Related Experiment Videos
Predicting lettuce canopy photosynthesis with statistical and neural network models.
J Frick1, C Precetti, C A Mitchell
1National Aeronautics and Space Administration Specialized Center of Research and Training in Bioregenerative Life Support, Purdue University, West Lafayette, IN 47907-1165, USA.
Journal of the American Society for Horticultural Science. American Society for Horticultural Science
|September 7, 2001
Summary
A statistical model more accurately predicted canopy photosynthetic rates (Pn) in lettuce than an artificial neural network (NN). The statistical model achieved 12.3% average difference, while the NN had 24.6% difference in predictions.
Area of Science:
- Plant Physiology
- Computational Biology
- Horticultural Science
Background:
- Optimizing crop yield in controlled environments requires precise monitoring of plant physiological processes.
- Canopy photosynthetic rates (Pn) are critical indicators of plant health and productivity.
- Predictive modeling can aid in managing environmental factors for enhanced crop growth.
Purpose of the Study:
- To develop and compare artificial neural network (NN) and statistical regression models for predicting canopy photosynthetic rates (Pn) in hydroponically grown lettuce.
- To identify optimal environmental setpoints for maximizing Pn.
- To evaluate model accuracy for short-term and long-term predictions.
Main Methods:
- Developed a simple three-layer, fully connected artificial neural network (NN) and a third-order polynomial statistical regression model.
- Used shootzone CO2 concentration, photosynthetic photon flux (PPF), and canopy age as independent variables.
- Validated models using hydroponically grown 'Waldman's Green' leaf lettuce under controlled environmental conditions.
Main Results:
- The statistical regression model demonstrated higher accuracy in predicting canopy Pn compared to the NN model.
- Average percent difference between predicted and actual Pn was 12.3% for the statistical model and 24.6% for the NN model over an 11-day validation period.
- Both models exhibited reduced accuracy for long-range predictions (≥ 6 days into the future).
Conclusions:
- Statistical regression models can be more effective than simple NNs for predicting canopy photosynthetic rates in controlled environment agriculture.
- Accurate Pn prediction is feasible for short-term management but challenging for long-term forecasting.
- Further research may explore more complex NN architectures or hybrid models to improve long-range prediction accuracy.