Encyclopedia of GIS

2017 Edition
| Editors: Shashi Shekhar, Hui Xiong, Xun Zhou

Patterns in Spatiotemporal Data

  • Hui Yang
  • Srinivasan Parthasarathy
Reference work entry
DOI: https://doi.org/10.1007/978-3-319-17885-1_966



Spatio-temporal data refer to data that are both spatial and time-varying in nature, for instance, the data concerning traffic flows on a highway during rush hours. Spatio-temporal data are also being abundantly produced in many scientific domains. Examples include the datasets in computational fluid dynamics that describe the evolutionary behavior of vortices in fluid flows, and the datasets in bioinformatics that study the folding pathways of proteins from an initially string-like 3D structure to their respective native 3D structure.

One important issue in analyzing spatio-temporal data is to characterize the spatial relationship among spatial entities and, more importantly, to define how such a relationship evolves or changes over time. In the traffic flow example, one might be interested in identifying and monitoring the automobiles that are following one another far too close. Such an issue is often summarized as finding interesting spatio-temporal patterns.

A spatio-temporal pattern characterizes the spatial relationship among a collection of spatial entities and the evolutionary behavior of such a relationship over time. As an example, Fig. 1 illustrates four types of spatial patterns, corresponding to four different types of spatial association. In this figure, each rectangle represents a spatial entity, and an edge indicates that the two involved entities hold a certain spatial relationship. For instance, an edge can mean that the Euclidean distance of the two entities is within a specified threshold. It can also mean that one entity is located to the left of the other. Or it can mean both of the above relations hold between the two entities. Assume that a collection of spatial entities E = (e0, e1, , ek) formed a star-like pattern (Fig. 1a) at time t1 and continued in this fashion until time t2. One can employ a spatio-temporal pattern in the form of (Star, E, t1, t2) to effectively model such an evolving process.
Patterns in Spatiotemporal Data, Fig. 1

Examples of spatial association patterns. (a) Star. (b) Clique. (c) Sequence. (d) minLink = 2

Due to the following reasons, spatio-temporal patterns are often multifaceted. Furthermore, the spatio-temporal characteristics captured by such patterns often vary from application to application (Morimoto 2001; Xiong et al. 2004; Yang et al. 2005):
  • Diversity of spatial relationship. For any pair of spatial entities, there exist a variety of spatial relations between them, such as directional relation, distance-based relation, and topological relation. Which of these relations should be captured in a spatiotemporal pattern is often specific to individual applications.

  • Complexity of temporal relationship. For instance, there exist 13 possible relations between two time intervals (Allen 1983). Again, it is often governed by the applications to decide what relations should be considered in the spatiotemporal patterns.

  • Representation of spatial entities: points or geometric objects?

  • Varying application-specific requirements. For instance, one application might require one to capture how the distances between entities change in time, whereas another application might be interested in investigating both the distance and relative directional arrangement between entities.

Note that evolving spatial clusters-collections of spatial entities that are similar to one another (e.g., entities within the same vicinity)-are another type of spatio-temporal patterns. The main difference between evolving spatial clusters and the above-described spatio-temporal patterns resides in the number of involved spatial entities. A spatial cluster often consists of much more spatial entities than a spatio-temporal pattern. Additionally, spatio-temporal patterns are more versatile in the sense that a variety of spatial and temporal relations can be considered simultaneously as needed, whereas spatial clusters are often concerned about only the distance-based relationship among entities.

Historical Background

The history of spatio-temporal association patterns is closely related to that of spatial association patterns, since the former is often derived by incorporating the temporal dimension into the latter. Spatial association patterns were first studied by Koperski and Han (1995). This early work focuses on extracting patterns specified in advance. Following this work, a considerable amount of work was conducted to detect spatial clusters (Ester et al. 2001). Such clusters mainly captured the spatial proximity among entities. Driven by the widespread location-based services at the turn of the century, researchers started to take a special interest in identifying spatial patterns that involve a smaller set of entities within a confined spatial neighborhood (Morimoto 2001). Such patterns were later termed as spatial collocation patterns (Huang et al. 2003). For instance, the collocation pattern (weather, airline schedule, Starbucks coffee shops) captures the phenomenon that the customers at Starbucks coffee shops tend to request weather information and airline schedules together through cellular phone. However, the research work up to this point often simplified spatial entities to point objects and mainly considered the Euclidean distance between objects. Recently, several studies were carried out to overcome such limitations (Xiong et al. 2004). In these studies, spatial entities are represented as geometric objects of different shape and size. In addition, the spatio-temporal patterns are capable to capture multiple spatial relations (e.g., both distance-based and directional relation). Consequently, the term spatial or spatio-temporal object association patterns were coined to emphasize such facts (Yang et al. 2005).

Another prominent development of spatio-temporal patterns analysis is that it has found more and more applications in scientific domains, such as astronomy, meteorology, biochemistry, and bioinformatics. This is in contrast to its earlier application mainly in geographic information systems.

Scientific Fundamentals

The process of identifying spatio-temporal patterns can be decomposed into three main phases. The first phase is data preprocessing. Main tasks in this phase include the following: (1) Determine the representation scheme of spatial entities: points or geometric objects? If it is the latter case, what geometric properties and domain specific attributes need to be considered? (2) Concretize the spatio-temporal patterns: what spatial and temporal relations should the patterns be modeling? (3) Identify and define the measurements that measure the “interestingness” of a pattern. For instance, support and prevalence have been proposed by Yang et al. to characterize the significance of a pattern (Yang et al. 2005). The second phase is to efficiently and effectively discover interesting spatio-temporal patterns. One main challenge is to achieve good scalability and performance in the presence of a large volume of data, which are often in the range of gigabytes and even terabytes. Efficient data structures and optimization strategies are often employed towards improving scalability and performance. The third and final phase is to evaluate the identified spatio-temporal patterns and put them into use. The nature and implementation of this phase is often application-specific.

In scientific domains, the discovery of spatio-temporal patterns often brings up new challenges. For instance, to discover spatio-temporal patterns of vortices in fluid flows, one needs to first detect and extract vortical objects at different time. This task by itself is still under intensive study currently. Readers are referred to Yang et al. (2005) for more details on a generalized framework for analyzing scientific spatio-temporal data. This framework is illustrated in Fig. 2.
Patterns in Spatiotemporal Data, Fig. 2

A generalized framework for analyzing spatiotemporal scientific data

Key Applications

Spatio-temporal association patterns have been used to address various issues in many domains. Below is a list of representative applications from different domains.

Traffic Management

Spatio-temporal association patterns can be used to identify and predict potential accidents by modeling automobiles within dangerous distance. Such patterns can also be used to redirect traffic flows, thereby avoiding potential traffic jam.

Behavior Tracking in Security Surveillance Systems

Surveillance systems track and record the behavior of human subjects aiming at identifying suspicious behaviors. One can use spatio-temporal patterns to model such behaviors by associating a person’s movement with objects in the surrounding area.


In astronomy, spatio-temporal patterns can be used to capture the evolution of interactions among astronomical objects in the vicinity by exploring the data accumulated in the past.

Transmissible Disease Control

To control and predict the spreading rate of transmissible diseases (e.g., SARS), one critical issue is to have a clear notion of how people in the infected areas regularly relate to each other and with people in the disease-free areas. Spatio-temporal association patterns can be applied to model such people-people interactions.

Computational Molecular Dynamics: Interaction and Evolution of Defects in Materials

It has been observed that multiple defects in materials often interact with each other. Such interactions eventually might lead to undesirable results, such as the amalgamation of small defects and the breakdown of large defects. Again, such behavior can be modeled and captured by identifying spatio-temporal association patterns of defects.

Computation Fluid Dynamics: Characterizing Vortical Flows

Vortices-swirling regions around a common center-in vortical flows can often produce undesirable effects, especially when such vortices interact with one another. For instance, vortices in the air flows surrounding an airplane can lead to audible noise and strong vibration. Therefore, designers often resort to computer simulations to study vortical flows around a certain model. Here one can use spatio-temporal patterns to characterize the evolving behavior of vortices at different locations of the model under study.

Bioinformatics: Protein Folding Trajectories Analysis

A protein folding trajectory describes the folding path of a protein from an initially string-like structure to its final native and often complex structure. Along this path, amino acids, the building blocks of a protein, interact with one another. Such interactions often result in a variety of folding events, such as nucleation and secondary structure formation. It has been demonstrated that spatio-temporal association patterns could be applied to address several issues: (1) summarizing a folding trajectory; (2) detecting and ordering folding events along a trajectory; and (3) identifying a consensus partial folding pathway across different trajectories of a protein Yang et al. (2007).

Future Directions

Discovering interesting and meaningful spatio-temporal association patterns is still a relatively new problem. Below are several potential research focuses related to this problem: (1) design scalable algorithms that can handle large volume of spatio-temporal datasets. Candidate solutions include the following: integrating efficient indexing schemes in the process and developing parallel or distributed algorithms; (2) implement effective approaches to incorporate domain-specific knowledge in the pattern discovering process; (3) utilize visualization techniques to facilitate an easier verification and a better understanding of the discovered spatio-temporal patterns; and (4) implement generalized software systems to discover spatio-temporal patterns similar application domains.


  1. Allen JF (1983) Maintaining knowledge about temporal intervals. Commun ACM 26(11):832–843MATHCrossRefGoogle Scholar
  2. Ester M, Kriegel HP, Sander J (2001) Algorithms and applications for spatial data mining. Geographic data mining and knowledge discovery, research monographs. In: GIS Chapter 7Google Scholar
  3. Huang Y, Xiong H, Shekhar S, Pei J (2003) Mining confident co-location rules without a support threshold. In: Proceedings of the 2003 ACM symposium on applied computing, Melbourne (SAC’03). ACM Press, pp 497–501Google Scholar
  4. Koperski K, Han J (1995) Discovery of spatial association rules in geographic information databases. In: Proceedings of the 4th international symposium on advances in spatial databases (SSD’95), Portland. Springer, pp. 47–66Google Scholar
  5. Morimoto Y (2001) Mining frequent neighboring class sets in spatial databases. In: Proceedings of the seventh ACM SIGKDD international conference on knowledge discovery and data mining, San Francisco. ACM Press, pp 353–358Google Scholar
  6. Xiong H, Shekhar S, Huang Y, Kumar V, Ma X, Yoo JS (2004) A framework for discovering co-location patterns in data sets with extended spatial objects. In: SIAM international conference on data mining (SDM), Portland, Apr 2004Google Scholar
  7. Yang H, Parthasarathy S, Mehta S (2005) A generalized framework for mining spatio-temporal patterns in scientific data. In: Proceeding of the eleventh ACM SIGKDD international conference on knowledge discovery in data mining (KDD’05). ACM Press, New York, pp 716–721Google Scholar
  8. Yang H, Parthasarathy S, Ucar D (2007) A spatio-temporal mining approach towards summarizing and analyzing protein folding trajectories. Algorithms Mol Biol 2(3)Google Scholar

Recommended Reading

  1. Mokbel MF, Ghanem TM, Aref WG. Spatio-temporal access methods. Technical report, Department of Computer Sciences, Purdue UniversityGoogle Scholar
  2. Neill DB, Moore AW, Sabhnani M, Daniel K (2005) Detection of emerging space-time clusters. In: Proceedings of SIGKDD 2005, Copenhagen, pp 218–227Google Scholar
  3. Rao CR, Suryawanshi S (1996) Statistical analysis of shape of objects based on landmark data. Proc Natl Acad Sci U S A 93(22):12132–12136MATHCrossRefGoogle Scholar

Copyright information

© Springer International Publishing AG 2017

Authors and Affiliations

  • Hui Yang
    • 1
  • Srinivasan Parthasarathy
    • 2
  1. 1.Department of Computer Science and EngineeringSan Francisco State UniversitySan Francisco, CAUSA
  2. 2.Department of Computer Science and EngineeringThe Ohio State UniversityColumbus, OHUSA