Methodology

How We Built the Tropical Cyclone Catalog

A comprehensive look at how archived TIGGE forecast data and advanced detection algorithms are transformed into an event-based catalog of tropical cyclone risk across the Northwest Pacific basin.

In simple terms

The Peak Re Tropical Cyclone Catalog combines large-scale forecast ensemble data with advanced detection algorithms. For each basin, we:

  • identify tropical cyclone signatures in TIGGE forecast data,
  • track storm evolution across time and space, and
  • calibrate intensity estimates against historical observations.

These elements are combined into event footprints: a consistent summary of when cyclones occurred, which regions were affected, and how intense each storm was.

At a glance

Forecast backbone

Control run forecasts from ECMWF and UKMO (2007–2024) provide physically consistent atmospheric states at regular intervals across the Northwest Pacific basin.

Historical validation

IBTrACS best-track data serves as the primary reference for training, calibrating, and validating detection and intensity models.

Hybrid detection approach

Enhanced Oliver (2018) algorithm combines contour detection with vortex identification, using MSLP and 10-m winds.

Event-based outputs

Each tropical cyclone track includes position, intensity (calibrated wind speeds), and affected administrative regions, enabling direct risk assessment and catastrophe modeling applications.

This page offers a conceptual overview. Specific model architectures, calibration parameters, and validation metrics remain proprietary to Peak Re.

From forecast data to tropical cyclone events

The catalog is built through several integrated stages: we extract atmospheric fields from TIGGE forecasts, detect and track tropical cyclones, correct intensity biases, and validate results against IBTrACS observations.

TIGGE forecast archive

Control run forecasts from ECMWF and UKMO (2007–2024) provide key atmospheric fields: mean sea-level pressure (MSLP), 10-meter wind components (u10/v10), and total precipitation, at 6-hourly intervals.

Stage 1 — TC detection

Enhanced Oliver algorithm searches for closed MSLP contours alongside iterative contour search and vortex detection, with physical constraints (latitude bounds, wind-pressure relationships) filtering false positives.

Stage 2 — Track formation & linking

Detected centers at consecutive time steps are linked into continuous tracks using proximity, intensity consistency, and steering flow criteria. Tracks shorter than minimum duration thresholds are excluded to ensure event robustness.

Stage 3 — Intensity calibration

Statistical bias correction models trained on IBTrACS best-track data adjust raw forecast intensities. Holland wind field profiles extend point estimates to full spatial footprints for impact assessment.

Peak Re Tropical Cyclone Catalog

Event-based catalog of TC tracks combining position time series, calibrated wind speeds, and affected regions (aligned with administrative boundaries), covering 2007–2024 across the Northwest Pacific basin.

IBTrACS best-track observations are used to train detection thresholds, calibrate intensity corrections, and validate track statistics throughout the entire modeling pipeline.

The schematic above illustrates the methodology's overall logic. Detailed detection parameters, calibration coefficients, and validation thresholds remain proprietary to Peak Re.

Key methodological themes

Four principles guide the design of the catalog:

Data foundation

TIGGE archive provides a single, well-documented forecast basis across all basins and years. Using uniform forecast data from leading operational centers avoids gaps and inconsistencies inherent in observation-only approaches while enabling expanded event sets beyond the historical record.

Event-based design

Rather than only cataloging individual time steps, we construct complete tropical cyclone tracks with continuous footprints: temporal evolution, spatial extent, calibrated intensity, and affected administrative regions. This makes the catalog immediately usable for risk quantification and catastrophe modeling.

Detection and intensity separation

Separating TC identification (contour + vortex detection) from intensity estimation (statistical calibration against IBTrACS) allows us to treat occurrence frequency and severity distribution as distinct but related problems—critical for tail risk assessment in reinsurance applications.

Training & validation

Detection thresholds and intensity corrections are validated against IBTrACS across the full 2007–2024 period. Track position errors, intensity bias metrics, and frequency distributions are systematically compared to best-track data to ensure catalog reliability for risk assessment.

The methodology builds on peer-reviewed tropical cyclone detection algorithms (Oliver 2018) combined with internal Peak Re research and validation. Technical documentation including validation statistics is available to professional counterparties on request.

Key References:

  • Bougeault, P., et al. (2010). The THORPEX Interactive Grand Global Ensemble (TIGGE). Bulletin of the American Meteorological Society.
  • Holland, G. J. (1980). An Analytic Model of the Wind and Pressure Profiles in Hurricanes.
  • Knapp, K. R., et al. (2010). The International Best Track Archive for Climate Stewardship (IBTrACS).
  • Nguyen, K. C., et al. (2014). TIGGE Tropical Cyclone Track Forecast Dataset.
  • Oliver, E. C. J. (2018). Tropical Cyclone Detection in CMIP5 Models. Journal of Climate.
  • Swinbank, R., et al. (2016). The TIGGE Project and Its Achievements. Bulletin of the American Meteorological Society.

Data Coverage & Updates

Temporal Coverage: 2007–2024 (18 years of TIGGE forecasts)
Spatial Coverage: Northwest Pacific basin
Update Frequency: Annual updates planned with new TIGGE data
Last Updated: December 2024