The Uncertainty Landscape Beyond Variance
Abstract
Modern science has become good at measuring uncertainty. Across statistics, artificial intelligence, economics, Earth observation, environmental modelling and decision science, researchers increasingly invoke uncertainty when interpreting evidence, evaluating predictions and supporting decisions. These disciplines often use the word uncertainty to describe conceptually different scientific situations. Different forms of uncertainty represent different states of knowledge and require different scientific responses. As uncertainty concepts move across disciplinary boundaries, their original meaning may be lost in translation, creating the misleading impression that they describe different amounts of the same uncertainty. Some forms of uncertainty arise from stochastic variability. Others reflect incomplete knowledge that can be reduced through learning. Still others reveal limitations of current models or even existing scientific concepts. Treating these different states of knowledge as merely different amounts of uncertainty can obscure both the nature of scientific evidence and the forms of inference needed for understanding and informed decision-making. We propose an uncertainty landscape whose regions are defined by what is known, what can be characterised probabilistically and what can be learned from additional evidence. Our aim is to reinterpret an existing five-level taxonomy as a cross-disciplinary landscape that links different states of knowledge to the scientific responses they require. Seen this way, uncertainty becomes a guide to where new scientific understanding is needed, rather than merely a limitation to be reduced.
Related
- Single-cell T-cell landscape in atherosclerosis: implications for targeted therapy and beyond
- Accounting for social feasibility in landscape-scale scenarios for restoring nature beyond protected areas
- Deconstructing overdispersion: a generative variance budget for luminescence dating from single grains to aliquots
- Structural and dynamic functional controllability: shared variance between network control metrics limits attribution to distinct domains of intelligence
- From Nagelkerke's R2 to Liability-Scale Variance Explained for Polygenic Scores
BibTeX
@misc{saarela2026uncertainty,
title = {{The Uncertainty Landscape Beyond Variance}},
author = {Svetlana Saarela and Sjur Baardsen and Julien Brajard and Laurent Bertino and Per Kristian Rørstad and Stephen Plummer},
howpublished = {EarthArXiv},
year = {2026},
month = sep,
doi = {10.31223/x5579s},
url = {https://eartharxiv.org/repository/view/14747/},
note = {CC BY-NC-SA 4.0},
}
RIS
TY - UNPB
TI - The Uncertainty Landscape Beyond Variance
AU - Svetlana Saarela
AU - Sjur Baardsen
AU - Julien Brajard
AU - Laurent Bertino
AU - Per Kristian Rørstad
AU - Stephen Plummer
PY - 2026
DA - 2026/09/02
PB - EarthArXiv
DO - 10.31223/x5579s
UR - https://eartharxiv.org/repository/view/14747/
AB - Modern science has become good at measuring uncertainty. Across statistics, artificial intelligence, economics, Earth observation, environmental modelling and decision science, researchers increasingly invoke uncertainty when interpreting evidence, evaluating predictions and supporting decisions. These disciplines often use the word uncertainty to describe conceptually different scientific situations. Different forms of uncertainty represent different states of knowledge and require different scientific responses. As uncertainty concepts move across disciplinary boundaries, their original meaning may be lost in translation, creating the misleading impression that they describe different amounts of the same uncertainty. Some forms of uncertainty arise from stochastic variability. Others reflect incomplete knowledge that can be reduced through learning. Still others reveal limitations of current models or even existing scientific concepts. Treating these different states of knowledge as merely different amounts of uncertainty can obscure both the nature of scientific evidence and the forms of inference needed for understanding and informed decision-making. We propose an uncertainty landscape whose regions are defined by what is known, what can be characterised probabilistically and what can be learned from additional evidence. Our aim is to reinterpret an existing five-level taxonomy as a cross-disciplinary landscape that links different states of knowledge to the scientific responses they require. Seen this way, uncertainty becomes a guide to where new scientific understanding is needed, rather than merely a limitation to be reduced.
ER -