Hamlet Architecture and Testing Workflow

Hamlet performs spatial-temporal hazard model checks and statistical evaluation of model consistency and performance against an observed earthquake catalog.

Hamlet follows this workflow:

  1. Read in YAML configuration file, that specifies:

    • Which tests and evaluations to run, and their parameters

    • Input files:

      • Seismic Source Model files

      • Seismic catalog (observed earthquakes)

      • GIS file of spatial bins (optional)

    • What to output:

      • Reports

      • GIS files

      • JSON results

  2. Read and process SSM:

    1. Load sources from a single logic tree branch (or iterate over all branches if branch: iterate is set)

    2. Sort sources based on their type

  3. Sort the ruptures from all sources by magnitude and into spatial bins:

    • Uses H3 hexagonal cells (by default) to spatially bin ruptures and observed earthquakes

    • Computes both model and empirical Magnitude-Frequency Distributions (MFDs) for each spatial bin

  4. Run the tests and evaluations:

    • Basic sanity checks (e.g., whether the observed earthquake maximum magnitude exceeds the model maximum magnitude in each spatial bin)

    • Statistical consistency tests (N-test, M-test, S-test, L-test, comparing count, magnitude, spatial, and likelihood consistency)

    • Evaluations (MFD comparisons, moment rate analysis, rupture matching, ground motion evaluation)

    • Multiple tests and evaluations can be run sequentially, and multiple frameworks can be used in a single run

  5. Write output:

    • HTML reports summarizing the results with maps, plots, and tables

    • GIS files (GeoJSON) with the test results for each spatial bin

    • JSON files with structured results

Branch Iteration

When branch: iterate is set in the configuration, Hamlet evaluates each logic tree branch independently in a single run:

  1. The earthquake catalog is loaded once (shared across all branches)

  2. The complete source model is loaded, and all branches are extracted

  3. For each branch:

    1. Ruptures are processed for that branch only

    2. Input data is grouped into spatial bins

    3. All configured tests and evaluations are run

    4. Results are collected and memory is freed

  4. Results from all branches are combined into a single report