Turn observations into maps

ALSAT-1B stores image data onboard and sends them to the ground during suitable communications opportunities. ASAL describes a processing chain with different product levels. A file on the ground is a necessary input, but its usefulness also depends on acquisition time, instrument settings, orbit information and the corrections already applied.

Radiometric correction addresses how recorded values relate to the instrument’s response, such as differences between detectors. Geometric correction addresses where observations belong on a map. These are distinct tasks. Correcting a stripe caused by detector response does not, by itself, align a road with its true coordinates. Conversely, aligning an image does not automatically remove sensor-response differences.

To compare two dates, first check that corresponding pixels refer to the same place. A small registration shift can make an unchanged field edge appear to move. Terrain and off-nadir views complicate this alignment. An orthorectified product uses geometric and terrain information to reduce displacement, but it still has uncertainty that should be appropriate for the intended map.

Our two contextual images show northern Algeria’s fires through Aqua/MODIS on 10 August 2021 and Sentinel-2 on 12 August. They are not Alsat acquisitions. Different dates, instruments, frames and processing prevent a controlled before-and-after comparison. NASA’s view emphasizes a regional smoke plume; ESA’s labelled view shows terrain and an annotated burn scar. Smoke drifting above land is not the same boundary as burned ground.

Keep two kinds of correction distinct

Imagine a fictional map labelling 100 test locations as burned. Independent reference checks agree at 85 of those locations. The 85 percent agreement among these positive predictions is useful, but it does not tell us how many actual burned locations the map missed. To investigate omissions, we also need reference locations outside the predicted burned class.

A complete assessment describes how reference samples were chosen, their dates and their uncertainty. Smoke-covered or cloudy locations should not silently be counted as unchanged land. The finished map should state its observation dates, projection, processing and limits. A visually persuasive image becomes a defensible mapping product through traceable processing and evidence of accuracy, not through colour alone.