Associate Professor, IIIT-B
There is a frequently used phrase, which says that “80% of all information contains some geo-reference.” Hahmann and Burghardt  have reformulated this phrase using evidence to a “60% assertion.” Nonetheless, the geo-referenced data remains to be relatively more ubiquitous than non-geo-referenced data. Consider the NISAR (NASAISRO Synthetic Aperture Radar) mission , which the two space organizations are going to jointly launch in 2021. The mission is expected to generate around 85 TB of data daily . The radar imaging satellite will be designed to use dual frequency for remote sensing for studying natural processes and estimating earth observations (e.g. biomass, surface deformation, soil moisture, etc.). The mission is estimated to gather about 140 PB in its 3-year mission.
Overall, the message is loud and clear that geospatial big data is going to stay and grow, more so. Hence, its analytics cannot be ignored. Li et al.  have discussed how visualization provides a class of methodologies to gain human insight to geospatial big data. Visualizations provide yet another channel for identifying outliers, building and verifying hypotheses, and most importantly, “seeing” dominant trends and patterns. Visualization as a domain has contributed towards techniques for summarization and exploration of data, as well as “computational steering”. Computational steering [29, 28] is the process of improving on computations by using intermediate data visualizations, which effectively influence the workflow. While computational steering is used in computational simulations, a more recent trend has been observed in the form of progressive visual analytics.
1.1 What is Visual Analytics and How is it Different from Visualizations?
Keim et al.  have defined visual analytics as a combination of conventional data mining and interactive visualizations in a data analytic workflow, for understanding, inferences, and decision-making support for big data. In all interactive applications today, visualizations have paved the way for visual analytics. This evolution has happened because a stand-alone visualization or a group of visualizations carefully assembled together is no longer sufficient for making sense of the big data. Visual analytics processes, as shown in Figure 1, could be non-deterministic, as well, and form a feedback loop which terminates only when “visualization requirements” are met.
Fig. 1: The visual analytics workflow, as defined by , has been modified to include (a) streaming data; and (b) data feed from data mining. Use of streaming data leads to incremental visual analytics  and use of data in (b) leads to progressive visual analytics [33, 4].
A more recent concept, “progressive visual analytics,” , entails the use of visualization to study the “meaningful partial results” given out during the execution of data analytic workflow. Fekete and Primet  have proposed progressive analytics to be a synonym for “progressive computations for data analytics,” with consideration for low latency. Including visualizations to support progressive analytics, Badam et al.  have differentiated between progressive and incremental visual analytics – where the latter caters to data streaming . Their prototype for “progressive visual analytics” combines both incremental as well as progressive visualization with algorithmic steering, i.e. iterative control over execution of a computational process .