Ensemble Kalman–Guided Model Predictive Path Integral Control for Spatially Localized Suppression of Extremes in Chaotic Geophysical Flows

Haru Kuroki, Kazumune Hashimoto, Yuki Uehara, Yohei Sawada, Duc Le, Masashi Minamide2026 · DOI

Introduction

The possibility of influencing extreme weather phenomena has been a subject of scientific discussion for decades. However, operational implementation remains elusive due to the inherent chaos of geophysical flows—often referred to as the "butterfly effect." In such systems, a small, poorly targeted intervention can lead to unpredictable and potentially detrimental outcomes. To make weather modification a feasible reality, researchers need a robust mathematical framework capable of identifying the optimal location, timing, and magnitude of perturbations to steer these complex flows.

The Challenge of Controlling Chaos

Current control methodologies face a significant trade-off between computational efficiency and the ability to handle nonlinear dynamics.

  1. Ensemble Kalman Control (EnKC): This method is statistically efficient and scales well to high-dimensional systems. However, it relies on linearizing the underlying physics, which often fails during the highly nonlinear phases of extreme weather events.
  2. Model Predictive Path Integral (MPPI) Control: MPPI can handle full nonlinear physics through sampling-based optimization. The drawback is that it is essentially "blind" in high-dimensional spaces; without a starting point or "warm-start," it requires an impractical number of samples to find an effective intervention.

A Hybrid Approach: EKG-MPPI

To overcome these limitations, researchers have developed a hybrid framework called Ensemble Kalman–Guided Model Predictive Path Integral (EKG-MPPI) control. This method combines the statistical guidance of Ensemble Kalman filters with the nonlinear optimization power of Path Integral control.

The EKG-MPPI process involves several key stages:

  • State Estimation: An Ensemble Kalman Filter (EnKF) is used to estimate the current state of the geophysical system.
  • Linear Guidance: The control objective (e.g., "keep regional wind speeds below a certain threshold") is treated as a "pseudo-observation." EnKC then computes a candidate perturbation that would satisfy this goal in a linearized version of the model.
  • Adaptive Sparsity: An adaptive thresholding procedure is applied to the EnKC result to identify the most influential actuator locations. This ensures the intervention is spatially localized and physically realistic.
  • Guided Sampling: Instead of sampling randomly, MPPI uses the EnKC result to center its Gaussian proposal distribution. It then runs multiple nonlinear "rollouts" (simulations) to refine the perturbation without needing to calculate complex gradients.
  • Weighted Optimization: The final control input is a weighted average of these rollouts, favoring the trajectories that best suppressed the targeted extreme event.

Experimental Validation

The EKG-MPPI framework was tested using two standard benchmarks for chaotic systems:

  • Lorenz–96 Model: A 40-variable dynamical system frequently used to study atmospheric predictability and control.
  • Surface Quasi-Geostrophic (SQG) Model: A more complex 2D fluid dynamics model (simulated on a 128x128 grid) that represents mesoscale features such as atmospheric fronts and eddies.

In both cases, the hybrid method was compared against a "nature run" (uncontrolled simulation) and standalone EnKC.

Key Results and Insights

The numerical experiments yielded several significant findings:

  • Superior Suppression: In the Lorenz-96 experiments, EKG-MPPI reduced the frequency of extreme value exceedances more effectively than EnKC alone.
  • Input Efficiency: Crucially, EKG-MPPI achieved better suppression while often using a smaller maximum control input compared to EnKC (e.g., 1.037 vs. 1.311 in specific trials).
  • Effective Wind Control: In the SQG model, EKG-MPPI consistently maintained lower maximum wind speeds in target regions across various scenarios.
  • Sample Efficiency: By centering the search on a physically informed guess from EnKC, MPPI became orders of magnitude more sample-efficient, making it viable for higher-dimensional models.
  • Robustness to Hyperparameters: The study found that a smaller "temperature" parameter ($\lambda$) in MPPI led to more aggressive and effective updates by placing higher weights on the most successful samples.

Future Directions

While EKG-MPPI is a promising building block for weather intervention, several hurdles remain. The method is computationally intensive due to the requirement for parallel nonlinear simulations at every time step. Future research will focus on improving computational efficiency, testing the framework's robustness against model-nature mismatch (where the simulation model differs from reality), and incorporating hard physical constraints on the magnitude and type of allowed interventions.

Summary updated

Abstract

The possibility of influencing extreme weather phenomena has been discussed for decades; however, it remains far from operational practice, and there is still no established framework for designing small, spatially localized perturbations that can reliably steer chaotic geophysical flows. In this study, we propose a hybrid control method, termed ensemble-Kalmanguided model predictive path integral control (EKG-MPPI), which combines ensemble Kalman control (EnKC) with model predictive path integral (MPPI) control. Within a control simulation experiment framework, an ensemble Kalman filter is first used for state estimation, after which EnKC computes a candidate perturbation by treating the control objective as a pseudo-observation. An adaptive thresholding procedure then enforces spatial sparsity, so that the EnKC perturbation identifies candidate actuator locations and their nominal amplitudes. This information is embedded into the mean and covariance of Gaussian proposal distributions for MPPI, which subsequently refines the perturbation through sampling-based optimization with nonlinear rollouts, without linearizing the dynamics or computing gradients. Numerical experiments with the Lorenz–96 model and the surface quasi-geostrophic (SQG) model demonstrate that EKG-MPPI can suppress extremes in state variables and regional wind speed more effectively than EnKC alone, while using comparable or smaller control inputs. These results highlight EKG-MPPI as a promising building block for simulation-based assessment of localized intervention strategies in geophysical flows.
BibTeX
@misc{kuroki2026ensemble,
  author = {Haru Kuroki and Kazumune Hashimoto and Yuki Uehara and Yohei Sawada and Duc Le and Masashi Minamide},
  title = {{Ensemble Kalman–Guided Model Predictive Path Integral Control for Spatially Localized Suppression of Extremes in Chaotic Geophysical Flows}},
  year = {2026},
  doi = {10.5194/egusphere-2026-419},
}
Figure