Learning Curves in Prospective Life Cycle Assessment

Abstract

Environmental learning curves have great potential to predict future changes in environmental footprints of technologies as part of prospective life cycle assessments. However, concrete guidance is currently missing on how to integrate environmental learning curves into prospective life cycle assessments. Here, we propose a method to combine (i) process-specific environmental learning curves for key technology parameters and (ii) projections from integrated assessment models to include relevant changes in background processes, such as expected decarbonization of the electricity grid. Our method enables process contribution analyses, uncertainty and sensitivity analyses, and flexibility in the assessment of various impact categories. Application of our proposed method is demonstrated in a case study assessing various environmental footprints of producing monocrystalline silicon photovoltaic panels. We showed that environmental footprints reduce 21-80% between 2020 and 2050 through a synergy of (i) and (ii). Footprint reductions were mostly driven by background changes when decarbonization is extensive, whereas process-specific environmental learning curves become the major driver for footprint reductions when developments in background processes follow a similar trajectory as charted by the past. Our method may also be used in the assessment of emerging technologies by applying process-specific environmental learning curves to mature parts of their supply chain.

Keywords: LCA; emerging technology; environmental footprint; ex ante; experience curve; photovoltaics; technological change; technological learning.

1

Flowchart of the proposed method for application of learning in foreground and background processes in prospective LCA.

2

Sankey diagram of the supply chain for the production of 1 W_p of PERC solar panel capacity. Values behind each colon represent the GHG footprint of that product in kg CO₂-eq per W _p. MG-Si: metallurgical grade silicon; poly-Si: poly silicon; Cz-sc-Si: Czochralski single-crystalline silicon.

3

11 process-specific learning curves for which data were obtained. Individual empirical observations are represented with black dots. Cumulative installed capacities, as well as these reported observations were log₁₀ transformed to enable ordinary least-squares fitting. The dark line represents the fitted learning curve and the shaded area represents the 95% confidence intervals, quantifying the uncertainty in the learning curve fit. Colored dots represented sampled values obtained through Monte Carlo simulation. a ₀: intercept, β: slope; LR: learning rate; cons.: consumption; prod.: production; MG-Si: metallurgical grade silicon; poly-Si: poly silicon; Cz-sc-Si: Czochralski single-crystalline silicon; SSP: shared socio-economic pathway; and RCP: representative concentration pathway.

4

Results for the ReCiPe 2016 (H) climate change impact category. Empirical learning curve based on Louwen et al. for the GHG footprints of monocrystalline silicon PV systems reported in the literature, with colored violin plots superimposed that represent the values predicted using process-specific learning curves for the foreground and IAM projections for the background. kg CO₂-equiv: kilogram carbon dioxide equivalent; W _p: Watt-peak; MW: megawatt; a ₀: initial GHG footprint; β: learning parameter; LR: learning rate; SSP: shared socio-economic pathway; and RPC: representative concentration pathway.

5

One-at-a-time sensitivity analyses showing how sensitive the impact reductions between 2020 and 2050 are to modeled developments in only the foreground system (F), only the background system (B), or both (F + B). kg CO₂-equiv: kilogram carbon dioxide equivalent; W _p: Watt-peak; F: foreground; B: background; SSP: shared socio-economic pathway; and RPC: representative concentration pathway.

6

Uncertainty analyses showing the Spearman’s rank correlation coefficients relating the 1000 GHG footprints obtained for each scenario against the 11 process parameters adapted in the foreground system using learning curves. SSP: shared socio-economic pathway; RPC: representative concentration pathway; MG-Si: metallurgical grade silicon; poly-Si: poly silicon; and Cz-sc-Si: Czochralski single-crystalline silicon.