Researchers Mark Bolinger, Ryan Wiser, and Eric O’Shaughnessy collected data on renewables from 908 wind farms and 822 solar operations in the United States, all of them larger than five megawatts in capacity. For the wind farms, they collected data from today back to 1982 when modern, utility-scale wind farms first appeared in the US. For solar, the data goes back to 2007, which is when the first utility-scale solar photovoltaic projects larger than five megawatts were built in the US. The research was published recently in the journal iScience.
What the researchers found is that the people operating renewables like solar and wind farms are learning to do so more efficiently, lowering the levelized cost of electricity (LCOE). Mark Bolinger, a research scientist at Lawrence Berkeley National Laboratory, tells Ars Technica the same is true for financing those renewable energy facilities.
When an industry is young, it’s considered to be risky, he says, so the cost of financing is high. But as experience in the industry grows, lenders and investors become more comfortable with the assets and are willing to offer more competitive rates, which also leads to lower costs. “It’s not just the upfront capital costs that can benefit from learning. Instead, it’s really all five or six LCOE inputs can benefit from learning by doing over time. All of them can contribute to lower costs.”
Bolinger says LCOE is made up of several components. The most important is the upfront installed cost of the plant. This is followed by the capacity factor, the measure of how much energy the plant can produce each year. LCOE also factors in operating costs, government tax rates, financing costs, and expected useful lifetimes of the plants. “[LCOE] is essentially spreading costs across the full lifetime of those plants,” he says.
The researchers determined LCOE-based learning rates were 15% for wind and 24% for solar. “Once you calculate those historical learning rates, you can then apply them to forward-looking deployment projections,” Bolinger says. For example, if by 2030 the cumulative deployment of wind doubled twice, it would imply a 30% reduction in the LCOE. “The whole thing about learning curves is that you’re looking at historical relationships, and then you’re extrapolating them.”
Previously this kind of research was largely based on capital costs, which only focused on the upfront installment cost, probably because it’s easier to find empirical data on installment costs, but harder to gather additional data about the other costs involved in LCOE, according to Bolinger.
“Your data intensity escalates quite a bit if you’re hoping to do this based on LCOE. But if you think about it, LCOE is really the correct metric you want to be using here because capital cost is just one of five or six inputs into the LCOE equation. And the industry, over time, has historically focused on optimizing or reducing LCOE rather than having a narrow focus on capital costs.”
Historical Decline In Cost Of Renewables
In their summary, the researchers focus on the historical declines in the cost of renewables. From an average of $440 per MWh over the first few years of the market (1982–1984), the LCOE for wind declined by 93% through 2020, to an average of $32 per MWh in 2020 dollars.. Utility scale solar LCOE declined by 85% over a much shorter period from more than $230 per MWh during the first few years of the market (2007–2010) to $34 per MWh in 2020.
“Though we portray these LCOE reductions over time, it is worth clarifying that within a learning curve framework, it is cumulative deployment, rather than time itself, that drives the cost reduction. In other words, a decline in LCOE over time is a manifestation, but not a direct measurement, of learning,” the researchers report.
Mark Bolinger points out that there is a large amount of literature on learning rate and learning curve theory. For instance, Moore’s Law predicts that the number of transistors per silicon chip doubles each year. Bolinger says that the learning rate — a measure of how much cost declines for each doubling of cumulative output — is similar.
What this research does is bring some carefully calculated predictability to future development of renewables. The more comfortable lenders feel about the safety of their investments, the more money will be available to make them a reality. The predictability is the most important benefit of this research.