SHARP: An Online, Real-Time Sugarcane Harvest Prediction System based on Crop Growth Simulations and PLS Regression

Pierre Todoroff, Mickaël Mezino, Lionel Le Mézo, and Jean-Baptiste Laurent

Keywords

Sugarcane, Harvest prediction, Crop growth model, Partial least squares, web, Information system

Abstract

This paper presents a sugarcane harvest prediction system at the level of a sugarcane growing area based on simulations of a semi-mechanistic daily time-step crop growth model and a multivariate regression. The regression’s predictors are the output values of the model’s simulations, and the predicted variable is the yield translated into production by multiplication with the cultivated area. The regression is computed by partial least squares regression on historical observed yield values. All the components of the prediction system are embedded in an open source web information system, connected to a network of automatic weather stations. The data needed by the crop growth model and the regression procedure are automatically extracted from the information system databases. The calculations are triggered and displayed on a web user interface. We present the performance of this prediction system applied in Reunion Island on the five sugarcane growing areas with 14 years of production history. We show that it is more accurate than the conventional sugarcane sampling-based prediction method, with less than 5% error at the entire island level, and provides earlier reliable predictions. The SHARP system is cost effective and designed to be used as a turnkey tool for the sugarcane industry decision makers.

Important Links:



Go Back