Skip to content

API Reference

Every public function and class, with its signature, parameters and an example. Signatures on these pages are generated from the code, so they always match the installed version.

How to import

import zeit gives you the most used functions at the top level (zeit.landtrendr, zeit.save_raster, …) and registers the .zeit xarray accessor. Everything else lives in submodules (zeit.twdtw, zeit.segmentation, zeit.ai, …), shown in each signature.

  • Data & I/O

    STAC and Earth Engine cubes, local files, compositing, GeoTIFF I/O, QA weights.

  • Embeddings

    TESSERA and AlphaEarth yearly embeddings as cubes; similarity search; change from year to year.

  • Pre-processing

    Tmask cloud masking, Whittaker and Savitzky-Golay smoothing, spike removal.

  • Change Detection

    LandTrendr, CCDC, BFAST, BFAST Monitor, BFAST Lite, CODED, event extraction and agreement.

  • Time-Series Analysis

    Mann-Kendall trends, phenology, TWDTW, SNIC segmentation, SOM clustering.

  • Post-processing

    Spatial filters, CCDC classification, water masks.

  • Validation

    Stratified samples, labelling points by eye, accuracies and error-adjusted areas with confidence intervals.

  • Visualisation

    zeit.plot: a fast time-series viewer, pixel inspector, basemaps and static figures.

  • Deep Learning

    TempCNN, LightTAE, U-TAE, Siamese and foundation models; datasets and losses.

  • Xarray Accessor

    DataArray.zeit.*: every algorithm on lazy Dask cubes, and Zarr output.

  • Many Tiles & Helpers

    A folder of tiles on this computer and on others of the local network, resumable.

  • Command Line

    The core algorithms as zeit subcommands, for scripts and HPC jobs.

All functions

Function What it does
Data & I/O
build_time_series Lazy cube from a STAC catalog
build_annual_composites Cloud-masked annual composites from a STAC catalog (LandTrendr input)
download_gee_timeseries Harmonised Landsat composites from Earth Engine
download_gee_image / resolve_roi Download any ee.Image; turn tile ids, files or boxes into an area
regularize_time_series Median or medoid composites at a fixed step
load_raster / save_raster Read any time series as a georeferenced cube / write GeoTIFFs
get_georef CRS and transform of a raster or cube
qc_sentinel2_scl, qc_modis_summary, qc_modis_state QA bands to observation weights
Embeddings
load_embeddings TESSERA or AlphaEarth embeddings of a region as a (time, band, y, x) cube
similarity Where (and when) pixels look like reference samples
embedding_change How far each pixel's embedding moved from year to year (events with extract_events)
Pre-processing
bandpass_adjust Sentinel-2 (and Landsat 5/7) on Landsat 8's bands: HLS's bandpass step, for L2A cubes (HLS itself: build_time_series)
tmask Time-series cloud and shadow detection of a cube
smooth Whittaker (uneven dates, gaps, weights) or Savitzky-Golay smoothing of a cube or series
unmix Fractions of green vegetation, NPV, soil, shade, cloud; the NDFI
desawtooth LandTrendr spike removal
Change detection
landtrendr LandTrendr on a file, cube, array or single pixel
extract_events One event per pixel, the same maps for every change algorithm (and for embedding change)
agreement Where and when several change maps agree
apply_vertices Fit another band to the same vertices
ccdc CCDC on a file, cube, array or single pixel
predict_synthetic_image Evaluate CCDC models on any date
bfast_monitor Near-real-time monitoring, on a file, cube, array or single pixel
bfast_lite Optimal multiple breakpoints
bfast Trend and seasonal breaks
coded Forest degradation and deforestation from the NDFI (CODED)
Time-series analysis
mann_kendall Trend test and Theil-Sen slope
phenology 19 phenology metrics per year or season
twdtw Classification by time-weighted DTW, on a file, cube or single pixel
snic / snic_to_polygons / snic_grid Superpixel segmentation of a map or cube, and its polygons
som Self-organizing map clusters of a map or cube, with each neuron's prototype
clean_samples Flag labelled samples whose class is not their SOM neuron's
SOM The SOM engine, online and batch (bit-exact minisom port)
Post-processing
apply_mmu_filter Remove patches below a minimum size
apply_majority_filter / apply_bayesian_filter Smooth class maps
train_classifier / classify Train at sample points, classify any cube, metrics or CCDC models
extract_water_mask Water mask from CCDC
Validation
sampling_design Sample size and allocation per stratum (Olofsson et al. 2014)
stratified_sample Stratified random points of a map
interpret Label reference points by eye: chips, series, keyboard, review
accuracy Accuracies and error-adjusted areas with confidence intervals
Visualisation
plot Any cube, map, result or pixel: interactive viewer in a notebook or window, or a matplotlib figure
Deep learning (zeit.ai)
ai.samples, ai.train, ai.predict From a cube and labelled samples to a map with a deep learning model
ai.save, ai.load Keep a trained model with what prediction checks
TempCNN, LightTAE, LTAE Pixel time-series classifiers
UTAE Spatio-temporal patch segmentation
SiameseChangeDetector Two-date change detection
GeoFoundationViT Foundation-model fine-tuning
STACCubeDataset Windows that cover a lazy cube, for a loop of your own
FocalLoss, TverskyLoss, ContrastiveSiameseLoss Losses for imbalanced data
Many tiles & helpers
run_tiles A command on every tile of a folder, here and on helpers; resumes where it stopped
run_worker Lend this computer to the group: wait for tiles and process them
invite_helpers A code with which other computers join the group
tile_urls, stac_tiles, gee_tiles Tiles each computer downloads or builds for itself