georisk-methodology-v0.3.1 · August 2026

Preliminary methodology and validation plan.

Canonical public summary for GeoRisk Engine Phase A. The technical Markdown specification remains normative. Formula expected-loss-v0.3 is unchanged in v0.3.1; meaning, assumptions and limitations are made explicit here.

Version georisk-methodology-v0.3.1 Published August 2026 Active modules Flood · extreme heat Output Phase A scenario-loss proxy
External positioning (v0.3.1). GeoRisk Engine v0.3.1 is a synthetically calibrated physical-climate-risk methodology prototype designed for pilot testing. Its current Phase A output is a scenario-loss proxy rather than fully probability-weighted EAL. It produces decision-support indicators and reproducible calculation records; its loss estimates have not yet been validated against sufficient real-world loss data.

Preliminary Methodology and Validation Plan — Asset-Level Physical Climate Risk and Financial Impact Assessment

Synthetic Calibration v0.3.1 — Phase A Scenario-Loss Proxy

What GeoRisk Engine is intended to support

GeoRisk Engine is a decision-support engine intended to translate physical climate hazards into financial risk metrics that can inform lending, collateral, investment and adaptation-planning processes. It goes beyond hazard mapping by combining asset-level hazard, exposure and vulnerability data.

The current v0.3.1 release is a synthetically calibrated methodology prototype for flood and extreme-heat modules. The Phase A output is a scenario-loss proxy, not a fully probability-weighted Expected Annual Loss (EAL). Its outputs are not validated loss estimates and must not be used as the sole basis for automated lending or investment decisions.

Five-layer analysis chain

  1. Hazard: Hazard type, intensity, annual occurrence or exceedance probability, climate scenario and time horizon.
  2. Exposure: Geolocation, asset value, construction/use class and the applicable valuation date.
  3. Vulnerability: Hazard- and asset-class-specific damage or financial-impact functions.
  4. Financial translation: In Phase A, application of a scenario damage ratio to a defined value base, with direct physical damage and operational effects tracked through separate loss channels. Probability-weighted annualisation is not implemented in the current release.
  5. Adaptation assessment: Baseline risk, post-measure residual risk, avoided expected loss, adaptation cost and prioritisation indicators.
Hazard × Exposure × Vulnerability → Phase A Scenario-Loss Proxy → Adaptation Scenario → Residual Risk and Avoided Loss

Phase A loss proxy and target EAL approach

In v0.3.1, the Phase A output displayed under the “EAL” label is a scenario-loss proxy produced by applying the combined damage ratio to a defined value base for the selected scenario. The current calculation does not include an exceedance-probability curve, return-period conversion or ∫ loss dF integration. It must therefore not be treated as equivalent to fully probability-weighted EAL used in actuarial or catastrophe-risk modelling.

In the target methodology, genuine EAL will be defined as EAL = Σ pᵢ Lᵢ, or an equivalent loss-distribution integral, where sufficient event-probability data exist. That transition will not be represented as complete until hazard-specific return periods, scenario weights, direct-damage/business-interruption treatment, value caps and calibration evidence are documented and validated.

Multi-hazard treatment

combined MDR = 1 − Π(1 − MDRh)

The v0.3.1 expression 1 − Π(1 − MDR) is a provisional combination rule applicable only to compatible, explicitly defined damage ratios measured against the same financial value base. It does not establish hazard independence and is not a general solution for correlation, concurrent or cascading events, or different loss channels.

Flood-related physical damage will not be mechanically combined on the same MDR basis with extreme-heat effects such as energy cost, productivity, downtime or reduced equipment life. Those effects will be calculated separately and aggregated only after dependency and double-counting controls.

The extreme-heat financial channel currently uses the same MDR × value placeholder path as flood. Calibrated business-interruption or productivity modelling is not implemented.

Reproducibility and calculation checking

Each calculation record is intended to retain methodology, data and damage/impact-curve versions, parameters, valuation date, timestamp and input references. Historical outputs can then be replayed under the version and parameters with which they were produced.

A replay test demonstrates that the same inputs and versions reproduce the stored result within a documented rounding policy and numerical tolerance. It does not, by itself, demonstrate model accuracy or agreement with observed losses. In the product demo, the replay result is labelled Recomputed EAL — a calculation-replay label only, not a validation verdict.

A separately implemented calculation checker recalculates sampled outputs against a reference specification. Code-isolated recompute is not the same as independent model validation. The term “independent model validation” will be reserved for a review of data, assumptions, performance, limitations and governance by a person, function or third party independent from model development.

Validation plan

Pre-event model inputs → observed hazard intensity → observed damage/financial impact → predicted loss → error and calibration metrics

The initial event study for the flood module is the 2021 Western Black Sea floods. A suitable historical dataset linking heat or heat stress to energy demand, operational disruption, productivity or equipment performance will be selected for the extreme-heat module. The Manavgat wildfires will not be presented as an active-module backtest unless a wildfire module and its data model are formally brought into scope.

Subject to sample adequacy, candidate performance measures include MAE, RMSE, MAPE/SMAPE, bias, calibration error and prediction-interval coverage. Applicable metrics will be fixed in the model specification based on the loss distribution and data quality.

Regulatory relevance

The outputs are intended to support selected physical-risk assessment and evidence needs relevant to ECB/EBA supervisory expectations, ESRS E1, EU Taxonomy climate-adaptation/DNSH assessment and Türkiye’s TSRS 2.1–5

This positioning does not, by itself, constitute regulatory compliance, reporting sufficiency or supervisory acceptance. Institution-specific materiality, governance, scenario analysis, data controls, credit-risk transmission, disclosure and assurance requirements remain applicable.

Use and limitations

  • Damage/impact curves are synthetically calibrated and have not yet been calibrated against real customer loss data.
  • Sample portfolios contain no real customer data.
  • All outputs must be labelled “Methodology v0.3.1 - Synthetic Calibration / Phase A Scenario-Loss Proxy”.
  • Outputs must not substitute for a credit decision, property valuation, insurance pricing, engineering safety assessment or regulatory filing.
  • Asset-level results must not imply spurious precision; data quality and model uncertainty should accompany the output.
  • A calibrated v1.0 will be released only after design-partner pilots and documented performance thresholds have been completed.

Evidence gaps before v1.0 / bank pilot

v1.0 / banka pilotu öncesi kanıt boşlukları

Minimum topics to add or close before calling a release bank-pilot ready (methodology note §12):

Bir sürümün banka-pilota hazır sayılmasından önce eklenmesi veya kapatılması gereken asgari konular (metodoloji notu §12):

  1. 01Data sources, licences, and citationsVeri kaynakları, lisanslar ve atıflar
  2. 02Spatial resolution and geocoding toleranceMekânsal çözünürlük ve geocoding toleransı
  3. 03NGFS / SSP / RCP scenario selectionNGFS / SSP / RCP senaryo seçimi
  4. 04Historical baseline periodTarihsel baz dönem
  5. 05Return periods and event probabilitiesReturn period’lar ve olay olasılıkları
  6. 06Damage-curve sources and calibrationHasar eğrisi kaynakları ve kalibrasyon
  7. 07Classical annualised EAL under a probability measureOlasılık ölçüsü altında klasik yıllıklaştırılmış EAL
  8. 08Inter-hazard dependence estimationTehlikeler arası bağımlılık tahmini
  9. 09Uncertainty intervals and sensitivity analysisBelirsizlik aralıkları ve duyarlılık analizi
  10. 10Backtesting sample and error metricsBacktesting örneklemi ve hata metrikleri
  11. 11Organisational model-change approvalKurumsal model değişikliği onayı
  12. 12Override, exception, and data-quality controlsOverride, istisna ve veri kalitesi kontrolleri
  13. 13Asset damage to PD / LGD / collateral translationVarlık hasarının PD / LGD / teminat çevirisi
  14. 14Intended-use and misuse controlsAmaçlanan kullanım ve kötüye kullanım kontrolleri

Release roadmap (themes)

Sürüm yol haritası (temalar)

Brief themes only — not a project plan. Full technical workstreams live in the normative methodology and product directive.

Yalnızca kısa temalar — proje planı değildir. Tam teknik iş akışları normatif metodolojide ve ürün direktifinde yer alır.

ReleaseObjective SürümHedef
v0.3.1 Correct public positioning and evidence level (this page) Kamuya açık konumlandırmayı ve kanıt düzeyini düzeltmek (bu sayfa)
v0.4 Freeze methodology specification (hazard-specific EAL, loss channels, versioning) Metodoloji spesifikasyonunu dondurmak (tehlike bazlı EAL, kayıp kanalları, sürümleme)
v0.5 Establish technical verification (tests, replay tolerances, checker coverage) Teknik doğrulamayı kurmak (testler, replay toleransları, checker kapsamı)
v0.6 Complete historical-event dataset design (flood and heat) Tarihsel olay veri seti tasarımını tamamlamak (sel ve sıcaklık)
v0.7 Pilot with design partners under data-governance controls Veri yönetişimi kontrolleri altında tasarım ortaklarıyla pilot
v0.8 Calibrate, benchmark, and produce uncertainty bands Kalibre etmek, benchmark yapmak ve belirsizlik bantları üretmek
v0.9 Independent challenge and remediation Bağımsız challenge ve iyileştirme
v1.0 Controlled production release after thresholds and governance Eşikler ve yönetişim sonrası kontrollü üretim sürümü

Regulatory reference notes

Düzenleyici referans notları

Use these references as context for selected evidence support, not as a claim that the product satisfies the complete framework:

Bu referansları ürünün çerçeveyi bütünüyle karşıladığı iddiası olarak değil, seçilmiş kanıt desteği bağlamı olarak kullanın:

Aşağıdaki resmi belge başlıkları İngilizce bırakılmıştır.

  1. EBA Guidelines on the management of ESG risks — institution-wide governance, identification, measurement, management and monitoring expectations: — kurum genelinde yönetişim, tanımlama, ölçüm, yönetim ve izleme beklentileri: eba.europa.eu
  2. Commission Delegated Regulation (EU) 2023/2772 — ESRS, including ESRS E1: eur-lex.europa.eu
  3. ECB economy-wide climate stress test methodology/context: ecb.europa.eu (PDF)
  4. EU Taxonomy Climate Delegated Act — climate adaptation and DNSH technical screening criteria: eur-lex.europa.eu
  5. Türkiye Sustainability Reporting Standards (TSRS) — cite the current KGK publication/version applicable at the document’s publication date: — belgenin yayımlanma tarihinde geçerli KGK yayını/sürümünü atıf gösterin: kgk.gov.tr

Canonical sources

Kanonik kaynaklar

This HTML page is the canonical public summary for v0.3.1.

Bu HTML sayfa, v0.3.1 için kanonik kamuya açık özetten oluşur.

Normative technical specification: docs/methodology/GEORISK_ENGINE_METHODOLOGY.md · raw Markdown

Normatif teknik spesifikasyon: docs/methodology/GEORISK_ENGINE_METHODOLOGY.md · ham Markdown

Layer-4 formula detail: app/loss/METHODOLOGY.md. Publication hierarchy: docs/methodology/PUBLICATION.md.

Katman-4 formül detayı: app/loss/METHODOLOGY.md. Yayın hiyerarşisi: docs/methodology/PUBLICATION.md.

Any PDF snapshot should be generated from the approved source and must not become an independently edited source of truth.

Herhangi bir PDF anlık görüntüsü onaylı kaynaktan üretilmeli ve bağımsız biçimde düzenlenen bir gerçeklik kaynağı haline gelmemelidir.