coverage

Regions covered by SolarMap.PH.

Greater Metro Manila is the calibrated, holdout-validated, SAM-segmented region (v1.0). SolarMap.PH also publishes cross-domain detections for seven additional Philippine franchise areas: Cebu, Davao, Iloilo, Cagayan de Oro, Legazpi, the Calabarzon belt south of Meralco, and Bacolod / Negros Occidental.

v1.2 retrained the canonical classifier region-stratified (clf_v5, sha256 5cc0a093) and added a per-domain calibration fit on a scan-realistic holdout. Cebu, Iloilo and Calabarzon now carry a conservative per-domain precision; Davao, CDO, Legazpi and Bacolod have too few labels and ship as honest candidate inventory (no precision claim). The published region detections themselves are unchanged from v1.1 (clf_v4-scored; v1.2 does not re-scan); clf_v5 is canonical for future scans. The same residential-suppression policy applies in every region.

v1.0 calibrated

Greater Metro Manila

Franchise: Meralco

High-confidence detections: 280
Candidate detections: 235
Cities with detections: 41
Aggregate kWp identified: 69.9 MWp
Calibration: F1 = 0.87 on holdout

Open the interactive map →

v1.2 per-domain calibrated

Cebu Metro

Franchise: VECO (Visayan Electric)

High-confidence detections: --
Candidate detections: --
Cities with detections: --
Served LGUs in franchise: 11
Calibration: Per-domain (v1.2), conservative

Data files load on demand. Direct GeoJSON link /data/rooftop_solar_cebu.geojson.

cross-domain candidate

Davao City

Franchise: DLPC (Davao Light and Power Company)

High-confidence detections: --
Candidate detections: --
Cities with detections: --
Served LGUs in franchise: 2
Calibration: Uncalibrated — candidate inventory

Data files load on demand. Direct GeoJSON link /data/rooftop_solar_davao.geojson.

v1.2 per-domain calibrated

Iloilo Metro

Franchise: MORE (MORE Electric and Power Corp.)

High-confidence detections: --
Candidate detections: --
Cities with detections: --
Served LGUs in franchise: 7
Calibration: Per-domain (v1.2), conservative

Data files load on demand. Direct GeoJSON link /data/rooftop_solar_iloilo.geojson.

cross-domain candidate

Cagayan de Oro

Franchise: CEPALCO (Cagayan Electric Power and Light Co.)

High-confidence detections: --
Candidate detections: --
Cities with detections: --
Served LGUs in franchise: 6
Calibration: Uncalibrated — candidate inventory

Data files load on demand. Direct GeoJSON link /data/rooftop_solar_cdo.geojson.

cross-domain candidate

Legazpi (Albay)

Franchise: ALECO (Albay Electric Cooperative)

High-confidence detections: --
Candidate detections: --
Cities with detections: --
Served LGUs in franchise: 5
Calibration: Uncalibrated — candidate inventory

Data files load on demand. Direct GeoJSON link /data/rooftop_solar_legazpi.geojson.

v1.2 per-domain calibrated

Calabarzon (south of Meralco)

Franchise: Multiple (BATELEC, FLECO, QUEZELCO et al) (Batangas Electric Cooperatives + First Laguna Electric + Quezon I and II + LUELCO)

High-confidence detections: --
Candidate detections: --
Cities with detections: --
Served LGUs in franchise: 13
Calibration: Per-domain (v1.2), conservative

Data files load on demand. Direct GeoJSON link /data/rooftop_solar_calabarzon.geojson.

cross-domain candidate

Bacolod / Negros Occidental

Franchise: CENECO (Central Negros Electric Cooperative)

High-confidence detections: --
Candidate detections: --
Cities with detections: --
Served LGUs in franchise: 7
Calibration: Uncalibrated — candidate inventory

Data files load on demand. Direct GeoJSON link /data/rooftop_solar_bacolod.geojson.

Honesty about the cross-domain regions (v1.2)

NCR is the calibrated reference (holdout precision 95.9%, F1 0.87). A v1.2 domain-shift measurement found that pure geographic shift between NCR and the seven regions is small (within NCR's own training-vs-scan envelope); the real problem was a calibration gap — the curated training set is ~0.88-separable from the natural scan distribution even within NCR, so the v1.1 implied cross-region precision was an overestimate everywhere.

v1.2 retrained region-stratified (clf_v5) and fit a per-domain calibration on a scan-realistic holdout. Cebu, Iloilo and Calabarzon have enough labels for a conservative per-domain precision (small holdouts, treat as lower bounds). Davao, CDO, Legazpi and Bacolod do not and remain an honest candidate inventory — no precision claim rather than a fabricated one.

A visual audit of the top cross-domain detections (32 tiles across all regions, see docs/screenshots/qa-2026-05/region-spot-check/) confirmed 28/32 rooftop solar, 3 ground-mount solar farms, 1 blue-roof false positive — the same failure modes as NCR's audit, now added to the v1.2 training set as hard negatives.

v1.2 also adds a per-detection SAM kWp estimate as a coarse scalar (each mask verified by a CLIP + classifier check before its area counts; Legazpi's lone false-positive correctly estimates 0 kWp). Detections remain at tile granularity (240 m) — no per-building polygons for these regions, so there is no individual-building exposure.

Same residential-suppression policy applies in every region: tile-granularity detections do not reveal individual buildings, per RA 10173.