Mean Squared Error of Yield Prediction by SOYGRO
- Josianne Colson,
- Daniel Wallach ,
- Andrée Bouniols,
- Jean-Baptiste Denis and
- James W. Jones
- S tation d'Agronomie
S tation de Biométrie et d'Intelligence Artificielle Institut National de la Recherche Agronomique (INRA), BP 27, 31326 Castanet-Tolosan Cédex, France
L aboratoire de Biométrie, INRA, Route de Saint-Cyr, 78026 Versailles Cédex, France
A gricultural Engineering Dep., Univ. of Florida, ,Gainesville FL 32611
Yield prediction is often one of the major intended uses of a crop simulation model. It is therefore important to evaluate how well a model performs as a predictor. The purpose of this study was to evaluate and analyze how well the SOYGRO model predicts soybean yield, using as a criterion the mean squared error of prediction (MSEP). The four target populations for prediction were irrigated or unirrigated plots at one location in France, for each of two varieties. The model parameters are estimated from the irrigated plots. The estimated MSEP values are on the order of 1(t ha−1)2 for all the target populations. For comparison, we defined an AVERAGE model. This model uses the average observed irrigated yield for each cultivar as the predictor of unobserved yields. AVERAGE was a better predictor than SOYGRO for the irrigated populations, while SOYGRO was better for the unirrigated populations. It seems that SOYGRO has sufficient built-in biological realism to extrapolate more reasonably than the AVERAGE model from irrigated to unirrigated conditions; however SOYGRO does not make as effective use of the data used for parameter estimation as does AVERAGE.
Copyright © . .