Data‐Driven Exploration of Tropical Cyclone's Controllability

Yohei Sawada, Masashi Minamide, Yuyue Yan, Kazumune Hashimoto, Le DucGeophysical 2026 · DOI

TL;DR

This paper introduces a novel data-driven approach, Ensemble Kalman Control (EnKC), to identify small-scale atmospheric perturbations that can mitigate the intensity of tropical cyclones (TCs). By leveraging the chaotic nature of the atmosphere, EnKC successfully identified targeted water vapor reductions that weakened simulated TCs, offering a promising new avenue for exploring TC mitigation strategies.

What Problem Does This Paper Address?

Tropical cyclones (TCs) are powerful and destructive weather phenomena. While it's understood that the atmosphere's chaotic nature means small initial changes can lead to significant future differences, effectively controlling TCs has remained a challenge. Previous attempts at TC mitigation, like cloud seeding, have had limited success or relied on large-scale interventions. The core problem this paper addresses is the lack of data-driven methods to precisely identify where, when, and how to apply small, targeted interventions to effectively weaken TCs by exploiting their inherent chaotic dynamics. This is particularly difficult because the atmosphere is both sensitive to small changes (controllable) and inherently unpredictable.

Methods

The authors employ a data-driven, "top-down" approach using Ensemble Kalman Control (EnKC). This method builds upon the Ensemble Kalman Filter (EnKF) and Ensemble Kalman Smoother (EnKS) techniques.

Here's a breakdown:

  • EnKF: Used to assimilate observational data and create an ensemble of possible atmospheric states, providing an estimate of the current system state and its uncertainty.
  • EnKS: Extends the EnKF by performing a smoother analysis over a prediction horizon. This is used to estimate optimal control perturbations.
  • Control Target: A desired future state of the system (e.g., a minimum sea-level pressure) is defined.
  • Minimization Problem: EnKC formulates a problem to find the smallest perturbation to the initial state that will lead to the control target being met. This is analogous to Model Predictive Control (MPC).
  • Sparsity Enforcement: To make interventions more realistic, a thresholding function is applied to the calculated perturbations. This function enforces sparsity, meaning only perturbations with a sufficiently large signal-to-noise ratio are applied, effectively limiting the spatial scale and number of interventions. The hyperparameter $\lambda$ controls this sparsity.

The novelty lies in applying this EnKC framework to a realistic atmospheric model for TC simulation, moving beyond simpler "toy models" used in previous chaos control research.

Datasets

The paper uses a simulated dataset generated by the SCALE (Scalable Computing for Advanced Library and Environment) regional atmospheric model. This model was configured to simulate an idealized tropical cyclone. The "nature run" serves as the ground truth, and the ensemble members of EnKC are generated from perturbed initial conditions derived from this nature run. No external real-world observational datasets were used for training or evaluation in this specific study.

Key Insights and Ideas

  • Chaos as a Tool, Not Just a Hindrance: The paper demonstrates that the chaotic nature of TCs, often seen as a source of unpredictability, can be actively leveraged as a mechanism for control.
  • Targeted Water Vapor Reduction is Effective: The most effective interventions identified by EnKC involved reducing water vapor in specific areas around the TC's eyewall. This suggests that manipulating moisture content is a viable strategy.
  • Small-Scale Interventions Can Be Powerful: Unlike many previous TC modification studies that considered large-scale interventions, EnKC identified that relatively small spatial interventions (potentially less than 10³ km²) can be effective when adaptively updated over time.
  • Data-Driven Discovery: EnKC offers a powerful way to discover how to intervene without needing prior explicit knowledge of TC dynamics. The algorithm learns the optimal intervention strategy from the simulated data.
  • Sparsity Matters for Realism: The sparsity enforcement mechanism is crucial for making the proposed interventions more plausible and potentially feasible, as it avoids modifying the entire atmospheric state.

Key Results

  • EnKC successfully identified small-scale perturbations that mitigated TC intensity in simulations.
  • The most effective interventions involved reducing water vapor approximately 250 km from the TC center, which suppressed convective activity and latent heat release in the eyewall.
  • Experiments with smaller $\lambda$ values (less sparsity) resulted in broader interventions, while larger $\lambda$ values (more sparsity) led to more localized interventions.
  • Most tested $\lambda$ values resulted in a significant increase in minimum sea-level pressure (weakening of the storm), with substantial reductions in maximum wind speed.
  • The study showed that EnKC could efficiently identify perturbations that mitigate TC intensity, with most values of $\lambda$ achieving similar levels of weakening, except for $\lambda = 0.95$ which showed less consistent results.

Limitations and Future Work

  • Idealized Simulation: The study used an idealized TC simulation, not real-world data. Future work should validate these findings with more complex and realistic TC simulations and eventually with observational data.
  • Technological Feasibility: While the identified interventions are small-scale, the magnitude of water vapor reduction (around 25-50% reduction in surface evaporation) might still be substantial and require further investigation into technological feasibility.
  • Operational Challenges: The adaptive updating of intervention locations is operationally difficult and needs further exploration.
  • Robustness: The paper acknowledges that the probabilistic nature of the approach means the overall success rate of interventions needs to be assessed in future works to confirm robustness for real-world applications.
  • Model Resolution and Hyperparameters: The authors suggest that improving model spatial resolution, combining multiple intervention strategies, fine-tuning hyperparameters (like prediction horizon and control weights), and refining the algorithm for non-linear dynamics and realistic constraints could further enhance intervention efficiency.

Summary updated

Abstract

Abstract Although the chaotic nature of the atmosphere may enable efficient control of tropical cyclones (TCs) via small‐scale perturbations, few studies have proposed data‐driven optimization methods to identify such perturbations. Here, we apply the recently proposed Ensemble Kalman Control (EnKC) to a TC simulation. We show that EnKC finds small‐scale perturbations that mitigate TC. An EnKC‐estimated reduction in surface water vapor, located approximately 250 km from the TC center, suppresses convective activity and latent heat release in the eye wall, leading to a reduction of TC intensity. To advance the discovery of feasible TC mitigation strategies, we discuss the potential of this data‐driven method for leveraging chaos, as well as its remaining challenges.
BibTeX
@inproceedings{sawada2026datadriven,
  author = {Yohei Sawada and Masashi Minamide and Yuyue Yan and Kazumune Hashimoto and Le Duc},
  title = {{Data‐Driven Exploration of Tropical Cyclone's Controllability}},
  booktitle = {Geophysical},
  year = {2026},
  doi = {10.1029/2025gl120393},
}
Figure