accuracy of a tangent linear model, Tellus A, 45, 462–477, 1993. a
Frerix, T., Kochkov, D., Smith, J. A., Cremers, D., Brenner, M. P., and Hoyer, S.: Variational Data Assimilation with a Learned Inverse Observation Operator, in: Proceedings of the 38th International Conference on Machine Learning (ICML), Proceedings of Machine Learning Research (PMLR), 139, 3449–3458, https://proceedings.mlr.press/v139/frerix21a.html (last access: 27 April 2022), 2021. a
Glimm, J., Hou, S., Lee, Y., Sharp, D., and Ye, K.: Sources of uncertainty and error in the simulation of flow in porous media, Comput. Appl. Math., 23, 109–120, 2004. a
Gustafsson, N., Janjić, T., Schraff, C., Leuenberger, D., Weissmann, M., Reich, H., Brousseau, P., Montmerle, T., Wattrelot, E., Bučánek, A., Mile, M., Hamdi, R., Lindskog, M., Barkmeijer, J., Dahlbom, M., Macpherson, B., Ballard, S., Inverarity, G., Carley, J., Alexander, C., Dowell, D., Liu, S., Ikuta, Y., and Fujita, T.: Survey of data assimilation methods for convective-scale numerical weather prediction at operational centres, Q. J. R. Meteor. Soc., 144, 1218–1256, https://doi.org/10.1002/qj.3179, 2018. a
Hansen, J. A.: Accounting for model error in ensemble-based state estimation and forecasting, Mon. Weather Rev., 130, 2373–2391, 2002. a
Hochreiter, S. and Schmidhuber, J.: Long short-term memory, Neural computation, 9, 1735–1780, 1997. a
Holmes, P., Lumley, J. L., Berkooz, G., and Rowley, C. W.: Turbulence, Coherent Structures, Dynamical Systems and Symmetry, Cambridge University Press, p. 386, ISBN 9781107008250, 2012. a
Le Dimet, F. and Talagrand, O.: Variational algorithms for analysis and assimilation of meteorological observations: theoretical aspects, Tellus A, 38, 97–110, 1986. a
Le Guen, V. and Thome, N.: Disentangling physical dynamics from unknown factors for unsupervised video prediction, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 13–19 June 2020, Seattle, WA, USA, 11474–11484, https://doi.org/10.1109/CVPR42600.2020.01149, 2020. a
Lorenc, A. C. and Rawlins, F.: Why does 4D-Var beat 3D-Var?, Quarterly J. Roy. Meteorol. Soc., 131, 3247–3257, 2005. a
Lynch, P.: The origins of computer weather prediction and climate modeling, J. Comput. Phys., 227, 3431–3444, 2008. a
Maulik, R., Egele, R., Lusch, B., and Balaprakash, P.: Recurrent neural network architecture search for geophysical emulation, in: SC20: International Conference for High Performance Computing, Networking, Storage and Analysis, Atlanta, Georgia, IEEE, p. 14, ISBN 9781728199986, 2020. a
Moritz, P., Nishihara, R., Wang, S., Tumanov, A., Liaw, R., Liang, E., Elibol, M., Yang, Z., Paul, W., Jordan, M. I., and Stoica, I.: Ray: A distributed framework for emerging AI applications, in: 13th USENIX Symposium on Operating Systems Design and Implementation, 561–577, ISBN 9781931971478, 2018. a
Palmer, T., Shutts, G., Hagedorn, R., Doblas-Reyes, F., Jung, T., and Leutbecher, M.: Representing model uncertainty in weather and climate prediction, Annu. Rev. Earth Planet. Sci, 33, 163–93, 2005. a
Penny, S. G., Smith, T. A., Chen, T.-C., Platt, J. A., Lin, H.-Y., Goodliff, M., and Abarbanel, H. D. I.: Integrating recurrent neural networks with data assimilation for scalable data-driven state estimation, arXiv preprint, arXiv:2109.12269, 14, e2021MS002843, https://doi.org/10.1029/2021MS002843, 2021. a, b
Rao, V. and Sandu, A.: A posteriori error estimates for the solution of variational inverse problems, SIAM/ASA, Journal on Uncertainty Quantification, 3, 737–761, 2015. a
Reidmiller, D., Avery, C., Easterling, D., Kunkel, K., Lewis, K., Maycock, T., and Stewart, B.: Fourth national climate assessment, Volume II: Impacts, Risks, and Adaptation in the United States, U.S. Global Change Research Program, Washington, DC, USA, 1515 pp., https://doi.org/10.7930/NCA4.2018, 2018. a
Sandu, A. and Chai, T.: Chemical data assimilation – An overview, Atmosphere, 2, 426–463, 2011. a
Sandu, A., Daescu, D. N., Carmichael, G. R., and Chai, T.: Adjoint sensitivity analysis of regional air quality models, J. Comput. Phys., 204, 222–252, 2005. a
Schmidt, O. T., Mengaldo, G., Balsamo, G., and Wedi, N. P.: Spectral empirical orthogonal function analysis of weather and climate data, Mon. Weather Rev., 147, 2979–2995, 2019. a
Wang, J. and Kotamarthi, V. R.: Downscaling with a nested regional climate model in near-surface fields over the contiguous United States, J. Geophys. Res.-Atmos., 119, 8778–8797, 2014. a
Zupanski, D. and Zupanski, M.: Model error estimation employing an ensemble data assimilation approach, Mon. Weather Rev., 134, 1337–1354, 2006. a
The requested paper has a corresponding corrigendum published. Please read the corrigendum first before downloading the article.
In numerical weather prediction, data assimilation is frequently utilized to enhance the accuracy of forecasts from equation-based models. In this work we use a machine learning framework that approximates a complex dynamical system given by the geopotential height. Instead of using an equation-based model, we utilize this machine-learned alternative to dramatically accelerate both the forecast and the assimilation of data, thereby reducing need for large computational resources.
In numerical weather prediction, data assimilation is frequently utilized to enhance the...