Is a store's weekly sales table spatiotemporal?

Yes, if the store locations form part of the data's spatial meaning and the weeks supply the temporal dimension.

Must all observations have identical intervals?

No. Irregular data are possible, but methods must handle their timing appropriately.

Is spatial proximity enough to establish a relationship?

No. Nearby events can share influences or occur by chance; interpretation requires a suitable analysis and context.

AI Glossary / Data science and analytics / Intermediate

Spatiotemporal data

Definition

Spatiotemporal data describe phenomena with both a spatial component and a temporal component: where something occurs or is measured, and when. Examples include vehicle trajectories, weather measurements across locations and events recorded with positions and times. The observations need not involve objects that move.

What it means

A map answers questions about location, while a time series describes change over time. Spatiotemporal data connect those dimensions so that patterns can be examined together. A traffic problem may occur only on particular roads during certain hours; either dimension alone can hide the pattern.

The spatial component might be coordinates, a route, an area or a grid cell. The temporal component might be an instant, an interval or a repeated observation period. A stationary sensor recording temperature each hour therefore produces spatiotemporal data when its location is part of the measurement context.

Meaningful analysis requires more than columns named latitude and time. Coordinate references, time zones, measurement precision and the meaning of the recorded timestamp must be understood.

How it works

First define the observation unit. It might be one event, one location during one interval, or one position along an object's trajectory. Preserve identifiers when observations belong to the same object or sensor.

Align spatial references and temporal conventions before combining sources. Event time, reporting time and processing time can differ. A location may be approximate, and observations may arrive irregularly or contain gaps.

Analyses can aggregate events into space-time bins, follow trajectories, compare nearby observations or estimate changes across regions. The size of a spatial cell and length of a time interval affect what patterns are visible. A space-time cube is one representation, not a requirement for every dataset.

Nearby observations in space or time may be related. For prediction, evaluation should reflect the intended use: forecasting later periods, transferring to unseen locations, or both. A random row split can make performance look easier when closely related observations cross the split.

Three practical examples

A delivery team combines vehicle positions with event times to identify routes and periods where delays recur.
A facilities team analyses hourly readings from fixed room sensors to distinguish a local ventilation issue from a building-wide daily pattern.
A retailer studies demand by store and week, using the geographic and temporal context rather than assuming all stores share the same trend.

Think of it this way

Think of a flipbook of maps. Each page shows a different time, making it possible to see where a pattern appears, moves or disappears.

The analogy is useful but incomplete. Real observations may be irregular, uncertain or continuous rather than neatly spaced frames. A line drawn between two recorded positions also represents an interpolation assumption; it does not prove the exact path traveled between them.

Worked example

Avoiding double counting at boundaries

Consider four invented delivery events: A in the West zone at 09:05, B in the West zone at 09:20, C in the East zone at 09:20 and D in the East zone at 09:30.

Use two time bins with inclusive starts and exclusive ends: 09:00-09:30 and 09:30-10:00. The first bin contains two West events and one East event. The second contains one East event and no West events. D belongs only to the second bin because its timestamp equals that bin's start.

Across both zones, the counts are three and one. Across both periods, the counts are two per zone. The full space-time view retains a distribution that either total alone hides.

These counts do not establish event rates unless the relevant exposure is known. Twice as many deliveries, vehicles or observation hours could explain a larger count without indicating a higher underlying risk or demand intensity.

Illustrative case study

Fictional case: Northbank Logistics combines location records from two depots to analyze morning congestion. One source stores local timestamps and the other stores UTC. The initial map appears to show a delay pattern moving between depots.

The team checks timestamp meaning and aligns the observations to a common reference before comparing them. It also marks gaps in vehicle reporting instead of drawing every gap as a confirmed straight journey.

The revised analysis uses explicit time windows, location accuracy and exposure measures. Managers can now investigate a defined pattern without assuming that all apparent movement came from operations. The example is fictional and illustrates data alignment, not a claimed logistics improvement.

Common misconceptions and mistakes

Spatiotemporal data require moving objects. Measurements at fixed locations over time also qualify.
A timestamp always records when an event happened. It may record arrival, entry or processing instead.
Connecting recorded positions reveals the exact route. The path between observations may be unknown.
Changing the map grid or time interval changes only appearance. Aggregation choices can change the apparent pattern and its interpretation.

Business use

Spatiotemporal analysis supports routing, capacity planning, environmental monitoring and local demand forecasting. Begin with the operational question and the resolution needed to answer it. Extremely precise coordinates are not useful if the decision concerns broad areas, and coarse timestamps may be inadequate for a rapidly changing process.

Record missingness and uncertainty, and preserve the distinction between observations and estimates. Location histories can also reveal personal routines, so limit identifiable detail to what the legitimate task requires. A persuasive animated map should not conceal weak coverage or incompatible measurement conventions.

Check your understanding

Two vehicle records show the same coordinates ten minutes apart. Does that prove the vehicle remained stationary throughout the interval?

Answer: No. It establishes the recorded positions at those two times, subject to measurement accuracy. The vehicle might have stayed still or traveled and returned. Intermediate movement cannot be inferred with certainty from those observations alone.

Related terms

Time series
Trajectory
Aggregation
Spatial data
coordinate reference system
event time
spatial autocorrelation
space-time cube

Learn next

Read event time and spatial autocorrelation. Build a small fictional table of locations and times, define nonoverlapping time bins, and compare the full space-time counts with totals by location or time alone. State which patterns each summary loses.

Sources

Esri, Space-time cube terminology: https://pro.arcgis.com/en/pro-app/3.6/tool-reference/space-time-pattern-mining/space-time-cube-terminology.htm. Open Geospatial Consortium, Moving Features overview: https://ogcapi.ogc.org/movingfeatures/overview.html. OGC, Moving Features Access, trajectories and coordinate references: https://docs.ogc.org/is/16-120r3/16-120r3.html. OGC, Environmental Data Retrieval: https://docs.ogc.org/is/19-086r6/19-086r6.html. Event counts and Northbank Logistics are original illustrations.

Previous: Spatial verification Next: Speaker diarization