Improving pasture intake predictions by variable weighting of plant marker concentrations

Abstract

It is difficult to measure pasture feed intake. A common method is based on naturally occurring, indigestible plant markers, such as long chain alkanes. Least-squares procedures are used to estimate diet composition and intake. If actual intake of a supplement is known, then total intake and the intake of all dietary components can be estimated. This ‘labelled-supplement’ approach requires an estimate of the faecal recoveries of the markers. The accuracy and precision of intake solutions for each animal is also affected by the sampling and measurement precision of the plant and faecal marker concentrations. This work was conducted to study whether weighting each marker's sums of squares in the least-squares procedure could be used to provide a more robust solution.Cluster and discriminant analyses of a plant marker database determined the contribution of each marker to discrimination between categories of plants. The markers’ cluster or discriminant weights were used to weight the sums of squares in the least squares procedures. The actual individual dry matter intakes (DMI) of 20 cattle were arbitrarily assigned for three different diets. Measurement and sampling variations in marker concentrations and/or faecal recoveries were simulated to generate predicted total pasture intakes around the actual values.Six marker weighting methods were compared for their DMI prediction error values and correlations between predicted and actual DMI: (A) all markers weighted by one; (B) separate cluster analyses of z scores for alkanes and alcohols; (C) combined cluster analyses for alkanes and alcohols; (D) discriminant analyses of z score marker data for plants categorized into grasses, legumes, shrubs and trees; (E) discriminant analyses of plants categorized on origin and plant, photosynthesis and reproduction type; and (F) discriminant analyses of plants categorized on plant, photosynthesis and reproduction type.The standard approach of weighting all markers by one (A) was satisfactory when marker concentration error was set at zero, however intake predictions were poor when the error was non-zero, which is likely. The weighted least-squares intake solutions that were more robust to variance in measured marker concentrations or in assumed faecal recovery rates were those using weights derived by methods D and F. Marker weights from Methods D, E and F resulted in similar intake prediction error variances and correlations. Methods E and F required more botanical information about plant species and method D was simpler, so method D is recommended rather than other methods studied here, including the standard method A. There are problems with using weights derived from an analysis of all published marker data, so better weighting methods may still be found for specific plant and marker datasets.