Combining the 5 factors into the MCE resulted in the cost surface pictured above. Each cell was assigned a value of suitability based on cost. In this instance, darker cells were more suitable than lighter cells as darker cells were less costly to cross. Note the variability in cost across the North Atlantic Ocean. This cost surface was used as the base for both the LCP and DFL analyses.
Least Cost Pathway
The routes identified by the LCP analysis were very mechanical and many followed the same overlapping route. The majority of routes leaving Britain tended to go around the Northern part of Ireland, through the southern tip of Greenland, then continued towards their destination in BNA. Overall, there was an expectation that shipping routes should follow the North Atlantic Jet Stream due to the high velocity currents, geographical location, and strong winds (Foster, 2014). The LCP appeared to have followed this route. Due to the fact that each route was individual, it was not possible to amalgamate routes to display them as proportional symbols. The visual aesthetics of the maps were poor, thus another method of path mapping was used.
See maps for all time periods here.
Distributive Flow Lines
The DFL analysis performed better than the LCP. The outputs remained quite mechanical; however, flow lines ran through the cost surface in a different pattern than the LCP. The major difference between the location of routes in the results was in the center of the Atlantic Ocean. Cheshire (2012) developed a map that showed routes taking a wider route to avoid the center of the ocean, while the DFL ran routes directly through it. This could indicate that distance from shore is a factor directing shipping routes; it was noted in the research that ships would keep close to shore as much as possible (Toghill, 2005). This factor should be included in future models to develop an increasingly accurate model.
In terms of overall impact, the DFL is a much stronger tool. Routes are more visible and the cost surface is used more effectively across the entire Atlantic Ocean. Additionally, it offers more options for display. Routes can be visualised based on routes per year (as shown above) or by origin throughout the years. This makes the output much more flexible.
Click the following links to see the final interactive maps created using CartoDB:
Strengths and Limitations
The MCE was effective in incorporating factors of varying significance into the final cost layer. It was ideal for this project as it can be used in situations in which the criteria are heterogeneous as it uses a common metric within the algorith in identifying potential routes based on the criteria. This was beneficial for this analysis as it combines both quantitative and qualitative data as rasters. It does have the potential to oversimplify the interactions between factors which could introduce error into the model. Furthermore, weighting for each variable can be difficult and arbritrary and a simple error could skew findings substantially (Dobes et al. 2009). However, it is a well known analysis that is straightforward and simple for people to understand. Additionally, there is extensive online support for this analysis. In the context of this class, this is an important feature as this project could be taken up by others in the future. In the context of the Empire Trees Climate project, others may wish to expand on the project in the future so an easy to follow and expandable approach is ideal.
The LCP had the capability of mapping routes across the ocean according to the least costly path. The framework is based on a well-understood theoretical foundation. LCP findings can be easily reproduced which makes future research on historical timber tracking more accessible. The limitations of this process were in the data used to create the cost surface: the ocean layer did not extend to the shoreline. This resulted in an area that was weighted less costly around Greenland which was the area that most of the routes took. Additionally, most routes overlapped each other which made visualisation poor. Furthermore, it was not possible to amalgamate them to show them through proportional symbols. It was not an efficient tool to use, as there was a large time requirement to map each route individually. Due to the fact that the data held thousands of routes to be mapped, using the LCP approach was impractical and time-consuming.
The DFL resulted in an output that was much more visually pleasing. All routes could be seen and symbolized effectively. The DFL also allows for more flexibility in the display options which is ideal for the interactive output. In terms of a general tool, the DFL was much more efficient and would not require as much of a time requirement to map all of the data. The output is a vector line which prevents the Raster to Polyline conversion that is necessary in the LCP. One limitation of the DFL tool includes arguments from scholars that the DFL approach provides information to map users that is still too cluttered and confusing (Phan, 2002).
Neither method resulted in outputs that were clearly ship-like. They were very mechanical outputs and while their general routes were fairly similar to expected results, there were limitations to the degree of similarity. This can be accounted for by the fact that not all factors that influence ship movements were represented in the MCE. Factors such as distance from shore, local knowledge, storm events, and other unknown factors should be explored and included in the MCE. Oversimplifying the dynamics of ship movements weakens the overall ability to accurately recreate routes. Data quality, human error, coarse resolution, and arbitrary scaling have all introduced error in the final results of each path mapping method.