What We Publish

Implications of uncertainty in technology cost projections for least-cost decarbonized electricity systems

Lei Duan and Ken Caldeira · iScience 27, 108685 · 2024

Key finding. Across 243 combinations of the low, middle, and high 2050 cost projections in the National Renewable Energy Laboratory's 2021 Annual Technology Baseline, it is not possible to say with confidence which technologies will dominate a least-cost zero-carbon electricity system, and cutting the cost of a single technology does not reliably increase either its deployment or its share of system spending.

Four stacked-area plots, one above another, of system cost in dollars per kilowatt-hour from 0 to 0.12 against carbon emission reduction from 0 to 100 per cent, coloured by technology. Panel A, all technologies at high projected cost, ends dominated by wind, solar, and biopower near 0.088. Panel B, only geothermal reaching its low cost, is overwhelmed by dark red geothermal above 90 per cent reduction. Panel C, wind and geothermal both low, is mostly light blue wind with a much smaller geothermal wedge. Panel D, all technologies low, reaches only about 0.06 and contains no geothermal at all.
The same model, the same weather, the same demand — only the assumed 2050 technology costs differ between panels, and the composition of the least-cost zero-carbon system changes completely. Geothermal dominates when it alone gets cheap (B), shrinks when wind gets cheap too (C), and disappears when everything gets cheap (D). Figure 2 from Duan and Caldeira (2024), iScience 27, 108685. Reproduced under CC BY 4.0. Extracted from the published PDF and resized for web display.

What question did this research address?

Every plan for a decarbonized grid rests on projections of what each technology will cost decades from now. Near-term cost estimates broadly agree; projections for 2050 diverge sharply, because they encode different beliefs about how far innovation will go.

Modelling studies typically pick one set of those projections, often an ad hoc set chosen by the authors, and report the resulting technology mix. This paper asked what happens if you stop picking — if you run the whole spread of published projections and look at how much the answer moves.

What did we find?

Six technologies — solar, onshore wind, offshore wind, geothermal, battery storage, and gas with carbon capture — each carry three published cost projections, giving 729 combinations, or 243 once fossil generation is excluded under a 100% emission-reduction constraint. Each is run through a greenfield, hourly, least-cost model of the contiguous United States.

The cheapest system found costs $0.06 per kilowatt-hour and contains every technology except nuclear and geothermal at its lowest projected cost. But systems within 5% of that cost include geothermal, within 10% include solar, offshore wind, and battery storage at their middle projections, and within 20% include solar at its high projection. Many different technology mixes are near-optimal.

Cheaper does not mean more money spent on it. Under the high-cost projection, solar accounts for 5.6 to 19.0% of total system cost; under the low-cost projection the range is 3.9 to 23.3%. Solar's median price elasticity is below 1.0 — each per cent of cost reduction brings less than a per cent more capacity — while onshore wind, offshore wind, and battery storage are at or above 1.0.

Technologies compete for the same job. Geothermal enters least-cost systems only if it reaches its low projection, 27% of current costs, and then dominates — but if onshore wind also reaches its low projection, 32% of current costs, geothermal's share falls sharply.

Two technologies do more than the rest to narrow the uncertainty. System costs span 30% when onshore wind follows its high projection but only 22% at its low projection; for geothermal the span falls from 46% to 22%. Cheap wind or geothermal is deployed so heavily that the costs of competing technologies stop mattering much.

Cheaper solar and offshore wind, by contrast, barely shift the distribution of system costs at all — they slightly increase their own capacity without changing what the system costs.

The analysis is deliberately stylized. It takes all cost numbers from the NREL dataset rather than generating any, treats the three projections as equally probable purely for illustration, assumes unlimited resource potential in the main cases, and claims robustness for the qualitative conclusions rather than the particular numbers.

Why does it matter?

It puts a bound on what this class of modelling can tell us. A study that picks one cost projection and reports the resulting technology mix is reporting a consequence of that choice, and the spread across published projections is wide enough to change which technology wins.

The result about elasticity is the counterintuitive one. Driving down the cost of a technology is usually assumed to increase its use, but in a system where technologies substitute for one another, a cost breakthrough can be absorbed by competitors instead — so a successful research programme need not produce the deployment it was aiming at.

It argues for a portfolio rather than a bet. Since no single technology's cost trajectory determines the outcome, and since relying on an uncertain breakthrough risks an expensive system if it fails to arrive, spreading cost-reduction effort and revisiting decisions as costs become known does better than committing early to one winner.

Citation

Lei Duan and Ken Caldeira (2024). Implications of uncertainty in technology cost projections for least-cost decarbonized electricity systems. iScience 27, 108685.

Related