Spatio-Temporal Networks: Analyzing Change Across Time and Place

June 9, 2014
Share

Sharing is Caring

Advertorial

Overview

Organizations are generating powerful insights by analyzing change in spatio-temporal networks in many applications, such as weather risk analysis. The rapid growth in size, variety and update rate of spatio-temporal data is creating new challenges and opportunities to efficiently store, validate process and analyze spatio-temporal networks with large time-series data. These challenges and trends in big data analytics for spatio-temporal networks are reviewed here. This article describes a proof-of-concept application for historical tornado event risk analysis that is implemented using Pitney Bowes® Spectrum™ Technology Platform software for configuring and running batch job and real-time web services integrated with R Density based Spatial Clustering and SQL Server Spatial. Pitney Bowes Spectrum™ Spatial for Business Intelligence software is used to provide a web-based map viewer to analyze and visualize tornado event spatial clusters over time, as well as the magnitude of and property damage caused by tornado events through surface density mapping.

Challenges and Trends:  Big Bata Analytics for Spatio-Temporal Networks

Spatio-temporal networks encompass the spatial relationship among locations and the time dimension of the data that those locations represent. Analyzing change in spatio-temporal networks is important in many applications in commerce, transportation, electricity and gas distribution, telecommunication networks, air/water/land quality monitoring and weather risk analysis, among many others. The size, variety and update rate of spatio-temporal data is growing fast, and quality is often an issue, as it comes from databases, web application logs, industry specific transaction data and location aware devices like mobile phones and many kinds of sensors. Spatio-temporal data is often massive in size. It is not uncommon to see visualization systems of spatio-temporal data, such as map viewers, that focus on the spatial dimension and do not effectively address the temporal dimension. These new challenges need to be addressed to efficiently integrate, store, validate, process and analyze spatio-temporal networks with large time-series data.

Global data growth, in general, and spatio-temporal data growth, specifically, are characterized by variety, volume, velocity and veracity. A new complexity is caused by many new data types that are now being collected in addition to master and transactional data such as semi-structured data (e.g., email, electronic forms, etc.) , unstructured data (e.g., text, images, video, social media, etc.), as well as sensor and machine-generated data. The volume of data that companies and governments are collecting is growing rapidly. They must now deal with data sources that are hundreds of terabytes or more that need to be stored and analyzed. The rate/velocity at which data needs to be created, processed and analyzed, often in real-time, is increasing in many applications, such as financial market and traffic data. The new variety, great volume and rate at which data needs to be processed and analyzed is creating even greater data veracity/data quality challenges. Data needs to be cleaned and data quality needs to be implemented before confidence can be established in the results of any analytics.

Big Data analytics typically involves analytical workloads where data variety, data volume and/or data velocity play a part. A variety of technology platforms are used for solution implementation, depending on the nature of the data and the analytical requirements. These technologies include relational and analytical DBMS (SQL Server and data warehouse appliances, respectively), non-relational data management platforms (e.g., Hadoop platform), NoSQL data store (e.g., graph database, such as Neo4j) and Stream processing software.Stream processing software, is used to support the automatic analysis of events as they happen in real-time or near real-time to identify significant patterns in data streams and trigger action to respond to them. In this case, analysis of the data, often using predictive or statistical models, usually takes place before the data is stored. A Rules engine can be used to automate decision-making and action taking.

Location intelligence often plays an important role in organizing and using the big data.Location is often used to relate and join disparate data sources that share a spatial relationship. Location intelligence visualization can help identify patterns and trends by seeing and analyzing data in a map view with spatial analysis tools such as thematic maps and spatial statistics. Location intelligence can help find data needles in a data haystack by using spatial relationships to filter relevant data.

Temporal location intelligence is being applied in many applications using spatio-temporal clusters, simulation and visualization; change detection; map animation and movement tracking. Temporal data is being integrated into location intelligence databases so that structured queries can executed against both spatial and temporal attributes. Providing temporal operators that can be used with spatial data operators is needed to effectively analyze spatio-temporal data. Location intelligence software, such as the Spectrum Platform, can provide data processing, visualization and analysis tools for both the time and geographic dimension of data that helps expose important insights and provide actionable information.

Use Case: Historical Tornado Event Risk Analysis POC

The Spectrum™ Technology Platform is service-oriented architecture (SOA) platform with a server that supports modules for a broad set of capabilities for data quality, master data management, analytics, location intelligence and data integration. The Spectrum Enterprise Designer client tool gives users drag-and-drop capabilities (e.g., Generate Time Dimension,  Geocode U.S. Address, Get Travel Boundary, Query Spatial Data, etc.) onto a workflow designer to construct business-process data flows  in the form of batch jobs or web services.

A proof-of-concept application for the analysis of historical tornado event data was implemented using the Spectrum Platform to visualize on a map tornado event spatial clusters in relation to their magnitude and property damage caused for a user defined study area (drive time or radius around a specified address) and for a user-defined time period (Jan 1, 2000 to Jan 1, 2010 and for the months of March, April and May/tornado season). Using the application, users can compare tornado event results for different time periods (tornado season vs. not tornado season) and/or different study areas.

The first implementation step involved adding the required time dimensions to the historical tornado events in order to prepare the historical tornado event records for analysis by time. Spectrum data flows were created using Spectrum’s Generate Time Dimension and Record Joiner capabilities, to generate a time dimension table and add the required time dimensions to the tornado events in an SQL Server table. Spectrum’s Time Dimension stage provides accurate time-based calculations on historic data without using complex SQL calculations. For example, time dimensions enable you to analyze your data by workdays versus holidays, weekdays versus weekends, by fiscal periods or in this case by tornado seasons.

Figure 1: The Generate Time Dimension dataflow and stage options create a table with the required time dimensions (click for larger image)

Figure 2: The Generate Tornado Event Time Dimension data flow adds the required time dimensions to the tornado event table 

A batch job and real-time web service data flows were configured and integrated with R Density based Spatial Clustering and SQL Server Spatial in order to:

  • Select historical  tornado events from SQL Server that took place within specified dates, within  specified months (e.g., tornado season) and within a specified drive distance or radius around a specified address (e.g., study area)
  • Calculate spatial cluster for each tornado event using R Density based DBScan Spatial Clustering

Figure 3: The Historical Tornado Event Risk Analysis data flow finds tornado events with a specified study area for a specified time period and calculates the spatial cluster value for each tornado event (click for larger image)

The Spectrum data flows were created to accept the following attributes as input for analysis.

  • StartDate & EndDate:Time period of tornado events to analyze (e.g., 1985-01-01 to 2011-01-01)
  • Months:Months during the year of tornado events to analyze, such as  March, April, May to represent  tornado season
  • AddressLine1,City,StateProvince, Postal Code:U.S. address that represents the starting point of the study area to analyze
  • SearchDistance:Distance in miles that will be used to create a drive distance or  radius around the specified address
  • SearchType:Drive distance or  radius around the specified address

The data flow processing of the specified locations include:

  • Standardizing and geocoding the input address
  • Generating a drive distance or radius boundary around the geocoded address using the specified search type and distance
  • Selecting  historical  tornado events from SQL Server that took place within the specified dates, within the specified months and within the specified drive distance or radius around the specified address using temporal and spatial operators
  • Calculating the spatial cluster for each tornado event using R Density based DBScan Spatial Clustering

For this application, Spectrum Spatial for Business Intelligence is used for web mapping visualization and analysis of the tornado event spatial clusters and attribute output. The web mapping application is used to automatically:

  • Zoom to and display a map that includes the drive distance or radius boundary around the geocoded address ontop of Google Maps for context
  • Generate individual value thematic map of tornado events by the R Density based spatial cluster
  • Generate Surface Density Tornado Property Damage or Magnitude Heat Map
  • Generate Bar Chart to show Tornado Spatial Clusters by Property Damage
  • Display a table showing the attribute values for the mapped  tornado events
  • Provide inactive functionality to filter tornado events in the map, chart and table by tornado event attrbute, tornado map location, or bar chart spatial cluster selection

Figure 4: Tornado events within the study area are color coded by the R Density based spatial cluster and bar chart shows Tornado Spatial Clusters by Property Damage (click for larger image)

Figure 5: A  Surface Density Heat map of tornado property damage is shown below the tornado events (click for larger image)

 

Editor's Note: Jeremy will be presenting a session titled Spatio-Temporal Networks: Analyzing Change Across Time and Place on June 12, during the MapInfo User Conference in Memphis, TN. Visit mapinfo.com for more information.  

Share

Sharing is Caring

Explore more: Location Intelligence

Geospatial Newsletters

Keep up to date with the latest geospatial trends!

Sign up

Search DM

Get Directions Magazine delivered to you
Please enter a valid email address
Please let us know that you're not a robot by using reCAPTCHA.
Sorry, there was a problem submitting your sign up request. Please try again or email editors@directionsmag.com

Thank You! We'll email you to verify your address.

In order to complete the subscription process, simply check your inbox and click on the link in the email we have just sent you. If it is not there, please check your junk mail folder.

Thank you!

It looks like you're already subscribed.

If you still experience difficulties subscribing to our newsletters, please contact us at editors@directionsmag.com