sw_geo_10min_2000_2024_scaled

This dataset accompanies the MSc thesis "Predicting Auroral Electrojet Index using Deep Learning" (Astrid Pettersen, UiT – The Arctic University of Norway, 2026). It was compiled and preprocessed by Andreas Kvammen and is provided here in scaled form, exactly as used by the forecasting models in the thesis. The file is a single time series spanning 2000-01-01 to 2024-12-31 at a fixed 10-minute cadence (1,315,003 rows). It contains 19 variables plus a DateTime index: Solar wind / IMF (from NASA OMNIWeb; in-situ ACE and Wind measurements at L1, propagated to Earth's bow shock): IMF By and Bz, solar wind speed (vsw), proton density (nsw). Solar activity and geomagnetic indices (from NASA OMNIWeb): F10.7 (f107), Dst, ap, and the auroral electrojet indices AE, AL, AU. Polar Cap North index (PCN; from DTU Space). Regional electrojet indices IEL, IEU, IEE (from Tromsø Geophysical Observatory, derived from the IMAGE magnetometer network in Scandinavia). Deterministic time features: sine/cosine of day-of-year (DOY_sin, DOY_cos) and time-of-day (TOD_sin, TOD_cos), and solar zenith angle (SZ), computed for a fixed reference point (67°N, 15°E) as a proxy for local ionospheric conductivity. Scaling. The observational variables are robust-scaled: x_scaled = (x − median) / IQR, with the median and IQR computed over the full 2000–2024 series prior to the train/test split. The scaling parameters (median, IQR, physical units) are provided alongside the data, and code to reverse the scaling is in the accompanying repository. The cyclic time features and SZ are bounded deterministic encodings and are not robust-scaled. Gaps. Missing-value gaps of two hours or less in the solar wind and PCN columns were filled by linear interpolation during preprocessing; gaps longer than two hours were left unfilled and remain as NaNs. Note on 2020. During 2020 the OMNIWeb AE/AL/AU values were missing and were replaced in preprocessing using the SuperMAG electrojet indices (SME/SML/SMU) as a proxy. These SuperMAG indices are not stored as separate columns.