Articles
USING SCALED GROWTH CURVES TO MAKE PREDICTIONS: PITFALLS AND SOLUTIONS
Article number
584_15
Pages
133 – 139
Language
English
Abstract
A common procedure for making predictions of harvest quantities (e.g. mean fruit size) is to scale a standard curve so that it passes through the mean of early-season observations, then project the scaled curve forward.
This procedure can work well late in the season, but early-season predictions can be misleading.
By using the fruit growth of two commercial cultivars of kiwifruit as examples, we show how standard curves, when combined with an appropriate scaling method, can be used to provide a series of improving predictions as the season progresses.
At the start of the season, before any fruit are measured, the best estimate available may simply be the historical average; whereas towards the end of the season the best estimate may come from the scaled standard curve alone.
We show how, by treating prediction as a matter of changing the weightings of these two estimates, we get a natural progression of improving predictions throughout the season.
Comparison is made with regression methods, showing that significant intercepts in linear regressions using early-season data correspond to placing substantial weightings on the historical average.
Our approach provides a natural way to interpolate between the usually discrete prediction times and parameters of the regression approach.
This procedure can work well late in the season, but early-season predictions can be misleading.
By using the fruit growth of two commercial cultivars of kiwifruit as examples, we show how standard curves, when combined with an appropriate scaling method, can be used to provide a series of improving predictions as the season progresses.
At the start of the season, before any fruit are measured, the best estimate available may simply be the historical average; whereas towards the end of the season the best estimate may come from the scaled standard curve alone.
We show how, by treating prediction as a matter of changing the weightings of these two estimates, we get a natural progression of improving predictions throughout the season.
Comparison is made with regression methods, showing that significant intercepts in linear regressions using early-season data correspond to placing substantial weightings on the historical average.
Our approach provides a natural way to interpolate between the usually discrete prediction times and parameters of the regression approach.
Publication
Authors
A.J. Hall, P.E.H. Minchin, W.P. Snelgar
Keywords
crop forecasting, kiwifruit, Actinidia deliciosa, Actinidia chinensis, fruit growth.
Online Articles (32)