Empirically grounded technology forecasts and the energy transition

•Empirically validated probabilistic forecasts of energy technology costs•Future system costs are estimated for three different scenarios•A rapid green transition will likely result in trillions net savings•Energy models should be updated to reflect high probability low-cost renewables Decisions about how and when decarbonize the global highly influenced by estimates cost. Most energy-economy have produced scenarios that overestimate due underestimating renewable cost improvements deployment rates. This paper generates technologies using a method has been statistically on data more than 50 technologies. Using this approach estimate future under scenarios, we find compared contuinuing with fossil fuel-based system, is savings. Hence, even without accounting climate damages or policy co-benefits, transitioning net-zero 2050 economically beneficial. Updating expectations could dramatically accelerate decarbonization systems. Rapidly decarbonizing critical addressing change, but concerns barrier implementation. historically underestimated rates overestimated their costs. These issues driven calls alternative approaches reliable forecasting methods. Here, use an based methods backtesting We generate solar energy, wind batteries, electrolyzers, conditional deployment. these explore uncertainty propagates through scenarios. Compared continuing overall savings many dollars—even co-benefits policy. Future determined combination produce, store, distribute energy. Their change time innovation, competition, public policy, other factors. To provide some perspective Figure 1 shows landscape evolved over last 140 years. 1A historical principal technologies, 1B gives deployment; both which logarithmic scale. As present 1A, diagram becomes congested, making it clear period unprecedented diversity, average around $100/MWh competing dominance. The long-term trends clue as competition may resolved: prices fuels such coal, oil, gas volatile, after adjusting inflation, now very similar what they were years ago, there no obvious long-range trend. In contrast, several decades photovoltaics (PV), wind, batteries dropped (roughly) exponentially at rate near 10% per year. PV decreased orders magnitude since its first commercial 1958.1Perlin J. From Space Earth: Story Solar Electricity. Aatech Publications, 1999Google Scholar 1880, coal passed traditional biomass. It documents slow exponential rise production oil natural century plateauing nuclear But perhaps most remarkable feature dramatic PV, electrolyzers transitioned from niche applications mass markets. increase 1970s, unlike all consistently experienced decreasing increasing anything observed any past, positions key challenge dominance within decade. How clean continue drop future? Under conditions happen, does imply transition? Is path forward can get us emissions cheaply quickly? address questions here applying empirically tested, state-of-the-art Historically, costs2Creutzig F. Agoston P. Goldschmidt J.C. Luderer G. Nemet Pietzcker R.C. potential mitigate change.Nat. Energy. 2017; 217140https://doi.org/10.1038/nenergy.2017.140Crossref Scopus (426) Google Scholar, 3Xiao M. Junne T. Haas Klein Plummeting - lagging?.Energy Strategy Rev. 2021; 35100636https://doi.org/10.1016/j.esr.2021.100636Crossref (42) 4Jaxa-Rozen Trutnevyte E. Sources photovoltaic technology.Nat. Clim. Change. 11: 266-273https://doi.org/10.1038/s41558-021-00998-8Crossref (39) 5Hoekstra Auke Steinbuch Verbong Creating agent-based management uncover profitable pathways mitigation.Complexity. 2017: 1-23https://doi.org/10.1155/2017/1967645Crossref (40) 6Shiraki H. Sugiyama Back basic: toward improvement technoeconomic representation integrated assessment models.Clim. 2020; 162: 13-24https://doi.org/10.1007/s10584-020-02731-4Crossref (8) 7Victoria Haegel N. Peters I.M. Sinton R. Jäger-Waldau A. del Cañizo C. Breyer Stocks Blakers Kaizuka I. et al.Solar ready power sustainable future.Joule. 5: 1041-1056https://doi.org/10.1016/j.joule.2021.03.005Abstract Full Text PDF (142) led 8Stern Economics: current grossly misleading.Nature. 2016; 530: 407-409https://doi.org/10.1038/530407aCrossref PubMed (172) 9Pindyck R.S. Climate policy: do tell us?.J. Econ. Lit. 2013; 51: 860-872https://doi.org/10.1257/jel.51.3.860Crossref (445) 10Farmer J.D. Hepburn Mealy Teytelboym A third wave economics change.Environ. Resour. 2015; 62: 329-357Crossref (140) 11Gambhir Butnar Li P.-H. Smith Strachan review criticisms proposed these, lens beccs.Energies. 2019; 12: 1747https://doi.org/10.3390/en12091747Crossref (84) 12McCollum D.L. Gambhir Rogelj Wilson Energy modellers extremes systematically scenarios.Nat. 104-107https://doi.org/10.1038/s41560-020-0555-3Crossref (47) 13Lovins A.B. Ürge-Vorsatz D. Mundaca L. Kammen D.M. Glassman J.W. Recalibrating prospects.Environ. Res. Lett. 14120201Crossref (14) 14Pye S. Broad O. Bataille Brockway Daly H.E. Freeman Geden Rogan Sanghvi al.Modelling systems requires approach.Clim. Policy. 21: 222-231https://doi.org/10.1080/14693062.2020.1824891Crossref (55) 15Stern Stiglitz J.E. Taylor immense risk, urgent action radical change: towards new change.J. Methodol. 29: 1-36Google Scholar. Recent efforts made progress direction16Grubler Bento Boza-Kiss B. Krey V. McCollum Rao N.D. Riahi K. De Stercke al.A low demand scenario meeting 1.5°C target development goals negative emission technologies.Nat. 2018; 3: 515-527https://doi.org/10.1038/s41560-018-0172-6Crossref (552) 17Victoria Zhu Brown Andresen G.B. Greiner Early decarbonisation European pays off.Nat. Commun. 6223https://doi.org/10.1038/s41467-020-20015-4Crossref (75) 18He Lin Sifuentes Liu X. Abhyankar Phadke Rapid decrease storage accelerates china’s system.Nat. 2486https://doi.org/10.1038/s41467-020-16184-xCrossref (135) 19Bogdanov Ram Aghahosseini Gulagi Oyewo A.S. Child Caldera U. Sadovskaia Farfan Souza Noel Simas Barbosa al.Low-cost electricity driver sustainability.Energy. 227120467https://doi.org/10.1016/j.energy.2021.120467Crossref (204) largely deterministic nature. probabilistic, allowing view placing bets After all, powering modern economies betting one way another, fuels—the best make good bets. Which bet on, pay off? focus solar, call “key technologies”, because play crucial roles strong well documented publicly available datasets. also consider major incumbent compare our projections influential models. investigate discuss implications whole pathways. provides glimpse into diverse nature technological fall dominance.20Arthur W.B. Nature Technology. What Evolves. Free Press, 2011Google 21Fouquet Heat, Power Light: Revolutions Services. Edward Elgar 2008Google 22Grübler Naki?enovi? Victor D.G. Dynamics change.Energy 1999; 27: 247-280https://doi.org/10.1016/S0301-4215(98)00067-6Crossref (487) reflects innovation learning produce outcomes diversity seen typical general.23Koh Heebyung Magee C.L. functional studying progress: application information technology.Technol. Forecasting Soc. 2006; 73: 1061-1083https://doi.org/10.1016/j.techfore.2006.06.001Crossref (97) 24Koh extension 2008; 75: 735-758https://doi.org/10.1016/j.techfore.2007.05.007Crossref (68) 25Nagy Farmer Bui Q.M. Trancik Statistical basis predicting progress.PLoS One. 8e52669Crossref (155) Roughly speaking, divided two groups improvement. For group, comprising vast majority inflation-adjusted remained roughly constant time. Fossil example: although enormous discovery extraction, easily accessible resources depleted, necessary extract less resources, creating “running-to-stand-still” dynamic (this true minerals26U.S. Geological SurveyMineral Commodity Summaries 2022. Technical report.http://pubs.er.usgs.gov/publication/mcs2022Date: 2022Google Scholar,27Trancik Jean Kavlak Klemun M.M. Edwards M.R. McNerney Miotti P.R. Mueller J.M. Needell Z.A. Technology Improvement Emissions Reductions Mutually Reinforcing Efforts: Observations Global Development Wind Working paper. MIT 2015Google Scholar). Another example non-improving carbon capture (CCS); despite significant effort, 50-year history enhanced recovery, not declined all.28Reiner Learning portfolio demonstration projects.Nat. 1: 1-7Crossref (165) Scholar,29Rubin E.S. Davison Herzog H.J. CO2 storage.Int. Greenhouse Gas Control. 40 (Special Issue commemorating 10th year anniversary publication Intergovernmental Panel Change Special Report Capture Storage): 378-400https://doi.org/10.1016/j.ijggc.2015.05.018Crossref (553) There cases, power, where increased. By select group improving while increased exponentially.23Koh Rates optical fibers transistors 40%–50% behaved similarly closer (see Document S1 section “The heterogeneity persistence change”). makes unit predictable, if specific innovations lead lower predictable. Because behavior so different, require tradable commodities, according efficient markets theory, follow random walk.30Malkiel B.G. Random Walk down Wall Street. Norton & Co, 1973Google useful approximation decade, longer spans display mean reversion.31Pindyck long-run evolution prices.Energy 2019449089http://www.jstor.org/stable/41322828Crossref Scholar,32Shafiee Topal worldwide fuel prices.Appl. 2010; 87: 988-1000https://doi.org/10.1016/j.apenergy.2009.09.012Crossref (81) autoregressive choice, them forecast Experimental procedures sections “AR(1) process,” “Oil,” “Coal,” “Gas”). improving, remarkably consistent.33Farmer Lafond predictable progress?.Res. 45: 647-665Crossref (146) dominant quantitatively data. generalized form Moore’s law, says function (i.e., fixed percentage year).23Koh Scholar,24Koh Scholar,34Moore G.E. Cramming components onto circuits.Proc. IEEE. 1998; 86: 82-85Crossref (1179) second Wright’s predicts law cumulative production.35Wright T.P. Factors affecting airplanes.J. Aeronaut. Sci. 1936; 122-128Crossref relationship called experience curve curve, experience. (For discussion challenges caveats concerning see “Wright’s caveats.”) Multifactor additional input variables, patenting activity research (R&D) expenditures, limited parameters. overfitting, resulting poor out-of-sample forecasts25Nagy “Bias-variance trade-off”). far properly here. Successful tend “S-curve” deployment, starting long phase growth eventually tapers off market saturation.22Grübler during time, slows, slows. Improving often spend phase, hard distinguish between law. Forecasts accuracy experiments.25Nagy brings up important question responsiveness investment. assumed exogenously independent depend Although directly cause drop, correlated factors do, level effort R&D, essential advantage being relatively easy measure.36Thompson quantity evidence organizational learning-by-doing.J. Perspect. 2012; 26: 203-224Crossref (48) Scholar,37Witajewski-Baltvilks Verdolini Tavoni Bending curve.Energy 52: S86-S99https://doi.org/10.1016/j.eneco.2015.09.007Crossref (19) comparison, series displayed plotted curves S17. same evident law—the fact means had effect—in stark contrast paper, satisfies basic intuition exerting greater induces effects. (We repeated modeling experiments found qualitative conclusions similar; “Moore’s results.” thorough causality, “Discussion causality.”) usually used point forecasts, meaning experience, error forecast. attempts introducing bars did priori forms, requirements testing prohibitive.25Nagy Scholar,38Alberth via curve—myth magic?.Technol. 952-983Crossref (50) More recently, derived predict volatility, number points forecasting.33Farmer Scholar,39Lafond Bailey A.G. Bakker Rebois Zadourian McSharry progress? distributional forecasts.Technol. 128: 104-117Crossref (30) Based comprehensive backtesting, was shown done selecting reference dates past then, only horizons 20 respect each date. 6,000 closely matched ahead.33Farmer Our main contribution apply method—which stochastic law—to transition. test applied polymer electrolyte membrane (PEM) electrolyzers; results 2. Data prior parameters, then subsequent conditioned reasonable: values lie 95% confidence interval (CI), consistent estimates. expected, decreases later Due short dataset accurate, intervals this. instructive 2 outputs inform (IPCC) guide policy.40IPCCMitigation change. Contribution working Group III sixth report Change.in: Cambridge University Integrated (IAMs) evaluate policies given targets assumption optimal decision-making economic agents. typically “projections” indicate intended forecasts. 3 emphasizes point. 3A International Agency’s (IEA) World Model IAMs compares correspond aggressive highest i.e., those optimistic declines. Nonetheless, projected much higher trends. inset histogram 2,905 annual investment would 2010 2020, reported nine separate IAM teams AMPERE comparison project.41Riahi Kriegler Johnson Bertram den Elzen Eom Schaeffer Edmonds Isaac al.Locked Copenhagen pledges — short-term feasibility goals.Technol. 90: 8-23https://doi.org/10.1016/j.techfore.2013.09.016Crossref (248) value reductions 2.6%, 6%. period, actually fell 15% bad idea treat produces known (and published 2020 costs, version indeed accurate IEA time44Ferguson C.D. Marburger L.E. Doyne Makhijani US future?.Nature. 467: 391-393https://doi.org/10.1038/467391aCrossref (9) One illustrate planning (Note better mature fuels, costs.).2Creutzig 45Wilson Grubler Bauer capacity technologies: evidence?.Clim. 118: 381-395https://doi.org/10.1007/s10584-012-0618-yCrossref (90) widely IAMs.46Anandarajah McDowall W. Ekins Decarbonising road transport hydrogen electricity: term scenarios.Int. Hydr. 38: 3419-3432Crossref (107) 47Heuberger C.F. Rubin Staffell Shah Mac Dowell expansion considering endogenous learning.Appl. 204: 831-845Crossref (110) 4