Explore, analyse and validate raster time series and data cubes.

Website · Download · Documentation · Roadmap

With Janus you explore satellite imagery pixel by pixel, find where, when and by how much the land changed, and measure how accurate your maps are. The change detection comes from Zeit, which ships with Janus: there is nothing else to install. Its command line is jn.

Features

• Opens any stack of rasters: a folder (one file per date), a multiband file (one band per date), a list of files or a wildcard; dates come from file names or band descriptions.
• Cloud-native: cloud-optimized GeoTIFFs straight from https://, s3://, gs:// or az:// (File → Open URL, or jn https://...), reading only the blocks and overview levels on screen.
• STAC catalogs: File → Search STAC catalog... finds the Sentinel-2, Landsat or Brazil Data Cube scenes of an area between two dates (Earth Search, Planetary Computer, BDC or any STAC API), below a cloud cover, and opens them as a series read in place (a mosaic per date, on one grid, the quality band masking clouds) or downloads them as GeoTIFFs. Step by step: Series from a STAC catalog.
• Climate cubes: NetCDF, Zarr and HDF variables with CF time (model calendars included), levels and a variable picker.
• Raster calculator and spectral indices: the 275 indices of the Awesome Spectral Indices (EVI, SAVI, NDRE, MNDWI...) or any expression over a series' bands, shown in place; Tools → Raster calculator combines the open layers by short names (a series with an image, series matched by date, layers of other grids and CRSs), reduces a series over time (max(ndvi), count(ndvi < 0.3), ndvi - mean(ndvi)) and reclassifies with rules (ndvi < 0.2 = 0); an editor that colours, completes and points at errors as you type, Ctrl+Enter to run, the result a layer like any other. Step by step: Raster calculator.
• 14 map modes on the GPU: value, anomaly, difference to a reference date, temporal statistics, trend, largest drop (date and magnitude), multitemporal RGB, temporal composites (median, mean, max... of a year or a moving window); a season (days of the year) for the statistics and composites; changing date, colormap or stretch is instant.
• Pixel series as you hover, pins to compare places, a region (a rectangle, a polygon or a freehand outline) with its mean and p10–p90 band; OLS and Sen trends.
• Vector layers (GeoJSON, Shapefile, GeoPackage, KML, CSV of points) over the map in any CRS: the series of every plot or parcel extracted to a CSV table, and a TimeSync-style interpretation mode to label samples with the keyboard.
• Map validation (Olofsson et al. 2014): strata from any map (classes, intervals, dates of change, the agreement of several maps, a buffer around change), sample size and allocation, a stratified random sample in a GeoPackage project, blind interpretation of its points with chips of every date, and the accuracy and error-adjusted areas ± confidence intervals of every map assessed with it (LandTrendr × CCDC × BFAST × MapBiomas...), live as labels come; several interpreters, kappa and adjudication; an HTML report and the matching zeit.accuracy code. Step by step, with screenshots: Validating a map.
• Compare: several series as layers, in any CRS (reprojected on the GPU), map panels side by side, a swipe divider, and space-time transects (distance × date) along a line, a polyline or a vector line.
• Basemap under the series: Esri World Imagery, Sentinel-2 cloudless, OpenStreetMap or your own XYZ tiles; off until you pick one.
• Categorical series (land cover, classifications) with class colours, names and change summaries; several bands per date with normalized differences, cloud masks and RGB composites (true colour, colour infrared...) while the chart keeps the index.
• Change detection built in: LandTrendr and the recovery after its disturbances (Y2R, R80P, RRI), Mann-Kendall, AR(1) trends and their regional test (PARTS, after remotePARTS), emerging hot spots (Getis-Ord Gi* of every date and its trend: new, intensifying, persistent... hot and cold spots, as ESRI's), BFAST, CCDC, EWMACD (monitoring every observation with control charts: disturbance, recovery, gain), CODED (forest degradation, with the NDFI), spectral unmixing with endmembers from pins (of every date's bands, or of a hyperspectral scene), phenology, TWDTW, SOM and k-means clustering (named by your samples), the post-processing of classifications (one-year flickers, majority, minimum mapping unit), SNIC segmentation, texture (the GLCM's properties, as scikit-image's), Tmask cloud screening and more from Zeit, fitted live on the chart (BFAST and CCDC models drawn over the series) or run on the image (also CCDC's cloud-free synthetic image of a date), plus the agreement of several change maps; deep learning (TempCNN, L-TAE, Presto, U-TAE) trained on your samples and applied on the GPU (optionally with sits' Bayesian smoothing of the probabilities, and the model's area of applicability: where the image is unlike its samples, as CAST's aoa), and Presto's pretrained embeddings, without Python; field delineation: Sentinel-2 fields as parcels by WRI's Trazo and Fields of The World's networks (their published accuracy), fine-tuned on fields you draw. The Python history keeps every finished run as Python (a script that reproduces it, to read, copy or save as a .py or a notebook), and the chart gives the code that reproduces a series and the tool's Zeit calls. Zeit's methods are part of Janus: no Python to install, and a fit takes under a millisecond.
• Spectra (hyperspectral, e.g. EnMAP): a file whose bands have wavelengths opens with the chart in nanometres, the spectrum of the cursor, pins and region, and true-colour, colour-infrared and SWIR composites; index maps (NDVI, red edge, PRI...), principal components and spectral similarity, and series of scenes by wavelength.
• Embeddings of foundation models (AlphaEarth, TESSERA): downloaded for the visible area over any series, shown through their principal components (also local to a region, refitted as you move it), the similarity to the cursor or a pin, the change between years and its direction (a PCA of the changes, the same colour for the same kind of change, or how like a pin's change each is), and classes from a few pins or a region (a logistic regression or centroids fitted in milliseconds, with their areas and the transitions between years); computed with SIMD kernels.
• Export: maps as PNG figures, timelapse animations (GIF, MP4, WebM) with the date, legend and scale bar, values, views, Zeit results and embedding views as georeferenced GeoTIFFs, series as CSV.
• Fast on any disk: about 0.15 s to the first frame, a disk cache of the cube, and a full-resolution cache that serves a pixel's series in about 1 ms even when the data sit on a spinning hard disk.

Install

Download the package for your system from the latest release. Each one carries GDAL, and Zeit is part of the program: nothing else to install.

System	Package
Windows	janus-<version>-setup.exe: per-user installer, optional jn on PATH and Open in Janus in the right-click menu
macOS 15+ (Apple Silicon)	janus-<version>-macos-arm64.dmg: drag Janus to Applications; not notarized, so open it the first time with right click → Open
Linux x86_64	janus-<version>-linux-x86_64.tar.xz: portable, extract and run jn

Details for each system: docs/install.md.

Quick start

jn D:\data\ndvi_annual\        # every raster in the folder, 1 file = 1 date
jn landsat_stack.tif           # 1 multiband file, 1 band = 1 date
jn "S2_*_NDVI.tif" --band 1    # a wildcard pattern

Or open a folder from the File menu, the Files panel, or by dropping it on the window. Then:

Hover	the pixel's series (exact once the mouse rests)
Click / Shift + drag / R	drop a pin / draw a region / draw a polygon
Alt + click	a polygon of a vector layer becomes the region
← → / P	previous or next date / play
Ctrl + drag / C	space-time transect / swipe
S / Space / F	satellite basemap / show or hide the active layer / fit the map
Ctrl+T / Ctrl+W	open / close a map panel

Everything else is in docs/usage.md and under Help → Shortcuts and usage.

Documentation

• Usage: command line, every interaction, the Zeit tools, performance notes
• Installation: details for Windows, macOS and Linux
• Building from source: Windows, Linux and macOS, the tests, packages
• Architecture: how the cube flows from the files to the GPU, and the crates
• RUST_MIGRATION.md (Portuguese): how Janus moved from C++ to Rust for 1.0, with the measurements
• IDEIAS.md (Portuguese): roadmap, design decisions and the measurements behind them

Building

Janus is written in Rust. It needs GDAL (e.g. from conda-forge); everything else comes from crates.io.

cargo build --release

See docs/building.md for pointing the build at GDAL on each platform, the tests and the packages.

License

Janus is free software under the GNU General Public License, version 2 or later (LICENSE), as is Zeit.

---

Built with egui, wgpu and GDAL; Direct3D 12 on Windows, Vulkan on Linux, Metal on macOS. Change detection by Zeit. Janus was written in C++ (Dear ImGui, ImPlot, GLFW) up to 0.39, and called tsv up to 0.16.