Researchers from the University of Adelaide, Sophia University in Japan, and South Australia-based commercial construction company Kennett have developed correlation and spatial validation models to characterise the relationships between economic, social, geographical, and climatic features and PV decommissioning dynamics, including PV waste generation and PV lifespan.
“Most studies on end-of-life solar panels focus on how much waste will be generated and when it will appear,” said corresponding author Jian Zuo to pv magazine. “Our research goes one step further by asking why these patterns differ across regions. Instead of only projecting future PV waste, we investigated how broader economic, social, geographical, and climatic conditions are associated with PV waste generation and PV lifespan.”
Zuo, from the University of Adelaide, added that by combining machine learning with spatial analysis, the group provides “a new way to understand the regional context behind PV decommissioning, offering evidence that can support more informed planning for future PV recycling and product stewardship.”
The researchers developed a three-stage machine-learning framework to identify the contextual factors associated with PV decommissioning in Australia. Using data from 2,634 postcodes, they trained machine-learning algorithms to predict previously modeled PV waste generation and PV lifespan under three Australian energy market operator decarbonisation scenarios: progress change (PC), step change (SC), and green energy export (GEE).
They used 15 higher-level contextual variables for their prediction, which comprised three economic indicators: household income, rent, mortgage repayments; four social indicators: population, families, household size, private dwellings; two geographical indicators: PV power output and optimum module tilt; and six climatic indicators: rainfall, temperature, direct normal irradiation, global horizontal irradiation, diffuse horizontal irradiation, and global tilted irradiation.
In the first stage of the framework, five machine-learning models -XGBoost, LightGBM, CatBoost, long short-term memory (LSTM), and support vector machines (SVM) were trained separately for PV waste generation and PV lifespan. XGBoost achieved the best performance for PV waste prediction, while LightGBM performed best for PV lifespan.
In the second stage, the researchers used SHapley Additive exPlanations (SHAP) to quantify how strongly each contextual variable contributed to the predictions and whether its influence was positive or negative. Finally, they applied multiscale geographically weighted regression (MGWR) to test whether these SHAP-derived relationships remained statistically significant and spatially consistent across Australia.
“One of the most interesting findings was that different types of factors matter for different aspects of PV decommissioning,” Zuo said. “We found that PV waste generation is more strongly associated with socioeconomic features, such as population, families, dwellings, income, and rent; while PV lifespan is more strongly associated with geographical and climatic conditions, including rainfall, temperature, and solar irradiation.”
Zuo said that these relationships vary considerably across Australia, suggesting that future PV recycling challenges are unlikely to be the same everywhere.
“My team is planning to conduct more studies examining the dynamic interactions among policy, economic conditions, technological advances, and stakeholder attitudes, as it is essential for understanding solar system retirement decisions,” he said.
The framework has appeared in “Spatiotemporal correlations of PV decommissioning: Machine-learning insights from socioeconomic and geoclimatic data,” published in Environmental Impact Assessment Review. Researchers from Australia’s Adelaide University, commercial construction company Kennett and Japan’s Sophia University contributed to the study.
This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: [email protected].
Comments
Please login to comment
The new issue of pv magazine Global is out now!
Available in print and digital – get your copy today!
Tuesday, August 11, 2026
3:00 pm – 4:00 pm CEST, Berlin, Paris, Madrid
Entries open in seven categories: Modules, Inverters, BoS, BESS, Manufacturing, Sustainability, Projects.
April 01 – August 31, 2026
Tuesday, August 25, 2026
10:00 am – 11:00 am CEST, Berlin, Paris, Madrid
Thursday, October 7, 2026
11:00 am – 12:30 pm CEST, Berlin, Paris, Madrid
You have no items in your basket.
