Chile: largest and smallest registered solar-plant module audits#
Two Analysis examples extend the Icor counter:
Neither establishes an automatic whole-plant image count. These are real-data transferability tests with explicit unresolved results, not successful repetitions of the Icor count. The gallery ZIP contains both scripts, exact static maps, metadata, registry selection and machine-readable audit exports.
Measurement |
CEME1 |
PMGD PFV Concón |
|---|---|---|
CEN site ID |
2010 |
2082 |
CEN |
412.9326 |
0.1 |
Documented DC nameplate, MWp |
480.19 |
0.118 |
Published module inventory |
882,720 |
Not established |
Image module count |
Unresolved: insufficient sampling |
Unresolved: no usable frame grid |
Whole-plant image count |
Not established |
Not established |
Selection and inventory evidence#
The 2026-10-31 CEN registry
has 1,384 records. Filter tipo_central_nombre == Solares,
estado == EN_OPERACION and finite positive potencia_maxima; sort by power
and ID. This leaves 757 entries. Coordinates do not affect ranking. Missing,
zero and negative power values are excluded. Ties remain in the ranking; the
pinned endpoints are unique in this snapshot. The CSV projects all records
onto seven selection fields; the original response hash is also retained.
This ranks the registry’s operating solar category, including solar thermal; both extrema are independently confirmed photovoltaic. It is not a census of private, off-grid or net-billing systems, or a minimum by surveyed ground area. Separate projects are not combined. Registry maximum power is not assumed MWp.
Generadora Metropolitana also identifies CEME1 as Chile’s largest PV plant. The CEN-hosted engineering report, section 3.2 and Table 3.1 (pp. 15–16), records 178,560 modules rated 540 W and 704,160 rated 545 W: 882,720 total, also consistent with 29,424 strings × 30 modules. The rating sum is 480.1896 MWp, rounded to 480.19 in the report. This is an inventory cross-check, not image recognition; it never enters detection.
For Concón, the commissioning form links the PMGD to Abastible and installer Sunbelt. The installer’s page reports 118 kWp with a Huawei string inverter and supplies photographs but no verified module total. Personal contact fields are not republished.
What the imagery supports#
Three attributed static Esri World Imagery exports are used, not calibrated bands or a newly open-licensed imagery dataset. Requests use EPSG:4326 and matching degree/pixel aspect ratios to prevent automatic extent expansion. Image hashes are checked before analysis. Every plot visibly credits the source.
At the CEME1 probe, metadata report 2025-08-28 acquisition and 0.34 m source resolution, but 1.2 m service sampling, with maximum map level 17. The delivered detail viewport samples about 1.33 m north–south. The report’s 1.134 m module edge occupies less than one effective sample. The counter needs at least three pixels per pitch and rejects the image before confusing array strips with modules. Finer export requests failed; a level-19 tile returned a “map data not yet available” placeholder. Neither failed response is counted or bundled.
The Concón probe reports 2026-07-02, 0.34 m source resolution and 0.3 m service
sampling. Strong glare obscures frames. The actual counter, using explicit
table-size/orientation parameters, finds no usable grid; three colour thresholds
return None. This is not zero panels or evidence of missing or faulty modules.
Metadata concern one probe, not guaranteed uniform coverage across a mosaic. CEME1’s diagnostic viewport is partial, with OSM way 1063236315 as context. Concón uses a manually selected inspection rectangle. Neither is a verified as-built boundary. No full-site extrapolation is made.
Reproduce and export#
uv sync --locked --all-packages
uv run --no-sync python examples/analysis/plot_chile_largest_module_counting.py
uv run --no-sync python examples/analysis/plot_chile_smallest_module_counting.py
from energy_analysis.module_counting.chile import export_chile_audit
export_chile_audit(
"examples/analysis/data/chile_modules",
"/tmp/chile-module-audit", # fresh directory
)
The export contains the eligible ranking and count_audit.json. Published
inventory, image candidates and whole-plant counts are separate fields. Unknown
counts serialize as null; no fake per-module CSV/GeoJSON is manufactured.
The complete gallery ZIP includes these outputs and runs offline on Python 3.12/3.13.
scripts/refresh_chile_module_inputs.py explicitly downloads the current registry
and saved image/metadata requests into a fresh directory. Changed bytes require
new visual review, ranking and hashes; refreshing never replaces the study silently.
Remaining work and reuse#
A defensible count needs licensed, georeferenced full-site imagery with visible module frames and reviewed boundaries. Large sites need bounded tiles, seam de-duplication, missing-coverage checks and held-out manual labels. Concón also needs an as-built module schedule. Upsampling or generative detail cannot replace missing optical measurements. Installer photos do not show every module unobscured.
The code license does not relicense the map images. They remain © 2026 Esri and its licensors (Esri, Vantor, Earthstar Geographics and the GIS User Community), all rights reserved, displayed as research-report illustrations under Esri’s static-map terms. Retain attribution and check provider terms for other uses. OSM geometry is © OpenStreetMap contributors, ODbL-1.0. Registry facts are attributed to CEN; no blanket license for its reports is asserted. Reports and installer photos are linked, not redistributed in the bundle.