Getting started¶

This page goes from a folder of DISP granules to a referenced displacement time series, a map and an export. Every block runs as written: the first one writes a small synthetic frame with disp_xr.tutorial, so nothing needs downloading. Replace the path with your own granules.

Make or find granules¶

import tempfile
from pathlib import Path

import disp_xr as dxr
from disp_xr.tutorial import make_disp_stack

folder = Path(tempfile.mkdtemp()) / "granules"
make_disp_stack(folder, "three_ministacks", shape=(128, 128))

table = dxr.find_products(folder)  # one row per granule, sorted by date
print(table[["frame_id", "reference_time", "secondary_time", "version"]])

find_products parses OPERA file names (and the /identification group when the name is not standard), drops duplicate granules keeping the latest version and production, and refuses unsupported product versions. On your own data it also accepts a glob, a list of paths or an s3:// prefix.

Open¶

ds = dxr.open_disp(table, corrections=True)
ds

ds is lazy: dims (time, y, x), coordinates time (secondary date) and reference_time (the ministack reference date of each epoch), the CRS from spatial_ref, and one variable per product layer. Integer layers keep their dtype and their no-data value; float layers use NaN.

Correct, mask and chain¶

The order is fixed: corrections and the quality mask are applied per granule, before the ministacks are chained.

ds = dxr.correct(ds)                  # ionosphere and solid-earth tide
ds = dxr.mask(ds, "recommended")      # NaN where recommended_mask is bad
ds = dxr.chain(ds)               # one series from the first date
ds.attrs["reference_date"]

The accessor gives the same operations as methods, so a chain reads in order: ds.disp.correct().disp.mask("recommended").disp.chain().

Re-reference¶

Displacement in DISP is relative to the product's reference point. Pick your own: a stable pixel chosen automatically, a point, the scene median, or an area.

ds = dxr.rereference(ds, "auto")          # deterministic stable pixel
# dxr.rereference(ds, "point", lon=-118.1, lat=34.2)
# dxr.rereference(ds, "area", bbox=(xmin, ymin, xmax, ymax))

Plot¶

ds.disp.plot_map()                     # last epoch, mm, symmetric scale
ds.disp.plot_point(row=64, col=64)     # series with reference changes marked

Export¶

dxr.to_zarr(ds, folder.parent / "frame.zarr")          # Zarr v3, sharded
dxr.to_geotiff(ds.isel(time=-1), folder.parent / "last.tif", layer="displacement")  # COG

Every output records the disp-xr version, the input granules and the operations applied (ds.disp.history). The gallery shows each format, including MintPy.

Large frames¶

A full frame is about 7,000 x 9,000 pixels and a few hundred epochs. open_disp keeps the layout of the files, one epoch and one 1,024-pixel block per chunk, so a map or a point reads only what it needs. For repeated per-pixel work, rechunk once with ds.disp.rechunk("timeseries"), or write a Zarr copy with preset="timeseries". See Chunks for maps and time series.