MODELING YIELD PATTERN OF GREENHOUSE-GROWN SWEET PEPPERS

The irregular yield pattern of greenhouse-grown sweet peppers has been a challenge to researchers and greenhouse producers.
Experimental data of four years, each consisting of 26 production weeks, were used for time series analysis, neural network (NN) modeling, and regression analysis.
Time series analysis revealed that current weekly yield (Yd) was influenced by yields of one or two preceding weeks (Yd_1 and Yd_2, respectively), along with cumulative light two and four weeks prior (L_2 and L_4, respectively) and average 24-h air temperature five weeks prior (T_5). Cumulative light (L) was transformed into kL by dividing by 1000 for subsequent neural network (NN) modeling and regression analysis.
These five identified inputs were used to establish NN models, which illustrated the positive influence of Yd_1, kL_4 and kL_2, and negative influence of Yd_2 and T_5 on weekly yield.
In regression analysis, these same five inputs illustrated the positive influence of Yd_1 and kL_2 and negative influence of Yd_2. Each input was further modified to include its squared value before entering the regression analysis, which resulted in the significant inputs of Yd_1 (positive), squared Yd_1 (negative) and squared Yd_2 (negative). Among these three analyses, the most consistent influencing parameters were Yd_1 and Yd_2, suggesting that the irregular yield pattern of greenhouse-grown peppers is mostly influenced by these biological inputs.
Our results also indicated that kL_2, kL_4 and T_5 gave contributions in time series and NN models, but not in regression models.
The NN approach employed in this study supported our previous approach in using NN models for commercial yield prediction.

International Symposium on High Technology for Greenhouse Systems: GreenSys2009

W.C. Lin, D. Frey, G. Nigh, C. Ying

Capsicum annuum, neural networks, regression, time series

https://doi.org/10.17660/ActaHortic.2011.893.83