| Sumario: | Accurate prediction of bodyweight (BW) from measurable animal characteristics serves
many practical purposes for livestock farmers in sub-Saharan Africa, where weighing scales
are often impractical or unaffordable. A volume of literature is devoted to BW estimation
using heart girth (HG) (circumference of chest diameter behind front legs) as a main
predictor, yet most published equations are derived from narrowly defined animal samples
(e.g., single breeds, age classes, or production systems), limiting their applicability across
Africa’s diverse livestock populations.. Based on a large and geographically diverse
repeated measurement sample of African bovines (n = 6,434; 13,790 total measurements)
and caprines (n = 2,113; 6,049 total measurements) we developed prediction equations
specific to ruminant species (bovines, sheep, goats), breed (indigenous caprines or Bos
indicus bovines vs. crosses thereof with exotics or B. taurus), and production system
(extensive vs. semi-intensive) by segmenting regression formulae according to tertiles (equal
thirds) of HG. Segmentation by HG improved model fit at upper extremities of HG where
residual errors are typically highest. Among four regression formulae: linear, square root,
quadratic, Box-Cox -- Box-Cox consistently performed best across sub-samples of species,
breed, and system, with the exception of indigenous African sheep breeds (best fit using a
quadratic formula). Lines-of-best fit were characterized as convex, with Box-Cox lambda
parameters ranging from 0.141-0.788, a finding in general agreement with other studies
performing non-linear regression. At the extremes of HG segmented regression had superior
predictive accuracy, in particular, with as much as a 25% reduction in NRMSE relative to
non-segmented regression at the upper HG tertile. The contributions of this study are thus:
(i) provision of the most comprehensive, genetically and geographically diverse set of BW
prediction equations for African domestic ruminants to-date, and (ii) for bovines, using
segmented regression, BW prediction robust to the extremities of BW/HG, yielding
substantially improved prediction accuracy (NRMSE < 5%) over existing published formulae.
As Box-Cox formulae are most accurate but structurally complex, authors propose formula
transformations embedded directly on weigh-tapes or charts to maximize ease-of-use and to
enable scalable, low-cost, and accurate BW estimation for African production systems.
|
|---|