Machine learning model for adaptive Geostationary Lightning Mapper parallax correction: Model and demonstration

Resource information

Format

ipynb, py

Original repository

18842183

DOI

10.5281/zenodo.18842183

Credit basis

Creators in Zenodo record metadata

Access

Open files available from the original Zenodo deposit

Last checked

Nov 5, 2026

License

cc-by-4.0

Last record update

Nov 5, 2026

About this resource

A Python-based Gradient Boosted Classifier model for adaptive parallax correction of Geostationary Lightning Mapper (GLM) group-level data is provided as a serialized binary file created using Python's pickle module (Python, 2026), named unified_lightning_gbc_model.p. The model is demonstrated in an included Jupyter notebook file (Beg et al., 2021), named demo_adaptive_parallax_correction.ipynb, written to download and use sample Level-2 GLM data from GOES-East on 17 July 2024 between 2346 and 2347 UTC (GOES-R Algorithm Working Group and GOES-R Series Program Office, 2018a). Supporting GOES-East Advanced Baseline Imager (ABI) Cloud Top Height (CTH) data (GOES-R Algorithm Working Group and GOES-R Series Program Office, 2018b) are used as well. GLM and ABI data are obtained for the demonstration from Amazon Web Services (AWS) through the NOAA GOES registry (https://registry.opendata.aws/noaa-goes/) where they are generally publicly available. Synthetic ground-based network lightning detection data are provided as well in the serialized binary file, 20240717_2346_demo_network_detections.p. These synthetic data are based on the lightning stroke dataset from the Vaisala U.S. National Lightning Detection Network (NLDN, https://www.vaisala.com/en/products/national-lightning-detection-network-nldn; Murphy et al., 2021) and the lightning flash dataset from the Earth Networks Total Lightning Network (ENTLN, https://www.earthnetworks.com/why-us/networks/lightning/; Zhu et al., 2022). The model workflow relies on the Bruning (2019) glmtools Python package and supporting script, make_GLM_grids.py, for conversion of Level-2 GLM data to imagery data over the conterminous United States (CONUS) sector (Bruning et al., 2019); the Fuchs et al. (2016) lmatools Python package for calculations; as well as the Virts and Koshak (2020) optimal emitter height model, the most recent version of which may be obtained from https://www.nsstc.uah.edu/data/katrina.virts/GLM/parallax/ . A minimal Python environment file to enable the demonstration notebook is also included, named adaptive_parallax_correction.yml. The demonstration and environment were tested using Python 3.10. The adaptive parallax correction model takes as 14 predictors the following information pertaining to GLM imagery observations: the Julian day number, month number, and solar hour of observation; the area of the group footprint and the length of the group footprint, calculated as the square root of the area; the lag time between the ABI and GLM observation; the minimum, maximum, median, and range of ABI CTH height within the GLM lightning footprint; the difference between median ABI CTH within the GLM lightning footprint and its corresponding optimal emitter height; the distance and bearing from satellite nadir of the GLM lightning footprint; and the distance ratio of the GLM lightning footprint, calculated as the distance between the median ABI CTH and optimal emitter height of the GLM lightning footprint dividied by its length. The resulting classifications are either 0="neutral, 1="VK best", or 2="ABI best." In the neutral and ABI best cases, the median ABI optimal emitter height should be used to parallax correct the GLM lightning footprint location. In the VK best case or in the absence of ABI data, the optimal emitter height should be used to parallax correct the GLM lightning footprint location. GLM lightning footprints can be either groups or flashes, where parallax correction of GLM event