User Guide

Step-by-step tutorials for every part of Zeit. Each one starts with what the method is for and a real output, explains how it works in plain language, then walks through the code and how to read the results.

New here?

Do the Quickstart first, and skim Core Concepts for the conventions (array shapes, dates, scale factors) that every tutorial uses. Unsure which method fits your question? See Choosing an Algorithm.

Change detection

Find where and when the land surface changed.

Annual dataLandTrendrDisturbance and recovery history from one image per year. Maps of year, magnitude and duration.

Dense dataCCDCHarmonic models of every clear observation. Dates changes and describes the land before and after.

Regular seriesBFASTSeparate breaks in the trend from breaks in the seasonal cycle.

Near real timeBFAST MonitorTest new observations against a stable history. The basis of alert systems.

Regular seriesBFAST LiteThe optimal number of breaks in a whole series, in a single fast pass.

Time-series analysis

Trends, seasons, patterns and objects.

TrendsMann-KendallSignificant greening and browning, with a robust slope per pixel.

SeasonsPhenologyStart, peak and end of every growing season, and 16 more metrics.

ClassificationTWDTWMatch each pixel to reference patterns, tolerating shifts in timing.

ClusteringSOMFind the typical trajectories in a landscape without labels.

SegmentationSNICGroup pixels with similar histories into fields and patches.

Deep learning

PyTorch models for classification and change detection, weight-compatible with their reference implementations.

Pixel classificationTempCNNA fast 1-D convolutional baseline for pixel time series.

Model	Task	Tutorial
Overview	Available models, data loading, loss functions	Deep Learning
TempCNN	Pixel time-series classification (convolutional)	TempCNN
LTAE / LightTAE	Pixel time-series classification (attention)	LTAE & LightTAE
U-TAE	Segmentation of whole image patches over time	UTAE
Siamese	Change detection between two dates	Siamese Change Detector
GeoFoundationViT	Fine-tuning pretrained foundation models	GeoFoundationViT

Getting and preparing data

Cloud catalogsSTAC Data CubesLazy cubes from Earth Search, Planetary Computer or Brazil Data Cube; compositing and smoothing.

Cloud maskingTmaskCatch the clouds and shadows that single-image masks miss.

• Plotting & Exploring Results: zeit.plot for maps, dense time series and pixel fits, in a notebook or a window.
• Google Earth Engine: harmonised Landsat composites prepared on Google's servers.
• Embeddings (TESSERA, AlphaEarth): the yearly embeddings of foundation models as a cube: classify from few samples, find similar places, follow change year to year.
• Parallel & Cloud Processing: from one core to a cluster, with Dask and Zarr.