1  Thematic Map Types

1.1 Overview

The thematic map type follows from the data structure, scale of measurement, and the visual variable best suited to encode the attribute:

1. Is the data a raster/continuous field or a set of discrete objects?
2. What is the scale of measurement (nominal, ordinal, interval/ratio)?
3. Which visual variable (in Bertin’s sense) is appropriate — and does this lead to a normalized or non-normalized representation?

All map type decisions ultimately reduce to the question of which visual variable is semantically appropriate for the data (Bertin, 1967):

Visual variable	Selective	Ordered	Quantitative	Appropriate for
Hue	Yes	No	No	Nominal categories
Lightness (value)	Yes	Yes	Partially	Ordinal, ratio
Saturation	Yes	Yes	Partially	Ordinal, ratio
Size	Yes	Yes	Yes	Ordinal, ratio (counts)
Shape (form)	Yes	No	No	Nominal
Orientation	Yes	No	No	Nominal (limited)
Texture	Yes	No	No	Nominal (limited)

hue alone should never encode order or quantity, as it has no intrinsic perceptual ordering. In sequential colormaps (e.g., viridis, YlOrRd), hue variation is acceptable because it reinforces a monotonic lightness gradient rather than substituting for it. The common error is using rainbow or jet color schemes, where hue varies without a monotonic lightness profile — creating false perceptual boundaries and implying structure that does not exist in the data (Borland & Taylor, 2007).

1.2 Decision tree

1.3 Continuous (Raster) vs. Discrete (Vector) Data

The first decision in the framework — raster field vs. discrete objects — determines more than just the data model. It has a direct visual consequence:

• Raster maps render as a continuous color surface. There are no polygon boundaries. The visual impression is of a smooth or pixelated field, and the reader’s eye interprets gradients and regions without hard edges.
• Discrete (vector) maps show explicitly bounded objects: polygon edges are visible, point symbols are distinct, and line features have defined geometry. The boundary is part of the visual encoding.

This distinction matters even when the color encoding logic is identical. A classified raster (ordinal, sequential lightness) and a shade map (ordinal polygon fill, sequential lightness) use the same visual variable — but they look fundamentally different because one shows boundaries and the other does not.

1.4 Map types

1.4.1 Raster path

For all raster map types, the visual output is a continuous color surface without polygon boundaries. The classification step (choosing break points and color scheme) is an explicit methodological decision with significant influence on the map’s message.

1.4.1.1 Classified map — categorical raster

A raster in which each cell belongs to a nominal class, encoded by a distinct hue. No order is implied between classes.

Examples: land cover classification (forest, urban, water, agriculture), soil type maps, habitat maps derived from remote sensing.

Note

Note: The visual encoding is analogous to a categorical (chorochromatic) map (polygon path, nominal data), but the absence of polygon boundaries gives it a distinctly different visual character.

1.4.1.2 Classified map — ordinal raster

A raster in which cells are assigned to ordered classes, encoded by sequential lightness steps (with optional hue shift).

Examples: slope steepness classes, earthquake hazard zones, classified NDVI (vegetation index) categories.

1.4.1.3 Heatmap / isoline map

Both map types visualize a continuous interval/ratio field — a variable that exists at every point in space without natural discrete boundaries.

• Isoline maps connect points of equal value with lines (isolines, contours). Classic examples include topographic contour maps and meteorological temperature maps (Slocum et al., 2009).
• Heatmaps render the field as a smooth color surface using a sequential or diverging color ramp to encode intensity.

Important

Important distinction: A heatmap can arise from two fundamentally different data pipelines that converge on the same visual form:

1. Direct field measurement: A ratio-scale variable measured continuously in space (e.g., air temperature from weather stations, interpolated to a raster surface). This is a true raster/field dataset.
2. Kernel density estimation (KDE) from point events: Nominal/binary point observations (e.g., accident locations, disease cases) are aggregated into a continuous density surface. The original data is a dot map — a discrete, nominal dataset. KDE transforms it into a derived raster with interval/ratio character, which then follows the raster path in the decision tree.

The term “heatmap” is used loosely in practice and often conflates these two cases.

1.4.2 Discrete path — nominal data

1.4.2.1 Dot map and connection map

Both map types use form and hue as visual variables to encode nominal attributes.

• Dot map (one-to-one): Each point symbol represents a single discrete object or event. The attribute is binary/nominal (present or absent). Examples: school locations, accident sites, bird observation records.
• Connection map (link map): Lines of uniform width connect pairs of locations to represent a nominal relationship. Direction may be indicated with an arrow but carries no quantitative meaning. Examples: airline routes, trade partnerships.

Note

Terminological note: The distinction between dot map and connection map is not always made explicit in textbooks. Some authors subsume both under “point symbol maps” (Kraak & Ormeling, 2020). The key differentiator is the geometry in the data: point features → dot map; line features connecting two locations → connection map.

Note

From dot map to heatmap via KDE: A dot map (nominal, one-to-one) can be transformed into a continuous density surface through kernel density estimation. The result is a derived raster dataset with interval/ratio character — which then follows the raster path in the decision tree, leading to a heatmap or isoline map. This is a deliberate analytical step, not a direct map type choice from the original data.

1.4.2.2 Categorical (Chorochromatic) Map

Polygon areas are filled with categorically distinct hues to indicate class membership. No order is implied between classes. Visual variable: hue (qualitatively distinct colors).

Examples: land use type, language region, dominant political party by district.

Note

Terminological note: The term chorochromatic is established in academic cartography (Kraak & Ormeling, 2020; Slocum et al., 2009) but rarely used in applied GIS contexts, where “categorical map” or “qualitative choropleth” is more common. The term “choropleth” should technically be reserved for normalized ratio data (see below); applying it to categorical data conflates two conceptually distinct map types.

1.4.3 Discrete path — ordinal data

1.4.3.1 Shade map

Polygon areas are filled with a sequential color scheme using discrete steps of lightness (with optional hue shift) to encode ranked categories. No arithmetic operations on the underlying values are meaningful — only rank order.

Examples: earthquake intensity zones (Mercalli scale), air quality index categories, education level by district.

Note

Relationship to choropleth: A classified choropleth map (ratio data, normalized, grouped into discrete classes) produces a visually identical result to a shade map: bounded polygons filled with sequential lightness steps. The difference is in the data origin: a shade map operates on data that is inherently ordinal, while a classified choropleth is a ratio variable that has been deliberately discretized. The visible polygon boundaries in both cases distinguish them from classified raster maps, which use the same color logic but show no boundaries.

Note

Terminological note: Several sources classify ordinal area-fill maps simply as “choropleth maps” (Dent et al., 2009; LibreTexts Geosciences, 2025). The stricter definition restricts choropleth to normalized ratio-scale data (Slocum et al., 2009). The term shade map is used here as a neutral label for the ordinal case.

1.4.4 Discrete path — interval/ratio data

1.4.4.1 Choropleth map

The choropleth map is the most widely used thematic map type. Polygon areas are shaded according to a normalized ratio-scale variable using a sequential (or diverging) color scheme, where lightness is the primary visual variable, typically reinforced by a secondary hue shift.

Normalization is essential. Mapping raw counts (e.g., total population per canton) to area fill creates a size bias: larger polygons appear more prominent regardless of the actual value. Normalization removes this by expressing the variable as a rate, density, or proportion (e.g., population per km², % unemployed).

Warning

Common error: Failing to normalize is one of the most frequent mistakes in choropleth mapping. Wikipedia contributors (2026) notes that during the COVID-19 pandemic, more than half of U.S. state government dashboards used non-normalized counts in choropleth maps, contributing to systematic misinterpretation.

Classed vs. unclassed: Most choropleth maps group values into discrete classes (classed choropleth), which produces a visual result identical to a shade map — the distinction lies entirely in the data. An unclassed choropleth assigns a continuous color proportional to each value, preserving more data detail but making it harder for readers to extract specific values (Axis Maps, 2020).

Color scheme: Sequential schemes are appropriate for variables with a natural direction. Diverging schemes are appropriate when values have a meaningful midpoint (e.g., deviation from average, population change).

Note

Terminological note: Some authors extend “choropleth” to include ordinal-level data (LibreTexts Geosciences, 2025). The stricter definition — ratio data, normalized to an enumeration unit — is preferred here and aligns with the original use of the term introduced by Wright in 1938 (Wikipedia contributors, 2026).

1.4.4.2 Proportional symbol map

A symbol is placed at or within each enumeration unit, where symbol size encodes the magnitude of a count or aggregate value. Unlike the choropleth, proportional symbol maps do not require normalization: absolute counts are valid because the symbol is geometrically independent of the polygon area.

Variants:

• Standard proportional symbol map: A single scaled symbol — typically a circle — where area ∝ value. Can be exactly proportional or range-graded (classed symbol sizes).
• Pictograph map (isotype map): Instead of a single scaled symbol, a number of identical icons is used, where count ∝ value (e.g., one person icon per 1 million inhabitants). The visual variable is still effectively size (total number of symbols), but the encoding is discrete and countable. This approach, rooted in Otto Neurath’s ISOTYPE system, is more accessible to general audiences but less precise for large value ranges.
• Pie chart map: A pie chart symbol replaces the simple circle, encoding the composition of a total at each location (e.g., vote shares by party). Useful when the part-whole relationship is the primary message, but increases cognitive load compared to a simple circle.
• Bar chart map: Bar charts placed at locations encode multiple variables simultaneously or show a distribution. Useful when comparing magnitudes rather than proportions.
• Flow map: Line width encodes the volume of flow between two locations (e.g., migration, trade, commuter flows). Direction is encoded by arrowheads. Conceptually a proportional symbol map applied to line geometry. Must be clearly distinguished from a connection map, where line width is uniform and the encoding is purely nominal (a connection exists or not).

Note

Pictograph maps and visual variables: In a pictograph map, the quantity is encoded by the count of symbols, not by any single symbol’s size, hue, or shape. This is arguably not one of Bertin’s graphical variables in the strict sense — similar to the density issue in dot density maps. The technique works perceptually because humans can subitize small counts and estimate larger ones, but precision decreases rapidly beyond ~10 symbols per location.

Note

Pie/bar chart maps: These are multivariate maps — useful when within-location composition is as important as between-location magnitude, but easily cluttered. They use size (of the whole symbol) plus angle or length (within the symbol) to encode a second variable dimension.

1.4.4.3 Cartogram

A cartogram distorts polygon area in proportion to a statistical variable, replacing geographic area with a data-driven area. Visual variable: size (of the geographic unit itself).

Cartograms are visually striking but cognitively demanding: readers must simultaneously recognize distorted geography and decode the data encoding. They work best when the geographic units are already well-known to the audience (e.g., countries, US states) (Monmonier, 2018).

1.5 Summary table

Map type	Data structure	Geometry	Scale	Visual variable	Boundaries visible	Normalization
Classified map (categorical)	Raster	Pixel	Nominal	Hue (categorical)	No	n/a
Classified map (ordinal)	Raster	Pixel	Ordinal	Lightness (+ hue)	No	n/a
Heatmap / isoline map	Raster	Pixel / line	I/R	Lightness (+ hue)	No	n/a
Dot map	Discrete	Point	Nominal	Form, hue	—	No
Connection map	Discrete	Line	Nominal	Form, hue	—	No
Categorical (Chorochromatic) map	Discrete	Polygon	Nominal	Hue (categorical)	Yes	No
Shade map	Discrete	Polygon	Ordinal	Lightness (+ hue)	Yes	No
Choropleth map	Discrete	Polygon	Ratio	Lightness (+ hue)	Yes	Required
Proportional symbol map	Discrete	Point/area	Ratio	Size	—	No
Pictograph map	Discrete	Point/area	Ratio	Count of symbols	—	No
Pie / bar chart map	Discrete	Point/area	Ratio (multivariate)	Size + angle/length	—	No
Flow map	Discrete	Line	Ratio	Size (line width)	—	No
Cartogram	Discrete	Polygon	Ratio	Area (distorted)	Yes	No

Axis Maps. (2020). Choropleth maps. https://www.axismaps.com/guide/choropleth

Bertin, J. (1967). Sémiologie graphique: Les diagrammes, les réseaux, les cartes. Gauthier-Villars.

Borland, D., & Taylor, R. M. (2007). Rainbow color map (still) considered harmful. IEEE Computer Graphics and Applications, 27(2), 14–17. https://doi.org/10.1109/MCG.2007.323435

Dent, B. D., Torguson, J. S., & Hodler, T. W. (2009). Cartography: Thematic map design (6th ed.). McGraw-Hill.

Kraak, M.-J., & Ormeling, F. (2020). Cartography: Visualization of geospatial data (4th ed.). CRC Press.

LibreTexts Geosciences. (2025). Map types. https://geo.libretexts.org/Bookshelves/Geography_(Physical)/GIS_Commons/06:_Output/6.03:_Map_Types

Monmonier, M. (2018). How to lie with maps (3rd ed.). University of Chicago Press.

Slocum, T. A., McMaster, R. B., Kessler, F. C., & Howard, H. H. (2009). Thematic cartography and geovisualization (3rd ed.). Pearson Prentice Hall.

Wikipedia contributors. (2026). Choropleth map. https://en.wikipedia.org/wiki/Choropleth_map