Geographic Profiling – a novel tool for locating tarsier sleeping sites
An analytical tool developed for use within criminology is being put to good use in ecological research. It has proved particularly useful in locating species in challenging habitats.
Geographic Profiling (GP) is an analytical tool developed originally to solve the problem of information overload when dealing with cases of serial crime. Here, it uses the spatial locations of connected crimes (such as serial rape, arson or murder) to predict the areas that will most likely include the criminal’s residence. It is a methodology that looks at the locations of connected crimes to determine the most probable anchor point (workplace or home for example) for that criminal. Investigations often result in large numbers of tip-offs, and large numbers of suspects rather than too few. Time and financial constraints will impede most investigations if every suspect has to be scrutinised. In the 1980s, police investigating the case of the Yorkshire Ripper accrued 268,000 suspect names and 5.4 million vehicle registration numbers. Clearly time constraints in this case meant that investigating each suspect would be unrealistic.
GP has been very successful within the field it was created for, and is widely used on a global scale, by police forces and investigative agencies. More recently this method has been applied to biological data sets such as animal foraging patterns, white shark dispersal patterns, invasive species and epidemiology. Human hunting behaviour observed in criminals involves searching and choosing a victim, method of attack by the offender and a target location. These behaviours are consistent with animals’ hunting and foraging behaviours, so applying criminology methodologies to ecological studies is a logical and valid step.
Using the tool within ecological research
these techniques originally used in criminology can be successfully applied to locating tarsier sleeping sites, and as such be further used within ecological research
SE Asia, specifically Sulawesi, houses a huge number of endemic species, and often habitat assessments of cryptic and elusive animals such as the tarsier are overlooked, primarily due to the difficulties of locating them in challenging habitats. The ability to accurately and reliably detect individuals as a basis for subsequent assessments of population growth, density and other variables is crucial for developing an understanding of change within populations and developing plans for management of both species and their habitat. Traditional assessment techniques are often limited by time constraints, costs and challenging logistics of certain habitats. In particular difficulties arise when studying elusive species with cryptic behaviours.
Did it work?
The package used is called Rgeoprofile, and it combines Google Maps using the R software package Rgooglemaps. This software was written by Dr Steven Le Comber, Dr Mark Stevenson and Dr Bob Verity of Queen Mary University London. A geoprofile is subsequently produced which is effectively a probability surface whereby every location on the geoprofile will have a probability attached to it of there being a source site there. Using the GPS location of tarsier vocalisations recorded on Buton Island, Indonesia, it was possible to see if the location of tarsier sleeping trees could be predicted from these vocalisation data alone.
We can assess the model’s performance using the hit score percentage. The hit score can be considered most intuitively by imagining a highly simplified game of ‘Battleships’. Let us suppose we are searching for a single battleship, occupying one cell in a 10×10 grid. If we call out grid squares at random (eg H2, C5, B7… etc), sometimes we will be lucky and the battleship will be in the first cell we call out. On the other hand, sometimes we will be unlucky and it will be in the 100th. In the first place, we have had to search one square out of 100; hence, the hit score percentage is 1%. In the second, the hit score is 100%. On average, we expect to have to search half of the 100 cells before locating the battleship. Thus, a hit score of 50% is the benchmark for a random, or non-prioritised, search and lower hit score percentages indicate a more efficient search strategy.
The model found 10 of the 26 known sleeping sites by searching less than 5% of the total area (3.4km2), or giving a hit score value of 5%. In addition the model located all but one of the sleeping sites by searching less than 15% of the area. We would expect a hit score percentage value of 50% from random search methods. The model also suggested that we searched for a tarsier sleeping tree at a particular GPS coordinate, and remarkably we found two trees within 5 metres of this, in an area of forest that previously we would not have surveyed. The results strongly suggest that these techniques originally used in criminology can be successfully applied to locating tarsier sleeping sites, and as such be further used within ecological research.