Modeling of surface conductance over sunn hemp by artificial neural network

Performances of an Artificial Neural Network (ANN), a multiple linear regression (MLR) and the Jarvis type model were compared to estimate the surface conductance which is a driving factor affecting evapotranspiration. It was modeled by ANN and MLR using various parameters including global solar radiation, temperature, soil water content, relative humidity, precipitation and irrigation, vapor pressure deficit, wind speed and leaf area index. The measurements were carried out during the growing season of sunn hemp in 2004. The best relationship (r²=0.73) between the surface conductance and all variables was estimated by the ANN when r² was 0.91 in the training period. The average absolute relative error was 26.54% for the ANN (r²=0.80), 51.07% for the MLR (r²=0.53) and 58.30% for Jarvis model (r²=0.26) when vapor pressure deficit, temperature, soil water content, global solar radiation and leaf area index were considered to model. The results showed that the ANN approach had a better modeling potential of the surface conductance compared to the MLR and Jarvis model.