Estimation of hourly global photosynthetically active radiation using artificial neural network models

Photosynthetically active radiation (PAR) reaching the earth's surface is a major parameter controlling many biological and physical processes related with the evolution of plant canopies, agricultural and environmental field s. Unfortunately, PAR is not often measured and therefore it must be estima ted. The unavailability of measurements of global solar radiation at the pl ace of interest and different factors affecting the linear relation between PAR and global solar radiation can preclude the estimation of PAR from glo bal solar radiation. In this paper, a novel approach based on a simple mult ilayered feedforward perceptron has been used to analyse the non linear rel ationships between PAR and different meteorological and radiometric variabl es in order to determine their relative relevance. An artificial neural net work based model for the estimation of the hourly PAR involving hourly glob al irradiance as only measured variable has been successfully developed. Th e model was tested using data recorded at six radiometric stations covering a wide range of climates. The model performance has been compared with oth er existing empirical complex models showing important improvements. Next, a second artificial neural network based model involving only sunshine dura tion measurements has been developed and proved to be an acceptable alterna tive to calculate hourly PAR when radiometric information is not available. (C) 2001 Elsevier Science B.V. All rights reserved.