Photosynthetic Active Radiation Forecast via Artificial Neural Network, Support Vector Regression and Multiple Linear Regression

Micrometeorological studies in urban and rural areas are required photosynthetically active radiation (PAR) data and its dependence on different sky conditions. Most of the biological and physiological processes are controlled by this variable. Although PAR is of great importance for these processes, it is not continuously measured variable. In this study, temporal PAR, PAR fraction (PAR/Global solar radiation (R_g)) is determined over two vegetation surfaces for two years from hourly measurements in the Kırklareli city where locates northwestern part of Turkey. Relationships between PAR and R_g, clearness index (K_t) and water vapor absorption (w) in visible band and dew point temperature (T_d) are investigated.

To predict PAR values, some powerful methods such as Artificial Neural Network (ANN), Support Vector Regression (SVR) and Multiple Linear Regression (MLR) is implemented separately for univariate and bivariate forecast. In addition to these methods, Chaotic Approach (CA) is used to prepare the input data matrix because SVR that is a machine learning algorithm requires a special input matrix. Most of the results are very promising and all model performance criteria show that SVR is the best method for forecast of PAR.