Statistical fan (ARIMA × ±0%, ±25%, ±50% rate perturbations)

ARIMA(0,1,1) point forecast yields near-linear future trajectory; consensus β collapses to 0 (constant ρ ⇒ slope of log|ρ| against log|Φ| is 0 by construction). σ_cross retains historical curvature and is informative.

IPCC AR6 SSP fan (Fox-Kemper et al. 2021 central trajectories)

SSPmean ρ (mm/yr)σ_crossβ GLSR² GLSβ Bayesian
SSP1-1.93.970.258−1.119−66.08−0.032
SSP1-2.64.770.256−0.945−36.71+0.058
SSP2-4.56.240.263+0.296+0.54+0.026
SSP3-7.07.770.257+0.541+0.45+0.461
SSP5-8.58.890.245+0.521+0.47+0.451

When AR6 SSP curve attaches to a long, mostly-flat historical record, gls_ar1's Cochrane-Orcutt transform can be near-degenerate (R² < 0). The Bayesian AR(1) MAP/median is more stable; both reported.

Frederikse 2020 component decomposition under each SSP

SSPStericGlaciersGrISAISTWS
SSP1-1.9+1.117+0.213+1.181+3.757−0.626
SSP1-2.6+1.176+0.382+1.281+3.664−0.312
SSP2-4.5+1.250+0.634+1.407+3.479+0.138
SSP3-7.0+1.293+0.806+1.478+3.316+0.427
SSP5-8.5+1.313+0.899+1.511+3.210+0.578

AIS β > 3 under EVERY SSP

Antarctic Ice Sheet is super-rate (β = 3.21 to 3.76) under all 5 SSPs including SSP1-1.9. Aggregate GMSL stays sub-rate under aggressive mitigation, but its dominant component does not. See locked-in risks. AIS post-tipping rate scales as dΦ/dt ~ Φ³·⁵ — super-fast acceleration once tipped.

Best pathway by aggregate: SSP1-1.9 (rank-sum 7 of 24 — tightest σ_cross + smallest mean ρ + β farthest below 1).

Source: scenario_fan_gmsl.py