Uncertainty & Sensitivity

Quantified uncertainty decomposition, Monte Carlo summary, and parameter sensitivity rankings for all CE models.

N = 10000 MC runs seed = 42 Converged at N = 8200
57%
Avg Epistemic
Knowledge / modelling uncertainty — reducible with more data
17%
Avg Aleatory
Irreducible natural variability
26%
Avg Structural
Model-form & specification uncertainty

Tornado Chart — One-at-a-time Sensitivity

Percentage change in total damage cost (USD bn, 2050) per ±1σ parameter perturbation.

Parameter Sensitivity Rankings

#ParameterModelTypeDistribution Impact LowImpact HighNotes
1 P-CLM-001 Equilibrium Climate Sensitivity (ECS) ClimateModelService epistemic lognormal -18.2% +28.4% Single largest uncertainty in temperature projection. IPCC AR6 likely range 2.5–4.0°C.
2 P-DAM-001 Damage function exponent (α) DamageModelService structural uniform -24.1% +31.7% Highly contested. DICE uses α=2; empirical estimates range 1.5–4.
3 P-ECO-002 Social discount rate (ρ) EconomicModelService epistemic uniform -32.0% +29.5% Ethically contested. 1.4% (Stern) vs 5%+ (Nordhaus) implies 10× NPV difference.
4 P-UNC-001 Monte Carlo sample size (N) MonteCarloEngine aleatory convergence -2.1% +2.3% N=10,000 achieves <2% convergence error on GDP impact distributions.
5 P-CLM-002 Carbon cycle feedback β ClimateModelService epistemic normal -8.4% +12.1% Land carbon uptake fraction. CMIP6 model spread: 0.28–0.52.
6 P-ECO-001 GDP growth baseline (g₀) EconomicModelService epistemic normal -11.3% +9.8% Baseline SSP scenario dependent. Range 1.5–3.5% annual for advanced economies.
7 P-UNC-002 Double-counting deduction factor (δ) GapAccountingEngine epistemic uniform -6.2% +7.9% Expert judgment. NDC overlap estimation poorly constrained.
8 P-DAM-002 Adaptation residual factor (φ) DamageModelService epistemic beta -5.8% +6.4% Fraction of damage that persists after adaptation. Expert range 0.4–0.85.
9 P-CLM-003 Permafrost CO₂ feedback γ ClimateModelService epistemic lognormal -3.1% +9.2% Highly asymmetric. Tipping points could amplify by 3×. Currently underweighted.
10 P-FIN-001 Climate WACC premium (Δr) FinancialStressService epistemic normal -4.2% +5.7% Limited empirical basis. Based on GFANZ consultation, 2024.
11 P-ADP-001 Adaptation cost factor (κ_adapt) AdaptationService epistemic normal -3.8% +4.6% % GDP for infrastructure adaptation. UNEP range 0.1–2.0% GDP.
12 P-ECO-003 Intertemporal elasticity ψ EconomicModelService epistemic normal -2.9% +3.1% Standard DSGE value 1.5. Climate models use 1.0–2.0.

Uncertainty Decomposition by Model

ModelEpistemic %Aleatory %Structural %Dominant Source
ClimateModelService 62% 18% 20% ECS range
EconomicModelService 55% 12% 33% Discount rate choice
DamageModelService 40% 15% 45% Damage function form
MonteCarloEngine 70% 25% 5% Input parameter distributions
GapAccountingEngine 65% 10% 25% Double-counting methodology
FiscalModelService 50% 20% 30% Revenue-expenditure coupling assumptions

Confidence Intervals — Key Outputs

Outputp5p25p50 (median)p75p95 UnitsBenchmarkSource
Global mean temperature 2050 (°C) 1.92 2.31 2.87 3.41 4.23 °C above pre-industrial 2.7 IPCC AR6 WG1
GDP impact 2050 (% of GDP) -6.4 -4.6 -3.2 -1.8 -0.3 % GDP -2.9 DICE-2023
Physical damage cost 2050 (USD bn) 1820 3240 4820 6590 8910 USD billion (2020 prices) 4500 Swiss Re Institute, 2023
Remaining carbon budget (GtCO₂) 102 178 245 318 412 GtCO₂ from 2023 250 IPCC AR6 WG3 Table SPM.2
Mitigation cost (% GDP) 0.8 1.4 2.2 3.1 4.8 % GDP annual 2.5 NGFS Phase IV / IEA NZE

Temperature Fan Chart (2025–2100)

Monte Carlo Summary

Sample Size
10,000
Random Seed
42
Converged at N
8200
Convergence Threshold
≤ 2.0%
OutputDistributionMeanStd Devp5p95
Temperature 2050 lognormal 2.87 0.61 1.92 4.23
Gdp Impact Pct normal -3.2 1.8 -6.4 -0.3
Damage Cost Bn lognormal 4820 1940 1820 8910
Carbon Budget Remaining Gt normal 245 88 102 412

Methodology Note

Epistemic uncertainty arises from incomplete knowledge — e.g. ECS range, discount rate ethics. Reducible with better data and models.

Aleatory uncertainty is irreducible natural variability — e.g. decadal climate oscillations, stochastic GDP shocks.

Structural uncertainty reflects model-form choices — e.g. damage function exponent α, DSGE calibration choices.

The tornado chart shows one-at-a-time (OAT) sensitivity. For correlated parameters, Monte Carlo sampling provides the joint distribution. Convergence is assessed via the Gelman–Rubin diagnostic (R̂ < 1.01).

Export JSON