davecook1985/solar-panel-yield-2026

Solar Panel Cleaning Yield Recovery — Datasets

Open data companion to the Solar Panel Cleaning Yield Recovery working paper and reference calculator. Seven CSV datasets covering the four technical domains that determine when and how a PV array should be cleaned:

1. Soiling physics — how fast transmittance drops as dust accumulates, by climate zone and panel tilt.
2. Water-fed pole (WFP) engineering — deionized-water resin capacity as a function of inlet TDS, and PV geometry → pole length reach.
3. Regional solar resource — monthly irradiance for six Ontario cities used in the cleaning ROI model.
4. Cleaning economics — annual yield recovered as a function of cleaning interval for residential, commercial, and utility-scale systems, with full payback arithmetic.

A seventh file is a plain-language regulation crosswalk (OSHA, CSA, IEC, ASTM, Health Canada, NRCan, ECCC, CanREA).

Configurations

Config	Rows	Description
soiling_loss_rates	660	Days-since-cleaning × climate zone × tilt × soil class → transmittance loss %
tds_resin_capacity	50	Inlet TDS × DI cartridge size → gallons treated to ≤ 10 ppm outlet
pv_geometry_reach	12	PV installation archetypes → minimum and recommended WFP pole length (ft)
regional_irradiance_ontario	78	Six Ontario cities × 12 months + annual avg → kWh/m²/day
cleaning_frequency_roi	81	System class × city × cleaning interval → annual yield recovered, cleaning cost, net ROI
tap_water_tds_profiles	28	Canadian municipalities → mean/low/high tap water TDS (ppm)
regulation_crosswalk	14	Topic → authority → citation for OSHA, CSA, IEC, ASTM, Health Canada, NRCan, ECCC, CanREA

Multi-config rationale

These seven CSVs have deliberately different schemas (different primary keys, different dimensions). The Hugging Face dataset viewer's default behaviour is to concatenate every CSV in the repo into a single table, which fails with DatasetGenerationCastError when schemas don't match. The configs: block in this card declares each CSV as its own configuration so the viewer loads each one independently. See: https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations

Source and methodology

All values are either drawn from published sources or computed from the reference model in the companion working paper. The bibliography and synthetic-vs-empirical labelling are in SOURCES.md at the project root. Key sources:

• Soiling: Kimber 2006; Mejia & Kleissl 2013; Ilse et al. 2019; Sarver, Al-Qaraghuli & Kazmerski 2013; NREL technical report series
• Performance monitoring: IEC 61724-1:2021 soiling ratio definition
• WFP / DI chemistry: Unger, Tucker, Xero, Streamline manufacturer documentation; WQA mixed-bed DI technical sheet; ASTM D1193 reagent water classification
• Irradiance: NRCan Photovoltaic Potential data (cross-checked with NASA POWER)
• Water quality: Health Canada TDS aesthetic objective (≤ 500 mg/L); municipal annual water quality reports
• Fall protection: OSHA 29 CFR 1926 Subpart M; Ontario O. Reg. 213/91 s. 26; CSA Z259 series

Regenerating the data

All seven CSVs are produced by generate_datasets.py in this directory:

python generate_datasets.py

The generator is pure Python 3 stdlib — no pandas, no numpy.

License

• Data: Creative Commons Attribution 4.0 International (CC BY 4.0)
• Generator script: MIT License (see repository root)

Cited standards (IEC, CSA, ASTM, OSHA CFR, Health Canada guidelines) are referenced but not redistributed — readers must obtain their own copies from the issuing bodies.

Companion artifacts

• Working paper (PDF): see the pdfs/ directory in the source repository
• Reference calculator: eight language bindings published to PyPI, npm, Crates.io, Maven Central, RubyGems, Hex.pm, Packagist, and pkg.go.dev
• Source repository: maintained by Binx Professional Cleaning (see repository README.md for all mirror URLs)