Abstract

Introduction

Material and methods

Results and discussion

Conclusions

Acknowledgments

Supplementary materials

Reference

Abstract

     Air pollution is a major obstacle to future sustainability, and traffic pollution has become a large drag on the sustainable developments of future metropolises. Here, combined with the large volume of real-time monitoring data, we propose a deep learning model, iDeepAir, to predict surface-level PM2.5 concentration in Shanghai megacity and link with MEIC emission inventory creatively to decipher urban traffic impacts on air quality. Our model exhibits high-fidelity in reproducing pollutant concentrations and reduces the MAE from 25.355 μg/m3 to 12.283 μg/m3 compared with other models. And identifies the ranking of major factors, local meteorological conditions have become a nonnegligible factor. Layer-wise relevance propagation (LRP) is used here to enhance the interpretability of the model and we visualize and analyze the reasons for the different correlation between traffic density and PM2.5 concentration in various regions of Shanghai. Meanwhile, As the strict and effective industrial emission reduction measurements implementing in China, the contribution of urban traffic to PM2.5 formation calculated by combining MEIC emission inventory and LRP is gradually increasing from 18.03% in 2011 to 24.37% in 2017 in Shanghai, and the impact of traffic emissions would be ever-prominent in 2030 according to our prediction.

Introduction

     Air pollution is a large obstacle to the world’s future sustainable developments, and millions of people die from air pollution-related diseases every year around the world (Zheng et al., 2017). This is seriously severe in some developing countries like China (Lelieveld et al., 2015), which has the highest country-level values globally for the population-weighted annual average concentration of PM2.5 (Tichenor and Sridhar, 2019; Zhang et al., 2012) and has been a major public health concern in recent years (Li et al., 2019a) . Shanghai, one of the most developed and populous cities in China, has suffered severe increasing haze episodes mostly attributed to the severe particle pollution especially fine particles (particles ≤2.5μm in aerodynamic diameter; PM2.5)(Han et al., 2020) since 1990s with the rapid urbanization and industrialization (Wang et al., 2015) . Under these circumstances, air pollution-related diseases have emerged gradually, such as respiratory diseases in the elderly and preterm birth and low birth weight for birth when maternal exposure to PM2.5 in Shanghai(Li et al., 2019a; Liu et al., 2017) .

Conclusions

     In this article, we propose a novel method to quantify the influence of anthropogenic emissions on PM2.5 concentration with a novel and traceable deep learning model. Our experiment results indicate that the proposed model could achieve better fitting and prediction performances than the LSTM, GBRT, Seq2Seq, DA-RNN model and other deep learning models. Furthermore, we output the contributions of input parameter to the prediction in LRP method and visualize and analyze the spatial correlation between traffic flow and PM2.5.

     In addition, we discover that transportation emissions will play the most dominate role in future urban air pollution, and how to reduce traffic emissions are an unavoidable issue on achieving sustainable developments in modern cities. Meanwhile, to some extent, new energy vehicles can be considered as an effective way to reduce traffic emissions. However, the current promotion policies and efforts are far from enough. To further improve urban air quality, we need some more effective and powerful measurements in response to the rapid growth of urban traffic emissions.

Abstract

Introduction

Experiments and methodology

Results

Discussions

Conclusions

References

Acknowledgements

Funding

Author information

Ethics declarations

Additional information

Supplementary Information

Rights and permissions

About this article

Advertisement

Abstract

Growing evidence indicates that the atmospheric and oceanic circulation experiences a systematic poleward shift in a warming climate. However, the complexity of the climate system, including the coupling between the ocean and the atmosphere, natural climate variability and land-sea distribution, tends to obfuscate the causal mechanism underlying the circulation shift. Here, using an idealised coupled aqua-planet model, we explore the mechanism of the shifting circulation, by isolating the contributing factors from the direct CO2 forcing, the indirect ocean surface warming, and the wind-stress feedback from the ocean dynamics. We find that, in contrast to the direct CO2 forcing, ocean surface warming, in particular an enhanced subtropical ocean warming, plays an important role in driving the circulation shift. This enhanced subtropical ocean warming emerges from the background Ekman convergence of surface anomalous heat in the absence of the ocean dynamical change. It expands the tropical warm water zone, causes a poleward shift of the mid-latitude temperature gradient, hence forces a corresponding shift in the atmospheric circulation and the associated wind pattern. The shift in wind, in turn drives a shift in the ocean circulation. Our simulations, despite being idealised, capture the main features of the observed climate changes, for example, the enhanced subtropical ocean warming, poleward shift of the patterns of near-surface wind, sea level pressure, storm tracks, precipitation and large-scale ocean circulation, implying that increase in greenhouse gas concentrations not only raises the temperature, but can also systematically shift the climate zones poleward.

Introduction

An increasing amount of evidence suggests that the atmospheric and oceanic circulation is shifting towards the poles under climate change (Thompson et al. 2000; Marshall 2003; Fu et al. 2006; Hu and Fu 2007; Lu et al. 2007; Seidel et al. 2008; Screen et al. 2018). For example, poleward migration of the patterns of storm tracks (Yin 2005), winds (Thompson et al. 2000; Chen et al. 2008), jet streams (Archer and Caldeira 2008), precipitation (Scheff and Frierson 2012), tropical cyclones (Kossin et al. 2014), atmospheric frontal activity (Rudeva and Simmonds 2015), cloud (Norris et al. 2016), atmospheric rivers (Ma et al. 2020) and large-scale ocean circulation (Wu et al. 2012; Yang et al. 2016b, 2020a; Wu et al. 2021) have been identified based on various observations and climate simulations. These climate pattern shifts redistribute the natural resources, such as water, vegetation and the related ecosystems, thus having broad implications for our societies (Heffernan 2016). Understanding the underlying causes does not only help us to understand why it happens, but also serve to better predict and boost our confidence in the global warming induced climate changes.

1- Introduction

2- Methods

3- Datasets

4- Results

5- Discussion

6- Conclusions

References

Abstract

We introduce two Convolutional Neural Network (CNN) classifiers optimized for inferring brain states from magnetoencephalographic (MEG) measurements. Network design follows a generative model of the electromagnetic (EEG and MEG) brain signals allowing explorative analysis of neural sources informing classification. The proposed networks outperform traditional classifiers as well as more complex neural networks when decoding evoked and induced responses to different stimuli across subjects. Importantly, these models can successfully generalize to new subjects in real-time classification enabling more efficient brain–computer interfaces (BCI).

Introduction

Deep Neural Networks have revolutionized many domains such as image recognition and natural language processing. To date, their application in the analysis of electro- and magnetoencephalographic (EEG and MEG) data has been limited by several domain-specific factors. First of all, electromagnetic brain signals are characterized by very low signal-to-noise ratio (SNR). Here, the term “noise” is understood widely and includes external interference, physiological (e.g. cardiac or oculomotor) artifacts as well as background brain activity unrelated to the studied phenomena. SNR in single-trial EEG and MEG measurements is typically assumed to be < 1 for evoked responses and
1 for oscillatory activity, which puts these data to stark contrast with those in traditional applications of deep learning. Typical EEG/MEG analysis employs a wide range of techniques to increase the SNR, e.g. spatial and temporal filtering, averaging a large number of observations, sourceseparation algorithms, and other complex feature extraction methods (e.g. wavelet transform). Thus, efficient noise suppression is required for high-accuracy classification of EEG/MEG signals. Second, these data have a complex, high-dimensional spatiotemporal structure. Modern EEG and MEG systems comprise several hundreds of sensors capable of sampling brain activity with sub-millisecond temporal resolution. In the case of MEG, these sensors may also measure different components of the neuromagnetic field. On one hand, this multitude of data points enables sophisticated analysis methods to extract finer details of brain function. On the other hand, manual analysis and interpretation of these data become increasingly complex and time-consuming. Machine-learning algorithms can be of great help in such tasks but the mere classification result is often not sufficient; ideally, the experimenter should understand why the algorithm is able to classify the data, i.e., the learned model should be interpretable in neurophysiological terms. A model able to reliably identify those neural sources that contribute to the discrimination between given experimental conditions could enable efficient exploitative analysis of these complex data sets and ultimately allow more complex experimental designs. Finally, deep-learning models require large numbers of training samples to perform optimally. In a typical EEG/MEG or BCI experiment, however, time constraints and data acquisition costs limit the sample sizes severely. Open datasets usually comprise data from a large number of individuals, providing a promising strategy to overcome this limitation, but in this case, the classifier needs to be robust to high interindividual variability stemming from differences in cortical anatomy, mappings of function to structure and physiological state. Taken together, these factors may easily lead to over-fitting (especially in more complex models) and poor interpretability of the findings. To address these challenges, we propose a Convolutional Neural Network (CNN) whose architecture is based on a generative model of non-invasive electromagnetic measurements of the brain activity (Daunizeau and Friston, 2007).

I. INTRODUCTION

II. MOTIVATION

III. HEVC INTRA PREDICTION

IV. PROPOSED VLSI ARCHITECTURE

V. IMPLEMENTATION RESULT

VI. CONCLUSION

REFERENCES

In the High Efficiency Video Coding (HEVC), a variety of CU sizes and intra prediction modes significantly improve coding efficiency, but also bring higher computational complexity. This paper proposes a new compact VLSI architecture for HEVC intra prediction, which is geared towards 8K video decoding. It supports all the transform unit (TU) sizes and 35 HEVC intra prediction modes. First, this paper introduces a TU-oriented intra predictor with a throughput of 32 pixels, which can be newly arranged with the TU size. It can be a line of 32 pixels, two lines of 16 pixels, 4 lines of 8 pixels or four lines of four pixels. This TU-oriented architecture allows intra-prediction and inverse discrete cosine transform (IDCT) to be computed in parallel, removing the memory between them. In addition, a horizontal and vertical line buffer for reference sample is proposed, which only cost 0.8K bit and is implemented by register files with SRAM-free. Finally, to further reduce hardware consumption, multipliers can be shared in the prediction. The implementation result shows that the compact architecture supports 8K video application and costs 66.2K logic gates, which is synthesized with the TSMC 65nm process under 400MHz.

INTRODUCTION

The Joint Collaborative Team on Video Coding (JCT-VC) has proposed the latest video coding standard, High Efficiency Video Coding (HEVC) [1]. It can achieve about a 50% bit-rate reduction than the previous video coding standard, H.264/AVC [2], but with higher computational complexity. Like AVC, HEVC also uses block-based coding. Each video frame is divided into multiple CTUs, and each CTU is further divided into smaller CUs. And in HEVC, the CU is further divided into PUs and TUs, where PU is the basic prediction unit and TU is for transform and quantization. The maximum CU size in HEVC is 64×64, not 16×16 in AVC, but the maximum size of TU is only 32 × 32. The PU size of intra prediction is suitable for TU, so the largest PU is 32 × 32. In addition, the number of HEVC intra prediction modes is 35, not 9 in AVC, in order to describe the texture better [4]. Although the complex block partition scheme and the prediction mode can improve the coding quality, the VLSI implementation is more challenging for high-throughput and high-resolution video coding. Due to the higher complexity and data dependencies, it is difficult to propose a compact high-performance intra prediction architecture. Previously, there was some related work on the intra prediction VLSI architecture, and some were optimized for the reference pixel preparation and prediction process.