Planned Research

A01 Harmonic Prediction Group

Research Task
A01: Creation of harmonic prediction technology for micro-meteorology based on AI-integrated simulation
Principal investigator
Ryo Onishi
Associate Professor, Global Scientific Information and Computing Center, Tokyo Institute of Technology


To realize a future city where autonomous drones provide a variety of services in response to the constantly changing environment, it is essential to integrate real-life meteorology with digital simulations and let them work interactively. While the micro-meteorological prediction data is provided to the multitude of autonomous systems, the massive volume of sensor data obtained by the autonomous systems should also be used for prediction. Micro-meteorology here means the micro-scale meteorology just above the earth surface strongly affected by the built environment and human activities.

To attain this integration, the research group intends to create harmonic prediction simulation technology that is adaptive to the ever-changing society and environment and is capable of real-time execution.

Specifically, a huge volume of environment sensor data obtained by ultrahigh-density monitoring and socioeconomic data including artificial exhaust heat are assimilated into a simulation. For this, our group uses its original distributed multi-scale data assimilation method to attain prediction with adaptability. Meanwhile, the group also intends to materialize a real-time micro-meteorological prediction system that can bring out ultrahigh-speed prediction information by integrating AI technology with the prediction simulation technology.

Background and Academic Question

An autonomous society in harmony with nature is where people can live safely and creatively with the reality and data linked together to a high degree. To realize this future society, integration of reality and data (cyberspace) is essential. With respect to meteorological phenomena, this integration has been realized to a certain degree; weather forecast services, for example, are already available and familiar to us. When it comes to micro-meteorology, however, it is directly connected with human life and safety and is also related to various social issues. The feasibility of micro-meteorological prediction is not self-evident to begin with, nor are the limitations of “predictable time-space scale and accuracy.” With uncertainties deriving from natural phenomena combining with those from socio-economic activities, the question, especially regarding the micro-meteorological scale, is profoundly interesting from the interdisciplinary viewpoint.

Objective, Originality, and Creativity

Our research intends to answer this question and realize the integration of real micro-meteorological-scale atmospheric conditions with data. To this end, the group develops harmonic prediction technology for micro-meteorology, which provides prediction data adaptive to a constantly changing environment and society on a real-time basis.

Our research group leads the world in technologies for building-level resolution and micro-meteorological simulation (Fig. 1). The group intends to attain adaptability to the environment and society by assimilating a huge amount of environmental sensor data obtained by high-density monitoring and socioeconomic data into the micro-meteorological simulation. The group also integrates the rapidly advancing AI technology with its predictive ability to enable real-time operation. Thus, we build the cyber meteorological space that harmonizes with the real space to materialize harmonic prediction for micro-meteorology. For instance, we developed the world’s first real-time prediction technology that updates the prediction every minute up to 30 minutes ahead of time while creating a micro-meteorological analysis product at a resolution of 5 m in a 2 square kilometer area. It is a grid dot data set obtained by data assimilation of simulation results and observation values. The ultrahigh-definition city-block micro-meteorological simulation at the several-meter mesh level requires a huge computing load, which can never be executed in real-time with the technology currently available. The group will resolve this bottleneck on the basis of our newly developed original AI-integrated simulation technology (Reference Paper 1).

In 1922, when computing was still done by human power, numerical weather forecasting was attempted for the first time in the U.K. It is called “Richardson’s dream.” One hundred years later, numerical weather forecasting is part of our daily life. Meanwhile, there has been no attempt to predict micro-meteorology in a building block up to 30 minutes ahead of time. Therefore, this research is the first challenge in the world to predict micro-meteorology at the city block level. The group will resolve this challenging problem by using its unique technology integrating simulation science and data science.

Fig. 1 An Example of the Results of Building Resolution and Urban Micro-meteorological Simulation conducted by Onishi et al.
It shows three-dimensional distribution of atmospheric temperature around Tokyo Station. The related video is on permanent exhibit at Miraikan, the National Museum of Emerging Science and Innovation in Japan, under the title “Tokyo Heat Island”.

Reference Paper 1:
Onishi et al., Super-Resolution Simulation for Real-Time Prediction of Urban Micro-meteorology, Scientific Online Letters on the Atmosphere, Vol. 15, 178−182 (2019)