Large-Scale Spatial Optimization for Industrial Decarbonization

Spatiotemporal optimization, 2024 to 2026. Manuscript under second-round review at Nature Energy.

The problem

Low-carbon hydrogen is technically capable of replacing fossil fuels and feedstocks in steel, ammonia, and methanol production, but technical potential does not establish a deployment pathway. Existing plants compete for unevenly distributed renewable electricity; strict off-grid rules increase storage and transmission requirements; sector demand changes at different rates; and carbon prices determine whether hydrogen can compete with incumbent production, carbon capture, or scrap-based electric arc furnaces. The decision is therefore not simply whether to use hydrogen, but where it should enter first, how electricity should reach each plant, and which alternative remains least costly during the transition.

What I built

I built a plant-level retrofit pathway model covering 548 operating facilities: 203 steel plants, 158 ammonia plants, and 187 methanol plants. The framework (1) matches each facility to hourly offshore wind, onshore wind, utility solar, and distributed solar resources on a 25 × 25 km grid; (2) co-optimizes generation, electrolysis, storage, grid purchases, and electricity sourcing distance; and (3) compares hydrogen retrofits with carbon capture, scrap-EAF, and unabated incumbent routes at five-year steps from 2025 to 2050. A policy module adds grid decarbonization, technology learning, emissions caps, and carbon pricing, while joint multi-sector optimization captures competition for the same nearby renewable resources.

What I found

Grid access is the largest near-term deployment lever because a small amount of grid electricity can replace disproportionately expensive storage and bring more plants within a practical renewable sourcing distance. Coordination also matters: optimizing steel, ammonia, and methanol together performs better than planning each sector independently because the three industries draw on the same regional resource base.

The transition follows a clear sequence. Ammonia moves first because hydrogen is already a feedstock, so decarbonization changes the hydrogen source rather than the production process. Steel follows through hydrogen-based direct reduced iron in the late 2030s, alongside scrap-based electric arc furnaces. Methanol moves last, and carbon capture remains a transitional option because increasing demand keeps unabated coal-based production competitive for longer. North, Northwest, and Northeast China lead because nearby renewable resources shorten electricity sourcing distances; resource-constrained regions require longer connections or later deployment.

Carbon pricing determines the pace. In the modeled pathway, pricing advances cost parity by 10 to 15 years, raises the low-carbon share of industrial production to 74% by 2035, and supports a 94% reduction in annual CO2 emissions by 2050. Hydrogen becomes the least-cost route for most plants by 2050, but it does not replace every alternative at the same time.

What this supports

The analysis supports a staged industrial strategy: permit hybrid renewable and grid supply rather than requiring strictly off-grid hydrogen; direct early support to ammonia and renewable-rich regions; move next to steel where hydrogen-based reduction can deliver large absolute reductions; and use credible carbon prices, industrial cluster planning, and demand guarantees to determine when individual retrofits clear the cost threshold. The central conclusion is that hydrogen deployment is a spatial and policy coordination problem as much as a technology problem.

Limits

The analysis optimizes supply and technology choices for existing plants. It does not model the wider power system response to new industrial electricity demand, interregional hydrogen transport, or the capital replacement cycle that determines the exact retrofit year for an individual facility. The manuscript remains under review, so the website reports the strategic findings without presenting unpublished source data.