Florian Hartig

← Publications

Full prediction uncertainty quantification: A plea from science and decision making

Carsten F. Dormann, Yannek Käber & Florian Hartig

Advances in Ecological Research, 191–212 (2026)
Cite this
@incollection{dormann2026full,
  author = {Carsten F. Dormann and Yannek Käber and Florian Hartig},
  title = {Full prediction uncertainty quantification: A plea from science and decision making},
  booktitle = {Advances in Ecological Research},
  publisher = {Elsevier},
  pages = {191--212},
  year = {2026},
  doi = {10.1016/bs.aecr.2026.02.003},
}

DOI: 10.1016/bs.aecr.2026.02.003
Citation statistics: 0 (Google Scholar) · 0 (OpenAlex), as of 29 October 2026

Google Scholar citations per year, as of 29 October 2026 (yearly counts are fetched separately from the total)

Funding

This study was supported through funding by Deutsche Forschungsgemeinschaft (CRC 1537).

View article (DOI)

Abstract

It is often tacitly assumed that the usefulness of a model depends primarily on the accuracy of its predictions. We disagree with this viewpoint and argue that in both applied decision making and theory development, understanding prediction uncertainty is often equally important. Unfortunately, many researchers still see the quantification and description of uncertainties as a nuisance that is unnecessary at best and counterproductive to the success of a model or theory at worst. Here, we argue that many objections against full prediction uncertainty quantification (FPUQ) are incoherent, or are reflecting a non-probabilistic tradition of process-oriented modelling communities. However, there are some real, at times substantial, problems when attempting a FPUQ. Those include quantifying structural model uncertainty, weighting of non-independent model predictions, estimating the fat-tailed prediction distributions and the challenge to obtain truly independent data. We discuss these challenges and argue that nevertheless these issues are outweighed by the reward of FPUQ, in particular the identification of real gaps in knowledge, and the establishment of credibility in decision making.

What the paper shows and why it matters

(AI-generated)

Models are often judged chiefly by how accurately they predict. The paper notes that this is frequently assumed tacitly, and that many researchers regard quantifying and describing the uncertainty of predictions as a nuisance: unnecessary at best, damaging to the success of a model or theory at worst. Part of this attitude reflects a non-probabilistic tradition in process-oriented modelling, that is, modelling built around the processes believed to drive a system. The paper takes on the assumption that a model’s usefulness depends mainly on the accuracy of its predictions.

The paper pleads for full prediction uncertainty quantification (FPUQ), arguing that understanding the uncertainty of predictions is often as important as their accuracy, both for applied decision making and for developing theory, and that FPUQ is worthwhile because its rewards outweigh its problems. It holds that many objections to FPUQ are incoherent or reflect the modelling tradition above. It also names real, at times substantial, problems: quantifying structural model uncertainty (doubt about whether a model has the right form), weighting the predictions of models that are not independent of each other, estimating prediction distributions with fat tails (extreme outcomes that are more likely than a normal distribution implies), and obtaining truly independent data.

The paper discusses these challenges and argues that they are outweighed by the reward of FPUQ. It highlights two rewards in particular: identifying real gaps in knowledge, and establishing credibility in decision making.