Geospatial, Hierarchical and Network Visualization

Introduction and Context

Data visualization becomes substantially more demanding when data are organized by location, containment, or relationships. A bar chart can usually treat observations as independent categories, but a map, tree, or network encodes structure that is not merely decorative. Geographic position, parent–child membership, adjacency, direction, and connectivity influence both what can be shown and what viewers are likely to infer.

Geospatial visualization represents phenomena located on or near the Earth. Hierarchical visualization represents nested structures such as organizational units, file systems, taxonomies, and product categories. Network visualization represents entities and relationships, including communication networks, supply chains, citation networks, and social interactions. These forms overlap: a transportation network is spatial and relational; a corporate ownership structure is hierarchical and network-like; a regional supply-chain map may combine all three.

At postgraduate level, the central design problem is not selecting a visually attractive chart. It is constructing a representation whose geometry, topology, aggregation, interaction, and semantics are appropriate to the analytical question. A technically precise visualization may still mislead if its projection exaggerates area, its hierarchy conceals cross-links, or its network layout suggests causality that the data do not establish.

Learning Objectives

By the end of this chapter, a learner should be able to:

1. Select and justify geographic coordinate systems and map projections.

2. Distinguish spatial reference, topology, geometry, and thematic encoding.

3. Design and critique choropleth, proportional-symbol, flow, and point-density maps.

4. Choose suitable tree layouts for hierarchy size, depth, and analytical purpose.

5. Model networks using nodes, edges, attributes, direction, weights, and temporal states.

6. Explain how layout algorithms and visual encodings affect network interpretation.

7. Integrate geospatial, hierarchical, and network views in an analytical workflow.

8. Identify uncertainty, ethical risks, accessibility issues, and governance requirements.

1. Geospatial Visualization: From Coordinates to Meaning

1.1 Spatial data concepts

A spatial dataset contains at least one geometry and an associated attribute table. Common geometries are:

• Point: a discrete location, such as a sensor or hospital.
• Line or polyline: a route, river, transmission line, or boundary.
• Polygon: an area such as a district, land parcel, or ecological zone.
• Raster: a regular grid of cells, commonly used for satellite imagery, elevation, temperature, and population surfaces.

A coordinate reference system (CRS) specifies how positions are expressed. A geographic CRS uses angular coordinates—typically longitude and latitude—on a reference ellipsoid. A projected CRS transforms the curved Earth onto a planar surface, usually expressing coordinates in linear units such as metres.

The distinction between geometry and topology is essential. Geometry concerns position, shape, distance, and area. Topology concerns relationships such as adjacency, containment, intersection, and connectivity. Two polygons may have slightly different coordinate boundaries but still share the same topological relationship. Topological validity matters in operations such as spatial joins, routing, and boundary aggregation.

1.2 Map projections and distortion

No flat map can preserve every property of a curved surface. A projection necessarily introduces distortion in one or more of area, shape, distance, direction, or scale. Projection selection should therefore begin with the task, not with visual habit.

• Equal-area projections preserve relative area and are appropriate for comparing population, land cover, or resource extent by region.
• Conformal projections preserve local angles and shapes, making them useful for navigation and some engineering applications, although area may be strongly distorted.
• Equidistant projections preserve distance from selected points or along selected lines, not all pairwise distances.
• Azimuthal or directional projections preserve direction from a selected center and can support air-route or communication analyses.
• Compromise projections balance several distortions for general reference maps but should not be treated as analytically exact.

A common error is to calculate distance or area directly from longitude and latitude as though degrees were uniform linear units. The length represented by one degree of longitude changes with latitude, while one degree of latitude is approximately—but not exactly—constant. For reliable measurement, use an appropriate projected CRS or geodesic calculations on the reference ellipsoid.

For a small region, a local projected CRS often minimizes distortion. For a global view, an equal-area projection may be preferable when comparing countries, while a compromise projection may be acceptable for contextual display. The map should state the projection when it materially affects interpretation.

1.3 Thematic mapping methods

A choropleth map shades enumeration units according to an attribute. It is appropriate when the attribute is meaningfully associated with the entire area or when the map communicates rates, proportions, or densities. Raw counts can be misleading because large or populous regions tend to appear important. A rate such as cases per 100,000 population may be more appropriate, provided the denominator is relevant and sufficiently reliable.

A proportional-symbol map places symbols at locations and scales them according to a quantitative value. It is suitable for counts associated with points or centroids. Symbol area, rather than radius, should represent magnitude. If a circle's radius is multiplied by a factor, its area increases approximately by the square of that factor; viewers may therefore perceive a larger change than intended if radius is mapped linearly.

A dot-density map represents counts through repeated dots. Each dot denotes a specified quantity, but dots are often randomly distributed within polygons and do not identify exact individual locations. This can communicate broad spatial distribution while protecting privacy.

A flow map displays movement between origins and destinations. Width may encode volume, while direction may be shown using arrows or tapered strokes. Overlapping flows can make interpretation difficult; filtering, bundling, small multiples, or origin–destination matrices may be required.

A cartogram intentionally changes geographic area to represent a variable such as population or electoral weight. Cartograms are useful when the analytical objective is comparison by the variable rather than geographic shape. They should be clearly labelled because familiar spatial recognition is reduced.

1.4 Classification, normalization, and uncertainty

Classifying continuous values into colour intervals changes the visual argument. Equal-interval classes are easy to explain but may produce empty or crowded categories. Quantile classes contain similar numbers of areas but can make very different values appear adjacent. Natural-break methods seek gaps in the distribution but may be unstable across time. Classification should reflect the question, distribution, and audience; it should not be selected solely to produce visually balanced colours.

Normalization is the process of relating a count to an appropriate denominator. A rate can be expressed as

\[

r = \frac{y}{n} \times k,

\]

where \(y\) is the observed count, \(n\) is the relevant population or exposure, and \(k\) is a reporting constant such as 1,000 or 100,000. This equation assumes that the denominator represents the population at risk and that the rate is meaningful for comparison. Small denominators may produce unstable rates. Confidence intervals, empirical-Bayes smoothing, or suppression rules may be needed when rates are used for decisions.

Spatial aggregation creates the modifiable areal unit problem (MAUP). Results may change when the same observations are grouped into different zones or when zone boundaries are redrawn. Aggregated maps can also generate the ecological fallacy: an association between regions does not prove the same association among individuals. Analysts should test alternative spatial scales and avoid individual-level claims from area-level data.

Spatial autocorrelation means that nearby observations may be more similar than distant observations. Standard statistical assumptions of independent observations may therefore fail. A visual cluster is a descriptive pattern, not by itself evidence of a causal process. Hotspot methods, spatial regression, and permutation tests may support inference, but they require explicit assumptions and appropriate spatial weights.

2. Hierarchical Visualization

2.1 Hierarchy as a data model

A hierarchy is an ordered structure in which each element may have a parent, except for one or more roots. In a strict tree, every non-root node has exactly one parent and there are no cycles. Real organizational and knowledge structures may be directed acyclic graphs, multiple-parent taxonomies, or general networks rather than trees.

Important attributes include depth, breadth, sibling order, node size, and aggregation values. A node's value may be directly measured or calculated as the sum of descendants. These meanings must not be confused. For example, revenue assigned to a department may differ from the sum of product revenues if shared costs or transfers are present.

2.2 Tree and space-filling layouts

A node-link tree displays parent–child edges explicitly. It supports tracing ancestry and is effective for small or moderately sized hierarchies. Radial trees use angular space to reduce vertical height, but deep trees can become crowded.

An indented tree uses indentation and alignment. It is highly readable for labels and interaction, especially in file systems or policy documents, but it consumes vertical space.

A treemap recursively partitions a rectangle among leaves or internal nodes. Area represents a quantitative value, colour may represent a second variable, and nesting is represented by enclosure. Treemaps scale well to many leaves but make precise area comparisons difficult, particularly for small or irregular rectangles. Squarified algorithms improve aspect ratios but do not make all comparisons perceptually equal.

A sunburst or icicle plot emphasizes depth through rings or bands. These layouts can reveal broad structure, but outer segments are difficult to compare because their angular position and radius vary. Hierarchical edge bundling can show relationships among leaves in a radial hierarchy, although excessive bundling may obscure individual links.

The appropriate layout depends on the task. Use node-link or indented designs for path tracing, treemaps for space utilization and composition, and sunburst or icicle layouts for depth and partition structure. A single display should not be expected to optimize all tasks.

2.3 Aggregation and interaction

Hierarchical visualization often relies on progressive disclosure. Overview, zoom, filter, search, and details-on-demand help users manage scale. However, interaction can conceal context. A user should be able to identify the current path, understand which values are aggregated, and return to an overview.

When a parent represents an aggregate, the visual encoding should make the aggregation rule clear. If categories overlap, a simple additive tree is invalid. A taxonomy in which an item belongs to multiple categories requires a DAG or a linked multi-membership view. Forcing such data into a tree duplicates values or arbitrarily assigns ownership.

3. Network Visualization

3.1 Network representation

A network is commonly represented as \(G=(V,E)\), where \(V\) is the set of nodes and \(E\) is the set of edges. A node may represent a person, server, place, or organization. An edge represents a relation such as communication, transaction, co-authorship, or physical connection.

Edges may be directed or undirected, weighted or unweighted, and static or temporal. A directed edge \((u,v)\) indicates an orientation from \(u\) to \(v\); it does not automatically establish causation. A weighted edge may represent frequency, capacity, strength, or cost, and the semantic meaning must be stated.

Common node measures include degree, in-degree, out-degree, and centrality. For an undirected graph, degree is the number of incident edges. For a directed graph, in-degree counts incoming edges and out-degree counts outgoing edges. Betweenness centrality measures the proportion of shortest paths that pass through a node:

\[

C_B(v)=\sum_{s\ne v\ne t}\frac{\sigma_{st}(v)}{\sigma_{st}},

\]

where \(\sigma_{st}\) is the number of shortest paths from node \(s\) to node \(t\), and \(\sigma_{st}(v)\) is the number of those paths passing through \(v\). The sum is over distinct source and target nodes, excluding \(v\). This measure assumes that shortest paths are analytically meaningful and that edge weights, if used, have an appropriate interpretation as distance or cost. If weights represent strength, they may need transformation before shortest-path analysis.

Other measures include closeness, eigenvector centrality, PageRank, density, modularity-based communities, and assortativity. Each answers a different question. High degree indicates many direct connections; high betweenness may indicate brokerage; neither necessarily implies power, trust, importance, or resilience.

3.2 Network layouts and visual encoding

Force-directed layouts model nodes as repelling objects and edges as attracting links. They are useful for exploratory analysis, but positions are algorithmic and can vary with initialization, parameters, and software. Spatial proximity in a force layout does not necessarily represent geographic proximity or stronger relation unless explicitly designed to do so.

Matrix views encode nodes on rows and columns and relationships in cells. They are often more scalable than node-link diagrams for dense networks and support identifying blocks, reciprocity, and missing links. Adjacency matrices are less intuitive for tracing paths, so coordinated node-link and matrix views can be valuable.

Geographic network maps preserve location but may produce severe overlap when many relations cross the same corridor. Hybrid methods—such as curved edges, edge bundling, aggregation, or separate schematic and geographic views—can address this tension.

Visual variables should have explicit meanings. Node size may encode degree, capacity, or population; colour may encode community or type; edge width may encode volume; opacity may encode confidence or recency. Redundant encoding can improve accessibility, but too many simultaneous variables increase cognitive load.

3.3 Dynamic and multilayer networks

A temporal network changes as nodes or edges appear, disappear, or change weight. Animation can show sequence but may hinder comparison because earlier states leave working memory. Small multiples, timelines, trails, and selectable time windows often support more reliable comparison.

A multilayer network contains different relation types or contexts, such as friendship, advice, and financial exchange. Combining layers into one undifferentiated edge set can create false conclusions. Layer controls, edge styles, separate panels, or formal multiplex models should preserve relation semantics.

4. Integrated Design Framework

A rigorous design process begins with the analytical question. Ask whether the user needs location, comparison, containment, path tracing, community detection, change over time, or anomaly identification. Then define the data model and uncertainty before selecting a layout.

A practical workflow is:

1. Specify entities and relations. Determine whether records are points, areas, nodes, edges, or nested categories.

2. Validate structure. Check CRS, invalid geometries, duplicate edges, cycles, orphan nodes, missing parents, and inconsistent identifiers.

3. Define analytical semantics. State what counts, rates, weights, colours, and centrality measures mean.

4. Choose an appropriate coordinate or layout system. Use projection for spatial measurement and a network algorithm for relational exploration.

5. Select scales and aggregation. Consider zoom level, hierarchy depth, network density, and temporal granularity.

6. Encode uncertainty and provenance. Show confidence, missingness, data date, source, and transformation rules.

7. Test interpretation. Conduct usability testing, sensitivity analysis, and alternative-layout comparisons.

8. Document limitations. Explain what the display cannot establish, especially causality, individual behaviour, and precise boundaries.

5. Illustrative Case: Emergency-Service Access and Hospital Transfers

Background

This illustrative case concerns a metropolitan health authority assessing emergency access and hospital transfer capacity. The authority has ambulance-call locations, district populations, hospital locations, road travel times, transfer records, and hospital service categories.

Problem

Senior managers want to know which districts have poor emergency access, which hospitals act as transfer hubs, and whether service categories are concentrated in a small number of facilities. A single map is insufficient because the problem combines spatial distribution, a network of transfers, and a hierarchy of health services.

Analysis and approach

First, the team uses a locally appropriate projected CRS for travel-distance calculations and retains the original geographic CRS for web display. District-level emergency counts are converted to rates using population at risk. Districts with small populations receive uncertainty warnings rather than being ranked solely by unstable rates.

A choropleth displays standardized emergency-call rates, while proportional symbols show hospital capacity. The team avoids shading raw counts because large districts would dominate. A separate flow map shows transfers between hospitals; edge width represents the number of transfers and arrows show direction. Because flows overlap in the central city, an origin–destination matrix is provided as a companion view.

The hospital service hierarchy is displayed as an indented tree: authority, hospital group, facility, department, and service. A hospital may provide multiple services, so service categories are represented as a linked multi-membership structure rather than being forced into a strict tree.

For the transfer network, betweenness centrality is calculated using travel time as edge cost. The team explicitly avoids interpreting high betweenness as clinical superiority. It indicates that a facility lies on many shortest modeled transfer routes. A sensitivity analysis compares results using travel time, transfer frequency, and capacity constraints.

Outcome

The integrated views reveal that one district with a high call rate also has relatively good hospital proximity, while a lower-rate peripheral district has poor access and limited specialist capacity. A central hospital is a transfer broker because of its location and service profile, not necessarily because it receives the most patients. Managers prioritize an additional specialist agreement and improved ambulance routing rather than simply expanding the highest-volume hospital.

Learning points

The case demonstrates that rates, projections, topology, and network metrics answer different questions. It also shows why linked views are preferable to one overloaded graphic. The conclusions remain conditional on data quality, travel-time assumptions, reporting completeness, and the chosen geographic units.

6. Limitations, Ethics, and Governance

Location data can reveal sensitive information about health, movement, residence, or vulnerable populations. Aggregation, spatial masking, minimum-count suppression, differential privacy, and access controls may be necessary. Masking can itself reduce analytical accuracy, so governance should document the privacy–utility trade-off.

Maps can stigmatize communities when colour scales imply danger or deficiency without socioeconomic context. Boundary choices may reinforce political narratives. Network diagrams can expose individuals or infer relationships that participants did not intend to disclose. Analysts should assess consent, purpose limitation, re-identification risk, and the consequences of publication.

Accessibility requires more than colour selection. Use patterns, labels, line styles, adequate contrast, keyboard-accessible interaction, textual alternatives, and screen-reader descriptions. Avoid red–green dependence and ensure that important distinctions remain visible in grayscale.

AI-assisted geospatial and network systems can classify imagery, infer missing links, recommend routes, or generate layouts. Such systems require validation against representative data, uncertainty reporting, provenance, and monitoring for geographic bias. An inferred relationship should not be displayed as an observed fact. Automated hotspot detection can amplify uneven reporting or historical enforcement patterns.

7. Contemporary Developments and Best Practices

Modern systems increasingly combine web maps, vector tiles, GPU rendering, spatial databases, graph databases, and interactive dashboards. Hexagonal binning, multiscale representations, and server-side aggregation improve performance for large point datasets. Knowledge graphs and graph neural networks extend network analysis, but their outputs still require interpretable visual summaries and careful validation.

Best practices include:

• Select projections according to analytical purpose and geographic extent.
• Use rates only with meaningful denominators and uncertainty assessment.
• Distinguish observed location from estimated or geocoded location.
• Treat spatial units as analytical choices, not natural facts.
• Use trees only when parentage is structurally valid.
• Label network edge direction, weight, time period, and relation type.
• Do not interpret force-layout distance as geographic or causal distance.
• Provide coordinated views for overview, comparison, and detail.
• Test alternative classifications, layouts, and aggregation levels.
• Record data lineage, transformation steps, software versions, and parameter settings.
• Design for accessibility, privacy, reproducibility, and responsible interpretation.

Chapter Summary

Geospatial, hierarchical, and network visualizations encode structure as well as values. Geospatial design requires attention to coordinate reference systems, projection distortion, topology, normalization, classification, spatial scale, and uncertainty. Choropleths, proportional symbols, flow maps, dot-density maps, and cartograms serve different analytical purposes.

Hierarchical displays depend on valid parent–child structure and should distinguish strict trees from DAGs and general networks. Node-link trees, indented trees, treemaps, sunbursts, and icicle plots support different tasks involving paths, composition, depth, or scale.

Network visualization requires explicit definitions of nodes, edges, direction, weight, time, and relation type. Layouts are analytical aids rather than neutral depictions. Centrality measures, including betweenness, are meaningful only under their assumptions and should not be equated automatically with influence or causality.

Reliable practice combines appropriate geometry or layout, transparent semantics, uncertainty communication, privacy protection, accessibility, sensitivity analysis, and documented provenance. Integrated views are often more effective than a single overloaded display.

Key Terms

Coordinate Reference System: A specification for expressing geographic positions.

Projection: A transformation of the curved Earth onto a plane.

Topology: Relationships such as adjacency, containment, and connectivity.

Choropleth Map: A map that shades areas according to an attribute.

Normalization: Relating a count to a relevant denominator.

MAUP: Variation in analytical results caused by spatial scale or zone definition.

Ecological Fallacy: Incorrectly inferring individual relationships from aggregate data.

Cartogram: A map that distorts area to represent a variable.

Hierarchy: An ordered structure of parent–child relationships.

Treemap: A space-filling display using nested rectangles.

Directed Acyclic Graph: A directed graph containing no cycles.

Network: A set of nodes connected by edges.

Degree: The number of edges incident to a node.

Betweenness Centrality: A measure based on shortest paths passing through a node.

Force-Directed Layout: A layout generated by simulated attraction and repulsion.

Multilayer Network: A network containing multiple relation types or contexts.

Spatial Autocorrelation: Dependence between observations based on spatial proximity.

Progressive Disclosure: Revealing detail incrementally through interaction.

Data Provenance: Documentation of data origins and transformations.

Review Questions

Short Answer Questions

1. Why can no planar map preserve area, shape, distance, and direction simultaneously?

2. When is a choropleth map preferable to a proportional-symbol map?

3. What distinguishes a strict tree from a directed acyclic graph?

4. What does betweenness centrality measure?

5. Why should force-directed layouts not automatically be interpreted geographically?

Long Answer Questions

1. Compare equal-area, conformal, equidistant, and compromise projections, including suitable applications and limitations.

2. Explain how classification, normalization, MAUP, and ecological fallacy affect choropleth interpretation.

3. Compare node-link trees, treemaps, sunbursts, and indented trees for different analytical tasks.

4. Discuss how direction, weight, temporal change, and multilayer structure should be represented in network visualizations.

5. Develop a complete design framework for an integrated spatial, hierarchical, and network dashboard.

Analytical / Application Questions

1. A map ranks districts by disease count, but the districts have very different populations. Redesign the analysis and explain how uncertainty should be communicated.

2. A company taxonomy allows products to belong to several categories. Explain why a strict tree is inappropriate and propose a suitable visualization.

3. A transport network identifies a station with high betweenness. Describe at least three alternative explanations for this result and design a sensitivity analysis.

Multiple Choice Questions with Answers

Question 1

Which projection property is most appropriate when comparing the relative areas of countries?

A. Equal-area

B. Conformal

C. Equidistant from every point

D. Perspective

Correct Answer: A. Equal-area

Question 2

A choropleth map most appropriately represents which measure?

A. A rate associated with geographic areas

B. Exact routes between facilities

C. Parent–child membership

D. Individual network degree

Correct Answer: A. A rate associated with geographic areas

Question 3

What is the main concern represented by MAUP?

A. Results may change with spatial units or scale

B. Maps cannot display categorical data

C. Networks must be undirected

D. Projection always preserves area

Correct Answer: A. Results may change with spatial units or scale

Question 4

In a strict tree, each non-root node has:

A. Exactly one parent

B. At least two parents

C. No descendants

D. A cycle to its root

Correct Answer: A. Exactly one parent

Question 5

Which visualization is generally well suited to showing hierarchical composition using area?

A. Treemap

B. Scatterplot

C. Flow map

D. Adjacency matrix

Correct Answer: A. Treemap

Question 6

In a directed network, out-degree counts:

A. Edges leaving a node

B. Edges entering a node

C. All shortest paths through a node

D. The number of connected components

Correct Answer: A. Edges leaving a node

Question 7

Betweenness centrality is based primarily on:

A. Shortest paths passing through a node

B. The geographic area of a node

C. The number of hierarchy levels

D. The colour assigned to an edge

Correct Answer: A. Shortest paths passing through a node

Question 8

Why can raw counts be misleading in a choropleth map?

A. They may reflect population or area differences rather than comparable intensity

B. They cannot be stored in databases

C. They always imply causation

D. They require a network layout

Correct Answer: A. They may reflect population or area differences rather than comparable intensity

Question 9

Which view is often effective for dense networks?

A. Adjacency matrix

B. Radial tree only

C. Cartogram only

D. Dot-density map only

Correct Answer: A. Adjacency matrix

Question 10

Which practice best supports responsible visualization of sensitive location data?

A. Aggregation or masking combined with privacy-risk assessment

B. Publishing exact coordinates by default

C. Removing all metadata

D. Treating inferred links as observed facts

Correct Answer: A. Aggregation or masking combined with privacy-risk assessment