arXiv subjects

Wansuo Duan

Publications and source records attributed to Wansuo Duan.

5 recordsLinked to original sources

arXiv2026

A Synergistic Approach: Dynamics-AI Ensemble in Tropical Cyclone Forecasting

This study addresses a critical challenge in AI-based weather forecasting by developing an AI-driven optimized ensemble forecast system using Orthogonal Conditional Nonlinear Optimal Perturbations (O-CNOPs). The system bridges the gap between computational efficiency and dynamic consistency in tropical cyclone (TC) forecasting. Unlike conventional ensembles limited by computational costs or AI ensembles constrained by inadequate perturbation methods, O-CNOPs generate dynamically optimized perturbations that capture fast-growing errors of FuXi model while maintaining plausibility. The key innovation lies in producing orthogonal perturbations that respect FuXi nonlinear dynamics, yielding structures reflecting dominant dynamical controls and physically interpretable probabilistic forecasts. Demonstrating superior deterministic and probabilistic skills over the operational Integrated Forecasting System Ensemble Prediction System, this work establishes a new paradigm combining AI computational advantages with rigorous dynamical constraints. Success in TC track forecasting paves the way for reliable ensemble forecasts of other high-impact weather systems, marking a major step toward operational AI-based ensemble forecasting.

physics.ao-ph↗

arXiv2026

Long-term prediction of ENSO with physics-guided Deep Echo State Networks

The El Niño-Southern Oscillation (ENSO) is a dominant mode of interannual climate variability, yet the mechanisms limiting its long-lead predictability remain unclear. Here we develop a physics-guided Deep Echo State Network (DESN) that operates on physically interpretable climate modes selected from the extended recharge oscillator (XRO) framework. DESN achieves skillful Niño3.4 predictions up to 16-20 months ahead with minimal computational cost. Mechanistic experiments show that extended predictability arises from nonlinear coupling between warm water volume and inter-basin climate modes. Error-growth analysis further indicates a finite ENSO predictability horizon of approximately 30 months. These results demonstrate that physics-guided reservoir computing provides an efficient and interpretable framework for diagnosing and predicting ENSO at long lead times.

physics.ao-ph↗

arXiv2024

A note on the role of the initial error structure in the tropics on the seasonal-to-decadal forecasting skill in the extratropics

The predictability of a coupled system composed by a coupled reduced-order extratropical ocean-atmosphere model forced by a low-order 3-variable tropical recharge-discharge model, is explored with emphasis on the long term forecasting capabilities. Highly idealized ensemble forecasts are produced taking into account the uncertainties in the initial states of the system, with a specific attention to the structure of the initial errors in the tropical model. Three main types of experiments are explored with random perturbations along the three Lyapunov vectors of the tropical model, along the two dominant Lyapunov vectors, and along the first Lyapunov vector, only. When perturbations are introduced along all vectors, forecasting biases are developing even if in a perfect model framework. Theses biases are considerably reduced only when the perturbations are introduced along the dominant Lyapunov vector. This perturbation strategy allows furthermore for getting a reduced mean square error at long lead times of a few years, and to get reliable ensemble forecasts on the whole time range. These very counterintuitive findings further underline the importance of appropriately control the initial error structure in the tropics through data assimilation.

nlin.CD↗

arXiv2020

The Deep Learning Galerkin Method for the General Stokes Equations

The finite element method, finite difference method, finite volume method and spectral method have achieved great success in solving partial differential equations. However, the high accuracy of traditional numerical methods is at the cost of high efficiency. Especially in the face of high-dimensional problems, the traditional numerical methods are often not feasible in the subdivision of high-dimensional meshes and the differentiability and integrability of high-order terms. In deep learning, neural network can deal with high-dimensional problems by adding the number of layers or expanding the number of neurons. Compared with traditional numerical methods, it has great advantages. In this article, we consider the Deep Galerkin Method (DGM) for solving the general Stokes equations by using deep neural network without generating mesh grid. The DGM can reduce the computational complexity and achieve the competitive results. Here, depending on the L2 error we construct the objective function to control the performance of the approximation solution. Then, we prove the convergence of the objective function and the convergence of the neural network to the exact solution. Finally, the effectiveness of the proposed framework is demonstrated through some numerical experiments.

math.NA↗

arXiv2020

On the use of near-neutral Backward Lyapunov Vectors to get reliable ensemble forecasts in coupled ocean-atmosphere systems

The use of coupled Backward Lyapunov Vectors (BLV) for ensemble forecast is demonstrated in a coupled ocean-atmosphere system of reduced order, the Modular Arbitrary Order Ocean-Atmosphere Model (MAOOAM). It is found that overall the best set of BLVs to initialize a (multiscale) coupled ocean-atmosphere forecasting system are the ones associated with near-neutral or slightly negative Lyapunov exponents. This unexpected result is related to the fact that these sets display larger projections on the ocean variables than the others, leading to an appropriate spread for the ocean, and at the same time a rapid transfer of these errors toward the most unstable BLVs affecting predominantly the atmosphere is experienced. The latter dynamics is a natural property of any generic perturbation in nonlinear chaotic dynamical systems, allowing for a reliable spread with the atmosphere too. Furthermore, this specific choice becomes even more crucial when the goal is the forecasting of low-frequency variability at annual and decadal time scales. The implications of these results for operational ensemble forecasts in coupled ocean-atmosphere systems are briefly discussed.

physics.ao-ph↗