The stochastic view used in climate sciences: (some) perspectives from (some of) mathematical statistics

Climate statistics is of course a very broad field, along with the many connections and impacts for yet other areas, with a history as long as mankind has been recording temperatures, describing drastic weather events, etc. The important work of Klaus Hasselmann, with crucial contributions to the field, along with various other connected strands of work,is being reviewed and discussed in other chapters. The aim of the present chapter is to point to a few statistical methodology themes of relevance for and joint interest with climate statistics. These themes, presented from a statistical methods perspective, include (i) more careful modelling and model selection strategies for meteorological type time series; (ii) methods for prediction, not only for future values of a time series, but for assessing when a trend might be crossing a barrier, along with relevant measures of uncertainty for these; (iii) climatic influence on marine biology; (iv) monitoring processes to assess whether and then to what extent models and their parameters have stayed reasonably constant over time; (v) combination of outputs from different information sources; and (vi) analysing probabilities and their uncertainties related to extreme events.

💡 Research Summary

The chapter titled “The stochastic view used in climate sciences: (some) perspectives from (some of) mathematical statistics” offers a comprehensive bridge between modern statistical methodology and the practical challenges of climate data analysis. Beginning with a cultural comparison between climate scientists and methodological statisticians, the author underscores the shared goal of extracting reliable inference from complex, non‑stationary, and often incomplete datasets, while also noting the differing vocabularies and expectations that can hinder collaboration.

Six thematic pillars are explored through concrete case studies. The first case examines a long‑run record of annual skiing days near Oslo (1896‑2022) that contains a 15‑year gap. A simple linear trend is extended by incorporating an AR(1) autocorrelation structure, yielding a four‑parameter model. Maximum‑likelihood estimation, AIC, and the Focused Information Criterion (FIC) are used to compare candidate models, and a confidence curve for the autocorrelation parameter (ρ) visualizes its uncertainty. The analysis shows that while ρ has little impact on the slope estimate, it substantially widens prediction intervals, highlighting the importance of accounting for serial dependence when forecasting.

The second case tackles global temperature anomaly series (monthly land‑sea averages from 1850 to the present). Beyond ordinary least squares, the author fits models that (i) include autocorrelation, (ii) allow for structural breaks via segmented regression, and (iii) predict the year at which a predefined warming threshold (e.g., 1.5 °C above the 1900‑2000 mean) will be crossed. By deriving a confidence distribution (CD) for the crossing year and its associated confidence curve, the author demonstrates that point estimates (e.g., 2056) can be accompanied by highly skewed intervals that may extend to infinity, implying a non‑zero probability that the threshold is never reached. This approach replaces the delta method with an exact formulation based on the F‑distribution, providing a more accurate quantification of uncertainty for small slopes.

The third example investigates the quality index of the North Arctic cod (skrei) spanning 1859‑present and its potential dependence on Kola region temperature anomalies. Multivariate regression and cross‑correlation analyses are employed, while also considering heteroscedasticity and heavy‑tailed error distributions (t‑distribution extensions). The study illustrates how marine biological time series can be linked to climatic drivers, and how model extensions (time‑varying variance, non‑Gaussian errors) can be tested without substantial gains in this particular dataset.

Section five shifts focus to the combination of disparate information sources—observational records, climate model outputs, and reanalysis products. The author sketches a Bayesian hierarchical framework that treats each source’s bias and variance explicitly, and also mentions simple weighted‑average or optimal linear combination schemes. Such fusion techniques are positioned as essential for improving predictive skill in multi‑model ensembles and for delivering coherent uncertainty quantification across data streams.

The final methodological theme addresses extreme‑event probability estimation. Using a hypothetical severe event with a point estimate of 3.5 % occurrence, the author shows that the 90 % confidence interval can span the full 0‑19 % range, again visualized via a confidence curve. This stark illustration underscores that point probabilities can be misleading without accompanying uncertainty bounds, a crucial consideration for risk assessment and policy making.

Throughout the chapter, the concept of confidence distributions (CDs) and confidence curves (ccs) is championed as a unifying language for conveying both point estimates and their asymmetric, often skewed, uncertainties. The author argues that traditional ±1.96 σ intervals are insufficient for many climate‑related quantities, especially when the underlying estimator’s distribution deviates from normality. By integrating CDs with model‑selection criteria tailored to specific tasks (AIC for overall fit, FIC for focused prediction, etc.), the work provides a flexible toolkit for climate statisticians.

In conclusion, the chapter calls for deeper collaboration between climate scientists and methodological statisticians. It highlights promising research directions such as spatio‑temporal hierarchical models, non‑linear dynamical systems, and the synthesis of machine‑learning predictors with rigorous statistical inference. By adopting the stochastic view and the statistical tools presented, the climate community can achieve more transparent, reliable, and actionable insights from its ever‑growing data repositories.