A heatmap is one of the most persuasive maps you can make, turning a cloud of points, crime reports, shop locations, or wildlife sightings, into a glowing surface that instantly shows where things cluster. It is also one of the easiest maps to get wrong, because two settings, the search radius and the weighting, decide whether the result reveals a real pattern or manufactures one. Done carefully in QGIS, a heatmap is a powerful exploratory tool; done carelessly, it is a colourful way to mislead yourself and your audience.

QGIS's heatmap renderer does not move your data; it draws a smooth surface over your points, where each point contributes a mound of influence that fades with distance, and overlapping mounds add up to peaks. The search radius sets how far each point's influence reaches, so a small radius shows tight, local clusters while a large radius smooths everything into broad regional trends. The weighting, if you use one, lets some points count more than others, for example weighting accident points by severity rather than treating every incident equally, which changes the story the map tells.

This is kernel density estimation in plain clothes, and understanding the mound idea prevents the commonest mistake, which is reading a heatmap as a count of events in a place. A bright spot means high density of influence, not a precise number, and the values depend entirely on the radius you chose. Two maps of the same data with different radii can look like different datasets, which is why you should always state your radius and explain it when you publish a heatmap, just as you would state the scale of any other map.

Load your point layer, open its symbology, and switch the renderer to Heatmap, then set the radius in map units, not pixels, if you want the result to stay consistent when you zoom. Start with a radius that matches the scale of the pattern you are investigating: a few hundred metres for neighbourhood patterns, several kilometres for regional ones, and experiment with two or three values to see how the pattern changes. Choose a colour ramp that is honest, with a transparent or pale colour for low values so empty areas do not glow, and avoid rainbow ramps that imply sharp boundaries where the data are smooth.

Set the maximum value manually rather than leaving it automatic, so that maps you compare share the same scale, and consider the weight field if your points carry a meaningful magnitude. Render, then check the result against the raw points: toggle the point layer on and off and ask whether each hotspot corresponds to a real cluster and whether any real cluster is missing. If a single isolated point creates a bright blob, your radius is too small or your maximum too low; if the whole map is one undifferentiated wash, your radius is too large.

A heatmap shows where points are dense, not why they are dense, and the why often matters more: a cluster of reported incidents may reflect where reporting is good, not where incidents are worst, and a cluster of shops may reflect zoning rather than demand. Always pair a heatmap with context, population, road networks, or survey effort, and consider normalising by the underlying population if you are mapping rates rather than raw counts. A map of disease cases that ignores population density will simply rediscover where people live.

For final products, a heatmap is best as an exploratory or presentation layer, supported by the underlying point data or by a statistical test if the claim matters. Export it with a legend, a stated radius, and a note about weighting, so readers can judge it fairly. Used that way, a heatmap earns its reputation: it turns a table of coordinates into an insight you can see at a glance, without pretending to be more precise than the data allow.