ISHS Acta Horticulturae 718: III International Symposium on Models for Plant Growth, Environmental Control and Farm Management in Protected Cultivation (HortiModel 2006)
UNCERTAINTY OF PREDICTED GREENHOUSE TOMATO PRODUCTION: MECHANISTIC VERSUS STOCHASTIC MODELS? |
| Authors: | E. Schrevens, C. Bojaca, A. Cooman |
| Keywords: | tomato yield prediction, TOMGRO2, General Linear Mixed Models, uncertainty propagation |
| Abstract:
In the first part of this paper an overview is given of the possibilities of the TOMGRO2 model for tomato yield predictions, emphasizing on the prediction of both the mean and the variance of the production at any point in time. The variance prediction is based on uncertainty propagation. The highest uncertainty on the simulated total fruit dry weight was caused by photosynthesis parameters. The uncertainty caused by the parameters related to dry matter distribution on the estimation of fruit weight was small. The development-parameters had nearly no effect on the uncertainty of the estimation of fruit weight. The initial conditions or characteristics of the planting material caused little uncertainty. To estimate the uncertainty caused by climate data, solar radiation, air temperature and CO2 were varied in a range that can occur in greenhouse conditions along to a 23 factorial design. The uncertainty of the model to the three climate variables was very similar. Over the space of uncertainty inducing parameters on yield predictions, a Monte Carlo simulation to estimate overall uncertainty was carried out and the corresponding variance structure was computed for selected harvest data. In the second part, a classical General Linear Mixed Model is calculated on the calibration sets. This model is able to accurately fit the time evolution of the mean as well as the heterogeneity of the variance structure over harvest times. For the mean yield prediction a first order linear model is elaborated, but extensions to higher order models are straightforward. Between-plant variability per harvesting time is estimated through Restricted Maximum Likelihood as a random factor, nested in time. Predictions of mean and variance of fruit DW between TOMGRO2 and Mixed Models are compared. Root Mean Square Errors against observed data are calculated as goodness of fit statistic. | |
Download Adobe Acrobat Reader (free software to read PDF files) | |
