Algeria's most fertile plain is being built over. This dashboard answers two questions a planner or an agricultural office has to act on: where is farmland turning into built-up land, and how fast, and where are this season's crops doing worse than their own normal.
Click a commune for its figures. Background: our own cloud-free Sentinel-2 composites (10 m, shown at 20 m).
The main findings, each marked with how far the evidence supports it, with a link to the section that shows the evidence.
Each 20 m cell of cropland or orchard is compared with its own 2019–2025 record (summers whose imagery is hazy are left out). Spring uses peak greenness (NDVI), which tracks rain-fed winter crops; summer uses canopy water content (NDMI), which tracks irrigated plots and orchards. A cell counts as stressed when 2026 is more than one standard deviation below its own normal; in an ordinary year about one cell in five is.
Ranked by the average of each commune's rank on the spring and on the summer stressed share (communes with at least 200 ha of cropland and orchards in the study area, so that percentages rest on enough pixels).
The same pipeline was run with a gradient-boosted tree model (LightGBM, per-pixel features) and a deep-learning model (U-Net, which also sees the shapes around each pixel). Where they agree, the result does not hinge on one modelling choice.
Three screening products built on the same maps: land cleared but not yet built (act before construction), the kind of development that is taking farmland (choose the right policy lever), and where conversion is most likely over the next five years.
Every site on the 2026 watchlist: farmland that was bare in both 2026 seasons and is not yet built-up, where a visit can still come before construction. Filter the list, show a site on the map (the commune's figures appear beside it; try the 2018 | 2026 comparison), and download the selection for a field visit.
Screening results, meant to point limited staff and budget to the right places first. They do not replace field checks.