Identifying Best Practice Management Strategies in the Australian Dairy Industry using Partial Least Squares Models

Abstract

Abstract The Australian dairy industry is undergoing significant change in order to ensure an industry which is not only sustainable on a short-term basis, but which also offers environmental sustainability in the long term. Dairy farmers in Australia are under increasing pressure to demonstrate that their production systems are environmentally friendly (Armstrong et al., 2000). They need assistance in evaluating the impact of a range of alternative management practices on profitability.

In 2000, the Dairy Research and Development Corporation conducted a large-scale telephone survey of 1826 Australian dairy farms to examine current on-farm natural resource management practices. Full details of the survey and initial analysis can be found in DRDC (2001). Farms were selected at random from eight major geographical regions, each of which has specific climatic and geological conditions. The importance of climate and geological conditions on milk production has been well documented by authors such as Kerr et al (1995), Armstrong et al. (2000) and Herrero et al. (2000). All analyses in the current study were carried out separately for each region. The survey aimed to identify environmentally sound, best management practices (EBMP), to quantify their relationships with milk production, and to assess relationships between individual EBMPs and their effects on milk production.

The aim of the research reported here was twofold: firstly, to identify the EBNPs and their relationships with each other and with milk production, and secondly to demonstrate to the Australian dairy industry the value of multivariate techniques in farm management decision making. An important, and often overlooked, component of the latter aim is the ease with which interpretation can be conveyed to a dairy farm manager. One of the multivariate techniques used was Partial Least Squares Regression (PLSR), and the use of this method as a management tool for dairy farmers is the subject of this paper.

The response variable used was milk yield measured as litres per cow over a 300-day lactation period. Variables identified as contributing to variation in milk yield were considered in seven groups: nutrition, farm/farmer demographics, land management issues, farm environmental practices, attitude to dairying and the environment, knowledge sources, and future expectations. Each group consisted of a matrix of variables including information relating to specific practices as well as farmer attitudes. Selection of the groupings and variables within was based on information provided by several experts in the dairy industry and personal experience.

A number of studies examine relationships between farm management practices and milk production (Kristensen et al., 1996; Gourley et al., 1998; Armstrong et al., 2000; Bergez et al., 2002). In general, however, these studies limit their work to a single area of management. (for example, fertiliser). Studies which consider whole farm management (Herrero et al., 1999; Ramsden et al., 1999) focus primarily on biological inputs to the farm system.

The component of milk production attributable to nutritional effects was identified using multiple regression following the procedure reported by Kerr et al. (1999). Included in the model were supplementary feeds, fertiliser, irrigation, stocking rate and area used as milk pasture. Residuals from these analyses represent the milk production attributable to factors other than nutrition; they were classified into three groups, below average, average and above average, where average was defined as a farm which had a milk yield within ±500 litres of that expected given the nutritional inputs. This measure of milk due to causes other than nutrition was then subjected to PLSR using various combinations of the six variable groups.

Partial Least Squares Regression (PLSR) was chosen because of the large number of variables and the suspected colinearities between variables within and betwe