A fully automated urban PV parameterization framework for improved estimation of energy production profiles

TL;DR

This paper presents an automated framework that uses remote sensing and GIS data to accurately parameterize urban PV systems, significantly improving energy production estimates and supporting large-scale solar deployment.

Contribution

It introduces a novel, fully automated method for extracting detailed PV system parameters from remote sensing data, enhancing the accuracy of energy modeling in urban environments.

Findings

Achieved R^2 of 0.92 with DSO records in Eindhoven.

Capacity estimates within ±25% for 73% of neighborhoods.

Reduced MAPE of energy forecasts by up to 160%.

Abstract

Accurate parameterization of rooftop photovoltaic (PV) installations is critical for effective grid management and strategic large-scale solar deployment. The lack of high-fidelity datasets for PV configuration parameters often compels practitioners to rely on coarse assumptions, undermining both the temporal and numerical accuracy of large-scale PV performance modeling. This study introduces a fully automated framework that innovatively integrates remote sensing data, semantic segmentation, polygon-vector refinement, tilt-azimuth estimation, and module layout inference to produce a richly attributed GIS dataset of distributed PV. Applied to Eindhoven (the Netherlands), the method achieves a correlation ($R^2$) of 0.92 with Distribution System Operator (DSO) records, while capacity estimates for 73$\%$ of neighborhoods demonstrate agreement within a $\pm$25$\%$ margin of recorded data. Additionally, by accurately capturing actual system configuration parameters (e.g., tilt, azimuth, module layout) and seamlessly linking them to advanced performance models, the method yields more reliable PV energy generation forecasts within the distribution networks. Centering our experiments toward a high PV-penetration community, configuration-aware simulations help to reduce Mean Absolute Percentage Error (MAPE) of energy generation modeling by up to 160$\%$ compared to the conventional assumption-based approaches. Furthermore, owing to its modular design and reliance on readily available geospatial resources, the workflow can be extended across diverse regions, offering a scalable solution for robust urban solar integration.