Shining a light on rooftop solar adoption: investigating key drivers in the City of Cape Town using agent-based modelling

Pande, Puja

Shining a light on rooftop solar adoption: investigating key drivers in the City of Cape Town using agent-based modelling

Thesis / Dissertation

2026

Publisher

University of Cape Town

Department

Department of Statistical Sciences

Faculty

Faculty of Science

Abstract

As global electricity markets evolve in response to rising energy demands and environmental concerns, renewable energy technologies, including rooftop solar, are becoming better integrated into national and local systems. While transitions in the Global North are environmentally motivated, those in the Global South, particularly in rural areas, are focused on improving supply security. In South Africa, ongoing load-shedding (which are scheduled blackouts) has accelerated the shift to renewable energy, with rooftop solar emerging as a viable and increasingly adopted solution. This study addresses a critical gap in the literature by examining rooftop solar adoption in the South African context, a topic that remains under-explored both locally and more broadly within the Global South. The research adapts and extends the Behavioural change in ENergy Consumption of Households (BENCH) agent-based model, tailoring it to the City of Cape Town's socio-economic context and behavioural dynamics. Similarly to the BENCH agent-based model, this model uniquely integrates non-economic variables, such as socio-demographic factors, household perceptions, and environmental attitudes alongside economic factors into a behavioural simulation of household decision-making. The model is calibrated using data from Statistics South Africa's General Household Survey and the Afrobarometer dataset, yielding a localised and nuanced representation of potential adopters. The model is applied to three empirical scenarios (i.e., base case, no trust in utility reliability, and reduced utility threshold) and one proof-of-concept policy scenario i.e., a rooftop solar rebate), each simulated under five learning scenarios that vary the degree and nature of social interaction. Results reveal no consistently superior learning strategy, with the no-learning configuration often performing comparably or better than socially informed ones, suggesting that social learning can at times inhibit adoption. Adoption levels were particularly sensitive to the utility threshold parameter, and the inclusion of a trust-in-utility attribute reduced uptake, underscoring the behavioural complexity of energy decisions. Among predictor variables, economic comfort emerged as the strongest correlate of adoption, suggesting targeted financial incentives may enhance adoption if combined with other enabling conditions. Despite its insights, the model is limited by data constraints, especially the absence of fine-grained, context-specific information on household-level behavioural and socio-environmental factors for the City of Cape Town. As such, the results—particularly from the policy scenarios—should be interpreted as illustrative rather than predictive. Nevertheless, the model provides a robust proof-of-concept that demonstrates the potential of agent-based approaches to inform decentralised energy transitions. These constraints highlight the need for improved data availability in the South African energy context to refine models and support evidence-based energy policy decisions.