AI image models render landscapes, weather and geological texture convincingly but cannot draw an accurate map, correct rock strata or a real coastline, so geography should always come from cartographic data rather than generation.
Earth science splits cleanly between the photographic and the cartographic. A storm front, a canyon wall, a glacier β excellent. A map of anywhere real β hopeless, and hopeless in a way that is unusually easy to miss, because a wrong coastline still looks like a coastline.
Maps are the hard failure here. Generated maps have invented coastlines, borders that do not exist and place names that are not words. Never use one to show a real location. Geological cross-sections have the same problem with strata order β layers appear in sequences that no depositional history could produce. Use real cartographic data and published sections.
Geology and earth science headers.
Close-up of layered sedimentary rock face, horizontal bands in ochre, rust and pale grey, strong side lighting emphasising texture, sharp detail, natural colour, no annotations
Things to change
Weather, climate and atmospheric topics.
Dramatic storm front advancing over flat open landscape, towering cumulonimbus with defined anvil, dark base and brilliant sunlit top, wide horizon, high dynamic range, natural colour, no text
Things to change
Climate and cryosphere content.
Aerial photograph of a glacier winding between mountains, deep blue crevasse fields, meltwater channels on the surface, cold flat overcast light, sense of enormous scale, documentary style
Things to change
Hydrology and water-quality sections, and backgrounds generally.
Abstract close-up of moving water surface, refracted light forming shifting caustic patterns, deep teal and turquoise, high shutter speed freezing the motion, no horizon, no context, purely textural
Things to change