Significance of Undertaking This Research
In recent years, climate change has advanced, and meteorological disasters such as typhoons and heavy rains have frequently occurred throughout Japan. Disasters impose enormous impacts upon us, including human casualties and infrastructure destruction. Against this backdrop, research on meteorological control aimed at mitigating disaster damages is progressing.
In meteorological control, it is necessary to seek appropriate interventions that steer us toward a desirable future. Such appropriate interventions are determined through meteorological intervention optimization. Meteorological intervention optimization refers to the process of optimizing parameters, such as the location and intensity of interventions, in order to achieve control objectives such as minimizing precipitation. However, due to the large scale and complexity of meteorological phenomena, strict optimization is practically impossible, and the use of approximate methods, heuristic approaches, and sequential or local optimization techniques is required.
At present (March 2025), attention is focused on black-box optimization methods, and research is being conducted on applying these methods to meteorological intervention optimization problems. Black-box optimization methods are those that search for input values that maximize or minimize the value of an evaluation function when neither the evaluation function nor the constraints are explicitly provided, with only the input and output of the function available. These methods do not require gradient information of the objective function and exhibit high search performance even with a limited number of function evaluations. Therefore, they are considered to be useful in meteorological intervention optimization.
Specific Research Content
Using a numerical model called SCALE-RM (SCALE-Regional Model) based on the software library SCALE (Scalable Computing for Advanced Library and Environment), simulation experiments are conducted to evaluate the effects of interventions determined by black-box optimization methods. Based on the SCALE USERS GUIDE (https://scale.riken.jp/archives/scale_users_guide.v5.5.4.pdf) and codes developed by predecessors, black-box optimization methods such as Bayesian optimization and particle swarm optimization are implemented, and the control effects (i.e., the degree of reduction in total rainfall) for each method are calculated.
The experimental settings addressed in previous research are introduced next.
Warm Bubble Experiment
The Warm Bubble Experiment is an experimental setting that reproduces the convective process of cumulus clouds under idealized conditions. At the initial time, warm air is placed in the lower central part of the computational domain. This warm air acts as a trigger for the development of updrafts, leading to precipitation over time. The computational domain (area of 10 km², height of 20 km) is discretized into grid cells divided into 1 cell in the x-direction, 40 in the y-direction, and 97 layers in the z-direction, and various atmospheric state variables in each discretized cell are computed with fine time steps to simulate meteorological phenomena.
The video below shows the time variation of precipitation intensity [mm/h] when no intervention is applied.
The figure below displays the cumulative precipitation over one hour.
The effect of intervention is evaluated using this experimental setting. The intervention is represented by modifying the values of atmospheric state variables, such as atmospheric momentum and potential temperature. In this study, an intervention is applied to the atmospheric momentum at the initial time (Time: 0 min) to formulate a control problem aimed at reducing the precipitation over one hour. The position and intensity of the intervention are determined by three variables. These three variables are optimized by a black-box optimization method to minimize the total one-hour cumulative precipitation (as represented by the total area of the bar graph above).
Under this experimental setting, particle swarm optimization was able to achieve a reduction of up to 43.6% in the one-hour precipitation over the computational domain.
Realistic Atmospheric Experiment
The Realistic Atmospheric Experiment is an experimental setting that conducts numerical simulations based on actual atmospheric states and surface conditions to reproduce and forecast real meteorological phenomena. By appropriately setting the initial time of the experiment, meteorological phenomena directly related to disasters such as typhoons can be reproduced.
The video below shows the time variation of precipitation intensity [mm/h] when no intervention is applied. Strong rainfall is observed in the Tohoku region and off the coast of the Kanto region. In this study, with the aim of mitigating heavy rainfall in the Tohoku region, an intervention is applied to the atmospheric momentum at the initial time to formulate a control problem aimed at reducing the cumulative precipitation over six hours within the region indicated by the red box on the right of the slide above. In this case, the position and intensity of the intervention are determined by four variables. These four variables are optimized by a black-box optimization method.
Under this experimental setting, using either Bayesian optimization or particle swarm optimization, the six-hour precipitation in the heavy rainfall area was reduced by up to 4.1%.
Control Theory
In approaching meteorological control, control theory—which is the theory for manipulating objects and events according to one’s will—is indispensable. However, existing control theory has not dealt with large-scale phenomena such as meteorology. Therefore, by incorporating artificial intelligence techniques and data technologies such as ensemble forecasting, efforts are being made to develop meteorological control methods that transcend the framework of conventional control theory. At present (March 2025), research is progressing with a focus on model predictive control.
Conclusion
This research is being conducted as one of the research and development challenges in the Moonshot project, under Goal 8, which envisions “a future liberated from the damage of concentrated heavy rainfall realized through maritime heavy rain generation.” For further details on the project, please refer to the following link (https://beyond-predictions.com/moonshot/ms_item01/).