SARLO-80: Worldwide Slant SAR Language Optic Dataset at 80 Cm Resolution

Why It Matters

By providing a ready‑to‑use, multimodal SAR‑optical corpus, SARLO‑80 lowers the barrier for developing advanced AI models that can exploit radar’s all‑weather capabilities alongside visual context, accelerating innovation in Earth observation and geospatial analytics.

Key Takeaways

• 2,500 Umbra SAR scenes standardized to 80 cm resolution
• 1,024 × 1,024 pixel patches paired with optical imagery
• English captions describe radar phenomena and scene content
• Dataset released under permissive license on Hugging Face
• Enables training SAR‑optical fusion and language models

Pulse Analysis

Synthetic aperture radar offers unique all‑weather, day‑and‑night imaging, yet its complex signal processing and scarcity of labeled data have limited widespread AI adoption. SARLO‑80 directly addresses this bottleneck by delivering a massive, uniformly processed SAR collection at 80 cm resolution, paired with co‑registered optical imagery. The inclusion of expert‑crafted and LLM‑assisted English captions adds a language dimension rarely available for radar data, turning raw backscatter into a rich, multimodal learning resource.

The dataset’s construction involved refocusing and resampling raw Umbra SICD files—originally spanning 20 cm to 2 m resolutions and diverse incidence angles—into consistent 80 cm slant‑range patches. Each patch was geometrically aligned with Sentinel‑2, Landsat‑8, or commercial optical tiles, ensuring pixel‑level correspondence despite differing sensor geometries. Captions were generated through a hybrid workflow that combined domain expert annotations with large‑language‑model assistance, capturing subtle radar effects such as layover, foreshortening, and speckle texture. All assets are openly licensed, encouraging reproducibility and community‑driven enhancements.

Researchers can now train and benchmark models for SAR‑optical fusion, multimodal retrieval, and language‑grounded remote sensing tasks without the overhead of custom preprocessing. Potential applications span disaster monitoring, infrastructure mapping, and climate analytics, where radar’s penetration through clouds complements optical detail. By lowering entry barriers and standardizing data formats, SARLO‑80 is poised to accelerate breakthroughs in geospatial AI, fostering more resilient and insightful Earth observation solutions.

SARLO-80: Worldwide Slant SAR Language Optic Dataset at 80 cm Resolution

Community Article · Published December 1, 2025

Authors: Solène Debuysère¹, Nicolas Trouvé¹, Nathan Letheule¹, Elise Colin¹, Georgia Channing²

Affiliations:

1 ONERA – The French Aerospace Lab

2 Hugging Face

---

Satellite imagery has transformed the way we observe our planet. Most of the time, these images come from optical sensors, which capture the world in visible light, just like our eyes. But there is another way to observe the planet: Synthetic Aperture Radar (SAR). SAR uses microwaves instead of visible light and can capture images at any time of day, even through clouds or bad weather.

We curated raw Umbra SAR acquisitions to create the SARLO‑80 (Slant SAR Language Optic, 80 cm) dataset, a structured, high‑resolution multimodal resource optimized for AI and machine‑learning applications. By pairing SAR imagery with geometrically aligned optical data and natural‑language descriptions, it bridges radar and vision–language domains.

Before outlining the processing steps, it’s helpful to briefly recall how SAR differs from conventional optical sensing.

Dataset repo: https://huggingface.co/datasets/ONERA/SARLO-80

---

Optics vs Radar: Two Different Views of Earth

Optical and radar imaging provide two fundamentally different ways of observing the Earth’s surface. While optical imagery resembles natural photographs formed by visible light, Synthetic Aperture Radar (SAR) imagery is constructed from microwave echoes that interact with the physical and electromagnetic properties of the terrain. This difference affects every aspect of image acquisition, resolution, geometry, and interpretation.

1. Active and Passive Sensing

Unlike optical sensors that depend on sunlight and clear skies, SAR actively emits microwaves and can image the Earth even through clouds — a key advantage when over 60 % of the planet is covered by clouds at any given time.

Figure 1: Example of Capella Image where Sequoia satellite of Brazil demonstrates how our high‑resolution SAR (left) can provide a clear view of deforestation, even when clouds obscure optical images (right).

2. Image Formation Principles

An optical image is a direct projection of light through a lens onto a sensor array. Radar imagery, by contrast, is reconstructed computationally from a sequence of radar echoes collected as the satellite moves along its orbit. By combining measurements over time, the system synthesizes a large “virtual” antenna — the synthetic aperture — which enables fine spatial resolution (see Figure 3).

In optical systems, spatial resolution depends primarily on the aperture size of the lens. In radar systems, it depends instead on signal frequency, bandwidth, and the distance traveled by the sensor during data acquisition. This distinction allows SAR satellites to achieve high resolution even with relatively compact antennas. This resolution is encoded in the size of the bright points, with each point corresponding approximately to the smallest distinguishable feature the radar can resolve.

3. Radar Geometry and Distortions

Optical and radar sensors observe the Earth from fundamentally different geometries. Optical systems capture images in a ground‑projected plane (green plane in Figure 2), where each pixel corresponds directly to a point on the surface. In contrast, SAR acquires data in slant‑range geometry (orange plane in Figure 2), measuring distances along the radar’s line of sight. To make SAR and optical images geometrically comparable, one of them must be reprojected into the geometry of the other—or both into a common reference geometry—to achieve approximate geometric superposability, since perfect geometric superposition is physically impossible due to their distinct viewing geometries.

Figure 2: SAR geometry acquisition with slant‑range and ground‑range planes.

This oblique acquisition causes elevated terrain and tall structures to appear displaced toward the sensor, introducing geometric distortions such as:

• Layover – Tall structures (mountains, buildings) appear to lean toward the radar because their upper parts return signals before their bases.

• Foreshortening – Slopes facing the radar appear compressed because their top and bottom are illuminated almost simultaneously.

• Shadowing – Areas hidden from the radar beam appear dark or unmeasured.

Figure 3: Comparison of optical vs. SAR image formation and distortions.

These effects are inherent to radar imaging and carry useful information about surface topography and orientation.

Figure 4: Example of layover in Copenhagen.

Figure 5: Example of volcano foreshortening.

4. Coherence and Speckle Characteristics

SAR sensors record not only the amplitude of the backscattered signal but also its phase — the precise timing of the returned wave. This property makes radar data coherent, enabling advanced techniques such as polarimetry and interferometry (InSAR).

Coherence also produces a characteristic speckle pattern, visible as granular texture in SAR images. Speckle results from the constructive and destructive interference of radar signals scattered by multiple small targets within a single resolution cell. Although it may resemble noise, speckle is a deterministic phenomenon that contains information about the surface’s physical structure and scattering behavior.

5. Interpretation and Applications

Interpreting SAR imagery requires understanding that brightness corresponds to backscattering intensity rather than optical brightness or color. Highly reflective surfaces (e.g., rough terrain or metallic structures) appear bright, while smooth surfaces (e.g., calm water or flat soil) appear dark. Despite its more abstract appearance, SAR provides unique observational capabilities that complement optical data:

• Surface deformation monitoring using interferometry

• Mapping of soil moisture, vegetation, and ice dynamics

• Detection of infrastructure, ships, and flood extents

Together, optical and radar observations form a comprehensive view of the Earth — optical systems providing intuitive visual context, and radar systems revealing structural, dynamic, and geophysical properties invisible to the human eye.

---

Creating the Umbra Dataset

Figure 6: Worldwide map of Umbra data.

Open data source: https://umbra.space/open-data/

Although radar offers remarkable sensing capabilities, it remains challenging to process. To make this data more accessible, we curated and transformed the open‑source radar imagery collected by the Umbra satellite constellation into a machine‑learning‑ready format.

We started from around 2 500 Umbra SICD images acquired across the globe. These SAR scenes, captured in complex format and VV or HH polarization, span resolutions from 20 cm to 2 m and incidence angles between 10° and 70°. To standardize them, we refocused the spectrum and resampled all data to 80 cm × 80 cm in slant‑range geometry, then split each large scene into overlapping 1 024 × 1 024‑pixel patches.

To make the dataset multimodal, each SAR patch was paired with a high‑resolution optical counterpart (from Sentinel‑2, Landsat‑8, or commercial providers) that was geometrically co‑registered to the SAR geometry. Finally, we generated natural‑language descriptions for each pair using a combination of expert annotation and large‑language‑model‑assisted captioning, ensuring that the textual modality captures both visual and radar‑specific characteristics (e.g., layover, speckle, backscatter intensity).

The resulting SARLO‑80 dataset therefore contains:

• SAR patches (80 cm slant‑range, 1 024 × 1 024 px)

• Co‑registered optical patches (same spatial extent, comparable resolution)

• English captions describing scene content, radar phenomena, and notable features

All assets are released under a permissive license, enabling researchers to develop and benchmark models for SAR‑optical fusion, multimodal retrieval, and language‑grounded remote‑sensing tasks.