Dataset: On Generalized Additive Models for Representation of Solar EUV Irradiance
Authors/Creators
- 1. Michigan Tech Research Institute
- 2. University of Michigan
- 1. Michigan Tech Research Institute
- 2. University of Michigan
Description
This dataset includes software for representing solar EUV irradiance using Generalized Additive Models (GAMs). These GAMs represent solar EUV irradiance in multiple wavelength bands using smooth functions of three solar indices: the F10.7 solar index, revised sunspot number (SSN), and the Lyman-alpha solar index. The performance of the GAMs are fit in two steps: (1) GAMs are fit between the solar indices and outputs of the FISM2 empirical solar EUV model in each band (Yi), and (2) GAMs are fit between the residuals of Yi and the native measurements of FISM2 in each band (\(\zeta_i\)). The resulting GAMs \(F_i=Y_i-\zeta_i\) are used to model solar EUV irradiance in Solar Cycle 24. The GAMs are driven with known historical inputs (models \(F^K_i\)) and inputs hindcasted with (a) an autoregressive model and (b) the novel Dynamic Superposed Epoch Analysis (DSEA) technique (models \(F^P_i\)). The performance of models \(F^K_i\), \(F^P_i\), and uncalibrated FISM2 estimates are evaluated in terms of relative error, skew normal distribution parameters, and assessed for dependency on solar activity and season.
All of the code for this project is written in Python 3.8.
Notes (English)
This dataset is composed of several .zip files that must be extracted before any code is run. The first set of files have 'indexmodeling' as their prefixes, and their contents must all be extracted to a single location. Within the data of that set, 'solarPrepare.py' must first be run: it loads in all of the irradiance data, removes bad values, and creates a training and testing set. 'solarEval.py' is then run, which performs the fitting of the GAMs and evaluates the outputs, generating a series of figures.
'indexPrediction' should be extracted separately, to a file whose location is at the same level as the file containing the contents of all of the 'indexmodeling' folders. Within this folder, 'predict.py' is what should be run, and it will perform autoregressive modeling of irradiance during two periods of time in SC24 (a 30-day period in solar minimum and a 30-day period in solar maximum), and it will also perform 'dynamic' superposed epoch analysis to hindcast SC24 from preceding cycles.