pith. sign in

arxiv: 2605.02738 · v1 · submitted 2026-05-04 · 💻 cs.AI

AI and Open-data Driven Scalable Solar Power Profiling

Shiliang Zhang , Sabita Maharjan , Damla Turgut This is my paper

Pith reviewed 2026-05-08 18:40 UTC · model grok-4.3

classification 💻 cs.AI
keywords solar panel detectionsatellite imageryfoundation modelsopen datasolar power profilinggeoreferenced polygonsrenewable energy mappingAI for energy systems
0
0 comments X

The pith

Foundation vision AI models detect solar panel geometries from open satellite imagery to build scalable city solar power profiles without manual labeling.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows how pre-trained vision AI models can identify rooftop solar panels in publicly available satellite photos, turning raw images into precise georeferenced polygon maps. These maps are then combined with open weather records to estimate how much solar power each area can produce. The method works across different image sources and cities because it skips the usual steps of hand-labeling data or training new models for every location. This matters for energy planners and researchers who need up-to-date solar inventories but lack access to costly proprietary tools or large labeled datasets. The authors also release an API so anyone can query solar details for any chosen building or neighborhood.

Core claim

Foundation vision AI models applied to open-source satellite imagery detect solar panel geometries, which are converted into georeferenced polygons; these polygons are then integrated with open weather data to generate spatially explicit and incrementally extensible regional solar power profiles, all without manual data labeling or case-specific model training.

What carries the argument

Foundation vision AI models that detect solar panel geometries directly from open-source satellite imagery and output georeferenced polygons for combination with weather data.

If this is right

  • Users can query any building location via the released API to receive detected solar panel polygons and associated power estimates.
  • The resulting inventories support analysis of distributed solar integration, local power flow optimization, energy tariff design, and infrastructure planning.
  • The approach eliminates reliance on proprietary imagery, manual labeling, and closed-source models while remaining transparent and extensible.
  • City-level solar profiles can be updated incrementally as new open imagery becomes available.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar open-data pipelines could be adapted to track other distributed energy resources such as battery storage or heat pumps.
  • Regular refreshes of the satellite imagery would enable near-real-time monitoring of new solar installations.
  • The method could help quantify solar potential in regions where official statistics are sparse or outdated.

Load-bearing premise

Foundation vision AI models will accurately detect solar panels across varied satellite imagery sources and urban environments without any extra training or validation.

What would settle it

A large-scale test on satellite imagery from multiple new cities showing frequent missed detections or false positives for solar panels on roofs with different materials or angles would falsify the generalization claim.

Figures

Figures reproduced from arXiv: 2605.02738 by Damla Turgut, Sabita Maharjan, Shiliang Zhang.

Figure 1
Figure 1. Figure 1: Example of solar panel polygons (right) from OSM and the view at source ↗
Figure 3
Figure 3. Figure 3: Example of building geometries retrieved from OSM. view at source ↗
Figure 2
Figure 2. Figure 2: Workflow for solar panel detection and geolocation for a single building. GIS denotes geographic information system. view at source ↗
Figure 4
Figure 4. Figure 4: Sources of aerial images and example imagery for the same building view at source ↗
Figure 5
Figure 5. Figure 5: Example of georeferencing solar panel pixels for a single building. view at source ↗
Figure 6
Figure 6. Figure 6: Relationship between sun irradiation and solar panel orientation. view at source ↗
Figure 7
Figure 7. Figure 7: OSM inventory of solar panels for Bulach. Only six panels (right) ¨ are publicly available in OSM. For Berg am Irchel, no panels are labeled, indicating very limited OSM coverage for this area view at source ↗
Figure 9
Figure 9. Figure 9: Cleaned solar panel detection result by SAM3 for B view at source ↗
Figure 10
Figure 10. Figure 10: Example yearly solar power profile (hourly resolution) for the view at source ↗
read the original abstract

Solar photovoltaic (PV) deployment is expanding rapidly, yet detailed, up-to-date information on the spatial distribution and capacity of rooftop PV remains limited. This paper presents an open, scalable framework for detecting solar panels from open data and generating city-level solar power profiles. We leverage foundation vision AI models to detect solar panel geometries from open-source satellite imagery. This avoids manual data labeling and case-specific model training while maintaining robustness across heterogeneous imagery. Detected solar panels are converted into georeferenced polygons, yielding spatially explicit and incrementally extensible inventories. By integrating open weather data, we translate panel footprints into regional solar power profiles. The framework reduces dependency on proprietary imagery, manual labeling, and closed-source models, and offers a transparent and scalable approach for solar planning and analysis. We released the data and an API resulted from this work. For any user-specified building location, our API retrieves aerial imagery, detects rooftop solar panels, and returns georeferenced polygons. This empowers researchers and developers to scan user-defined areas to build solar panel maps and associated solar production profiles, thus facilitating advanced analysis like distributed solar production integration, local power flow optimization, energy tariff design, and infrastructure planning.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper presents an open, scalable framework for detecting solar panels from open-source satellite imagery using foundation vision AI models to generate georeferenced polygons and city-level solar power profiles by integrating open weather data. It avoids manual labeling and proprietary tools, and releases an API that allows users to query aerial imagery, detect panels, and obtain polygons and production profiles for specified locations.

Significance. Should the approach prove reliable, it would provide a significant advancement in open and accessible solar energy data generation, facilitating research and planning in distributed solar systems, power flow optimization, and infrastructure development without the barriers of proprietary data or extensive labeling efforts. The public release of data and API is a strength that enhances reproducibility and usability for the community.

major comments (2)
  1. [Abstract] Abstract: The assertion that the method 'maintains robustness across heterogeneous imagery' and 'avoids ... case-specific model training' is presented without any supporting quantitative evidence such as precision, recall, IoU scores, validation datasets, or failure cases.
  2. [Abstract] Abstract: No description is provided of the exact foundation model(s) employed (e.g., SAM or equivalent), the prompting strategy, or the post-processing steps used to convert detections into georeferenced polygons.
minor comments (1)
  1. The abstract could be strengthened by briefly noting any preliminary qualitative observations or planned validation steps even if full results appear later in the manuscript.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful and constructive review. The comments highlight opportunities to strengthen the abstract, and we address each point below with specific revisions to the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The assertion that the method 'maintains robustness across heterogeneous imagery' and 'avoids ... case-specific model training' is presented without any supporting quantitative evidence such as precision, recall, IoU scores, validation datasets, or failure cases.

    Authors: We acknowledge that the abstract, due to its brevity, does not include quantitative metrics to support the claims of robustness and avoidance of case-specific training. The full manuscript demonstrates these properties through applications to satellite imagery from multiple cities and sources with varying resolutions and conditions (detailed in the results and validation sections), without retraining the foundation models. To directly address this concern, we will revise the abstract to include key quantitative indicators such as average IoU, precision, and recall from our multi-city validation, along with a brief reference to the datasets used. revision: yes

  2. Referee: [Abstract] Abstract: No description is provided of the exact foundation model(s) employed (e.g., SAM or equivalent), the prompting strategy, or the post-processing steps used to convert detections into georeferenced polygons.

    Authors: We agree that the abstract would benefit from greater technical specificity. The manuscript employs the Segment Anything Model (SAM) as the core foundation vision model in a zero-shot setting. The prompting strategy uses bounding-box and point prompts derived from initial coarse detections, followed by post-processing that includes mask refinement, polygon simplification, and georeferencing via coordinate transformation to produce vector polygons. These elements are described in the methods section; we will add a concise summary of the model, prompting approach, and polygon conversion pipeline to the revised abstract. revision: yes

Circularity Check

0 steps flagged

No circularity: descriptive framework without derivations or fitted predictions

full rationale

The paper presents a methodology and tool release for solar panel detection via off-the-shelf foundation vision models on open imagery, followed by georeferencing and weather-data integration for power profiles. No equations, parameter fitting, uniqueness theorems, or prediction steps are described. Central claims concern practicality, scalability, and avoidance of manual labeling; these are not shown to reduce to self-definitions, self-citations, or inputs-by-construction. The work is self-contained as an applied framework with released API/data rather than a mathematical derivation chain.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework depends on the unproven generalization of foundation models to this domain and on the accuracy of open weather data for power translation; no free parameters or new entities are introduced in the abstract.

axioms (1)
  • domain assumption Foundation vision AI models detect solar panel geometries robustly from heterogeneous open-source satellite imagery without case-specific training or labeling.
    Explicitly invoked in the abstract as the mechanism that avoids manual labeling and maintains robustness.

pith-pipeline@v0.9.0 · 5504 in / 1158 out tokens · 68131 ms · 2026-05-08T18:40:04.094520+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

24 extracted references · 24 canonical work pages

  1. [1]

    Global photovoltaic solar panel dataset from 2019 to 2022.Scientific Data, 12(1):637, 2025

    Anqi Li, Luling Liu, Shijie Li, Xihong Cui, Xuehong Chen, and Xin Cao. Global photovoltaic solar panel dataset from 2019 to 2022.Scientific Data, 12(1):637, 2025

    work page 2019

  2. [2]

    Transformer approach to nowcasting solar energy using geostationary satellite data.Applied Energy, 377:124387, 2025

    Ruohan Li, Dongdong Wang, Zhihao Wang, Shunlin Liang, Zhanqing Li, Yiqun Xie, and Jiena He. Transformer approach to nowcasting solar energy using geostationary satellite data.Applied Energy, 377:124387, 2025

    work page 2025

  3. [3]

    The road to fair energy data: Motivations, shortcomings, and use cases.International Informa- tion & Library Review, 57(2):177–192, 2025

    Rawia Awadallah and Anna Maria Tammaro. The road to fair energy data: Motivations, shortcomings, and use cases.International Informa- tion & Library Review, 57(2):177–192, 2025

    work page 2025

  4. [4]

    Performance monitoring of photovoltaic modules using machine-learning-based solutions: A survey of current trends.Energy Science & Engineering, 14(3):1663–1682, 2026

    Nangamso Nathaniel Nyangiwe, Abraham Dimitri Kapim Kenfack, Nicolas Thantsha, and Mandla Msimanga. Performance monitoring of photovoltaic modules using machine-learning-based solutions: A survey of current trends.Energy Science & Engineering, 14(3):1663–1682, 2026

    work page 2026

  5. [5]

    Zoubir Barraz, Imane Sebari, Nassim Lamrini, Kenza Ait El Kadi, and Ibtihal Ait Abdelmoula. Fast and automatic solar module geo-labeling for optimized large-scale photovoltaic systems inspection from uav thermal imagery using deep learning segmentation.Cleaner Engineering and Technology, page 101048, 2025

    work page 2025

  6. [6]

    Development of a smart cloud-based monitoring system for solar photo- voltaic energy generation.Unconventional Resources, 6:100173, 2025

    Challa Krishna Rao, Sarat Kumar Sahoo, and Franco Fernando Yanine. Development of a smart cloud-based monitoring system for solar photo- voltaic energy generation.Unconventional Resources, 6:100173, 2025

    work page 2025

  7. [7]

    Leveraging large- scale aerial data for accurate urban rooftop solar potential estimation via multitask learning.Solar Energy, 290:113336, 2025

    Alessia Boccalatte, Ankit Jha, and Jocelyn Chanussot. Leveraging large- scale aerial data for accurate urban rooftop solar potential estimation via multitask learning.Solar Energy, 290:113336, 2025

    work page 2025

  8. [8]

    Detection of solder paste defects with an optimization-based deep learning model using image processing techniques.Soldering & Surface Mount Technology, 33(5):291–298, 2021

    Ali Sezer and Aytac ¸ Altan. Detection of solder paste defects with an optimization-based deep learning model using image processing techniques.Soldering & Surface Mount Technology, 33(5):291–298, 2021

    work page 2021

  9. [9]

    Automated detection and tracking of photovoltaic modules from 3d remote sensing data.Applied Energy, 367:123242, 2024

    Andressa Cardoso, David Jurado-Rodr ´ıguez, Alfonso L ´opez, M Isabel Ramos, and Juan Manuel Jurado. Automated detection and tracking of photovoltaic modules from 3d remote sensing data.Applied Energy, 367:123242, 2024

    work page 2024

  10. [10]

    Boost-up efficiency of defective solar panel detection with pre-trained attention recycling.IEEE Transactions on Industry Applications, 59(3):3110– 3120, 2023

    YeongHyeon Park, Myung Jin Kim, Uju Gim, and Juneho Yi. Boost-up efficiency of defective solar panel detection with pre-trained attention recycling.IEEE Transactions on Industry Applications, 59(3):3110– 3120, 2023

    work page 2023

  11. [11]

    Ruiqing Yang, Guojin He, Ranyu Yin, Guizhou Wang, Xueli Peng, Zhaoming Zhang, Tengfei Long, Yan Peng, and Jianping Wang. A large- scale ultra-high-resolution segmentation dataset augmentation frame- work for photovoltaic panels in photovoltaic power plants based on priori knowledge.Applied Energy, 390:125879, 2025

    work page 2025

  12. [12]

    Solar photovoltaic assessment with large language model.Applied Energy, 402:126835, 2025

    Muhao Guo and Yang Weng. Solar photovoltaic assessment with large language model.Applied Energy, 402:126835, 2025

    work page 2025

  13. [13]

    GeoPixel: Pixel grounding large multimodal model in remote sensing

    Akashah Shabbir, Mohammed Zumri, Mohammed Bennamoun, Fa- had Shahbaz Khan, and Salman Khan. GeoPixel: Pixel grounding large multimodal model in remote sensing. InProceedings of the 42nd Inter- national Conference on Machine Learning, volume 267 ofProceedings of Machine Learning Research, pages 54095–54111. PMLR, 13–19 Jul 2025

    work page 2025

  14. [14]

    Extracting and analysing geographic information from natural language texts, 2026

    Xuke Hu, Ross S Purves, Ludovic Moncla, Jens Kersten, and Kristin Stock. Extracting and analysing geographic information from natural language texts, 2026

    work page 2026

  15. [15]

    Col- laborative inspection of solar panel farms using yolov5-based computer vision and ugv-uav integration.Journal of Intelligent & Robotic Systems, 111(2):66, 2025

    Johann SJC Amorim, Accacio FS Neto, Rafael S Chaves, Alessandro RL Zachi, Josiel A Gouv ˆea, Fabio AA Andrade, and Milena F Pinto. Col- laborative inspection of solar panel farms using yolov5-based computer vision and ugv-uav integration.Journal of Intelligent & Robotic Systems, 111(2):66, 2025

    work page 2025

  16. [16]

    Data-centric artificial intelligence: A survey.ACM Comput

    Daochen Zha, Zaid Pervaiz Bhat, Kwei-Herng Lai, Fan Yang, Zhimeng Jiang, Shaochen Zhong, and Xia Hu. Data-centric artificial intelligence: A survey.ACM Comput. Surv., 57(5), January 2025

    work page 2025

  17. [17]

    Distributed solar photovoltaic array location and extent dataset for remote sensing object identification.Scientific data, 3(1):160106, 2016

    Kyle Bradbury, Raghav Saboo, Timothy L Johnson, Jordan M Malof, Ar- jun Devarajan, Wuming Zhang, Leslie M Collins, and Richard G Newell. Distributed solar photovoltaic array location and extent dataset for remote sensing object identification.Scientific data, 3(1):160106, 2016

    work page 2016

  18. [18]

    Descriptor: Norwegian electricity in geographic dataset (noregeo).IEEE Data Descriptions, 3:82–92, 2026

    Shiliang Zhang, Sabita Maharjan, Kai Strunz, and Jan Christian Bryne. Descriptor: Norwegian electricity in geographic dataset (noregeo).IEEE Data Descriptions, 3:82–92, 2026

    work page 2026

  19. [19]

    Crowdsourced data for urban planning: A critical evaluation of openstreetmap accuracy and completeness

    Federica Deri, Federico Mara, and Chiara Anselmi. Crowdsourced data for urban planning: A critical evaluation of openstreetmap accuracy and completeness. InInternational Conference on Computational Science and Its Applications, pages 403–420. Springer, 2025

    work page 2025

  20. [20]

    Analysing the use of openaerialmap images for openstreetmap edits.Geo-spatial Information Science, 28(3):1179–1194, 2025

    Ammar Mandourah and Hartwig H Hochmair. Analysing the use of openaerialmap images for openstreetmap edits.Geo-spatial Information Science, 28(3):1179–1194, 2025

    work page 2025

  21. [21]

    The astronomical almanac’s algorithm for approx- imate solar position (1950–2050).Solar energy, 40(3):227–235, 1988

    Joseph J Michalsky. The astronomical almanac’s algorithm for approx- imate solar position (1950–2050).Solar energy, 40(3):227–235, 1988

    work page 1950

  22. [22]

    Modeling daylight availability and irradiance compo- nents from direct and global irradiance.Solar energy, 44(5):271–289, 1990

    Richard Perez, Pierre Ineichen, Robert Seals, Joseph Michalsky, and Ronald Stewart. Modeling daylight availability and irradiance compo- nents from direct and global irradiance.Solar energy, 44(5):271–289, 1990

    work page 1990

  23. [23]

    Assessing the outdoor operating temperature of photo- voltaic modules.Progress in Photovoltaics: Research and Applications, 16(4):307–315, 2008

    David Faiman. Assessing the outdoor operating temperature of photo- voltaic modules.Progress in Photovoltaics: Research and Applications, 16(4):307–315, 2008

    work page 2008

  24. [24]

    United States

    David L King, Jay A Kratochvil, and William Earl Boyson.Photovoltaic array performance model, volume 8. United States. Department of Energy, 2004

    work page 2004