Visualizing the generation process and intervention effects of heavy rains using deep learning and visual information processing techniques
This study will utilize information science to promote social research and dissemination necessary to realize weather control. Using deep learning and visual information processing techniques, we will accelerate weather control research by visualizing the generation process and intervention effects of heavy rains in four dimensions, and by analyzing the effects of weather control operations. Furthermore, this visualization system will be used to contribute to the project’s outreach promotion.
Item 10-1Advection model on deep neural fields
Principal investigator: Takuya Funatomi
Outline
To accelerate the meteorological study of heavy rainfall generation processes and to validate intervention results, this study develops techniques for visualizing four-dimensional data, including internal variables from numerical weather models and observation data from weather radars. This involves modeling temporal changes within three-dimensional space using advection modeling techniques. We conduct precipitation prediction using data-driven methods such as deep neural fields, while evaluating the effects of weather control on advection.
Method
First, we develop techniques for advection modeling that target clouds and precipitation regions using meteorological observation data, applicable at both local and global scales. We collaborate with Item 10-2 to develop the visualization techniques that contribute to weather control research of the generation processes of heavy rainfall. Additionally, we select four-dimensional data that contribute to the verification of intervention results and develop techniques for advection modeling based on that data. Subsequently, we visualize the modeling outcomes.
Importance
Using visual information processing techniques to visualize the generation processes of heavy rainfall and the effects of intervention is essential for analyzing the impacts of weather intervention. This approach is expected to accelerate weather control research. In addition, such visualization techniques for public outreach promote understanding among citizens regarding weather control, making it essential for achieving societal acceptance of these methods.
Visualizing the generation processes requires representing the continuous temporal changes of weather phenomena. However, numerical weather models we target for visualization consist only of discrete datasets that store physical quantities related to weather at specific locations and times. We aim to obtain functional representations that describe continuous temporal changes by applying techniques based on deep neural fields to discrete datasets stored at each location and time. This technology focuses on extracting continuous function representations necessary for visualizing the generation processes of heavy rainfall and intervention effects. This study is important for examining effective intervention strategies in meteorology.
Expected problems and solutions
We plan to implement advection modeling based on the latest and most noteworthy technology, “deep neural fields.” However, the pace of innovation in deep learning is rapid, and it is not uncommon for cutting-edge techniques to be replaced by new methods within just 2 to 3 years. We conduct ongoing technology trend investigation regarding the foundational techniques, allowing us to flexibly revise our plans to incorporate the latest advancements at any given time.
Members
Item 10-2Developing visualization methods for evaluating intervention effects
Principal investigator: Hiroyuki Kubo
Outline
This study utilizes visual information processing techniques to visualize the generation processes of heavy rainfall and intervention effect in four dimensions, aiming to accelerate weather control research. Specifically, we visualize the physical quantities from numerical weather models using four-dimensional volume rendering to enhance the visual understanding of effective interventions for the generation of maritime heavy rainfall. This will accelerate weather control research. In addition, we utilize the visualization system to communicate the results of this scenario to the public.
Method
First, we conduct a technology trend investigation on methods for displaying large-scale volume data and the associated display devices.
Next, we develop a prototype for visualizing small scale four-dimensional data within a limited area and time span. Then we visualize representative events demonstrating weather control effects. Moreover, we develop technologies for efficiently processing large-scale four-dimensional data. After that, we visualize multiple events demonstrating weather control effects, creating a display system to deepen visual understanding.
Importance
By visually displaying the generation processes of heavy rainfall and the effects of weather control interventions as four-dimensional data, we aim to accelerate weather control research. Additionally, by visualized images generated by the developing techniques are utilized in outreach activities for this project, fostering understanding and trust in the technologies being developed.
Expected problems and solutions
The four-dimensional weather data obtained from measurements consists solely of discrete information at specific areas and times, which can pose challenges when visualizing actual weather changes. To address this, we collaborate with Item 10-1 to visualize the data as continuous information.
Members
