FORECASTING OF PHOTOSYNTHETICALLY ACTIVE RADIATION USING RECURRENT NEURAL NETWORK BASED LSTM ARCHITECTURE
12th Global Conference on Global Warming (GCGW-2024), Şanlıurfa, Türkiye, 16 - 19 Mayıs 2024, ss.261-264, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Basıldığı Şehir: Şanlıurfa
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.261-264
- Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
- İstanbul Üniversitesi-Cerrahpaşa Adresli: Evet
Özet
The escalation of greenhouse gases due to global warming has expedited the progression of the climate crisis through the heightened accumulation of atmospheric carbon dioxide. Photosynthetically active radiation (PAR) is important in applications related to plant physiology and the carbon cycle. PAR directly affects ecosystem productivity and carbon stock estimates. However, despite this importance, a global network has not yet been established for measuring PAR. Nevertheless, this phenomenon is being achieved more rapidly, with less labor, and at a lower cost using reanalysis data provided by satellite-based sources for use in projects related to climate change. The choice of architecture in deep learning based RNN architectures and the design of this architecture linearly affect the accuracy of forecasts conducted for the future. In this study, an RNN-based LSTM architecture was established on PAR data using the MERRA-2 reanalysis method, and forecasting were made. The accuracy of the architecture's predictions was measured by RMSE. The RMSE value of 0.0515 was obtained, indicating a successful forecasting process.