Cumulonimbus clouds develop at altitudes exceeding 10 kilometers and can sometimes cause catastrophic damage. Shin-ichiro Shima of the University of Hyogo, a Performer of the Yamaguchi Project in the Moonshot Goal 8, is working to understand the "particles" of these clouds in order to improve weather prediction accuracy and achieve weather control. We spoke with him about the simulation methods that have attracted attention through cumulonimbus videos posted on YouTube, and their significance.
What are the simulations that reproduce the interior of cumulonimbus clouds using "super-droplets"?
What kind of cumulonimbus simulations are you working on?
Shima: Cumulonimbus clouds are dynamic clouds commonly seen in summer that cause thunder and intense rainfall. From a distance they appear as a single mass, but in reality, they are made up of countless water droplets and ice particles. Our research uses computer simulations to replicate how clouds form and evolve by tracking these "particles." However, the number of particles in a cloud can reach several million to several billion per cubic meter, and calculating the behavior of all of them is not practical even with supercomputers. Therefore, the "Super-Droplet Method" I developed replaces the enormous number of particles in a cloud with a small number of virtual particles in the computer—called "super-droplets"—and tracks their behavior. By using representative samples, we reduce the computational load and enable precise simulation of the overall structure and properties of clouds.
How does this differ from the cloud simulations we see in weather forecasts?
How does cloud simulation using the "Super-Droplet Method" differ from the conventional cloud models used in weather forecasting?
Shima: Cloud models used in weather forecasting employ a method called the "bulk method." The bulk method handles cloud droplets and raindrops collectively as "average properties per grid cell." For example, it roughly calculates "what is the total mass in this grid cell" or "how many droplets are there." While the calculations are fast, the state of individual particles is ignored, making it difficult to reproduce fine-scale phenomena. The bin method is somewhat more advanced, dividing particles into groups called "bins" for each grid cell. For example, bins are created for each particle size, and changes in the number of particles and mass within each bin are calculated. However, if you try to consider properties beyond particle size—such as shape, density, or the amount of water adhering to the surface—the number of bins increases explosively, and computational demands become very high.
Yes, when you consider not just particle size but also the shape and density of ice particles, and whether there are frozen droplets on the surface (supercooled droplets that collided and froze), the types of information multiply accordingly.
Shima: Exactly. For example, ice particles have complex states and even approximating their shape as spheroids requires information about the major and minor axes, plus density and the amount of frozen droplets on the surface. If you try to express all of this with the bin method, dimensions increase—2D, 3D, 4D—and the number of bins per grid cell grows to tens of thousands, hundreds of thousands, or even millions. With the "Super-Droplet Method," many particles in a 100-meter-square grid can be represented by about 100 "super-droplets" in total. This allows faithful tracking of complex particle state changes without the explosive increase in bin numbers that makes calculation difficult in the bin method.
So, the bulk method is simple but struggles to reproduce fine-scale phenomena, and while the bin method divides things more finely, computational demands become enormous. The Super-Droplet Method can reproduce complex particle states while keeping computational loads down.
Shima: Exactly. For example, inside a cumulonimbus cloud, water droplets, ice particles, and aerosols (fine particles that serve as cloud nuclei) coexist, merging, splitting, freezing, and melting—with various phenomena occurring simultaneously. With the Super-Droplet Method, we can track each of these changes through virtual particles in the computer, allowing more faithful reproduction of phenomena.
The significance of improving simulation accuracy: Contributing to Moonshot Goal 8
That said, even the bulk method can predict "whether tomorrow will be rainy or sunny." Where is the necessity for high-precision simulation using the Super-Droplet Method?
Shima: It's true that bulk method calculations are fast and remain mainstream in weather forecasting today. However, when trying to understand in-depth what is happening inside clouds and the mechanisms involved, there are limits to the bulk and bin methods. To reproduce phenomena such as "how small droplets become large raindrops," "how quickly rainfall begins," and "to what extent cloud properties change depending on ice particle shape and state," we need to track the detailed state of each particle.
In recent years, due to the increase in weather disasters caused by climate change and extreme weather events (such as guerrilla heavy rainfall and typhoons), there is growing social demand for more accurate prediction of "when, where, and how much rain will fall." With conventional bulk methods, inaccuracies remain in predicting the timing of rainfall onset and cessation, and the scale and extent of heavy rain.
Furthermore, "aerosols," which are fine particles in the atmosphere, play a major role in cloud and rain formation. To understand how industrial activities and anthropogenic aerosol emissions affect clouds and the climate, precise simulations that delve into the origins and properties of particles are essential.
Could the Super-Droplet Method be used in weather forecasting in the future?
Shima: I think there is considerable potential for that to happen. However, at present, calculations using the Super-Droplet Method still take a lot of time. This cumulonimbus video depicts a two-hour phenomenon, but the calculation took roughly one week, even when using a supercomputer. This is because we spend about 60 times longer on cloud calculations compared with the bulk method, and we're computing at very high resolution—but that doesn't make it a weather "forecast" (laughs).
In the future, with improvements in supercomputer performance, algorithm innovations, and having AI learn from Super-Droplet Method calculation results as "training data," I believe weather forecasting using the Super-Droplet Method will become possible. We are aiming for practical application within a 5-10-year timeframe.
You are participating as a member of the Yamaguchi Project in the Moonshot Goal 8: "Heavy Rainfall Control for Living Together with Isolated-Convective Rainstorms and Line-Shaped Rainbands." What contribution is this research expected to make toward Goal 8?
Shima: Goal 8 is advancing research and development to "control typhoons and heavy rainfall". In other words, to prevent disasters by artificially modifying extreme weather. So far, computer simulations have begun to show the possibility of disaster mitigation by using seeding technology (dispersing aerosols that serve as cloud nuclei into the air) to change cloud growth and precipitation distribution.
To accurately predict the effects of such interventions in real time, models that can faithfully calculate cloud changes at the particle level are needed. With conventional bulk and bin methods, it was difficult to finely express the effects of aerosols. The Super-Droplet Method can directly calculate the state of each particle, including aerosols, enabling faithful tracking of responses to seeding.
For weather control, you need to obtain simulation results faster than real time, don't you?
Shima: Exactly. To decide on countermeasures and intervene before heavy rainfall actually arrives, and then receive feedback on the results, ultra-fast and high-precision simulation is essential. We are advancing development through various means—including algorithms, utilization of supercomputers, and AI acceleration—so we can calculate cloud simulations using the Super-Droplet Method or one with equivalent precision "faster than reality" by 2031.
Within Goal 8, outdoor observations and experiments are also being conducted, and we plan to further improve accuracy by comparing measured data with Super-Droplet Method simulation results. To this end, we have established an organization called the "MS8 Joint Unit for the Super-Droplet Method" and are advancing model development and application in collaboration with all four projects under Moonshot Goal 8, not just the Yamaguchi Project.
Discoveries and challenges in the research field, and toward the future
How well do simulations using the Super-Droplet Method match actual observational data? Have there been any unexpected discoveries or challenges during the research?
Shima: Research from other institutions has reported that simulations using particle-based cloud models nearly match observations of "warm clouds" that do not contain ice particles. This was a study comparing aircraft observation data off the Philippines with Super-Droplet Method calculation results, showing that the particle size distribution of particularly large raindrops nearly matches. For clouds containing ice particles and complex crystal structures, we plan to continue further verification.
A major challenge in bringing the Super-Droplet Method to practical use was that when actual water droplets collide and merge to form raindrops, the number of "super-droplets" as samples decreases dramatically, losing representativeness. We addressed this by devising computational algorithms to prevent sample numbers from decreasing, thereby enabling us to faithfully track phenomena across the entire cloud.
I understand you were not originally a meteorology specialist. What was the inspiration and catalyst for developing the Super-Droplet Method?
Shima: I obtained my doctorate in the field of nonlinear science, particularly researching the mathematics of synchronization phenomena. After that, I moved to the Japan Agency for Marine-Earth Science and Technology (JAMSTEC) knowing nothing about meteorology and joined a group developing new simulation models with future supercomputers in mind.
My supervisor at that time, Kanya Kusano (currently Professor Emeritus at Nagoya University), had the idea of applying the "Particle-In-Cell method (PIC method)" used in plasma physics to meteorology, and developing that into cloud simulation became my Super-Droplet Method. I also coined the name "super-droplet." In the PIC method they are called "super-particles," but I named them "super-droplets" to make them more intuitive as a meteorological cloud model.
The future of cumulonimbus simulation—understanding clouds and protecting society
What are your prospects for achieving Goal 8 and realizing a weather-controlled society?
Shima: There is still much we don't understand about clouds and weather phenomena. Our research may provide clues not only for "weather control intervention" but also for understanding how the "unintentional interventions that humanity has been making in the weather", such as aerosol emissions from industrial activities, have affected natural clouds and climate.
In the future, if we can control clouds and rain to some degree, it will not only reduce damage from heavy rainfall and floods but also help address drought and water resource management and adaptation to climate change. However, we need to carefully discuss ethical and social issues such as how far intervention should go and what the social impacts would be.
Clouds are transient, ephemeral entities
Understanding phenomena at the "particle" level of clouds might reveal the future of weather control, right?
Shima: Cumulonimbus clouds tower from the ground to high altitudes and are the most dynamic clouds, with many types of internal particles and intense changes. That is precisely why we need simulations to understand them hierarchically, from particles to the entire cloud.
Just as clouds consist of water droplets and ice particles, gases are also composed of molecules and atoms. However, clouds are not stable like gases. Their state changes moment by moment, and they fall to the ground as rain. Gases maintain their state, but clouds are transient—they are "ephemeral" entities that gradually disappear. How to describe and understand such clouds—that connects to my fundamental interest.