A tool for satellite image preprocessing and composition

The availability of remote sensing big data and cloud computing services provides new opportunities for the preprocessing, analysis, and visualization of satellite images. But accessing and using such data and services is challenging and usually requires considerable expertise in remote sensing and scripting languages like JavaScript or Python. To lower this barrier, CDE researchers have developed a user-friendly online tool that requires only little user input to produce ready-to-use image composites.

Use of Landsat and Sentinel data

The tool uses existing high-level remote sensing data products – Landsat “surface reflectance” and Sentinel “top of atmosphere” data – and performs a topographic correction to remove any differences in the reflectance that result from illumination conditions. In addition, it applies a user-defined image composition strategy that deals with noise (e.g. clouds, haze, or shadows) and maximizes the amount of “valid pixels” in the output composite image.

Topographic illumination correction

In mountainous areas, the varying illumination conditions resulting from slope, aspect, and the position of the sun lead to variations in the reflectance within the same land cover type. For example, the same forest type reflects differently on a sunlit slope than on a shaded slope. The tool reduces differences in solar irradiance related to slope and aspect and produces an image which approximates reflectance values that would have been recorded over a flat surface.

Image composition

The tool creates a new, composite image by selecting the “best version” of each pixel from a series of multiple images. This function enables users to create image composites tailored to their study focus (e.g. covering a specific season) without facing trade-offs in the selection of individual images from the point of view of clouds and seasonality. The tool selects the most suitable pixel for the image composite based on multiple criteria:

1. Comparison of pixel reflectances based on user-defined criteria (e.g. focus on high greenness or low greenness)
2. Distance to clouds and cloud shadows (pixels further away from clouds are prioritized in the composite as they show less noise)
3. Distance to the “target day” of the composite (pixels closer to the target day are prioritized in the composite to reduce noise related to phenology)
4. Weights for the different years if the composition period spans multiple years (e.g. focus on recent images when the aim is to analyse recent land cover change)

Users can adjust parameters related to criteria 1, 3, and 4 according to their needs. This enables them to produce tailor-made image composites quickly and efficiently.