Amirpasha Mozaffari

Amirpasha Mozaffari

Auto-generated profile, based on publicly available information

Amirpasha Mozaffari

Jülich Supercomputing Centre, Forschungszentrum Jülich, 52425 Jülich, Germany

Research Focus & Metrics

Academic Career Length

12Years (2015-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
15 February 2021 in Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences
Verified
Journal Article
Open access
MSMartin Schultz
Martin Schultz
CBClara Betancourt
Clara Betancourt
BGBing Gong
Bing Gong
FKF. Kleinert
F. Kleinert
MLM. Langguth
M. Langguth
SSScarlet Stadtler
Scarlet Stadtler
The recent hype about artificial intelligence has sparked renewed interest in applying the successful deep learning (DL) methods for image recognition, speech recognition, robotics, strategic games and other application areas to the field of meteorology. There is some evidence that better weather forecasts can be produced by introducing big data mining and neural networks into the weather prediction workflow. Here, we discuss the question of whether it is possible to completely replace the current numerical weather models and data assimilation systems with DL approaches. This discussion entails a review of state-of-the-art machine learning concepts and their applicability to weather data with its pertinent statistical properties. We think that it is not inconceivable that numerical weather models may one day become obsolete, but a number of fundamental breakthroughs are needed before this goal comes into reach. This article is part of the theme issue ‘Machine learning for weather and climate modelling’.

13 December 2022 in Geoscientific Model Development
Verified
Journal Article
Open access
PDF file
BGBing Gong
Bing Gong
MLMichael Langguth
Michael Langguth
YJYan Ji
Yan Ji
AMAmirpasha Mozaffari
Amirpasha Mozaffari
SSScarlet Stadtler
Scarlet Stadtler
MSMartin Schultz
Martin Schultz
Numerical weather prediction (NWP) models solve a system of partial differential equations based on physical laws to forecast the future state of the atmosphere. These models are deployed operationally, but they are computationally very expensive. Recently, the potential of deep neural networks to generate bespoke weather forecasts has been explored in a couple of scientific studies inspired by the success of video frame prediction models in computer vision. In this study, a simple recurrent neural network with convolutional filters, called ConvLSTM, and an advanced generative network, the Stochastic Adversarial Video Prediction (SAVP) model, are applied to create hourly forecasts of the 2 m temperature for the next 12 h over Europe. We make use of 13 years of data from the ERA5 reanalysis, of which 11 years are utilized for training and 1 year each is used for validating and testing. We choose the 2 m temperature, total cloud cover, and the 850 hPa temperature as predictors and show that both models attain predictive skill by outperforming persistence forecasts. SAVP is superior to ConvLSTM in terms of several evaluation metrics, confirming previous results from computer vision that larger, more complex networks are better suited to learn complex features and to generate better predictions. The 12 h forecasts of SAVP attain a mean squared error (MSE) of about 2.3 K2, an anomaly correlation coefficient (ACC) larger than 0.85, a structural similarity index (SSIM) of around 0.72, and a gradient ratio (rG) of about 0.82. The ConvLSTM yields a higher MSE (3.6 K2), a smaller ACC (0.80) and SSIM (0.65), and a slightly larger rG (0.84). The superior performance of SAVP in terms of MSE, ACC, and SSIM can be largely attributed to the generator. A sensitivity study shows that a larger weight of the generative adversarial network (GAN) component in the SAVP loss leads to even better preservation of spatial variability at the cost of a somewhat increased MSE (2.5 K2). Including the 850 hPa temperature as an additional predictor enhances the forecast quality, and the model also benefits from a larger spatial domain. By contrast, adding the total cloud cover as predictor or reducing the amount of training data to 8 years has only small effects. Although the temperature forecasts obtained in this way are still less powerful than contemporary NWP models, this study demonstrates that sophisticated deep neural networks may achieve considerable forecast quality beyond the nowcasting range in a purely data-driven way.

23 May 2023 in Geoscientific Model Development
Verified
Journal Article
Open access
PDF file
YJYan Ji
Yan Ji
BGBing Gong
Bing Gong
MLMichael Langguth
Michael Langguth
AMAmirpasha Mozaffari
Amirpasha Mozaffari
XZXiefei Zhi
Xiefei Zhi
The prediction of precipitation patterns up to 2 h ahead, also known as precipitation nowcasting, at high spatiotemporal resolutions is of great relevance in weather-dependent decision-making and early warning systems. In this study, we are aiming to provide an efficient and easy-to-understand deep neural network – CLGAN (convolutional long short-term memory generative adversarial network) – to improve the nowcasting skills of heavy precipitation events. The model constitutes a generative adversarial network (GAN) architecture, whose generator is built upon a u-shaped encoder–decoder network (U-Net) and is equipped with recurrent long short-term memory (LSTM) cells to capture spatiotemporal features. The optical flow model DenseRotation and the competitive video prediction models ConvLSTM (convolutional LSTM) and PredRNN-v2 (predictive recurrent neural network version 2) are used as the competitors. A series of evaluation metrics, including the root mean square error, the critical success index, the fractions skill score, and object-based diagnostic evaluation, are utilized for a comprehensive comparison against competing baseline models. We show that CLGAN outperforms the competitors in terms of scores for dichotomous events and object-based diagnostics. A sensitivity analysis on the weight of the GAN component indicates that the GAN-based architecture helps to capture heavy precipitation events. The results encourage future work based on the proposed CLGAN architecture to improve the precipitation nowcasting and early warning systems.

28 April 2020 in Geophysics
Verified
Journal Article
Closed access
AMAmirpasha Mozaffari
Amirpasha Mozaffari
AKAnja Klotzsche
Anja Klotzsche
CWCraig Warren
Craig Warren
GHGuowei He
Guowei He
AGAntonios Giannopoulos
Antonios Giannopoulos
JKJan van der Kruk
Jan van der Kruk
Full-waveform inversion (FWI) of cross-borehole ground-penetrating radar (GPR) data is a technique with the potential to investigate subsurface structures. Typical FWI applications transform 3D measurements into a 2D domain via an asymptotic 3D to 2D data transformation, widely known as a Bleistein filter. Despite the broad use of such a transformation, it requires some assumptions that make it prone to errors. Although the existence of the errors is known, previous studies have failed to quantify the inaccuracies introduced on permittivity and electrical conductivity estimation. Based on a comparison of 3D and 2D modeling, errors could reach up to 30% of the original amplitudes in layered structures with high-contrast zones. These inaccuracies can significantly affect the performance of crosshole GPR FWI in estimating permittivity and especially electrical conductivity. We have addressed these potential inaccuracies by introducing a novel 2.5D crosshole GPR FWI that uses a 3D finite-difference time-domain forward solver (gprMax3D). This allows us to model GPR data in 3D, whereas carrying out FWI in the 2D plane. Synthetic results showed that 2.5D crosshole GPR FWI outperformed 2D FWI by achieving higher resolution and lower average errors for permittivity and conductivity models. The average model errors in the whole domain were reduced by approximately 2% for permittivity and conductivity, whereas zone-specific errors in high-contrast layers were reduced by approximately 20%. We verified our approach using crosshole 2.5D FWI measured data, and the results showed good agreement with previous 2D FWI results and geologic studies. Moreover, we analyzed various approaches and found an adequate trade-off between computational complexity and accuracy of the results, i.e., reducing the computational effort while maintaining the superior performance of our 2.5D FWI scheme.

1 April 2022 in Data Intelligence
Verified
Journal Article
Open access
PDF file
AMAmirpasha Mozaffari
Amirpasha Mozaffari
MLMichael Langguth
Michael Langguth
BGBing Gong
Bing Gong
JAJessica Ahring
Jessica Ahring
ACAdrian Rojas Campos
Adrian Rojas Campos
MSMartin G. Schultz
Martin G. Schultz
Machine learning (ML) applications in weather and climate are gaining momentum as big data and the immense increase in High-performance computing (HPC) power are paving the way. Ensuring FAIR data and reproducible ML practices are significant challenges for Earth system researchers. Even though the FAIR principle is well known to many scientists, research communities are slow to adopt them. Canonical Workflow Framework for Research (CWFR) provides a platform to ensure the FAIRness and reproducibility of these practices without overwhelming researchers. This conceptual paper envisions a holistic CWFR approach towards ML applications in weather and climate, focusing on HPC and big data. Specifically, we discuss Fair Digital Object (FDO) and Research Object (RO) in the DeepRain project to achieve granular reproducibility. DeepRain is a project that aims to improve precipitation forecast in Germany by using ML. Our concept envisages the raster datacube to provide data harmonization and fast and scalable data access. We suggest the Juypter notebook as a single reproducible experiment. In addition, we envision JuypterHub as a scalable and distributed central platform that connects all these elements and the HPC resources to the researchers via an easy-to-use graphical interface.

1 June 2016
Verified
Conference Paper
Closed access
AMA. Mozaffari
A. Mozaffari
AKA. Klotzsche
A. Klotzsche
GHG. He
G. He
HVHarry Vereecken
Harry Vereecken
JKJ. van der Kruk
J. van der Kruk
AGA. Giannopoulos
A. Giannopoulos
2D crosshole ground penetrating (GPR) full-1 waveform inversion (FWI) has shown superior resolution 1 compared to ray-based inversion tomograms for synthetic and experimental data. To invert measured 3D data with a 2D j model that has a reduced geometrical spreading and assumes infinite source and receiver dimensions perpendicular to the 2D inversion plane, the Bleistein transformation can be used to convert the measured 3D data to 2D. This far-field conversion consists of a phase shift and amplitude correction that is based on the first arrival travel time of each trace. In the case of late arrival and high amplitude events that 1 can occur due to low-velocity waveguides andfast propagating refracted waves, this transformation can introduce errors especially in the amplitude such that the inverted conductivity is less accurate. To overcome these problems, we have replaced the 2D finite difference time domain FDTD forward model in the FWI scheme with the well-known gprMax3D FDTD modeling program. In this way, we do not need to use the Bleistein 3D-to-2D filter with its far-field approximation, and we can deal with the correct geometrical spreading and approximate better realistic point source and receivers. Currently, the 2D FWI algorithm has been extended to 2.5D by replacing the 2D FDTD with the gprMax 3D FDTD modeling program. The first test of the 2.5D FWI is based on inversion results of a data set acquired at the Widen site in Switzerland that contained a low-velocity waveguide. The new 2.5D FWI performed well in reconstructing the models. The new 2.5D FWI enables a more reliable re-construction of the subsurface image, especially the electrical conductivity tomograms and to the full use of all the modeling possibilities of the gprMax modeling tool.