Harvest Plan's prediction widget

Thoughtpool Pty Ltd on LinkedIn

This widget is primed with real life samples of a Chardonnay block in the Eden Valley in 2004. It plots four types of predictions.

  • OPW (One Per Week): This rule of thumb assumes grapes ripen at one Baume per week, it requires adjustments with every sample.
  • XPW (X Per Week): This model uses the block's "best fit" of 1.43 Baume per week, it requires adjustments with every sample.
  • Allawah (variable speed): This model developed by Thoughtpool does not require adjustments with every sample.
  • Clayton (variable speed): This model developed by CSIRO does not require adjustments with every sample.

You can change, insert or remove samples on the left and click Redraw to see the effect.
You can also hide or show a model by clicking its name in the legend.
Hovering over a model name in the legend puts some emphasis on its graph line.
Hovering over a line shows the detail at this point in time.
There is also a menu icon on the upper right corner of the graph to download the graph.

Harvest-Plan now does not use linear models anymore as there is enough evidence of their bias.
The base model is now CSIRO's developed Clayton model in its "variety only" denomination.
The Clayton family of models will gradually be expanded to take into account other factors.

When data is available, the Allawah model is used.
The Allawah family of models will be expanded to be usable when less data is available.

In the case of this block, Allawah used eleven years of this block's samples to calculate a block specific equation and error margin. It will not change its mind easily; it is "confident" as it builds specific knowledge about this block.

Clayton by contrast does not use this block's historical samples, as it applies a general knowledge it acquires during the analysis of all historical samples.

When a model's line "breaks" to rejoin with the sample, it gives a different prediction date. In this case, Allawah only does that once with the second sample and then sticks to its date; whereas other models change their dates with every sample, overwhelming the planner with five times as many (and unnecessary) plan adjustments.

Select a variety Note: this widget works with measures in Baume