Synthetic Regional Coverage Dataset for Satellite Constellation Surrogate Modelling over Europe (500 Designs, SGP4-Labelled)

Synthetic training dataset for a machine learning surrogate that predicts regional satellite constellation coverage. It contains 500 constellation designs from four sampling groups (Walker, repeat ground track, 2-D lattice flower, Molniya), each sampled as one complete Latin hypercube. Because the lattice-flower designs coincide with Walker patterns, the four groups give three geometry types. Each design is labelled on a 1° grid over Europe (35–72°N, 10°W–40°E, 1,938 cells) using SGP4 propagation with a 10° elevation mask over a 30-day window at a 2-minute step. Each row is one design and one grid cell, giving 969,000 rows with 24 columns: the design parameters and three labels, the coverage percentage, the mean gap and the maximum gap (minutes). Parameters that do not apply to a group are left empty. A cell with no access in the window has a mean gap and a maximum gap of 43,200 minutes. Changes in version 2: the repeat-ground-track designs use the J2 repeat condition after Shin et al. (2021), are placed over a target point in Europe and share one common track; their tracks close to within 1.4 × 10⁻⁵ degrees per repeat cycle. The Molniya apogees are placed over the region and stay there. The Walker and lattice-flower phasing is part of the Latin hypercube. New columns: target_lat_deg, target_lon_deg and closure_residual_deg. Version 1 (training_data1.csv) belongs to the first draft of the thesis and is no longer used. File: training_data2.csv, MD5 5f71768e61e06161380e3729b41d2bad. Generated with generate_cells.py, commit d28c809 of https://github.com/JuliusPinsker/surrogate-pipeline-cubesats.