What resilience metrics miss: from perturbation curves to animal welfare

Resilience is among the most frequently invoked and least settled concepts in the life sciences, and in livestock science its relationship to animal welfare remains insufficiently resolved. Holling’s work^1,2 made clear that the term encompasses at least two distinct conceptions: engineering resilience, the speed of return to equilibrium, and ecological resilience, the capacity to persist despite disturbance across more than one possible system state. Wiig and colleagues³ crystallised the problem further, arguing that resilience only becomes operationally meaningful when four questions are answered: for what, to what, of what, and through what. Livestock science has often answered those questions by defining resilience as the capacity to cope with short-term perturbations and return rapidly to a prior or pre-challenge functional state. That definition is coherent and measurable. But it prompts a more fundamental question: does capturing perturbation and recovery equate to understanding an animal’s resilience and does it gives us reliable information about its welfare? This editorial argues it does neither with sufficient precision.

Deviation-based indicators derived from automated milk yield, activity, and feeding data are among some of the most widely used resilience indicators in livestock research. These indicators have been associated with health, longevity, and related functional outcomes⁴. A broader biological interpretation is already present in the work of Colditz and Hine⁵, and in Colditz’s later formulation^6,7 of “competence to thrive”, which treats resilience as a positive property of integrated regulation rather than merely the absence of failure. Yet even within that broader framework, one assumption deserves more scrutiny than it has received: that resilience is adequately captured by the degree to which an animal returns to its pre-challenge state. That benchmark is not universally appropriate. For shallow, reversible perturbations, return to baseline may be exactly the right criterion. For developmental transitions, cumulative challenges, or challenges deep enough to alter future system capacity, however, resilience may instead consist in adaptive restabilisation around a new functional state. Once recovery is recognised as challenge-dependent rather than universal, the observed perturbation curve becomes more ambiguous than it first appears. It can reasonably be understood as reflecting at least three influences: the quality of biological regulation before challenge, the depth of reserve available for that class of challenge, and the magnitude of the challenge itself. These components likely differ in genetic architecture, developmental origins, and management implications. A single deviation index cannot reliably separate these contributions. Nor can it determine whether apparent non-recovery reflects poor resilience, a challenge whose depth made full restoration biologically unrealistic, or a state transition in which return to the previous baseline was never the appropriate criterion. The next advance in resilience science will require methods that estimate these contributions separately rather than collapsing them into one score.

That challenge is not only methodological but also conceptual, and one the scientific literature has not yet confronted with sufficient clarity. In much livestock research, resilience has been operationalised primarily as a functional or performance trait: the maintenance or recovery of biological function under challenge. Welfare asks a different question. It concerns whether the animal is suffering, whether its motivations can be expressed, and whether its daily life is compatible with positive functioning affectively and behaviourally. The two constructs are related, but they are not interchangeable. An animal may recover efficiently from perturbation while experiencing sustained negative affective states. Equally, an animal may function well in daily life yet respond poorly to a challenge for which its specific biological reserve is limited. A science that connects them with precision is only beginning to emerge.

This distinction is also visible in practice. In our qualitative research recently published⁸ with experienced dairy farmers, animals described as resilient were often identified not primarily by rapid recovery from discrete events, but by a more continuous pattern of coherent biological function: sustained appetite, behavioural engagement, social ease, and apparent fit with their environment and management system. These observations were strikingly similar across farms that differed in breed and production systems. They point to resilience, in practice, as an enduring quality of day-to-day biological functioning, not only as recovery from acute challenge. That understanding may align more closely with some affective and functional dimensions of welfare than perturbation-based metrics were designed to assess. Dawkins⁹ has asked whether smart farming measures what animals want; the farmer evidence here suggests that resilience metrics, as currently formulated, may not capture it either.

Dynamical system theory offers one promising framework for expanding the science of resilience. Healthy biological systems often display structured variability across timescales, whereas declining regulatory capacity may be accompanied by altered temporal dynamics, reduced flexibility, or in some systems, signatures consistent with what Scheffer and colleagues¹⁰ termed critical slowing down. In this view, perturbation responses are not the only relevant signal; changes in the organisation of daily biological function may also provide early evidence of emerging loss of capacity before overt clinical thresholds are crossed. Precision Livestock Farming generates the continuous, individual-level time-series data needed to test this: patterns in feeding microstructure, activity rhythms, and behavioural regularity, quantified against each animal’s own longitudinal baseline rather than population averages. Combined with genomic, metabolomic, and epigenomic profiling, such approaches could help decompose perturbation responses by distinguishing regulatory quality from challenge-specific reserve, and by linking both to molecular correlates of life-course exposure and physiological regulation. The empirical tools are advancing faster than the conceptual framework needed to interpret them. Helping to build that framework is one role this journal can play.

That framework cannot be developed within any single discipline. Veterinary and animal scientists, data and systems scientists, quantitative geneticists, molecular biologists, welfare scientists, and social scientists must work on the same questions in genuine intellectual exchange. As a Nature Partner Journal spanning this full range, npj Veterinary Sciences is built precisely for that collaboration. We will publish work that retains the rigour of current resilience phenotyping while broadening its biological and ethical frame; work that links sensor phenotypes to mechanism; and work precise enough to distinguish resilience from welfare, and ambitious enough to relate them. The most important papers in this space will be those that require more than one disciplinary lens to evaluate well. Too often, those papers have had no obvious home.

Resilience matters. So does the animal whose resilience is being measured. The task ahead is not to collapse those concerns into one term, but to develop a science capable of relating them with greater precision.