What Are Heat Maps?

Heat maps are data visualizations in which the value of a variable is encoded as color across a regular grid or a continuous surface, so that magnitude can be read from hue or lightness rather than from position or length. The display trades the precision of a scatterplot or bar chart for density: a single figure can present thousands of values at once, and the eye picks out blocks, gradients, and outliers that would be invisible in a table. Two broad families share the name. Matrix heat maps shade the cells of a data matrix, with rows and columns representing observations and variables. Spatial heat maps shade a geographic or image plane, usually after smoothing scattered point measurements into a continuous density field.

The form has a long lineage in statistical graphics. Shaded matrix displays appear in nineteenth-century tables and were formalized through twentieth-century work on matrix permutation and clustering, a history traced in Leland Wilkinson and Michael Friendly's account of the cluster heat map in The American Statistician.

Color Encoding and Perceptual Scales

The color map is the part of a heat map that carries the data, so its perceptual properties determine whether the figure is readable or misleading. A sequential map should be monotonic in perceived lightness, so that ordering in the data survives translation into color, and it should be approximately uniform, so that equal steps in value produce equal perceived steps in color. Rainbow maps such as the classic jet scale fail both tests: they introduce false boundaries at yellow and cyan, compress real differences elsewhere, and become unreadable in grayscale or to viewers with color vision deficiency. Analyses of the misuse of color in science communication document how often these artifacts change the apparent structure of published figures, and they motivated the perceptually uniform maps now standard in scientific plotting libraries. Diverging maps, with a neutral midpoint and two hues, are the correct choice when the data have a meaningful zero such as a correlation or a difference from baseline.

Matrix Reordering and Cluster Heat Maps

A matrix heat map is only as informative as the order of its rows and columns, because permuting them changes nothing about the data but everything about the visible pattern. The cluster heat map solves this by attaching hierarchical clustering dendrograms to both margins and ordering rows and columns so that similar profiles are adjacent, which turns diffuse noise into visible blocks. This construction became the standard display for gene expression microarray and sequencing studies and is now routine for correlation matrices, confusion matrices, and adjacency matrices of networks. Scaling choices matter as much as ordering: row-wise standardization highlights relative patterns across samples, while raw values preserve absolute magnitude, and the two can support opposite conclusions from the same matrix.

Spatial and Density Heat Maps

Spatial heat maps convert scattered events into a continuous field, most often through kernel density estimation, where each observation contributes a smooth kernel and the bandwidth controls how much detail survives. Choosing a bandwidth is the central judgment: too narrow and the map shows sampling noise, too wide and genuine structure is smoothed away. Applied guides such as the Columbia Mailman School's overview of heat maps and quilt plots describe how these displays are used in epidemiology to summarize incidence over space and time. Related variants include eye-tracking and click maps in interface research, thermal and radiometric images from infrared sensors, and coverage maps of received signal strength in wireless network planning.

Applications

Heat maps have applications across a wide range of disciplines, including:

• Genomics and bioinformatics, for gene expression and sequence similarity matrices
• Epidemiology and public health surveillance
• Machine learning diagnostics, including confusion matrices and attention maps
• Web and interface analytics, for gaze and interaction tracking
• Wireless network planning and radio coverage assessment
• Integrated circuit design, for on-chip thermal and congestion analysis