1. Overview

This chapter gives the minimum conceptual background needed to use openAMUNDSEN-DA without scattering the same explanations across later pages.

openAMUNDSEN-DA prepares snow observations and uses them to update an ensemble of openAMUNDSEN model realizations in a step-wise data assimilation workflow. The tutorial uses the single-domain Rofental example; subdomain execution is an advanced workflow. It focuses on station snow-depth, fractional snow covered area (fSCA) and wet-snow observations.

The Workflow In One Paragraph

A setup contains the openAMUNDSEN model configuration, input grids, forcing, and observation files. Inside that setup, a project YAML defines the time period, observation mappings, data assimilation events, and data assimilation settings. The shipped Rofental example already includes baseline fSCA (scf) and wet-snow summary tables under obs/summaries/, while station observations and station data assimilation metadata live under obs/stations/. Observation rasters can still be reprocessed into project summaries and per-step observation CSVs. The project pipeline then runs the ensemble step by step, assimilates the prepared observations, and writes diagnostics and result products.

Setup And Project Layout

The schematic below shows where the main files live inside one setup and one project workspace. It separates setup-level inputs from the project-specific configuration, step folders, and outputs.

Shared layout of a setup root and one project workspace. Optional sub-domain folders are omitted for clarity.

The Core Terms

These four terms are the important ones for the tutorial:

• setup: the top-level case directory with model inputs, observations, and projects
• project: one data assimilation configuration inside a setup
• step: one time window between two data assimilation events
• member: one ensemble realization inside a step

Inside the ensemble workflow, three more terms matter:

• open_loop: the unperturbed reference simulation
• prior: the ensemble state before assimilating the current observation
• posterior: the ensemble state after weighting/resampling the current observation

How Observations Enter The Framework

The framework does not read observation rasters directly during the assimilation step. Instead it uses a two-stage preprocessing workflow:

1. observation rasters are summarized to project-level tables such as scf_summary.csv
2. openamundsen-da prepare aligns the summaries with assimilation_events and writes one-row per-step observation CSVs under steps/*/obs/

That separation is deliberate. It keeps preprocessing explicit, makes event alignment transparent, and gives you files that can be inspected before the model run starts.

Continue with 2. Installation.