Bringing Analytics to Time, Space Data
An experimental service unveiled by IBM Research looks to plumb the depths of mounting geospatial and temporal data generated by satellites, drones and sensor networks.
The cloud-based service called Physical Analytics Integrated Data Repository and Services, or PAIRS, Geoscope, seeks to glean insights from a growing roster of geospatial data sources. Once organized by the platform, users could apply their own analytical tools to historical and constantly refreshed data sets.
The IBM (NYSE: IBM) platform stems from collaboration with the E. & J. Gallo Winery in which the partners developed a precision irrigation system incorporating a cloud-based communications network along with sensors and actuators. The platform also combined satellite imagery with a model for estimating water loss to estimate future irrigation needs.
The result was the demonstration of drip irrigation technology that delivered a 26 percent increase in crop yield while reducing water usage by 22 percent.
PAIRS Geoscope incorporates lessons gleaned from the demonstration, including the reality that geospatial and temporal data sets are often too large to transfer for analysis when they are needed. Those delays are expected to grow as data sources like Internet of Things sensor networks generate an estimated 600 zettabytes of data annually by 2020.
The other problem is messy geospatial and temporal data sets collected in an array of complex formats.
IBM researchers propose reversing the situation through a service that allows users to bring analytics to geospatial and temporal data sets. The platform dispenses with conventional data acquisition and preparation tasks via access to a searchable catalog of geospatial and temporal data that is continuously updated.
The repository of historical data is growing at a rate of terabytes a day, IBM said this week. The heterogeneous data sets are indexed to speed queries and retrievals. According to an IBM blog post, the platform utilizes machine learning and artificial intelligence techniques “to make predictions based on a complex mix of parameters, models, and historical data.”
IBM said PAIRS Geoscope is currently in trial deployments with clients working in agriculture, finance, energy, manufacturing and meteorology. The service also is being used by analysts at the University of Chicago and University of Michigan.
Among the emerging applications is using geolocation data from satellites to improve long-term weather forecasts. Generally speaking, the more data points that can be plugged into forecasting models, the higher the resolution of those models. The result is more accurate forecasts that can translate into economic benefits.
IBM said one customer is tapping into satellite geolocation data that includes three-dimensional temperature, pressure and humidity profiles of the atmosphere. These data sets are blended with historical weather forecasts and machine learning techniques to improve weather forecasts extending out a month or more. The goal is a 30 percent improvement in forecast accuracy.