Numerical Weather Prediction
The foundational method in traditional weather forecasting that discretizes the atmosphere into a three-dimensional grid and advances atmospheric states forward in time by solving systems of fluid dynamics and thermodynamics equations, requiring substantial computational resources and energy.
Core Facts
AI weather models train deep neural networks on massive historical meteorological reanalysis data, achieving inference speeds hundreds to thousands of times faster than traditional NWP.
The upper time limit for deterministic weather forecasting is generally considered to be approximately two weeks
Timeline (last 90 days)
A single global high-resolution seasonal simulation in numerical weather prediction (NWP) often consumes millions of CPU hours.
Traditional numerical weather prediction is based on physical equations and integrates step by step in a time-stepping manner; the time step is limited by the CFL condition, and at high resolution the step is smaller and the number of steps greater, so the cost grows superlinearly with resolution.
AI weather models train deep neural networks on massive historical meteorological reanalysis data, achieving inference speeds hundreds to thousands of times faster than traditional NWP.
Traditional numerical weather prediction (NWP) relies on systems of atmospheric physics equations solved by supercomputers, with computation often taking hours and horizontal resolution typically in the range of several to tens of kilometers.
Before around 2018, all weather forecasting models were based on derivations from physical equations.
Numerical weather prediction (NWP) discretizes the atmosphere into a three-dimensional grid and solves the Navier-Stokes equations along with thermodynamic, humidity, and other equations at each grid point.
A large network of ground-based GNSS receiver stations worldwide assimilates data in real time into numerical weather prediction models, improving precipitation forecast accuracy.
Globally evenly distributed, high-frequency barometric pressure data is one of the most sought-after inputs for current numerical weather prediction models, especially in observation-sparse regions such as oceans and polar areas.
Numerical Weather Prediction (NWP) typically has spatial resolutions on the order of several kilometers to tens of kilometers, with time update intervals of 6 hours or longer
Traditional weather forecasting relies primarily on large-scale numerical forecast models published by government meteorological agencies, which have broad coverage but limited update frequency and relatively coarse spatial resolution
5 more timeline events
All Facts (15)
AI weather models train deep neural networks on massive historical meteorological reanalysis data, achieving inference speeds hundreds to thousands of times faster than traditional NWP.
75%VerifiedThe upper time limit for deterministic weather forecasting is generally considered to be approximately two weeks
75%UnverifiedCompared to traditional numerical forecasting, AI weather models demonstrate two core advantages: higher computational efficiency (prediction speeds can be thousands of times faster than traditional methods) and comparable or even superior forecast accuracy.
70%UnverifiedA single global high-resolution seasonal simulation in numerical weather prediction (NWP) often consumes millions of CPU hours.
50%UnverifiedTraditional numerical weather prediction is based on physical equations and integrates step by step in a time-stepping manner; the time step is limited by the CFL condition, and at high resolution the step is smaller and the number of steps greater, so the cost grows superlinearly with resolution.
50%UnverifiedTraditional numerical weather prediction (NWP) relies on systems of atmospheric physics equations solved by supercomputers, with computation often taking hours and horizontal resolution typically in the range of several to tens of kilometers.
50%UnverifiedNumerical weather prediction (NWP) discretizes the atmosphere into a three-dimensional grid and solves the Navier-Stokes equations along with thermodynamic, humidity, and other equations at each grid point.
50%UnverifiedBefore around 2018, all weather forecasting models were based on derivations from physical equations.
50%UnverifiedA large network of ground-based GNSS receiver stations worldwide assimilates data in real time into numerical weather prediction models, improving precipitation forecast accuracy.
50%UnverifiedGlobally evenly distributed, high-frequency barometric pressure data is one of the most sought-after inputs for current numerical weather prediction models, especially in observation-sparse regions such as oceans and polar areas.
50%UnverifiedNumerical Weather Prediction (NWP) typically has spatial resolutions on the order of several kilometers to tens of kilometers, with time update intervals of 6 hours or longer
50%UnverifiedAI weather models do not need to solve systems of partial differential equations during inference, and are therefore orders of magnitude faster than traditional NWP
50%UnverifiedTraditional weather forecasting relies primarily on large-scale numerical forecast models published by government meteorological agencies, which have broad coverage but limited update frequency and relatively coarse spatial resolution
50%UnverifiedThe spatial resolution of global numerical weather prediction (NWP) products is typically 9–25 kilometers
50%UnverifiedTraditional Numerical Weather Prediction (NWP) was born in the 1950s and simulates the future state of the atmosphere by solving partial differential equations such as the Navier-Stokes equations, thermodynamic equations, and water vapor conservation equations.
50%