Satellites face a basic trade-off: one sees Earth often but roughly, another sees detail but less often. This study asks whether their views can be combined into a clearer picture of a changing planet. Satellite images come from multiple sensors because one sensor cannot overcome every limitation on its own.

Combining their views can help sustain Earth observation across different applications. The study presents a public dataset built from Sentinel-two and Sentinel-three. It covers diverse landscapes, ecosystems, weather conditions, and changes over time and space.

The dataset is publicly accessible and is used to compare widely used methods for combining satellite images, so researchers can work toward more robust results. The field has made progress in combining satellite data, but available methods have not been examined thoroughly using shared benchmark datasets.

Comparisons often use different datasets and specific sensors, making consistent evaluation difficult when locations, landscapes, situations, and newer sensors vary. Few public datasets are suitable for data fusion, making consistent comparison difficult and leaving researchers without a common testing ground.

The dataset combines images from two European satellite systems with different timing, detail, and colour information. In everyday terms, it is like matching a frequent wide-angle snapshot with a less frequent close-up. It is large in both the detail of its images and the number of paired images, and it keeps a large difference in detail between the two sources.

It also includes different ecosystems, landscapes, weather conditions, and changes over time and space, then compares five commonly used image-combining methods. The benchmark covers 15 sites across different ecosystems and geographic locations, with 689 pairs of images from the two satellite systems.

The sites were chosen to challenge image-combining methods with rapid environmental changes, urban expansion, and changing coastal ecosystems. Each pair joins a broader image with 16 information channels at 300 metres of detail to a sharper image with 10 channels at details ranging from 10 to 60 metres.

The study tests representative image-combining methods on the benchmark dataset. The goal is to examine current performance when using these satellite images, while creating a foundation for future work. The comparison measures how closely each predicted image matches the real image across every information channel used in the test.

All the methods can combine different channels from both satellite systems, even when the difference in detail between the sources is large. One method produced results closer to the best possible values than the others across all six channels, while another also produced results comparable to it.

Across five dates at each site, four measures compare predicted images with the real images, revealing that performance changes with both location and timing rather than staying consistent across methods. The differences are most noticeable where the landscape changes more sharply over time.

When landscapes are fairly uniform and change slowly, the five methods show no significant differences. But their performance separates in more varied landscapes and places changing more quickly. The method called ESTARFM is more suitable for varied landscapes and sudden events when preserving structural detail matters, even though it does not always have the lowest errors.

When matching pairs of broad and sharp images are not always available, SFSDAF or STARFM still gives competitive results and can work with a minimum number of image pairs. The benchmark pairs images from Sentinel-two and Sentinel-three, two European satellites with varied temporal, spatial, and spectral characteristics across the dataset.

It spans distinct sites in different regions, with diverse spatial features and a wide range of spectral bands for testing image-combining methods. Its varied sites provide a difficult test of whether these methods can handle different land types and changes over time.

The study also compares the performance of five of the most widely used spatiotemporal fusion methods discussed in the literature. The results reveal the central limitation: all the methods struggle to capture fine spatial details in highly varied landscapes and sudden changes such as forest fires.

The new public benchmark makes satellite-image fusion easier to test fairly across many landscapes and changing conditions. It also shows that sudden events and complex places remain difficult to reconstruct reliably.