Philip 2020 - Extreme Event Attribution Protocol - Overview

Summary

Philip et al. (2020) document the eight-step World Weather Attribution (WWA) protocol: analysis trigger, event definition, observed probability and trend, model evaluation, multi-method multi-model attribution, hazard synthesis, vulnerability/exposure, communication. The statistical core is extreme-value distributions (GEV, GPD, Gumbel, Gaussian, gamma) whose location/scale shifts with smoothed global mean temperature, giving a probability ratio $PR=p_1/p_0$ and an intensity change $\Delta I$.

Overview

Question. How should one attribute an extreme event to climate change in a standardised, comparable, quickly executable way?

Contribution. A reproducible protocol built from ~two dozen rapid and slow studies (hot, cold, wet, dry, stormy extremes), emphasising probability-based (not storyline) framing, model evaluation before use, multi-model synthesis with uncertainty inflation, and explicit separate treatment of vulnerability and exposure.

Key result. Examples with numbers: Peru 2013 cold wave return period 17 y (95% CI 7-75) and $PR=0.08$ (0.02-0.21) for the normal shift fit; southern Louisiana August 2016 rain return period about 500 y (360-1400), $PR=2.8$ (1.7-3.8), intensity +17% (10-21%); De Bilt 2018 heat around 300 y (>30 y) and “formally impossible” in 1900 so $PR=\infty$ (they prefer “extremely unlikely”).

Motivating PyMC examples

E37 “Extreme-event attribution: how much did warming load the dice for the 2021 Pacific Northwest heatwave?” (NOAA GHCN-Daily annual maxima at Sea-Tac and Portland; NASA GISTEMP) - block maxima and the GEV (pymc_extras.GenExtreme checked against scipy), a geophysical shape prior, a reparameterisation removing support-wall divergences, non-stationary GEV on GMST - at https://redam94.github.io/pymc-challenges/examples.html. See Philip 2020 - Bayesian GEV Attribution in PyMC. E113 (log-Pearson III flood frequency) is in Concept Notes - Log-Pearson III Flood Frequency and Historic Floods.

Main Content — Source Map

Paper section	Pages	Covered in
Abstract, 1 Introduction (probability vs storyline; framings)	177-179	Philip 2020 - Probability Framing, Event Definition and Return Periods
2 Step 1: Analysis trigger	179-180	Philip 2020 - Probability Framing, Event Definition and Return Periods
3 Step 2: Event definition (goal, variable, area)	180-182	Philip 2020 - Probability Framing, Event Definition and Return Periods
4 Step 3: Observed probability and trend (4.1 data; 4.2 GEV, GPD, Gumbel, Gaussian, gamma; 4.3 shift/scale fit; 4.4 natural variability)	182-190	Philip 2020 - Extreme Value Distributions for Attribution, Philip 2020 - Trend Models and Natural Variability
5 Step 4: Model evaluation	190-192	Philip 2020 - Model Evaluation, Multi-Model Attribution and Hazard Synthesis
6 Step 5: Multi-method multi-model attribution (fixed vs transient forcing)	192-193	Philip 2020 - Model Evaluation, Multi-Model Attribution and Hazard Synthesis
7 Step 6: Hazard synthesis	193-195	Philip 2020 - Model Evaluation, Multi-Model Attribution and Hazard Synthesis
8 Step 7: Vulnerability and exposure	195-196	Philip 2020 - Vulnerability, Exposure and Communication
9 Step 8: Communication	196-198	Philip 2020 - Vulnerability, Exposure and Communication
10 Conclusions; data and acknowledgements	198	Philip 2020 - Vulnerability, Exposure and Communication
(concept translation to Bayesian GEV)	-	Philip 2020 - Bayesian GEV Attribution in PyMC

Relevance to marketing measurement

“Probability ratio under a counterfactual” is the same logic as incrementality: $P(\text{outcome}\mid \text{campaign})/P(\text{outcome}\mid \text{no campaign})$ for a tail event (a stock-out, a viral spike, a fraud surge). The protocol’s pre-registered event definition, refusal to tune the event to the answer, and model-evaluation gate are directly transferable to measurement governance. See Decision Analysis.

Connections

Hierarchical Models (pooling stations/models), Sensitivity Analysis in Observational Studies (robustness to the counterfactual).

See Also

Concept Notes - Log-Pearson III Flood Frequency and Historic Floods