Estimation of sunshine duration by using satellite data and geographic parameters over Turkey

In this study, support vector machine (SVM) method was applied to estimate the daily and monthly sunshine duration (SD) over Turkey. Firstly, daily and monthly datasets of atmospheric and geographical parameters which were assumed to affect the SD were constructed. These datasets were composed of 27 daily and 26 monthly different parameters which includes the thermal channels of the geostationary orbiting Meteosat SEVIRI sensor, the solar channels of the polar orbiting Suomi-NPP VIIRS sensor and extra-terrestrial solar radiation, day length, sun hour angle, declination angle, julien day number, location (latitude, longitude, altitude) and time (year, month, day). Inputs of daily and monthly models were determined by using recursive feature elimination (RFE) algorithm which selects the most suitable parameters among the dataset. Using the most appropriate function and its parameters of SVM models, SD estimations were performed for the considered pixels. Errors were calculated for the training and test phases and results were analysed. Finally, the daily, monthly and yearly SD maps of Turkey were created with a 600 meters spatial resolution.