CR_025: National 16-region solar layer validation 2026 (Perplexity Computer

CR_025: National 16-region solar layer validation 2026 (Perplexity Computer

National 16-region solar layer validation 2026

Validates RD_013 nationally — 13/15 regions within ±10%; PVGIS has a ~+5% positive bias; Wellington/Dunedin flagged

A Perplexity Computer validation (2026-06-11) cross-checking the PVGIS-ERA5 figures in RD_013 against NIWA ground-station solar normals (1991–2020) for each NZ region’s main centre. 13 of 15 centres agree within ±10%, confirming the model’s solar layer is sound nationally (previously only the single Nelson point was validated). PVGIS runs ~+5% high; Wellington (+16.4%) and Dunedin (+10.8%) are flagged for local cloud/topography. data_quality: medium (NIWA ground-truth is authoritative; PVGIS regeneration is AI-prepared).

Summary

A validation of the national solar-yield layer (RT_207, which replaced the retired single-site RT_099). PVGIS-ERA5 global horizontal irradiance — the basis of the model’s pv_yield_kwh_per_kwp via RD_013 — was compared against NIWA’s published “Mean daily global radiation (MJ/m²)” 1991–2020 normals (90+ pyranometer ground stations) for the main centre of each NZ region. The result confirms RD_013 is sound nationally, quantifies a small consistent PVGIS positive bias, and flags two regions where the coarse reanalysis grid overstates the resource.

Key claims

- claim: "PVGIS-ERA5 (the RD_013 solar layer) was validated against NIWA ground-station normals (1991–2020, 90+ pyranometer network) for 15 main centres covering all 16 NZ regions (Tasman and Nelson share the Nelson station). 13 of 15 centres agree within ±10% — confirming the layer is sound NATIONALLY, not just the single Nelson/Tasman ground point that was previously the only ~2%-validated cell. NIWA SolarView itself is login-gated / per-date / tilted and unusable for batch validation, so NIWA's published MJ/m² normals were used instead (the same instrument network), converted via × 365 ÷ 3.6."
  source_location: "TL;DR; Method note; 16-region validation table"
- claim: "PVGIS-ERA5 shows a consistent small POSITIVE bias of ~+5% vs NIWA ground truth nationally — recommended as a conservative haircut in generation modelling. Per-region GHI differences (PVGIS vs NIWA): mostly +0.0% to +9.7% — Hawke's Bay −0.0%, Waikato −0.5%, Bay of Plenty +0.3%, Marlborough +1.2%, Southland +2.5%, West Coast +2.6%, Nelson +2.8%, Gisborne +5.3%, Manawatū +6.4%, Northland +6.9%, Taranaki +7.3%, Auckland +7.5%, Canterbury +9.7%."
  source_location: "TL;DR; validation table (Diff % column)"
- claim: "Two regions exceed +10% and are flagged: Wellington +16.4% (NIWA ref = Kelburn, 125 m exposed hill in a cloudy/windy basin; PVGIS's ~30 km grid averages over Cook Strait and misses local cloud/orographic shading — its 1,571 kWh/kWp 'national high' is too optimistic, apply ~−15%); Dunedin +10.8% (Musselburgh coastal, frequent sea fog — apply ~−10%). Christchurch +9.7% is borderline (eastern-South-Island fog/inversions) and worth watching. For these, prefer the NIWA ground value or a larger haircut."
  source_location: "Flagged regions section"
- claim: "PVGIS optimal north-facing tilt yields (kWh/kWp/yr, free-standing crystalline-Si, 14% system loss) by region: Invercargill 1,247, West Coast/Hokitika 1,302, Otago/Dunedin 1,332 (flagged), Waikato 1,334, Manawatū 1,392, Northland 1,396, Hawke's Bay 1,430, Bay of Plenty 1,432, Nelson/Tasman 1,446, Canterbury 1,466, Auckland 1,473, Gisborne 1,484, Marlborough 1,512, Taranaki 1,545, Wellington 1,571 (flagged). Optimal tilt 34° (upper North Island) to 37° (rest). These are the per-region pv_yield_kwh_per_kwp values for the model."
  source_location: "validation table (PVGIS opt. yield column); solar-layer-validation.csv"

Retrieval provenance

• Upstream source: NIWA “Mean daily global radiation (MJ/m²)” 1991–2020 normals (90+ pyranometer ground stations); PVGIS v5.2 PVcalc API (ERA5).
• Prepared by: Perplexity Computer, 2026-06-11.
• Prompt: RT_207 — validate the PVGIS-ERA5 figures in RD_013 against NIWA for each region’s main centre; flag any region differing >~10%. (Full prompt in the _retrieval.md sidecar.)
• Verification: NIWA side is authoritative ground-truth; the PVGIS regeneration + station matching is the tool’s own, AI-prepared and not independently re-run. data_quality: medium. (Convention: wiki/CLAUDE.md → Retrieval provenance.)

Neobiome Intelligence relevance

Resolves RT_207 and firms RD_013 — the model’s pv_yield_kwh_per_kwp layer is now validated against NIWA ground truth across all 16 regions (13/15 within ±10%), upgrading it from “2% validated (Nelson only)” to nationally credible. Two model implications: (1) the consistent ~+5% PVGIS positive bias is a candidate conservative national haircut on pv_yield, or it can feed the P90 bad-year band (D16) rather than the P50 central; (2) Wellington, Dunedin and (watch) Canterbury need larger site haircuts (−10 to −15%) because the coarse reanalysis over-predicts where local cloud/topography drives the resource. The per-region optimal-tilt yields (1,247–1,545, excluding the two flagged) are the defensible regional pv_yield values; the pilot-region figure (Nelson 1,446) is consistent with CR_010 (1,350–1,380) and RD_013. It also confirms — independently — why NIWA SolarView was the wrong tool (login-gated/per-date/tilted), retroactively validating the retirement of RT_099. ⚠ AI-prepared; the NIWA comparison is sound, but treat the PVGIS regeneration as the tool’s own work.

Research targets

Research gaps

• RT_207 (RESOLVED → this page): RD_013’s 16-region solar layer validated against NIWA ground stations; +5% PVGIS bias quantified; Wellington/Dunedin flagged.

Connections

Links to

Sources (2): CR_010 · RD_013

Referenced by

Sources (2): LIT_071 · RD_013