How mgovorcin/venti works
Project Type and Purpose
• Venti is an open-source Python toolbox (not a web app) for scientific analysis of vertical land motion (VLM) using InSAR and GNSS data.
• It provides data-ingestion readers, processing pipelines (calibration, interpolation, LOS decomposition), statistical modeling (variogram), and inversion solvers (hector, midas).
File-Structure Analysis
• Top-level directories:
• data/ – sample plate boundary and metadata files
• notebooks/ – Jupyter workflows demonstrating GNSS & InSAR workflows
• src/venti/ – core library organized by functional subpackages: calibration, gnss, insar, losdecomp, solvers, utils
• Configuration and packaging: setup.py, MANIFEST.in, tox.ini, environment.yml (implied).
• Obsolete/ contains legacy code; ignore for the main architecture.
Key Components
a. Data Inputs
• Raw GNSS observations (src/venti/gnss/readers via gps.py)
• InSAR stacks (via mintpy wrapper in src/venti/insar/mintpy)
• Plate-boundary metadata (data/plates)
b. Data-Ingestion Layer
• Reader modules: src/venti/gnss, src/venti/insar/mintpy, plus calibration/utils for metadata
c. Processing Pipelines
• GNSS interpolation (hvlsc_py3 collocation, empirical covariance, grid)
• Insar referencing & correction (std_reference_correction, ts_spatial_referencing)
• LOS decomposition (src/venti/losdecomp)
• Calibration routines (src/venti/calibration)
• Variogram modeling (src/venti/insar/variogram)
d. Solvers & Inversion
• hector and midas implementations in src/venti/solvers
e. Utilities & Execution
• Dask-based clustering for scaling (src/venti/utils/dask_cluster)
• CLI entry points (setup.py) or notebook orchestration
f. Outputs
• VLM rate grids, error estimates, diagnostic plots
Relationships and Data Flow
• Merge paths into Calibration & Referencing module → LOS Decomposition → Variogram Model
• Output of variogram enters Solvers → Final VLM estimates → Results (files/plots)
• Utilities (dask_cluster, utils) underpin all heavy computations
Architectural Patterns & Principles
• Modular subpackages, each encapsulating one domain (single responsibility)
• Pipeline pattern: clear data-flow stages chained together
• Plugin-style interpolation backends (hvlsc_py3) and solver backends (hector, midas)
• Separation of I/O, processing, and analysis