A comparative analysis for a deep learning model (hyDL-CO v1.0) and Kalman filter to predict CO concentrations in China

Abstract

Abstract. The applications of novel deep learning (DL) techniques in atmospheric science are rising quickly. Here we build a hybrid DL model (hyDL-CO), based on convolutional neural networks (CNNs) and long short-term memory (LSTM) neural networks, to provide a comparative analysis between DL and Kalman filter (KF) to predict carbon monoxide (CO) concentrations in China in 2015–2020. We find the performance of DL model is better than KF in the training period (2015–2018): the mean bias and correlation coefficients are 9.6 ppb and 0.98 over eastern China and are −12.5 ppb and 0.96 over grids with independent observations (i.e., grids with CO observations that are not used in DL training and KF assimilation). By contrast, the assimilated CO concentrations by KF exhibit comparable correlation coefficients but larger negative biases. Furthermore, the DL model demonstrates good temporal extensibility in the test period (2019–2020): the mean bias and correlation coefficients are 95.7 ppb and 0.93 over eastern China and 81.0 ppb and 0.91 over grids with independent observations, while CO observations are not fed into the DL model as an input variable. Despite these advantages, we find a weaker prediction capability of the DL model than KF in the test period, and a noticeable underestimation of CO concentrations at extreme pollution events in the DL model. This work demonstrates the advantages and disadvantages of DL models to predict atmospheric compositions with respect to traditional data assimilation, which is helpful for better applications of this novel technique in future studies.

Rapid assimilations of O3 observations – Part 1: methodology and tropospheric O3 changes in China in 2015–2020

“…Open fire emissions are from the Global Fire Emissions Database (GFED4) (van der Werf et al, 2010). We employ the sequential KF to assimilate O3 observations, which has been used in recent studies to optimize tropospheric CO concentrations (Tang et al, 2022;Han et al, 2022).…”

“…For example, the tagged-Ox mode of the GEOS-Chem model has been used to analyze the sources and transport of tropospheric O3 (Zhang et al, 2008;Zhu et al, 2017;Han et al, 2018). However, it may not 2023)) via a sequential Kalman Filter (KF) assimilation system (Tang et al, 2022;Han et al, 2022) to investigate the performance of single tracer simulation on O3 assimilations. Furthermore, the rapid assimilation capability based on the tagged-O3 mode allows us to perform a convenient, comparative analysis to investigate the changes in tropospheric O3 in E.…”

Inverse modeling of Chinese NO<sub>x</sub> emissions using deep learning: Integrating in situ observations with a satellite-based chemical reanalysis

“…For western and northeastern China, the spatial coverage is lower. Han et al (2022) investigated the impact of observational coverage by removing 10 % of the grid-averaged observations from the training of a similar DL model and found no noticeable performance degradation in the evaluation of the DL model over these regions. However, the low station densities in some grids could lead to representation errors in the aggregated observations.…”

Ground-level gaseous pollutants (NO2, SO2, and CO) in China: daily seamless mapping and spatiotemporal variations

“…To date, we have combined the advantages of artificial intelligence and big data to construct a virtually complete set of major air quality parameters concerning both particulate and gaseous pollutants over a long period of time across China, including PM 1 (1 km, 2000-present) (Wei et al, 2019), PM 2.5 (1 km, 2000-present) (Wei et al, 2020, PM 10 (1 km, 2000-present) , O 3 (10 km, 1979-present) (Wei et al, 2022a;L. He et al, 2022), and NO 2 (1 km, 2019-present) (Wei et al, 2022b), serving environmental, public health, economy, and other related research.…”