Dataset for "A spatiotemporal feature-guided AI model for Medium-Scale Traveling Ionospheric Disturbances Nowcasting"

This dataset provides image-sequence samples for nowcasting Medium-Scale Traveling Ionospheric Disturbances (MSTIDs), built from high-resolution Detrended Total Electron Content (DTEC) Map products of the Japanese GEONET GNSS network. Each sample is a short DTEC-map sequence — 9 consecutive input frames (40 min of history) followed by 4 target frames (the next 20 min) — at a temporal resolution of 5 minutes and a spatial resolution of 0.15° × 0.15°. Every frame is a 160 × 160 grid covering 24°N–48°N latitude and 124°E–148°E longitude over Japan, stored as a comma-separated CSV file of detrended TEC values in TECU. Cells without GNSS coverage (sea areas and grid edges) are marked as NaN. The dataset merges MSTID events from three years spanning diverse solar-activity conditions — 2020 (solar minimum), 2022 (moderate), and 2024 (solar maximum) — into 2301 samples, split per event into 1835 training, 221 validation, and 245 test samples (approximately 8:1:1). Splitting is applied at the event level so that all frames of an event stay within the same split, with no temporal leakage. MSTID events were identified from the raw DTEC Maps using an AI detection and feature-matching procedure (YOLO-based detection followed by wave-pattern continuity checks); events lasting at least one hour (13 consecutive frames) were retained, and a 5-minute sliding window produced the 9-to-4 input/target samples. Contents (top-level folder "dataset/"):- train/, val/, test/, each containing input/ (9 frames) and target/ (4 frames)- tid_{train,val,test}_sample.csv : per-split sample manifests- global_stats.txt : global mean/std (TECU) for z-score normalization- README.md : full format specification, naming convention, and a Python loader Frame files are named tid_<sample_id>_t<k>_<HH_MM>.csv, where k = 0–8 are input frames and k = 9–12 are target frames, and HH_MM is the UT timestamp. This dataset accompanies the paper "A spatiotemporal feature-guided AI model for Medium-Scale Traveling Ionospheric Disturbances Nowcasting." The underlying DTEC Map product is provided by GEONET / NICT (https://aer-nc-web.nict.go.jp/GPS/GEONET/MAP/). Please cite the paper and acknowledge the GEONET / NICT DTEC Map product when using this dataset.