Fig. 1: Schematic of the three-stage framework for identifying key uncertainties.

Stage 1 (panel i - blue band) runs TEMOA optimization, yielding optimal design under deterministic treatment of uncertainties for each pathway--Business as Usual (BAU), Fully Renewable (FR), and Fully Decarbonized (FD)--over the transition horizon. Stage 2 (panel j - green band) conducts a probabilistic cost analysis for the 2050 optimized design using a nested Monte Carlo scheme. The outer loop (green box) samples epistemic inputs--technology costs (a) fuel price (b) long-term changes in hurricane trends (c, d) demand and tariff projections, and organizational inefficiency. For every outer draw, the inner loop (red box) generates many year-scale aleatoric realizations: daily wind/solar availability (e, f) hurricane frequency and intensity (g), component damage and restoration (h). Each inner sample computes operational, damage, and outage costs whose mean forms the expected annual cost for that epistemic draw. Stage 3 (panel k - orange band) trains a neural-network surrogate on the outer-loop results and applies Sobol global sensitivity analysis to rank the key epistemic uncertainties.