1:45 PM - 2:00 PM
[AAS09-13] Data-driven atmospheric substances transport modeling in the polar regions
★Invited Papers
Keywords:Trajectory Analysis, Data Assimilation, Machine Learning, Aerosol Atmosperic River
Because Antarctica is surrounded by the ocean and isolated from other continents, it has been considered a suitable site for monitoring the global environment, including long-term measurements of carbon dioxide concentrations. However, the harsh environment and inconvenience of transportation make it difficult to continue continuous field observations. In addition, we have studied ice cores collected over Antarctica, e.g., Dome Fuji station and other sites, to deepen our understanding of paleoclimate changes and predict the future by analyzing their components. On the Japanese route from the Syowa Station on the coast to the Dome Fuji inland, we have continued to measure the recharge rate using a snow scale and correct the surface snow cover since the 1990s. The results of the component analysis of surface snow suggest regional differences in atmospheric transport depending on the topography of Antarctica and the meteorological fields affected by the topography. This study aims to develop a data-driven aerosol transport model based on the data of various substances in multiple atmospheric aerosols, surface snow cover, and ice cores. Data-driven means that model decisions are made by machine learning and statistical methods based on the data, which is different from simulations using weather models that carefully solve physics. This model aims to achieve the same or better computational speed and prediction accuracy than meteorological simulations while requiring fewer computational resources than meteorological simulations.
Trajectory analysis has been mainly used to investigate the causes of rapid changes in atmospheric trace substance concentrations observed at Syowa Station. The analysis places virtual particles in an atmospheric field and tracks their advection by wind velocity fields. Since the predicted position at each time step is a point estimate, it is evident that the error increases exponentially as the prediction time increases. To solve this issue, in a previous study, multiple trajectories are calculated by adding perturbations to the initial positions of particles and creating frequency distributions, etc. This study uses a data assimilation method to estimate the error for each predicted position at each time step. By providing a probability distribution for the predicted positions of particles, which were previously point estimates, it is possible to determine how reliable the predictions are. In addition, we target black carbon as an atmospheric trace substance, and areas of high concentration in the atmospheric aerosol optical thickness observed by GCOM-C/SGLI and other satellites are captured as aerosol emission events and tracked to see if they are transported to Antarctica. We choose the atmospheric transport routes generated by trajectory analysis with data assimilation and the objective meteorological data as learning objects to predict the movement of high-concentration areas by satellite. To improve the accuracy of the prediction model, we use satellite images and aerosol concentrations in the atmosphere at Syowa Station as Ground Truth. Finally, the model automatically predicts aerosol transport when an aerosol emission event is detected by satellite observation and compares the observed aerosol concentration at Showa Station to evaluate the consistency. Once the model's accuracy is confirmed, we will conduct reproducible experiments on mass transport in surface snow cover and ice cores and make future predictions.
In this article, we introduce the study's framework, the event analysis results currently underway, and the trajectory model description.
This study was supported by JAXA's (The Japan Aerospace Exploration Agency) EORA3 (The 3rd Research Announcement on the Earth Observations) program.
Trajectory analysis has been mainly used to investigate the causes of rapid changes in atmospheric trace substance concentrations observed at Syowa Station. The analysis places virtual particles in an atmospheric field and tracks their advection by wind velocity fields. Since the predicted position at each time step is a point estimate, it is evident that the error increases exponentially as the prediction time increases. To solve this issue, in a previous study, multiple trajectories are calculated by adding perturbations to the initial positions of particles and creating frequency distributions, etc. This study uses a data assimilation method to estimate the error for each predicted position at each time step. By providing a probability distribution for the predicted positions of particles, which were previously point estimates, it is possible to determine how reliable the predictions are. In addition, we target black carbon as an atmospheric trace substance, and areas of high concentration in the atmospheric aerosol optical thickness observed by GCOM-C/SGLI and other satellites are captured as aerosol emission events and tracked to see if they are transported to Antarctica. We choose the atmospheric transport routes generated by trajectory analysis with data assimilation and the objective meteorological data as learning objects to predict the movement of high-concentration areas by satellite. To improve the accuracy of the prediction model, we use satellite images and aerosol concentrations in the atmosphere at Syowa Station as Ground Truth. Finally, the model automatically predicts aerosol transport when an aerosol emission event is detected by satellite observation and compares the observed aerosol concentration at Showa Station to evaluate the consistency. Once the model's accuracy is confirmed, we will conduct reproducible experiments on mass transport in surface snow cover and ice cores and make future predictions.
In this article, we introduce the study's framework, the event analysis results currently underway, and the trajectory model description.
This study was supported by JAXA's (The Japan Aerospace Exploration Agency) EORA3 (The 3rd Research Announcement on the Earth Observations) program.