Tutorials
GNSS vs InSAR validation (notebook)
The GNSS-InSAR validation notebook runs the complete comparison workflow on the real Sentinel-1 Kīlauea test dataset that ships with the repository: metric-buffer sampling, per-station time series, velocity scatter, the structure function with a requirement curve, per-epoch network misfit, and the effect of the sampling buffer radius. It is the executable companion to How to validate an OPERA DISP product against GPS.
Tour of the analysis modules (notebook)
The geepers tour notebook is a runnable, self-contained walkthrough — synthetic data with known truth, so every result can be checked against what went in:
- robust velocities with MIDAS;
- velocities with realistic uncertainties under power-law + white noise, and why OLS sigmas are 5-10× too optimistic;
- temporal velocity variability (sliding-window MIDAS);
- spotting bad stations with spatial variability;
- interpolating a velocity field two ways — robust GPS Imaging vs. geostatistical collocation — on a field with a sharp boundary and a corrupted station.
Run it locally with:
jupyter lab docs/notebooks/geepers_tour.ipynb
Detecting steps and testing linearity (notebook)
The steps and linearity notebook answers the question that comes before quoting any velocity: does one rate describe this series, and for how long can it be extrapolated? On synthetic series with known truth, then on a real station next to the 2019 Ridgecrest earthquakes, it shows how to
- find uncatalogued offsets with
detect_steps, and why the size check matters under temporally correlated noise; - run
linearity_test(linear vs quadratic vs piecewise) on 30-day means, with the steps declared; - read the verdict, the departure from a straight line and the
p-values, and turn them into a
validity_horizon; - see what goes wrong when a step is left out, and why the noise model is kept at flicker + white.
Sections 1-5 run offline; section 6 downloads one station from UNR.
jupyter lab docs/notebooks/steps_and_linearity.ipynb
Your first GPS-InSAR comparison
-
Gather inputs: a stack of displacement rasters (one per date, dates in the filenames) or a Zarr/NetCDF cube, plus a 3-band line-of-sight ENU raster on the same grid.
-
Run the workflow:
bash
geepers --los los_enu.tif --timeseries-files displacement_*.tif -o GPS
GPS stations inside the raster bounds are discovered and downloaded
automatically (cached under ~/.cache/geepers).
-
Look at the outputs in
GPS/: -
combined_data.csv— every GPS and InSAR LOS series, tidy format; relative_comparison.csv— both series referenced to a common station;-
station_summary.csv— per-station GPS (MIDAS) and InSAR rates, their difference, and quality metrics. -
Plot a station (pandas one-liner):
python
import pandas as pd
df = pd.read_csv("GPS/combined_data.csv", parse_dates=["date"])
(df[df.id == "P123"]
.pivot_table(index="date", columns="measurement", values="value")
[["los_gps", "los_insar"]]
.plot(style=".", ms=3))
For recipes on screening stations, estimating velocities, and gridding velocity fields, see the How-To Guides; for the underlying methods, see Background theory.