To realize weather control, it is necessary to predict how the weather will respond when control inputs are given and to estimate when, where, and how we should add signals to the atmosphere to achieve desirable weather. However, there is currently no system capable of handling these tasks. In item 4, we aim to develop a weather control computation system that can simulate the impact of control inputs on weather based on existing numerical weather prediction systems and estimate optimal control inputs.
In addition, item 4 addresses the acceleration of prediction and control computations. To achieve weather control, as mentioned above, these computations must be completed before a disaster occurs or before the appropriate timing for control is missed. However, running numerical models used for weather prediction generally requires significant computational cost. With current computational resources and mainstream computational algorithms, it is anticipated that it would be difficult to complete these computations in a realistic timeframe. Therefore, in item 4, we will introduce surrogate models and latent space representation techniques derived from mathematical research and deep learning into the prediction and control computation system. This will enable the calculation of control inputs in a realistic timeframe while evaluating effective elemental technologies for weather control computation. We also aim to utilize quantum computers, which are a cutting-edge computational technology, to optimize model predictive control and data assimilation calculations by mapping them to the Ising model and accelerating computations through quantum annealing.

Since the start of this research project in December 2023, item 4 has focused on developing a system that can predict the impact of control inputs on weather. In the first year, we selected floating domes as control inputs and implemented the actuator functions to simulate the impact within a regional weather model. Using SCALE, a regional weather model developed by RIKEN (Nishizawa et al., 2015; Sato et al., 2015), we modified the topography within the model to incorporate the effects of floating domes. We set up an environment to run this on the supercomputer Fugaku, conducted experiments with and without control, and compared the simulation results to examine the effects of control. In collaboration with Item 5, we selected the September 2015 Kanto-Tohoku heavy rainfall event as a test case. The results showed that installing floating domes generated an area of heavy rainfall of about 30mm over 12 hours downwind of the dome. Sensitivity analysis regarding the size of the domes indicated that larger domes tend to produce more precipitation downwind, but this effect diminishes as the dome size increases. Further investigation is planned to examine the reduction of precipitation downwind, although current effects are unclear.
Weather predictions are inherently imperfect due to the chaotic nature of the atmosphere and always contain errors. Accurately indicating the extent of these errors is a condition for good predictions. Appropriate representation of errors is also required when estimating control inputs. In item 4, we have initiated the development of ensemble prediction methods that accurately represent errors by combining ensemble data assimilation and multi-physics ensemble predictions. In 2023, we promoted the quantification of model errors and initial value errors in SCALE. Although currently these are estimated independently, combining them will allow for the generation of predictions that consider both initial condition errors and model bias. This task will continue in subsequent years, along with efforts to enhance prediction accuracy.
Regarding computational acceleration, we investigated the latest technological trends and identified techniques and methods related to surrogate models, latent space representations, and quantum computing technologies. Using mesoscale ensemble weather prediction information provided by the Japan Meteorological Agency, we promoted research on the dimensionality reduction of prediction information and found that there may be branching points that serve as control "hot spots" for typhoons. We also investigated the latest research trends on quantum annealing machines and developed methods to accelerate data assimilation calculations. This suggested that using quantum annealing machines for weather prediction calculations could potentially be over 100 times faster than using conventional computers. We are also investigating low-cost computational methods based on the latest technological trends. Specifically, we are exploring methods to represent weather information in latent space, focusing on generative AI technologies such as variational autoencoders and diffusion models.

We will continue to investigate the impact of control inputs on weather. In 2023, we focused solely on floating domes, but we plan to consider more methods and increase the number of case studies. This will clarify which types of control are effective. Additionally, we will start developing methods to estimate when, where, and how control should be implemented. First, we will establish methods using low-dimensional weather models that are easy to handle, and then reflect the knowledge gained in real atmospheric models to advance the development of the weather control computation system.