spadac_logo.png

 

VAST 2008 Challenge

Mini-challenge 2: Migrant Boats (geo-temporal analysis)

 

Authors and Affiliations:

 

Jason Dalton, SPADAC, jason.dalton@spadac.com
Lisa Kuchy, SPADAC

Ben Holland, SPADAC

 

Student team: NO

 

Tool(s):

 

Two-Page Summary:   NO

 

ANSWERS:

Boat-1: Characterize the choice of landing sites and their evolution over the three years.

 

Detailed Answer:

 

Our objective was to use a variety of visual analytic tools and methods to find geo-temporal patterns of migrations from Isla Del Sueño from 2005-2007. Initially, we used Google Earth to view the overall geospatial trends over the time period. We were able to dynamically view landings over given time frames by using the slider bar feature shown in Figure 1.



Figure 1: Google Earth with the slider bar option enabled for viewing landings over a specific time period.

 

We next used a temporal variogram (which shows to what extent a set of points becomes more or less similar in time as they get farther away in space) to analyze the overall space-time correlation in the landings data. It gave us an indication as to whether or not there is a significant relationship between time and space, and to what extent. The variogram in Figure 2 shows a low variance (i.e. dissimilarity) at short distances, and a progressively higher variance up until around 1.5 decimal degrees (about 165 km) when the slope of the curve starts to flatten out. We can therefore infer that on average there is a significant space-time correlation up to about 1.5 decimal degrees (165 km) away from a given landing site.



Figure 2: Temporal variogram of migrant boat landings, 2005-2007

 

A temporally sequenced geospatial dendrogram (which shows clusters of points with time stamps in geographic space) was created [using SPADAC’s Signature Analyst] to analyze local clusters of landings by year. In contrast with the variogram, here we are most interested in local (yearly, and by geographic region) patterns of space-time clustering. The time stamps are indices which correspond to a given time. In this case, an index of 1 corresponds with 01/01/05 while an index of 441 corresponds with 12/31/2007. By examining the time stamps on each individual node, we were able to discern whether or not a noticeably large proportion of values in a given time period are in a given region. For example, under the spatial cluster of Northern Cancun on the left side of Figure 3, the majority of index values for the individual nodes were greater than 300, which means that most of the landings in that region took place in 2007.

 


Figure 3: Geospatial dendrogram using temporally sequenced landing events from 2005-2007

 

The dendrograms indicate that there is a migration of landing sites over time. To determine the nature of this migration required us to show migration events progressing over time. We animated all encounters, landings only, and interdictions only using the temporal indexing capablilty of the KML specification. The video for landings is embedded below.

 

Get the Flash Player to see this player.

 

Another visual analytical tool that we used to find local geo-temporal patterns in the data, shown in Figure 4, is a proportional symbol migration map. The thicknesses of the arrows are proportional to the number of landings in a given area. This enabled us to coherently visualize the magnitude of the number of landings in comparison with where they occurred.

 

Geo-temporal characteristics of the landing sites

 

2005 Most landings in 2005 occurred around Miami and the Florida Keys, as seen by the arrows in the top left of figure 3. This is expected, as the Florida Keys are geographically closer to Isla Del Sueño than the rest of Florida, and are therefore more convenient and safer for attempted landings. There are a relatively low number of landings that year with only 46 landings occurring in 2005. That number would increase dramatically (129 in 2006, 266 in 2007) over the next 2 years as the migrants expanded their geographic extent of migration. This increase was most attributable to increasing pressure from the government on the movements of followers and leadership as well as the U.S. “wet foot, dry foot” policy for Isla Del Sueño migrants.

 


Figure 4: Geographic patterns of migration (landings) from Isla Del Sueño, 2005-2007

 

2006 There were fewer landings around Miami although much of the travel was still in the Florida Keys vicinity. From May 2006 to October 2006, landings began to take place farther north along the west coast of Florida around Tampa. This trend starts up again from February 2007 to October 2007. The landings gradually shift away from the Florida Keys due to the proportionally large number of interdictions there that same year. This forced the migrants to adjust their planned landing locations, assuming that they adjusted their routes to avoid the patrols. Furthermore, the USCG is probably less likely to patrol areas farther north like Tampa which are more difficult and dangerous for the migrants to get to from Isla Del Sueño.  There is a clustering of boat landings in Cancun, which occurred from April to November. The fact that this destination became popular may be because the USCG does not patrol there and Mexican immigration rules may be more permissive. It is also likely that migrants simply switched to using Mexico as a land bridge to the U.S. as they perceived the probability of interdiction greater on the water near Florida than crossing the U.S. - Mexican border. All landings during this time period occurred on Isla Muerjes, near Cancun.

 

2007 The majority of landings happened in Cancun, and farther north along the east and west Florida coasts as compared with previous years (especially 2005). Between March 2007 and September 2007, many landings took place along the east coast of Florida (not including Miami). It is only during this time that there is activity near Daytona Beach and the rest of the central east coast.  Between February 2007 and August 2007 there were widespread landings near Cancun (notice the thickness of the arrow pointing there on the bottom of figure 4), and after this period the activity there drops off significantly, as is consistent with the trend of increasingly fewer landings in the fall and winter as compared with the spring and summer months. In 2007, for instance, the average number of landings per month from October through March was 7.8, as compared with 36.5 from April through October.

 

Monthly and Weekly Patterns

The overall increase in landings during the warmer months (see the graph on the right in Figure 5) could be due to meteorological factors as well as seasonal fluctuations in ocean currents. In examining the number of landings by day of week, there is a noticeably higher number on Thursdays (as seen in the left graph in Figure 5). This may be due to the migrants’ speculation of fewer coast guard patrols on that day as compared to the other days.

 


Figure 5: Number of boat landings by week and month, 2005-2007

 

Boat-2: Characterize the geographical patterns of interdiction over the three years.

Short Answer:                                                                                                                                                                                                                              

In 2005, the majority of the interdiction sites were in the vicinity of the Florida Keys (likely due to the geographical convenience for the migrants). In 2006, an increasing number of interdictions were around the Isla Del Sueño, as the Coast Guard was likely increasing patrols there. There were also interdictions starting to occur off the southwest Florida coast, in response to the increasing number of landing attempts in those areas. In 2007, the interdiction locations did not change drastically from 2006, although the proportion of overall interdictions in Miami and in southwest Florida increased, and first interdictions near Tampa occurred. The proportional symbol migration maps in figure 6 were created in PowerPoint using Google Earth generated maps. Figure 7 shows a geospatial dendrogram of temporally sequenced interdictions (created in Signature Analyst using a hierarchical clustering algorithm), which helped us infer space-time correlation in the different geographical regions.

 


Figure 6: Geographic patterns of migration (interdictions) from Isla Del Sueño, 2005-2007

 


Figure 7: Geospatial dendrogram using temporally sequenced interdiction events from 2005-2007

 

 

 

Boat- 3 What is the successful landing rate over the time period?

Short Answer:

The migrants became progressively more successful (success_rate = [#landings]/[#total encounters]): 2005 - 30%, 2006 - 42%, 2007- 57%, with an overall success rate of 48%. In a spatial context, the map in figure 8 shows that there is a clear trend of an increasing success rate from south to north. Figure 8 was created using inverse distance weighted interpolation (IDW) on a set of sampled points. We chose IDW over other methods because it produced the lowest cross validation error.

 

In separating the encounters by month, we found that the migrants were less successful in the winter months when fewer trips are made. Using a Chi-squared test in R, we found that pattern to be statistically significant at the 0.05 level (p = 0.03). There also seems to be an unusually high success rate on Thursdays, as in figure 9. However, that abnormality is not statistically significant (p=0.69).

 


Figure 8: Interpolated map of landing success rate (from 2005-2007) using inverse distance weighting. * Please note that this is a trend surface – the actual rates away from land are assumed to be 0.

 


Figure 9: Success rate by day of week and by month, 2005-2007