MODELING UNCERTAINTY OF GREENHOUSE CROP LETTUCE GROWTH MODEL USING KALMAN FILTERING

The NICOLET model was developed to predict biomass growth and nitrate content of a greenhouse lettuce crop. However, several studies have shown that accu-rate prediction of nitrate concentration is rather difficult, due to cultivar differences and its sensitivity to changes of the environment. The present work explores possibilities of improving the prediction performance of the NICOLET model by incorporating information (data assimilation) coming from samples of destructive measurements of an actual crop. A lettuce crop was grown from March to May 2008 in a hydroponic system inside an unheated plastic greenhouse located at the University of Queretaro, Mexico. Air temperature and humidity, solar radiation and also carbon dioxide inside the greenhouse were measured at the center of greenhouse and recorded in a data logger every 5 min. At regular intervals a variable number of lettuce plants were harvested in order to measure both fresh matter and dry weight per plant, leaves nitrate content and plant leaf area index. The parameters of the NICOLET model were estimated with the mean data coming from the sampled plants. The unscented Kalman filter (UKF) was selected as assimilation method. Simulation results were improved by using the measured data on a different data set that the one used for par¬ameter estimation. The filters estimated properly the state variables, thus significantly improving the model fitting in comparison with the simulation without data assimila¬tion. The results suggest that Kalman filtering is a suitable method to provide the required automatic adaptation to time-varying phenomena in complex crop models.