Machine-Learning Based Lightning Nowcasting Data Archive

A dataset supporting a preprint on lightning nowcasting using machine learning. It contains prediction results from multiple models, including one informed by aerosol observations and satellite data, and training datasets for model development. The data was produced by researchers at Wuhan University and is associated with a 2023 preprint.

Use Cases

• Training machine learning models for lightning prediction based on aerosol and meteorological data.
• Comparing model performance between aerosol-informed and baseline models based on Probability of Detection (POD) and False Alarm Ratio (FAR).
• Analyzing the relationship between PM2.5 concentrations and lightning activity based on the included real-time PM2.5 dataset.

Strengths

• Includes results from three distinct model variants, allowing for comparative analysis.
• Contains a spatially continuous, hourly-level PM2.5 dataset derived from EPA AQS monitoring network data.
• Provides training datasets for public use in machine learning development.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Wuhan University

Collection Method

Derived data from machine learning models using satellite, meteorological, and ground-based observations.

Time Range

The demo uses data from the year 2020.

Geography

Likely focuses on the United States, given the use of the EPA AQS network.