Skip to content

Latest commit

 

History

History
55 lines (36 loc) · 1.76 KB

vision.md

File metadata and controls

55 lines (36 loc) · 1.76 KB

Vision

ModelSkill wishes to be your modelling assistant. It should be useful enough for you to use every time you do a MIKE 21/3 simulation.

Objective

We want ModelSkill to make it easy to

  • assess the skill of a model by comparing with measurements
  • assess model skill also when result is split on several files (2d, 3d, yearly, ...)
  • compare the skill of different calibration runs
  • compare your model with other models
  • use a wide range of common evaluation metrics
  • create common plots such as time series, scatter and taylor diagrams
  • do aggregations - assess for all observations, geographic areas, monthly, ...
  • make fast comparisons (optimized code)

And it should be

  • Difficult to make mistakes by verifying input
  • Trustworthy by having >95% test coverage
  • Easy to install (pip install modelskill)
  • Easy to get started by providing many notebook examples and documentation

Scope

ModelSkill wants to balance general and specific needs:

  • It should be general enough to cover >90% of MIKE FM simulations
  • But specific enough to be useful
    • Primarily support dfs files (using mikeio)
    • Handle circular variables such as wave direction
    • Handle vector variables such as u- and v-components of current
    • Tidal analysis

Limitations

ModelSkill does not wish to cover

  • Extreme value analysis
  • Deterministic wave analysis such as crossing analysis
  • Rare alternative file formats
  • Rarely used model result types
  • Rare observation types
  • Anything project specific

Future

Web app

Create a web app that wraps this library