9:30 AM - 9:45 AM
[MIS10-03] Automatic monitoring and classification of cloud dynamics in alpine regions using time-lapse cameras
Keywords:Alpine weather, Machine learning, Time-lapse camera
Alpine ecosystems are extremely vulnerable to climate change, and monitoring these ecosystems is essential for conserving alpine ecosystems with many endemic species and investigating the responses of ecosystems to climate change. In this study, we are developing an automatic method for observing localized and short-lived fog and clouds in alpine areas using a time-lapse camera. Temporal dynamics of clouds and fog are expected to affect animal behavior and plant photosynthesis. However, manual observation is difficult in alpine areas where human presence is difficult, and there have been few studies on this subject. The advantages of using a time-lapse camera include: 1) it is less affected by bad weather than satellite imagery, 2) unlike satellite images, it can distinguish between clouds in the sky and fog on the ground, and 3) it enables high-frequency observations at a low cost.
As a previous result of this study, we have developed a deep-learning-based method that divides images into tiles and then detects fog and clouds for each tile. By applying this method to time-lapse camera images taken at a high frequency (e.g., one image per hour), we can obtain time-series data showing changes in cloud positions over time. In this study, this time-series data were clustered to search for time-series patterns of clouds and fog automatically. This analysis allows us to obtain trends and changes in the temporal dynamics of clouds and fog, for example, in terms of the frequency of each pattern.
By applying the developed method to existing time-lapse cameras, we expect to clarify the dynamics of clouds and fog in the alpine zone, which have been poorly understood in the past, and to contribute to the detection of their changes, their relationship with climate change, and their impact on the ecosystem. In this presentation, we will report on the results of the proposed method using images taken at Mt. Chogatake, Nagano Prefecture, over nine years.
As a previous result of this study, we have developed a deep-learning-based method that divides images into tiles and then detects fog and clouds for each tile. By applying this method to time-lapse camera images taken at a high frequency (e.g., one image per hour), we can obtain time-series data showing changes in cloud positions over time. In this study, this time-series data were clustered to search for time-series patterns of clouds and fog automatically. This analysis allows us to obtain trends and changes in the temporal dynamics of clouds and fog, for example, in terms of the frequency of each pattern.
By applying the developed method to existing time-lapse cameras, we expect to clarify the dynamics of clouds and fog in the alpine zone, which have been poorly understood in the past, and to contribute to the detection of their changes, their relationship with climate change, and their impact on the ecosystem. In this presentation, we will report on the results of the proposed method using images taken at Mt. Chogatake, Nagano Prefecture, over nine years.