Hazard Aggregrations

Produce Hazard Aggregrations

flowchart LR
    classDef nshm stroke:lightgreen, stroke-width:3px
    classDef AWS stroke:orange, stroke-width:3px

    RHA[/run hazard aggregations/]:::nshm
    thp[toshi-hazard-post]:::nshm
    HA[(hazard aggregations)]:::AWS

    RHA -->|with| thp -->|produces| HA            

inputs:

• NSHM SRM logic tree that identifies the sources (Model, Source IDs) (from nzshm-model library)
• realizations from prior step (e.g. 15GB of THS hazard curves )
• identify the realisations (having the right permutations of site, IMTS, VS30s etc, using ToshiIDs). In spreadsheet now.
• user can choose any subset of the rlz permutations.
• config the aggregations (e.g. mean, cov, 0.1, 0.5...)

outputs:

• entire LT structure in json (200kb)
• caches composite realizatons to disk (many GB)
• produces aggregate hazard curves in THS DynamoDB (4K site, 20 periods, 44 IMTS, => 280MB per VS30)

resource/cost/metrics:

• runs locally or in AWS EC2 M5
• for each VS30 (NSHM has 18) we run either:
  • locally (CDC 32 core machine): ~2m preproc + 15s per site/IMT => ~ 10 hours per VS30
  • cloud (up to ~600cores), ~30s per site/IMT => ~2 hours per VS30.
• cloud scheduling within limits of Account DynamodDB rate limiting (reading rlzs).