Probabilistic projections of granular energy technology diffusion at subnational level.

Projections of granular energy technology diffusion can support decision-making on climate mitigation policies and infrastructure investments. However, such projections often do not account for uncertainties and have low spatial resolution. S-curve models of technology diffusion are widely used to project future installations, but the results of the different models can vary significantly. We propose a method to create probabilistic projections of granular energy technology diffusion at subnational level based on historical time series data and testing how various projection models perform in terms of accuracy and uncertainty to inform the choice of models. As a case study, we investigate the growth of solar photovoltaics, heat pumps, and battery electric vehicles at municipality level throughout Switzerland in 2000-2021 (testing) and until 2050 (projections). Consistently for all S-curve models and technologies, we find that the medians of the probabilistic projections anticipate the diffusion of the technologies more accurately than the respective deterministic projections. While accuracy and probabilistic density intervals of the models vary across technologies, municipalities, and years, Bertalanffy and two versions of the generalized Richards model estimate the future diffusion with higher accuracy and sharpness than logistic, Gompertz, and Bass models. The results also highlight that all models come with trade-offs and eventually a combination of models with weights is needed. Based on these weighted probabilistic projections, we show that, given the current dynamics of diffusion in solar photovoltaics, heat pumps, and battery electric vehicles in Switzerland, the net-zero emissions target would be missed by 2050 with high certainty.
(© The Author(s) 2023. Published by Oxford University Press on behalf of National Academy of Sciences.)