On the potential of a neural network-based approach for estimating XCO2 from OCO-2 measurements

Abstract. In David et al (2021), we introduced a neural network (NN) approach for estimating the column-averaged dry air mole fraction of CO2 (XCO2) and the surface pressure from the reflected solar spectra acquired by the OCO-2 instrument. The results indicated great potential for the technique as the comparison against both model estimates and independent TCCON measurements showed an accuracy and precision similar or better than that of the operational ACOS (NASA’s Atmospheric CO2 Observations from Space retrievals – ACOS) algorithm. Yet, subsequent analysis showed that the neural network estimate often mimics the training dataset and is unable to retrieve small scale features such as CO2 plumes from industrial sites. Importantly, we found that, with the same inputs as those used to estimate XCO2 and surface pressure, the NN technique is able to estimate latitude and date with unexpected skill, i.e. with an error whose standard deviation is only 7° and 61 days, respectively. The information about the date mainly comes from the weak CO2 band, that is influenced by the well-mixed and increasing concentrations of CO2 in the stratosphere. The availability of such information in the measured spectrum may therefore allow the NN to exploit it rather than the direct CO2 imprint in the spectrum, to estimate XCO2. Thus, our first version of the NN performed well mostly because the XCO2 fields used for the training were remarkably accurate, but it did not bring any added value. Further to this analysis, we designed a second version of the NN, excluding the weak CO2 band from the input. This new version has a different behaviour as it does retrieve XCO2 enhancements downwind of emission hotspots, i.e. a feature that is not in the training dataset. The comparison against the reference Total Carbon Column Observing Network (TCCON) and the surface-air-sample-driven inversion of the Copernicus Atmosphere Monitoring Service (CAMS) remains very good, as in the first version of the NN. In addition, the difference with the CAMS model (also called innovation in a data assimilation context) for NASA Atmospheric CO2 Observations from Space (ACOS) and the NN estimates are significantly correlated. These results confirm the potential of the NN approach for an operational processing of satellite observations aiming at the monitoring of CO2 concentrations and fluxes.

Download Full-text

Using neural networks to improve simulations in the gray zone

10.5194/npg-2021-20 ◽

2021 ◽

Author(s):

Raphael Kriegmair ◽

Yvonne Ruckstuhl ◽

Stephan Rasp ◽

George Craig

Keyword(s):

Neural Network ◽

Initial Conditions ◽

Water Model ◽

Training Dataset ◽

Gray Zone ◽

Low Resolution ◽

Time Step ◽

The Neural Network ◽

Highly Nonlinear ◽

The Difference

Abstract. Machine learning represents a potential method to cope with the gray zone problem of representing motions in dynamical systems on scales comparable to the model resolution. Here we explore the possibility of using a neural network to directly learn the error caused by unresolved scales. We use a modified shallow water model which includes highly nonlinear processes mimicking atmospheric convection. To create the training dataset we run the model in a high and a low-resolution setup and compare the difference after one low resolution time step starting from the same initial conditions, thereby obtaining an exact target. The neural network is able to learn a large portion of the difference when evaluated offline on a validation set. When coupled to the low-resolution model, we find large forecast improvements up to one day on average. After this, the accumulated error due to the mass conservation violation of the neural network starts to dominate and deteriorates the forecast. This deterioration can effectively be delayed by adding a penalty term to the loss function used to train the ANN to conserve mass in a weak sense. This study reinforces the need to include physical constraints in neural network parameterizations.

Download Full-text

Fully Automated DCNN-Based Thermal Images Annotation Using Neural Network Pretrained on RGB Data

Sensors ◽

10.3390/s21041552 ◽

2021 ◽

Vol 21 (4) ◽

pp. 1552

Author(s):

Adam Ligocki ◽

Ales Jelinek ◽

Ludek Zalud ◽

Esa Rahtu

Keyword(s):

Neural Network ◽

Large Scale ◽

Thermal Image ◽

Training Dataset ◽

Small Scale ◽

Thermal Images ◽

Data Annotation ◽

The Neural Network ◽

Improved Performance ◽

Image Datasets

One of the biggest challenges of training deep neural network is the need for massive data annotation. To train the neural network for object detection, millions of annotated training images are required. However, currently, there are no large-scale thermal image datasets that could be used to train the state of the art neural networks, while voluminous RGB image datasets are available. This paper presents a method that allows to create hundreds of thousands of annotated thermal images using the RGB pre-trained object detector. A dataset created in this way can be used to train object detectors with improved performance. The main gain of this work is the novel method for fully automatic thermal image labeling. The proposed system uses the RGB camera, thermal camera, 3D LiDAR, and the pre-trained neural network that detects objects in the RGB domain. Using this setup, it is possible to run the fully automated process that annotates the thermal images and creates the automatically annotated thermal training dataset. As the result, we created a dataset containing hundreds of thousands of annotated objects. This approach allows to train deep learning models with similar performance as the common human-annotation-based methods do. This paper also proposes several improvements to fine-tune the results with minimal human intervention. Finally, the evaluation of the proposed solution shows that the method gives significantly better results than training the neural network with standard small-scale hand-annotated thermal image datasets.

Download Full-text

A device employing a neural network for blood pressure estimation from the oscillatory pressure pulse wave and PPG signal

Sensor Review ◽

10.1108/sr-09-2020-0216 ◽

2021 ◽

Vol 41 (1) ◽

pp. 74-86

Author(s):

Jian Tian ◽

Jiangan Xie ◽

Zhonghua He ◽

Qianfeng Ma ◽

Xiuxin Wang

Keyword(s):

Neural Network ◽

Blood Pressure ◽

Diastolic Blood Pressure ◽

Pressure Pulse ◽

Pulse Wave ◽

Content Type ◽

The Neural Network ◽

The Difference ◽

Oscillatory Pressure ◽

Mercury Sphygmomanometer

Purpose Wrist-cuff oscillometric blood pressure monitors are very popular in the portable medical device market. However, its accuracy has always been controversial. In addition to the oscillatory pressure pulse wave, the finger photoplethysmography (PPG) can provide information on blood pressure changes. A blood pressure measurement system integrating the information of pressure pulse wave and the finger PPG may improve measurement accuracy. Additionally, a neural network can synthesize the information of different types of signals and approximate the complex nonlinear relationship between inputs and outputs. The purpose of this study is to verify the hypothesis that a wrist-cuff device using a neural network for blood pressure estimation from both the oscillatory pressure pulse wave and PPG signal may improve the accuracy. Design/methodology/approach A PPG sensor was integrated into a wrist blood pressure monitor, so the finger PPG and the oscillatory pressure wave could be detected at the same time during the measurement. After the peak detection, curves were fitted to the data of pressure pulse amplitude and PPG pulse amplitude versus time. A genetic algorithm-back propagation neural network was constructed. Parameters of the curves were inputted into the neural network, the outputs of which were the measurement values of blood pressure. Blood pressure measurements of 145 subjects were obtained using a mercury sphygmomanometer, the developed device with the neural network algorithm and an Omron HEM-6111 blood pressure monitor for comparison. Findings For the systolic blood pressure (SBP), the difference between the proposed device and the mercury sphygmomanometer is 0.0062 ± 2.55 mmHg (mean ± SD) and the difference between the Omron device and the mercury sphygmomanometer is 1.13 ± 9.48 mmHg. The difference in diastolic blood pressure between the mercury sphygmomanometer and the proposed device was 0.28 ± 2.99 mmHg. The difference in diastolic blood pressure between the mercury sphygmomanometer and Omron HEM-6111 was −3.37 ± 7.53 mmHg. Originality/value Although the difference in the SBP error between the proposed device and Omron HEM-6111 was not remarkable, there was a significant difference between the proposed device and Omron HEM-6111 in the diastolic blood pressure error. The developed device showed an improved performance. This study was an attempt to enhance the accuracy of wrist-cuff oscillometric blood pressure monitors by using the finger PPG and the neural network. The hardware framework constructed in this study can improve the conventional wrist oscillometric sphygmomanometer and may be used for continuous measurement of blood pressure.

Download Full-text

Real-Time Guitar Amplifier Emulation with Deep Learning

Applied Sciences ◽

10.3390/app10030766 ◽

2020 ◽

Vol 10 (3) ◽

pp. 766 ◽

Cited By ~ 2

Author(s):

Alec Wright ◽

Eero-Pekka Damskägg ◽

Lauri Juvela ◽

Vesa Välimäki

Keyword(s):

Neural Network ◽

Neural Networks ◽

Real Time ◽

Reference Signal ◽

Network Models ◽

Neural Network Models ◽

Desktop Computer ◽

The Neural Network ◽

Highly Nonlinear ◽

The Difference

This article investigates the use of deep neural networks for black-box modelling of audio distortion circuits, such as guitar amplifiers and distortion pedals. Both a feedforward network, based on the WaveNet model, and a recurrent neural network model are compared. To determine a suitable hyperparameter configuration for the WaveNet, models of three popular audio distortion pedals were created: the Ibanez Tube Screamer, the Boss DS-1, and the Electro-Harmonix Big Muff Pi. It is also shown that three minutes of audio data is sufficient for training the neural network models. Real-time implementations of the neural networks were used to measure their computational load. To further validate the results, models of two valve amplifiers, the Blackstar HT-5 Metal and the Mesa Boogie 5:50 Plus, were created, and subjective tests were conducted. The listening test results show that the models of the first amplifier could be identified as different from the reference, but the sound quality of the best models was judged to be excellent. In the case of the second guitar amplifier, many listeners were unable to hear the difference between the reference signal and the signals produced with the two largest neural network models. This study demonstrates that the neural network models can convincingly emulate highly nonlinear audio distortion circuits, whilst running in real-time, with some models requiring only a relatively small amount of processing power to run on a modern desktop computer.

Download Full-text

The Added Value of Surface Data to Radar-Derived Rainfall-Rate Estimation Using an Artificial Neural Network

Journal of Atmospheric and Oceanic Technology ◽

10.1175/2010jtecha1361.1 ◽

2010 ◽

Vol 27 (9) ◽

pp. 1547-1554 ◽

Cited By ~ 3

Author(s):

B. Root ◽

T-Y. Yu ◽

M. Yeary ◽

M. B. Richman

Keyword(s):

Neural Network ◽

Artificial Neural Network ◽

Radar Reflectivity ◽

Rainfall Rate ◽

Added Value ◽

Training Dataset ◽

Estimation Model ◽

Rate Estimation ◽

Surface Data ◽

Artificial Neural

Abstract Radar measurements are useful for determining rainfall rates because of their ability to cover large areas. Unfortunately, estimating rainfall rates from radar reflectivity data alone is prone to errors resulting from variations in drop size distributions, precipitation types, and other physics that cannot be represented in a simple, one-dimensional Z–R relationship. However, improving estimates is possible by utilizing additional inputs, thereby increasing the dimensionality of the model. The main purpose of this study is to determine the value of surface observations for improving rainfall-rate estimation. This work carefully designed an artificial neural network to fit a model that would relate radar reflectivity, surface temperature, humidity, pressure, and wind to observed rainfall rates. Observations taken over 13 years from the Oklahoma Mesonet and the KTLX WSR-88D radar near Oklahoma City, Oklahoma, were used for the training dataset. While the artificial neural network underestimated rainfall rates for higher reflectivities, it did have an overall better performance than the best-fit Z–R relation. Most importantly, it is shown that the surface data contributed significant value to an unaugmented radar-based rainfall-rate estimation model.

Download Full-text

Neural Network Based Hurdle Avoidance System for Smart Vehicles

International Journal of New Practices in Management and Engineering ◽

10.17762/ijnpme.v8i04.79 ◽

2019 ◽

Vol 8 (04) ◽

pp. 01-07

Author(s):

Dr. B. Maruthi Shankar

Keyword(s):

Neural Network ◽

Small Scale ◽

Direct Estimate ◽

Feed Forward ◽

Smart Vehicles ◽

Ir Sensor ◽

The Neural Network ◽

Feed Forward Network ◽

Positive Direct ◽

Matlab Programming

The structure of a self-ruling vehicle dependent on neural sophisticated network for route in obscure condition is proposed. The vehicle is equipped with an IR sensor for obstacle separation estimation, a GPS collector for goal data and heading position, L298 H-connect for driving the engines which runs the wheels; all interfaced to a controller unit. The smaller scale controller forms the data gained from the sensor and GPS to produce robot movement through neural based network. The neural network running inside the small scale controller is a multi-layer feed-forward network with back-engendering blunder calculation. The network is prepared disconnected with tangent-sigmoid and positive direct estimate as enactment work for neurons and is executed progressively with piecewise straight guess of tangent-sigmoid capacity. The programming of the miniaturized scale controller is finished by PIC C Compiler and the neural network is actualized utilizing MATLAB programming. Results have shown that up to twenty neurons can be actualized in shrouded layer with this method. The vehicle is tried with differing goal places in open air situations containing fixed as well as moving obstructions and is found to arrive at the set targets effectively and its yield exactness is about equivalent to that of the normal precision.

Download Full-text

Coupled online learning as a way to tackle instabilities and biases in neural network parameterizations: general algorithms and Lorenz96 case study (v1.0)

10.5194/gmd-2019-319 ◽

2020 ◽

Author(s):

Stephan Rasp

Keyword(s):

Neural Network ◽

Machine Learning ◽

Online Learning ◽

High Resolution ◽

Machine Learning Algorithms ◽

Training Dataset ◽

The Neural Network ◽

Training Stage ◽

Lorenz 96 ◽

Resolution Simulation

Abstract. Over the last couple of years, machine learning parameterizations have emerged as a potential way to improve the representation of sub-grid processes in Earth System Models (ESMs). So far, all studies were based on the same three-step approach: first a training dataset was created from a high-resolution simulation, then a machine learning algorithms was fitted to this dataset, before the trained algorithms was implemented in the ESM. The resulting online simulations were frequently plagued by instabilities and biases. Here, coupled online learning is proposed as a way to combat these issues. Coupled learning can be seen as a second training stage in which the pretrained machine learning parameterization, specifically a neural network, is run in parallel with a high-resolution simulation. The high-resolution simulation is kept in sync with the neural network-driven ESM through constant nudging. This enables the neural network to learn from the tendencies that the high-resolution simulation would produce if it experienced the states the neural network creates. The concept is illustrated using the Lorenz 96 model, where coupled learning is able to recover the "true" parameterizations. Further, detailed algorithms for the implementation of coupled learning in 3D cloud-resolving models and the super parameterization framework are presented. Finally, outstanding challenges and issues not resolved by this approach are discussed.

Download Full-text

Development of a generative adversarial neural network for identification of potential HIV-1 inhibitors by deep learning methods

Informatics ◽

10.37661/1816-0301-2020-17-1-7-17 ◽

2020 ◽

Vol 17 (1) ◽

pp. 7-17

Author(s):

G. I. Nikolaev ◽

N. A. Shuldov ◽

A. I. Anishenko, ◽

A. V. Tuzikov ◽

A. M. Andrianov

Keyword(s):

Neural Network ◽

Deep Learning ◽

Rational Design ◽

Binding Free Energy ◽

Viral Envelope ◽

Training Dataset ◽

Learning Methods ◽

The Neural Network ◽

Wide Range ◽

Hiv 1

A generative adversarial autoencoder for the rational design of potential HIV-1 entry inhibitors able to block the region of the viral envelope protein gp120 critical for the virus binding to cellular receptor CD4 was developed using deep learning methods. The research were carried out to create the architecture of the neural network, to form virtual compound library of potential anti-HIV-1 agents for training the neural network, to make molecular docking of all compounds from this library with gp120, to calculate the values of binding free energy, to generate molecular fingerprints for chemical compounds from the training dataset. The training the neural network was implemented followed by estimation of the learning outcomes and work of the autoencoder. The validation of the neural network on a wide range of compounds from the ZINC database was carried out. The use of the neural network in combination with virtual screening of chemical databases was shown to form a productive platform for identifying the basic structures promising for the design of novel antiviral drugs that inhibit the early stages of HIV infection.

Download Full-text

Bias in Deep Neural Networks in Land Use Characterization for International Development

Remote Sensing ◽

10.3390/rs13152908 ◽

2021 ◽

Vol 13 (15) ◽

pp. 2908

Author(s):

Do-Hyung Kim ◽

Guzmán López ◽

Diego Kiedanski ◽

Iyke Maduako ◽

Braulio Ríos ◽

...

Keyword(s):

Neural Network ◽

Neural Networks ◽

Land Use ◽

International Development ◽

Deep Neural Networks ◽

Satellite Image ◽

Fine Tuning ◽

Training Dataset ◽

Model Accuracy ◽

The Neural Network

Understanding the biases in Deep Neural Networks (DNN) based algorithms is gaining paramount importance due to its increased applications on many real-world problems. A known problem of DNN penalizing the underrepresented population could undermine the efficacy of development projects dependent on data produced using DNN-based models. In spite of this, the problems of biases in DNN for Land Use and Land Cover Classification (LULCC) have not been a subject of many studies. In this study, we explore ways to quantify biases in DNN for land use with an example of identifying school buildings in Colombia from satellite imagery. We implement a DNN-based model by fine-tuning an existing, pre-trained model for school building identification. The model achieved overall 84% accuracy. Then, we used socioeconomic covariates to analyze possible biases in the learned representation. The retrained deep neural network was used to extract visual features (embeddings) from satellite image tiles. The embeddings were clustered into four subtypes of schools, and the accuracy of the neural network model was assessed for each cluster. The distributions of various socioeconomic covariates by clusters were analyzed to identify the links between the model accuracy and the aforementioned covariates. Our results indicate that the model accuracy is lowest (57%) where the characteristics of the landscape are predominantly related to poverty and remoteness, which confirms our original assumption on the heterogeneous performances of Artificial Intelligence (AI) algorithms and their biases. Based on our findings, we identify possible sources of bias and present suggestions on how to prepare a balanced training dataset that would result in less biased AI algorithms. The framework used in our study to better understand biases in DNN models would be useful when Machine Learning (ML) techniques are adopted in lieu of ground-based data collection for international development programs. Because such programs aim to solve issues of social inequality, MLs are only applicable when they are transparent and accountable.

Download Full-text

Coupled online learning as a way to tackle instabilities and biases in neural network parameterizations: general algorithms and Lorenz 96 case study (v1.0)

Geoscientific Model Development ◽

10.5194/gmd-13-2185-2020 ◽

2020 ◽

Vol 13 (5) ◽

pp. 2185-2196

Author(s):

Stephan Rasp

Keyword(s):

Neural Network ◽

Machine Learning ◽

Online Learning ◽

High Resolution ◽

Learning Algorithm ◽

Training Dataset ◽

The Neural Network ◽

Training Stage ◽

Lorenz 96 ◽

Resolution Simulation

Abstract. Over the last couple of years, machine learning parameterizations have emerged as a potential way to improve the representation of subgrid processes in Earth system models (ESMs). So far, all studies were based on the same three-step approach: first a training dataset was created from a high-resolution simulation, then a machine learning algorithm was fitted to this dataset, before the trained algorithm was implemented in the ESM. The resulting online simulations were frequently plagued by instabilities and biases. Here, coupled online learning is proposed as a way to combat these issues. Coupled learning can be seen as a second training stage in which the pretrained machine learning parameterization, specifically a neural network, is run in parallel with a high-resolution simulation. The high-resolution simulation is kept in sync with the neural network-driven ESM through constant nudging. This enables the neural network to learn from the tendencies that the high-resolution simulation would produce if it experienced the states the neural network creates. The concept is illustrated using the Lorenz 96 model, where coupled learning is able to recover the “true” parameterizations. Further, detailed algorithms for the implementation of coupled learning in 3D cloud-resolving models and the super parameterization framework are presented. Finally, outstanding challenges and issues not resolved by this approach are discussed.

Download Full-text