Decision-Centred Modelling for Industrial Process Heat Planning under Deep Uncertainty: Design Principles and an Edendale–Southland Proof of Concept

Abstract

Industrial process heat is a pivotal yet under-served domain in energy transition analysis. Its decarbonisation is shaped by heterogeneous, site-specific equipment and operating constraints, long asset lifetimes, and tight dependencies on regional electricity headroom, renewable fuel availability, infrastructure delivery and evolving policy. In New Zealand these challenges are amplified by strong regional asymmetries and a data landscape that is improving in coverage but remains largely annualised and fragmented. Decision makers therefore face high-stakes, path-dependent technology choices, such as electrification versus biomass, whose consequences cross the boundary of any single model.

This thesis develops Decision-Centred Modelling (DCM), a modular site-to-region framework in which the analytical boundary is set by the consequences that can change the comparison of alternatives, not by the physical extent of the system being modelled. DCM rests on four commitments: a decision-first boundary principle; a thin-waist architecture in which modules exchange a small set of versioned, acceptance-gated artefacts (notably DemandPack and SignalsPack), so that methods may differ inside modules while interfaces stay fixed; robustness evaluation across ensembles of plausible futures using regret, satisficing rate, and win rate; and progressive, non-destructive refinement guided by the analysis's own diagnostics.

The framework is demonstrated through a proof of concept at Fonterra's Edendale dairy site in Southland. It is methodological rather than predictive. A synthetic hourly demand profile is met by proportional dispatch under an hourly first-law energy balance; the exported incremental electricity is screened against Grid Exit Point headroom and a reinforcement menu; and two 2035 pathways, electrification completion and biomass completion, are compared over 100 paired futures. Under biomass completion, the 97 MW reinforcement was sufficient in every future; under electrification, the least-cost option switched to 150 MW in 49 futures, a threshold-triggered consequence that a site-boundary analysis cannot represent. Biomass completion had the lower system cost in 93 of 100 futures, with the ranking driven mainly by relative fuel prices. The diagnostics also showed that the most influential inputs, biomass cost and the dispatch-driven electricity peak, were among the least evidenced, so the analysis identifies its own refinement priorities. The results are conditional on two technology pathways, first-law energy accounting, and synthetic demand and headroom.