CROP MODELLING AND YIELD PREDICTION FOR GREENHOUSE-GROWN LETTUCE

Abstract

Production of greenhouse lettuce requires high energy input and high capital cost. Modelling optimal growth and yield was used to determine factors to improve production efficiency. Year-round light levels are very variable in northern latitude areas. In winter, when natural daylight is short and light levels are low, supplementary lighting can be used. In summer, when high light intensity prevails, shading can be used. The market requires a certain minimal fresh weight per living lettuce plant, which is sold with its root. Artificial neural networks (ANN) to model the final fresh weight were investigated. Daily light, temperature, and relative humidity data were obtained with a greenhouse environment control computer. A video camera was used to capture a daily or hourly image of the plants to determine percentage canopy cover. Only weekly data were used in this study. Plant (fresh) weight at harvest at a specific age (days after transplanting to final space) was modelled by using these weekly data. As the crop grew, the most recent weekly data were added to predict final fresh weight. Eight lighting experiments were conducted over 18 months and generated 64 data points (cases) for modelling. Using ANN, the final weight can be predicted either by weekly average light, weekly percent canopy cover, or both, with similar levels of error. The predicted plant weight was close to observed values, with an average difference of about 10 to 11 %, although the root mean square (RMS) errors were high. The RMS errors can be reduced by grouping inputs of light so that these 64 cases can be increased to 380 cases. The resulting ANN models permit simulation by end-users to understand the interrelationship between final plant weight, crop age, and daily light level. Preliminary results indicated that the growth models can be used as a guide to avoid unnecessary supplementary lighting.