Forecasting coal power plant retirement ages and lock-in with random forest regression

Achmed Edianto; Gregory Trencher; Niccolò Manych; Kazuyo Matsubae

doi:10.1016/j.patter.2023.100776

Forecasting coal power plant retirement ages and lock-in with random forest regression

Patterns (N Y). 2023 Jun 21;4(7):100776. doi: 10.1016/j.patter.2023.100776. eCollection 2023 Jul 14.

Authors

Achmed Edianto¹, Gregory Trencher², Niccolò Manych^{3

4}, Kazuyo Matsubae¹

Affiliations

¹ Graduate School of Environmental Studies, Tohoku University, Miyagi, Japan.
² Graduate School of Global Environmental Studies, Kyoto University, Kyoto, Japan.
³ Mercator Research Institute on Global Commons and Climate Change, Berlin, Germany.
⁴ Department Economics of Climate Change, Technische Universität Berlin, Berlin, Germany.

Abstract

Averting dangerous climate change requires expediting the retirement of coal-fired power plants (CFPPs). Given multiple barriers hampering this, here we forecast the future retirement ages of the world's CFPPs. We use supervised machine learning to first learn from the past, determining the factors that influenced historical retirements. We then apply our model to a dataset of 6,541 operating or under-construction units in 66 countries. Based on results, we also forecast associated carbon emissions and the degree to which countries are locked in to coal power. Contrasting with the historical average of roughly 40 years over 2010-2021, our model forecasts earlier retirement for 63% of current CFPP units. This results in 38% less emissions than if assuming historical retirement trends. However, the lock-in index forecasts considerable difficulties to retire CFPPs early in countries with high dependence on coal power, a large capacity or number of units, and young plant ages.

Keywords: coal-fired power plants; emissions; forecasting; lock-in; phase-out; retirement.