Pascale Proulx, Oculus Info, pproulx@oculusinfo.com
Susan Li, Oculus Info, sli@oculusinfo.com
Rob Harper, Oculus Info, rharper@oculusinfo.com
GeoTime: GeoTime supports the visualization and analysis of entities and events over time and geography within a 3D space. Entity movements, event, relationships, and interactions over time within a spatial context can be easily seen and understood. GeoTime includes event animation, keyword search, link analysis, imagery display, geometry display, annotation and numerous other analytical functions. GeoTime supports the extension Configurable Spaces that extends GeoTime's X, Y, T coordinate space for temporal analysis to any arbitrary diagrammatic workspace by replacing a geographic map with a diagram. [See Kapler, Thomas, and William Wright, 2004. GeoTime Information Visualization, IEEE Symposium on Information Visualization.]
nSpace2: nSpace2 is the web version of nSpace, an environment supporting the whole analytical workflow from brainstorming, creating hypotheses, to querying, scanning, comparing, reading and annotating, evidence marshaling and reasoning, to evidence assessment, collaboration and reporting. It has two main components: TRIST, focused on information triage, and the Sandbox, for evidence marshaling and analytical sense-making. nSpace2 (still in beta) is revolutionary in terms of what it can do on the web; however, it currently has only a subset of the capabilities that its parent nSpace has. Nevertheless, its core initial capabilities and in particular its strength in supporting multiple analysts working on related projects were definitively key to this team's analytical process. [See Wright, William, D. Schroh, P. Proulx, A. Skaburskis and B. Cort, The Sandbox for Analysis - Concepts and Methods, paper accepted for ACM CHI 2006.]
Excel Visualizer: Oculus Excel Visualizer is an Oculus Microsoft Excel extension designed to give users immediate understanding of the data that drives their business intelligence. By leveraging the ubiquity, power and ease-of-use provided by Excel spreadsheets, Oculus has created a new paradigm in rapid data visualization. Users can now take advantage of our integrated charting capabilities to provide new views on data to provide further insight and comprehension.
The migrant records were imported into GeoTime using the GeoTime/Excel plugin. Records were mapped to events in GeoTime as either an interdiction or landing. The number of deaths was mapped to the quantity attribute of each event, which is represented as the event indicator size in GeoTime. The map image of Isla Del Sueno was georeferenced and imported into GeoTime's map cache. The result was a seamless merge of the Island image in the appropriate location on the default world map in GeoTime.
To get a high level picture of the choice of landing sites over the three years, only the landing events were made visible using simple visual filters. The Calendar View in GeoTime was selected to make it easier to focus on patterns over time. Some observations could be made right away (see Figure 1): the triangular shape that the landing events form in the Calendar View reveals that migrants are expanding their range of landing sites over time, landing further and further from Isla Del Sueno; also of interest are the noticeable gaps that roughly correspond to the winter months in North America. These observations were immediately captured in the Sandbox as annotations on top of a GeoTime report (a snapshot of the view of interest). These initial observations prompted more questions and hypotheses that were also captured in the Sandbox..
 |
|
|
Figure 1: Landings over time and space in GeoTime. The blue dots represent landing events, here displayed in the Calendar View. The GeoTime report (annotated snapshot of view of interest for analysis) was imported in a Sandbox where related observations, questions and thoughts were also captured. The empty assertion boxes and explicit 'TO DO' notes will act as visual prompts as to next steps throughout the working sessions. | |
To quantify the geographical distribution of the landings, events were aggregated by regions. This was accomplished by using the drawing tools to outline the regions and then associating the events in those regions with regional places; an association being done in one drag and drop gesture. Looking at the relative length of the charts in Figure 2, it looks like a considerable proportion of the landings occur in Mexico (43.3 %). The trends in time of landings are shown in Figures 3-5.
The choice of vessel was also analyzed to check for any potential correlation with the choice of landing site. Figures 3-5 shows that there are just a few rafts that get to land higher north in Florida. Distance may not be the only factor because many rafts also landed further in Mexico. This line of inquiry was not pursued any further.
 |
|
|
Figure 2: Quantification of landing events by region using aggregation charts. The colors correspond to the vessel type. Looking at the relative length of the charts, it looks like a considerable proportion of the landings occur in Mexico. The exact proportion was calculated using the information in the tooltip over a chart and the numbers on the histogram in the charts panel (e.g. 191 events in Mexico / (287 + 88 + 66) = 43.3 % of landings in Mexico). Another observation that can be made from this visualization is that more deaths seem to occur on Go Fast boats (see big green dots). A quick calculation revealed that the Go Fast has the highest portion of deaths, but this may be because it can carry more passengers than other vessel types. | |
 |
 |
 |
Figure 3: In 2005, almost all landings are on the Florida Keys (89.1%). |
Figure 4: In 2006, the Keys still receive the largest proportion of the landings (40.3%), but Mexico is a notable second (31.8%), and the rest land on the west coast of Florida (27.9%). |
Figure 5: In 2007, the proportion of landings on the Keys goes down dramatically (8.3%) and Mexico becomes the main landing site (56.6%). Unlike the previous years, some migrants (14.3%) also land on the east coast of Florida. |
In order to understand and analyze why migrants were choosing particular landing sites, it seemed crucial to simultaneously consider the location of interdiction events over time. Thus with all records visible and colour applied to distinguish the landings (blue) from the interdictions (red), the time slider was moved and a cat and mouse game started to emerge. Snapshots for each year can be seen in Figures 6-8 below.
The apex of the triangle in Figure 1 and the curved line formed by the blue dots in Figure 6 shows that in 2005, almost all the landings occur on the Florida Keys. An assumption as to why this may be is that the Keys are the closest US shores that the migrants can reach from their island. However, being the obvious destination is probably also why migrants get caught almost 70% of the time somewhere in that straight path. Of the 30% that made it to shore, about half landed on the west part of the Keys, which correspond to a relatively less direct path.
In 2006, there are still migrants landing on the Keys but others have spread to the west coast of Florida. Some have also landed in Mexico (see right side of Figure 1 and bottom left corner of Figure 7). These new landing sites could be the result of attempting to escape the USCG by taking new routes. The result is an increase in the migrants' rate of landing success, which is about 40% in 2006. Looking at how interdictions have spread out along the Keys and clustered right along the shore where the boats are departing from Isla Del Sueno, it seems like the USCG are aware of where the migrants landed and they adapt their tactics to curb the landings in those sites. One such tactic seems to be to intercept the migrants as soon as they leave their island.
In 2007, the Florida Keys has a greatly reduced number of landings, which is consistent with the idea that the USCG probably increased security on routes leading to previously chosen landing sites. The increasing geographical range of interdictions may have pushed the migrants to land even further north on Florida's west coast and also all along its east coast. Expeditions landing in Mexico increased in 2007, which is consistent with our hypothesis of migrants going out of their way to avoid being caught since there were no interdictions on that route yet. More than 57% of the migrant boats landed in 2007. Since Mexico gets the most landings that year, it appears that the most important factor in selecting the landing site is not how far they are willing to go as much as choosing a route and destination that has proved to be free of interdictions.
 |
 |
 |
|
Figures 6-8: From left right: This shows the interdiction and landings of the migrants from Isla Del Sueno for 2005, 2006 and 2007. They land on US and Mexican shores (blue points); however the US Coast Guard intercepts (red points) them in international waters. The yellow mark-up highlights the changes over time. It was added using the drawing tools in GeoTime. | ||
At this point, the locus of attention was turned to individuals of interest in the Migrants dataset. The xml version of the data was made available to TRIST. All events were loaded at once and dimensions (shown in Figure 9) were used to look for interesting relationships across record fields.
 |
||
|
Figure9: User-defined dimensions, manually created, except for the People dimension, which is automatically populated with the results of entity extraction (performed by FairIsaac/HNC Text Analytic Engine (TAE)). The bar charts (logarithmic scale where each segment represent 5 results) provide a quick understanding of the data content. Knowing that Catalano and Vidro are names of individuals of interest, they were included in a user-defined dimension called 'Key Actors'. | ||
Selecting Catalano for example reveals the content of the records associated with him (see Figure 10). Scanning over the results highlighted by the 'Catalano' selection, one could observe that Eduardo Catalano and Jesus Vidro tried to migrate in three occasions, once a year, until they succeeded at their last attempt, in Mexico.
 |
 |
 |
|
Figure10: In 2005, they left in August on a Go Fast boat and got caught by the USCG Assured. |
Figure11: In 2006, they tried in Aug, again on a Go Fast boat, and got caught by the USGC Pompano. |
Figure12: n 2007, they left in Feb on a Rustic boat and landed safely in Mexico. |
 |
|
|
Figure13: Jesus Vidro and Eduardo Catalano migration attempts. In 2005 and 2006, they were aiming at the Keys and got intercepted both times. |
|
They must have observed that there were no interdictions stopping migrants going to Mexico. Other potential factors that may have improved their chance of success in this third attempt are 1) since fewer expeditions were taking place in the winter months, the USCG would be less vigilant then; and 2) they might have realized that Rustic boats had more success than Go Fast so far.
 |
|
|
Figure14: Vidro and Catalano are not the only ones that tried multiple times. The passengers dimension in TRIST shows the list of passengers that were recorded as being on more than one of the expeditions. |
|
 |
|
|
Figure15: Main hypothesis in the Sandbox to explain choice of landing sites over the three years. Once the analysis was completed in the Sandbox, the files were dragged into the Shared folder to make them accessible to the other analysts for review and to be used in the Grand Challenge. |
|
In 2005, the interdictions occur on the south east side of the Florida Keys, in the passage that directly link the Isla Del Sueno to the US. Few interdictions happen outside that block.
In 2006, the interdictions move west along the Keys. There are less landings on the west part of the Keys and more interdictions on the direct path from the island to that landing site. Interdictions now also occur directly offshore of Isla Del Sueno. The location of these interdictions correspond to the main three departures sites. Interdictions also started to happen on the west side of Florida. The USCG seems to be reacting to the landings, following the migrants' new routes, except for the Mexican one.
In 2007, the interdictions spread to the west side of Florida, a popular new destination for the migrants. The interdictions are now occurring directly offshore of the island, surrounding it completely.
 |
|
|
Figure16: Geographical patterns of interdictions over time become more meaningful when looking at them in conjunction with the landing events. Thus with all records visible, and colour applied to distinguish the landings (blue) from the interdictions (red), a cat and mouse game becomes visible when animating the events through time. The one place where no interdiction occurs, on the route to Mexico, is the main reason for the increasing landing success rate. Snapshots for each year were captured and imported in the Sandbox where the analysis was captured using notes and assertions. |
|
To clearly see the ratio of landings versus interdictions, the events were aggregated in GeoTime. The histogram on the map visually summarizes this ratio across the years. In 2005, the USCG has the upper hand and less than a third of the boats landed. In the second half of 2006, successful landings have started to pick up. In 2007, the migrants outperformed the USCG and more than 57% of the boats landed.
 |
|
|
Figure 17: : Proportion of landings vs interdictions for each year. | |
The number of deaths should also be considered as we can hardly qualify a landing where all passengers died as a successful migration. The Excel Visualizer plugin was used on the Migrant Data.xsl file. New columns were created to focus on people instead of boats. Then, the visualization was created by simply selecting these columns. The increasing height of the boxes over time confirms that the proportion of migrants landing alive is increasing over the three years.
 |
|
|
Figure 18: Migrants' success over time, measured as a proportion of travelers landing alive. | |