Use of multivariate approaches in biomass energy plantation harvesting: logistics advantages

Abstract

Agricultural biomass supply chain consisting of multiple harvesting, storage, pre-processing, and transport operations.  This network operates in space and time coordinates and produces empirical data used for many purposes, including wood-flow planning, harvesting cost calculation and work rate setting.  The aim of this study was to explore and propose the use of a multivariate approach, namely, the Partial Least Squares (PLS) multivariate regression approach and compare its performance with the commonly used Ordinary Linear Regression (OLS).  In particular, the study aimed at comparing the main statistical significance of indicators attributed to models calculated with OLS and PLS regressions from the same original datasets, for the purpose of quantifying the eventual improvement, obtained with the new techniques.  The dataset is composed by a series of measurements (harvesting distance, load carried, plantation production, numbers of plants harvested, and tractor engine power) conducted in a harvesting yard of a poplar plantation, to forecast the demanded working times.  The technical analysis was accompanied by economic scenarios, based on three hypothetical harvesting yards.  The results indicated that the PLS innovative approach is better performing; model error indicators are 5%-6% lower than those estimated with the OLS method.  From an economic point of view the harvesting cost per ton ranges among 8.69-14.59 € t^-1, 12.10-16.56 € t^-1 and 13.18-16.31 € t^-1 referring to the different load capacity of the trailers, using the PLS model.  Based on these results the differences between PLS and OLS varied up to 40 € ha^-1.  PLS modeling and more in general the advanced multivariate approach, are getting increasingly popular, because they are very robust and are particularly suitable for modeling complex systems.

Keywords: harvesting, biomass, logistics, machine costs, multivariate statistics, ordinary linear regression, partial least square