← Back to Tools-Radar

MetNet: A Neural Weather Model for Precipitation Forecasting

Categories: Data Analysis, Research  |  Pricing: Freemium  |  Official Website ↗

MetNet is a deep neural network that predicts future precipitation at 1 km resolution over 2-minute intervals for up to 8 hours.

MetNet is a deep neural network developed by Google Research for precipitation forecasting. Unlike traditional physics-based models, MetNet learns directly from observed data using backpropagation, identifying patterns and complex transformations from inputs to desired outputs. It processes data from ground-based radar stations (MRMS) and NOAA's Geostationary Operational Environmental Satellite system, covering the continental US with image-like inputs. The model is designed to predict precipitation at 1 km resolution over 2-minute intervals for up to 8 hours ahead. It utilizes a spatial downsampler to manage memory, a convolutional LSTM for temporal encoding of past data snapshots, and axial self-attention for capturing long-range spatial dependencies. MetNet outputs a discrete probability distribution for precipitation rates, allowing for the inference of most likely rates with associated uncertainties. It has demonstrated superior performance compared to NOAA's HRRR system for predictions up to 7-8 hours ahead, making predictions for the entire US in seconds.

Key Features

• 8-hour precipitation forecasting
• 1 km resolution predictions
• 2-minute interval predictions
• Utilizes radar and satellite data
• Probabilistic output with uncertainty
• Faster prediction time (seconds for entire US)
• Outperforms physics-based models for short-term forecasts

Pros

• Predicts precipitation up to 8 hours ahead with high resolution.
• Processes data from radar and satellite networks automatically.
• Generates predictions for the entire US in seconds, unlike hours for physical models.
• Outperforms state-of-the-art physics-based models for 7-8 hour forecasts.
• Provides probabilistic predictions, accounting for atmospheric uncertainty.

Cons

• Uncertainty in predictions increases with longer forecast times.
• Predictions can appear smoother than ground truth due to probabilistic nature.
• Requires adequate radar and optical satellite data for regional extension.
• Not a commercial product, but a research model.
• Details on accessibility or direct use by external parties are not provided.

Use Cases

• Short-term weather prediction
• Precipitation nowcasting
• Climate impact studies
• Disaster preparedness planning

Best For

• Meteorological agencies
• Climate researchers
• Environmental scientists
• Organizations requiring short-term precipitation forecasts

Platforms: Web

Watch demo on YouTube ↗

View full MetNet: A Neural Weather Model for Precipitation Forecasting profile on Tools-Radar | Browse Data Analysis tools | Alternatives to MetNet: A Neural Weather Model for Precipitation Forecasting

Tools-Radar is a free directory of 10,000+ AI tools — discover, compare, and choose the right AI software for your needs. Visit tools-radar.com