Awesome
<!-- omit in toc -->
Satellite Image Time Series Datasets
This page presents a list of satellite imagery datasets with a temporal dimension, mainly satellite image time series (SITS) and satellite videos, for various computer vision and deep learning tasks. It covers multi-temporal datasets with more than two acquisitions but not bi-temporal datasets. We focus mainly on annotated datasets.
<!-- omit in toc -->
Table of Contents
Semantic and Instance Segmentation
Datasets are sorted by annotation granularity. We note that polygons annotations are reserved for crop-type identification tasks, while pixel annotations might be considered in more general tasks such as land cover mapping.
Pixel annotations for each image
Dataset name | Year | Image source | Spatial resolution | Temporal resolution | Number of classes | Acquisition |
---|
MultiEarth 2023 | 2023 | Sentinel-1 + Sentinel-2 + Landsat-5 + Landsat-8 | 10m + 10m + 30m + 30m | Weekly acquisitions depending on the source & Monthly annotation | 2 | Amazon (1984-2021) |
MultiEarth 2022 | 2022 | Sentinel-1 + Sentinel-2 + Landsat-5 + Landsat-8 | 10m + 10m + 30m + 30m | Weekly acquisitions depending on the source & Monthly annotation | 2 | Amazon (1984-2021) |
Dynamic World | 2022 | Sentinel-2 | 10m | Weekly acquisition and weekly automatic annotation without human verification | 9 | Global (2015-present) |
DynamicEarthNet | 2021 | PlanetFusion | 3m | Daily acquisition & Monthly annotation | 7 | Global (2018-2019) |
SpaceNet 7 | 2020 | PlanetScope | 4m | Monthly acquisition & annotation | 2 | Global (2017-2020) |
Pixel annotations for each time series
Dataset name | Year | Image source | Spatial resolution | Temporal resolution | Number of classes | Acquisition |
---|
MultiSenNA | 2024 | Sentinel-1 + Sentinel-2 | 5m + 10m | Daily + weekly acquisition | 14 | Southwestern France (2019-2020) |
DAFA-LS | 2024 | Planet | 3m | Monthly acquisition | 2 | Afghanistan (2016-2023) |
BraDD-S1TS | 2023 | Sentinel-1 | 10m | Weekly acquisition | 2 | Brazil (2020-2021) |
FLAIR #2 | 2023 | Sentinel-2 | 10m | Weekly acquisition | 13 | France (1-year aquisition) |
MultiSenGE | 2022 | Sentinel-1 + Sentinel-2 | 5m + 10m | Daily + weekly acquisition | 14 | Eastern France (2019-2020) |
PASTIS | 2021 | Sentinel-2 | 10m | Weekly acquisition | 18 | France (2018-2019) |
PASTIS-R | 2021 | Sentinel-1 + Sentinel-2 | 5m + 10m | Daily + weekly acquisition | 18 | France (2019) |
AI4EO Enhanced Sentinel 2 Agriculture | 2021 | Sentinel-2 | 10m | Weekly acquisition | 2 | Slovenia (2019) |
UTRNet | 2021 | Landsat-8 | 30m | Irregular acquisition | 2 | China (2013-2021) |
MTLCC | 2018 | Sentinel-2 | 10m | Weekly acquisition & Annual annotation for 2016 and 2017 | 17 | Munich, Germany (2016-2017) |
TiSeLaC | 2017 | Landsat-8 | 30m | Bi-monthly acquisition | 9 | Reunion Island (2014) |
Polygon annotations for each image
Dataset name | Year | Image source | Spatial resolution | Temporal resolution | Number of classes | Acquisition |
---|
Sen4AgriNet | 2022 | Sentinel-2 | 10m to 60m | Weekly acquisition & Annual annotation | 168 | Catalonia & France (2019-2020) |
Deep Crop Rotation | 2021 | Sentinel-2 | 10m | Weekly acquisition & Annual annotation | 10 | France (2018-2020) |
Campo Verde | 2018 | Landsat-8 + Sentinel-1 | 30m + 10m | Bi-monthly acquisition & annotation | 14 | Brazil (2015-2016) |
LEM | 2018 | Landsat-8 + Sentinel-1 + Sentinel-2 | 30m + 10m + 10m | Bi-monthly (L8+S1) + weekly (S2) acquisition & Monthly annotation | 14 | Brazil (2017-2018) |
MTLCC | 2018 | Sentinel-2 | 10m | Weekly acquisition & Annual annotation | 17 | Munich, Germany (2016-2017) |
Polygon annotations for each time series
Dataset name | Year | Image source | Spatial resolution | Temporal resolution | Number of classes | Acquisition |
---|
SICKLE | 2024 | Landsat-8 + Sentinel-1 + Sentinel-2 | 30m + 3m + 10m | Bi-monthly + 12d + weekly acquistion | 21 | India (2018-2021) |
AgriSen-COG | 2023 | Sentinel-2 | 10m | Weekly acquisition | 103 | Austria, Belgium, Spain, Denmark, Netherlands (2019-2020) |
TimeMatch | 2022 | Sentinel-2 | 10m | Weekly acquisition | 16 | Austria, Denmark, mid-west France, southern France (2017) |
DENETHOR | 2021 | Cloud-free fusion of images from various satellites | 3m | Daily acquisition | 10 | Germany (2018-2019) |
EuroCrops | 2021 | Sentinel-2 | / | Weekly acquisition | 43 | Europe (2015-2022) |
TimeSen2Crop | 2021 | Sentinel-2 | 10m | Weekly acquisition | 16 | Austria (2017-2018) |
Canadian Cropland | 2021 | Sentinel-2 | 10m | Monthly acquisition | 10 | Canada (2019) |
ZueriCrop | 2021 | Sentinel-2 | 10m | Weekly acquisition | 48 | Zurich, Switzerland (2019) |
Crop type in Western Cap | 2021 | PlanetScope + Sentinel-1 + Sentinel-2 | 3m + 10m + 10m | Bi-monthly (Planet+S1) + weekly (S2) acquisition | 5 | South Africa (2017) |
Spot the crop challenge | 2021 | Sentinel-1 + Sentinel-2 | 5m + 10m | Bi-monthly + weekly acquisition | 10 | South Africa (2016) |
BreizhCrops | 2020 | Sentinel-2 | 60m | Weekly acquisition | 9 | Brittany, France (2017) |
Crop type in Ghana | 2020 | PlanetScope + Sentinel-1 + Sentinel-2 | 3m + 10m + 10m | Bi-monthly (Planet+S1) + weekly (S2) acquisition | 4 | Ghana (2017) |
Crop type on South Soudan | 2020 | PlanetScope + Sentinel-1 + Sentinel-2 | 3m + 10m + 10m | Bi-monthly (Planet+S1) + weekly (S2) acquisition | 4 | South Soudan (2017) |
CV4A Kenya | 2020 | Sentinel-2 | 10m | Bi-monthly acquisition | 7 | Kenya (2019) |
Pixel-Set dataset | 2020 | Sentinel-2 | 10m | Weekly acquisition | 20 | France (2017) |
Image-level annotations
Dataset name | Year | Image source | Spatial resolution | Temporal resolution | Number of classes | Acquisition |
---|
fMoW-Sentinel | 2022 | Sentinel-2 | 10m | Irregular acquisition | 63 | Global (2015-2019) |
SEN12-FLOOD | 2020 | Sentinel-1 + Sentinel-2 | 10m + 10m | Bi-monthly + weekly acquisition | 2 | African, Iranian and Australian cities (2018-2019) |
fMoW-RGB | 2018 | DigitalGlobe constellation | multiple resolutions (0.3m to 3.7m) | Irregular acquisition | 63 | Global (2002-2017) |
Datacube-level annotations
Dataset name | Year | Image source | Spatial resolution | Temporal resolution | Number of classes | Acquisition |
---|
Sen4Map | 2024 | Sentinel-2 | 10m + 20m | Weekly acquisition | 119 | Europe (2018) |
Planted | 2024 | Sentinel-1 + Sentinel-2 + Lansat-7 + ALOS-2 + MODIS | 10m (S1+S2) + 30m (L7+A2) + 250m (M) | Seasonal (S1+S2+L7) yearly (A2) and monthly (M) acquisitions | 64 | Global (2013-2017) |
TreeSatAI-Time-Series | 2024 | Sentinel-1 + Sentinel-2 | 10m + 10m | Weekly acquisition | 20 | Germany (2017-2020) |
RapidAI4EO Corpus | 2023 | PlanetFusion + Sentinel-2 | 3m + 10m | 5-days + monthly acquisition | 44 (multi-label) | Europe (2018-2019) |
Regression
Dataset name | Year | Image source | Spatial resolution | Temporal resolution | Acquisition |
---|
Open Buildings 2.5D Temporal | 2024 | Sentinel-2 + ? | 10m (S2) + 50cm (?) + 50cm (GT) | Weekly + annual acquisition & Annual annotation | Africa, South Asia, South-East Asia, Latin America and the Caribbean (2016-2023) |
Wald5Dplus/Forest5Dplus | 2024 | Sentinel-1 + Sentinel-2 | 10m | Weekly acquisition | Germany (2020-2021) |
Multi-Modal Satellite Imagery Dataset | 2024 | Sentinel-2 + Multilabel metadata | 10m + municipality-level | Weekly (S2) acquisition | Colombia (S2: 2016-2018, metadata: 2007-2019) |
CropNet | 2024 | Sentinel-2 + WRF-HRRR | 9km + 9km | 14d + 1d & Annual annotation | USA (2017-2022) |
SICKLE | 2024 | Landsat-8 + Sentinel-1 + Sentinel-2 | 30m + 3m + 10m | Bi-monthly + 12d + weekly acquistion | India (2018-2021) |
BioMassters | 2023 | Sentinel-1 + Sentinel-2 | 20m + 10m | Monthly acquisition & Annual annotation | Finland (2016-2021) |
Forecasting
[!NOTE]
Here we list a few forecasting datasets, particularly for weather forecasting, but this list is by no means exhaustive. More weather forecasting datasets are listed here.
Dataset name | Year | Image source | Spatial resolution | Temporal resolution | Number of classes | Acquisition |
---|
GreenEarthNet | 2024 | Sentinel-2 + meteorological observations | 20m | Weekly (S2) + daily | / | Europe (2017-2022) |
SeasFire | 2023 | ERA5, MODIS, ... | 27km | 8d | / | Global (2001-2021) |
Digital Typhoon | 2023 | Himawari | 5km | 60min | / | Western North Pacific basin (1978-2022) |
SEN2DWATER | 2023 | Sentinel-2 | 10m | Every 2 months | / | Italy & Spain (2020-2022) |
EarthNet2021 | 2021 | Sentinel-2 + mesodynamic models | 20m + 1,28km | Weekly (S2) + daily | / | Europe (2016-2020) |
CloudCast | 2021 | Meteosat Second Generation | 3km | 15min | 11 | Europe (2017-2018) |
MeteoNet | 2020 | Ground station observations, satellite images, rain radar observations, weather forecasting models and land-sea and relief masks | Variable | Variable | / | France (2016-2018) |
SEVIR | 2020 | GOES-16 + NEXRAD | 2km + 1km | 5min | / | USA (2017-2019) |
Object tracking
Dataset name | Year | Image source | Spatial resolution | Temporal resolution | Number of classes | Acquisition |
---|
TMS | 2024 | Jilin-1 + SkySat + Synthetic | 1m | 1 frame per second | 1 | Cities |
AIR-MOT | 2022 | Jilin-1 | 1m | 5 to 10 frame per second | 2 | Cities |
VISO | 2021 | Jilin-1 | 1m | 10 frame per second | 4 | Cities |
SatSOT | 2021 | Jilin-1 + SkySat + Carbonite-2 | 1m | 10 to 25 frame per second | 4 | Cities |
Other tasks
Dataset name | Year | Task | Image source | Spatial resolution | Temporal resolution | Acquisition |
---|
BreizhSR | 2024 | Super-resolution | Sentinel-2 + SPOT-6/7 | 10m + 2.5m | Weekly (S2) acquisition | Brittany France (2018) |
SSL4EO-L | 2023 | Pre-training task | LandSat-4,5,7,8,9 | 30m | Seasonally acquisition | Global |
SSL4EO-S12 | 2023 | Pre-training task | Sentinel-1 + Sentinel-2 | 5m + 10m | Seasonally acquisition | Global |
SAT-MTB | 2023 | Detection, segmentation and object tracking | Jilin-1 | 1m | 10 frame per second | Cities |
TimeMatch | 2022 | Domain adaptation | Sentinel-2 | 10m | Weekly acquisition | Austria, Denmark, mid-west France, southern France (2017) |
WorldStrat | 2022 | Super-resolution | Spot-6 + Spot-7 + Sentinel-2 | 1,5m + 1,5m + 10m | Weekly (S2) acquisition | Global |
Jilin-189 | 2022 | Video super-resolution | Jilin-1 | 1m | 25 frame per second | Cities |
SEN12MS-CR-TS | 2022 | Cloud removal | Sentinel-1 + Sentinel-2 | 10m + 10m | Bi-monthly (S1) + weekly (S2) acquisition | Global (2018) |
NASA Harvest | 2022 | Field Boundary Detection | PlanetScope | 3.7m | Monthly acquisition & Time-independant annotation | Rwanda (2021) |
AI4Boundaries | 2022 | Field boundary detection | Sentinel-2 + aerial ortho-photo | 10m + 1m | Monthly acquisition & Yearly annotation | Europe (2019) |
Seasonal Contrast | 2021 | Pre-training task | Sentinel-2 | 10m | Seasonally acquisition | Global |
PROBA-V Super-Resolution | 2019 | Super-resolution | PROBA-V | 300m + 100m | Daily acquisition | Global |
Citation
The authors thank the French spatial agency (CNES) and the Brittany region for their financial support.
<sup>1</sup>Université Bretagne Sud, IRISA, UMR CNRS 6074, Vannes, France
<sup>2</sup>Centre National d’Études Spatiales (CNES), Toulouse, France
If you use this work, consider citing it as below.
@misc{dufourg2023sitsdatasets,
author = {Dufourg, Corentin and Pelletier, Charlotte and May, Stéphane and Lefèvre, Sébastien},
title = {Satellite Image Time Series Datasets},
howpublished = {\url{https://github.com/corentin-dfg/Satellite-Image-Time-Series-Datasets}}
}