Abstract:The initial value problem weather prediction is a classical deterministic problem in classical physics,there are a lot of difficulties in the method of initial value problem to predict long range weather and short range climate.Three characteristics of atmospheric motion are very important:nonlinearity,being forced and dissipation. It is necessary to combine the dynamical and statistical methods.The prediction problem could be looked as the inverse problem of the initial value problem,i.e. an evolution problem,so the multiple time data in the past could be used to modify initial values and/or numerical prediction model.The inverse problems are divided into there kinds,there are corresponding solutions. After the inverse problem of the initial value problem is introduced,the global analysis to the atmospheric dynamical equations(ADE) are reviewed. The global analysis is another method of climate research different from the simple initial value problem. The global analysis is also called global asymptotic analysis.It is used to study the long range behavior of deterministic systems.It is unnecessary to solve the equations,the properties of ADE are researched by means of studying the equations themselves.With the aid of the direct geometric pictures(the concept of phase space is introduced),the qualitative global analysis to the limit solution set of ADE could be done.It is proved that ADE is a special operator equation in Hilbert Space.The properties of operators must be maintained when ADE is simplified or discretized.It is revealed that the nonlinear adaptation process of a system to external source,so the system could be described by a small number of degrees of freedom sustaining attractors because of dissipative process.Cell to cell mapping is a powerful tool to study nonlinear systems,by use of the concept and method of cell to cell mapping,the global analysis to the general characteristics of numerical model of atmospheric system could be done.The global analysis to round off errors and observational errors could be done.With the aid of the language of probability theory,the statistical description of the characteristics could be given.It is revealed that the indeterminacy of individual state and the determinacy of global state.The restrict mathematical definition of the vague concept of “CLIMATE”is:climate could be defined as probability distribution function of state variables in chaotic attractors determined by controlling variables,and it is proved that it exists chaotic state in climate system really.The approach of quantitative studies to the predictability of the atmosphere could be given,the predictability is the time before the state of a system arrive the probability distribution of chaotic attractors.

Learning to infer weather states using partial observations

Spatiotemporally Coherent Probabilistic Generation of Weather from Climate

Deriving Accurate Surface Meteorological States at Arbitrary Locations Via Observation-Guided Continuous Neural Field Modeling.

Data driven weather forecasts trained and initialised directly from observations

Generative Data Assimilation of Sparse Weather Station Observations at Kilometer Scales

A Study of Impact of the Geographic Dependence of Observing System on Parameter Estimation with an Intermediate Coupled Model

Insights on the role of accurate state estimation in coupled model parameter estimation by a conceptual climate model study

Generative emulation of weather forecast ensembles with diffusion models

Identifying Opportunities for Skillful Weather Prediction with Interpretable Neural Networks

Further Insights on the Role of Accurate State Estimation in Coupled Model Parameter Estimation by a Simple Climate Model Study

Estimating the State of Large Spatiotemporally Chaotic Systems

Ensemble forecasting: A foray of dynamics into the realm of statistics

Towards Representation Learning for Atmospheric Dynamics

Global atmospheric state analysis using objective Gaussian probability density functions

Improving seasonal forecast using probabilistic deep learning

SEEDS: Emulation of Weather Forecast Ensembles with Diffusion Models

Decomposing weather forecasting into advection and convection with neural networks

Towards Unsupervised Segmentation of Extreme Weather Events

Data Assimilation using ERA5, ASOS, and the U-STN model for Weather Forecasting over the UK

PROGRESS IN THE GLOBAL ANALYSIS TO THE ATMOSPHERIC DYNAMICAL EQUATIONS

FuXi‐En4DVar: An Assimilation System Based on Machine Learning Weather Forecasting Model Ensuring Physical Constraints