Supported Models (Overview)¶

This page is the model selection entry point. Use it to answer one question quickly: which model IDs should I shortlist for this task?

Once you have a shortlist, use Advanced Model Reference for side-by-side preprocessing and temporal details, then open the linked detail page for the exact contract, caveats, and examples. If you are about to change input_prep, variant, fetch resolution, patch size, or image size, read Before You Start first, because those knobs affect both runtime cost and embedding semantics.

How To Read This Page¶

Start with the quick chooser, then scan the catalog table for input and temporal fit, and open the detail page before benchmarking or production use.

Canonical model IDs use the short public names shown on this page, such as remoteclip, prithvi, terrafm, and thor. Some detail-page filenames still use older names for compatibility, but the canonical IDs above are the names users should copy into code.

Quick Chooser by Goal¶

Goal	Good starting models	Why
Fast baseline / simple pipeline	`tessera`, `gse`, `copernicus`	Precomputed embeddings, fewer runtime dependencies
Simple S2 RGB on-the-fly experiments	`remoteclip`, `satmae`, `satmaepp`, `scalemae`	Straightforward RGB input paths
Time-series temporal modeling	`prithvi`, `olmoearth`, `galileo`, `anysat`, `agrifm`	Native multi-frame temporal packaging — see Temporal Sampling
Multispectral / strict spectral semantics	`satmaepp` (`modality="s2_10b"`), `dofa`, `clay`, `terramind`, `thor`, `satvision`	Strong channel/schema assumptions
Mixed-modality experiments (S1/S2)	`terrafm`, `thor`	Supports S2 or S1 path (per call)

Model Catalog Snapshot¶

Precomputed Embeddings¶

Model ID	Type	Primary Input / Source	Default Resolution	Dim	Temporal mode	Notes	Detail
`tessera`	Precomputed	GeoTessera embedding tiles	10m	128	yearly coverage product	Fast baseline, source-fixed precomputed workflow; tiles resampled onto the common EPSG:3857 grid	detail
`gse`	Precomputed	Google Satellite Embedding (annual)	10m	64	`TemporalSpec.year(...)`	Annual product via provider path	detail
`copernicus`	Precomputed	Copernicus embeddings	0.25°	768	limited (2021)	Coarse resolution product on fixed EPSG:4326 grid	detail

On-the-fly Foundation Models¶

Model ID	Primary Input	Dim	Default Resolution	Input size (px)	Temporal style	Notable requirements	Detail
`prithvi`	S2 6-band	768	30m	224	multi-frame (auto, ≤4)	required temporal + location side inputs	detail
`olmoearth`	S2 L2A 12-band / S1 VV/VH	128–1024	10m	256 (flexible)	multi-frame (auto, ≤12)	FlexiViT; 4 sizes (nano/tiny/base/large)	detail
`dofa`	Multispectral + wavelengths	768	10m	224	single composite	wavelength vector required	detail
`clay`	S2 L2A 10-band	1024	10m	256	single composite	metadata conditioning (latlon/time/gsd/wavelengths)	detail
`terramind`	S2 12-band	384	10m	224	single composite	ViT-S class; strict z-score normalization	detail
`terrafm`	S2 12-band or S1 VV/VH	768	10m	224	single composite	dual-modality by channel count	detail
`thor`	S2 10-band or S1 VV/VH	768	10m	288	single composite	dual-modality; grouped tokens; native-snap	detail
`galileo`	S2 10-band time series	128	10m	64	multi-frame (auto, ≤12)	nano default; month tokens	detail
`anysat`	S2 10-band time series	768	10m	24	multi-frame (fixed `T`)	JEPA; `s2_dates` DOY side input	detail
`agrifm`	S2 10-band time series	1024	10m	224	multi-frame (fixed `T`)	Video Swin; fixed `T` frame stack	detail
`fomo`	S2 12-band	768	10m	64	single composite	per-channel spectral modality keys	detail
`wildsat`	S2 RGB	256	10m	224	single composite	biodiversity training; image_head default	detail
`satvision`	TOA 14-channel (MODIS)	4096	1000m	128	single composite	SwinV2 Giant; strict channel calibration	detail
`remoteclip`	S2 RGB (`B4,B3,B2`)	512	10m	224	single composite	CLIP projection; RGB preprocessing	detail
`scalemae`	S2 RGB + scale	1024	10m	224	single composite	`sensor.scale_m` is a model input	detail
`satmae`	S2 RGB (`B4,B3,B2`)	1024	10m	224	single composite	ViT-L; MAE token/grid	detail
`satmaepp`	S2 RGB (`B4,B3,B2`) or S2 10-band	1024	10m	224 (rgb) / 96 (s2_10b)	single composite	`modality=rgb` (default) or `s2_10b`; ViT-L; fMoW eval preprocessing; 10-band uses strict band order + grouped-channel tokens	detail

Input size (px) is the fixed spatial size each model's encoder consumes: under the default input_prep="tile", inputs larger than it are cut into tiles of this size at native resolution and the outputs stitched (both for provider fetches and user-provided data — see User Data API); under input_prep="resize" they are downsampled to it in one step. Together with Default Resolution it gives the native footprint of one forward pass, e.g. galileo 64 px × 10 m ≈ 640 m. olmoearth (FlexiViT) accepts any size divisible by its patch size — 256 is its training tile size; user-provided data is consumed natively at its own size by default (warning above 512 px), and an explicit model_config image_size restores fixed-size behavior. anysat and prithvi sizes are env-tunable (RS_EMBED_ANYSAT_IMG, RS_EMBED_PRITHVI_IMG).

Temporal and Comparison Notes (What People Usually Miss)¶

TemporalSpec.range(start, end) is usually a compositing window rather than a single-scene selector, and OutputSpec.grid() may be a token or patch grid rather than a georeferenced raster, especially for ViT-like backbones. Cross-model comparisons are usually easiest with OutputSpec.pooled() plus fixed ROI, temporal, and compositing settings.

Provider-backed models and tessera share one output grid (EPSG:3857 at the model's scale_m); a ROI that straddles a UTM zone boundary is served with a warning about the resampling seam. copernicus keeps its coarse product-native EPSG:4326 grid, so check its detail page before comparing grid outputs directly against other models.

On this page, "Default Resolution" means the default source-side fetch resolution, not the final resized tensor shape sent into the backbone. Multi-frame models such as prithvi, olmoearth, galileo, anysat, and agrifm also need extra attention to frame count and temporal side inputs — how each one turns a TemporalSpec.range into frames is summarized in Temporal Sampling.

Read the details in Supported Models (Advanced Reference).

More Detail¶

For cross-model preprocessing, temporal packaging, and environment knobs, continue to Advanced Model Reference. For user-facing guidance on how to trade compute for quality, spatial detail, or temporal fidelity, read Before You Start. If you are adding a new adapter, use Extending to keep the implementation and documentation consistent.