Where did the time go ?! Seems like we just began this term and it is already over. While I feel pretty unproductive right now, especially after seeing the outstanding work my fellow grad students have accomplished, I am extremely appreciative of the opportunity to explore the statistical tools available in ArcGIS and elsewhere that this course has provided. I found the freedom to explore our own data sets and the advice and encouragement provided by Julia and everyone else in the class incredibly rewarding. Thanks Julia for a great class!
I began the term with a pretty undefined thesis project (its still fairly fuzzy but at least I can start to make out a faint outline now). My data set is small and limited to a single survey season. As depicted in Figure 1, I have GPS track logs and the encounter locations for 20 cetacean surveys conducted in the Marquesas Islands of French Polynesia.
My primary focus is on a poorly understood species, melon-headed whales (Peponocephala electra).
These members of the dolphin family form very large “herds” (50 to as many as 1000 individuals) and have been observed congregating near the shore of Nuku Hiva in very specific locations on a regular, daily basis (Figure 2).
I spent most of this term finding, accessing, downloading, importing, reclassifying, converting, re-downloading, cussing at, etc., etc., environmental and bathymetric data. Using this data and other environmental data such as information on currents and sea surface height, I hope to investigate the differences and similarities between melon-headed whale encounter locations in order to 1) characterize these resting/socializing areas and 2) develop a model to predict possible resting/socializing locations in areas that have not been surveyed.
I was able to explore some of the tools in the Spatial Statistics Toolbox but for this data many of the tools are not applicable. For example, Ordinary Least Squares and Geographically Weighted Regression assume that there is linearity in the data. My data does not show linearity, even after transformation, and my response variable is not continuous. Running the Average Nearest Neighbor Tool produced the results that one would predict after looking at the maps provided above – the encounter locations are more clustered than predicted (Figure 3).
All of these results brought me to a point where I needed to take a step back and reexamine my data and my objectives. I felt like I was attempting to ask questions that just aren’t going to be answered by my data. My main question involves the characteristics of the encounter locations that define melon-headed whale resting locations. To get at this question, I plan on defining encounter locations spatially, i.e. delineate polygons of a certain size around encounter locations, and statistically examine the similarities and differences between the polygons using the environmental and oceanographic data mentioned above. I will continue to explore the tools available in ArcMap as well as the plethora of non-ArcGIS tools to answer this question.
Hi John,
Great Job this term! I definitely understand your frustration with trying to get data and then finding out that its not the correct data or that your data may not be suitable to answer the questions that you are interested in.
Its very possible that I have no clue what I am talking about but…I was wondering if you can use other indicators/characteristics from the areas where your dolphins are being sighted, for instance are the dolphins feeding in these areas? If so, is it possible to track the food source and then possibly link area characteristics back to the dolphins?
Hi Candice,
These guys don’t appear to be feeding in the “resting” (for lack of a better name) locations. However, I will be looking at the proximity of the resting locations to potential feeding areas. Also, I want to look into using landscape ecology type tools to characterize these areas and maybe tease out what ecological characteristics of the resting areas the melon-headed whales seem to prefer.
Hi John,
Your plan to “delineate polygons of a certain size around encounter locations” made me think of a few things that you might want to explore. Thiessen polygons may be a useful method to employ for developing these areas around your points. Another method that I recently observed at a friend’s defense was Tobler’s Hiking Function. While this was originally developed for terrestrial analysis of hiking speed based on slope, I wonder if someone has applied this in an ocean setting – something to investigate perhaps? Lastly, along the same lines of using Thiessen polygons, the Grouping Analysis tool has a Delaunay Triangulation and Nearest Neighbor option within it that might be useful for you to explore.
Here are some links to check out.
Ianko’s GIS page:
http://www.sciencecentral.com/site/453413
Ianko’s Thiessen Polygons Wizard tool:
http://www.ian-ko.com/ET_GeoWizards/UserGuide/thiessenPolygons.htm
Grouping Analysis (with a brief mention of both the Delaunay Triangulation and Nearest Neighbor options):
http://resources.arcgis.com/en/help/main/10.1/index.html#//005p0000004w000000
Keep up the good work!
-Doug
Hi John,
Wanted to say good luck on your cetaceous exploits. What you lack in data you make up for in enjoyable photography.
Regarding you project, one thing I can’t recall seeing form your work yet was a histogram analysis of the frequency of values in your background data versus the frequency of values in your sightings. You could get lucky in finding that the occurrences all happen in a range of values that’s very rare for the map as a whole, which would help you with statistical justification.
The other thing to perhaps pursue, if your background layers aren’t getting you what you want, is to create new data layers from existing data. Distance from island, we’d talked about looking into persistence, getting creative seems a good way to go here.
Also, even though it’s perhaps not an amazing or awe-inspiring finding, simply completing the mind-numbingly frustrating data processing step is an achievement on its own. That basic overlay data should have some value in and of itself, even if it can’t make the perfect habitat distribution model.
Cheers,
– Max
John,
You did a great job this term, and produced a clear, useful final presentation. As you note, the small size of the dataset makes it difficult to provide rigorous interpretations. But it can be done, and you are moving in the right direction. One useful approach for a small dataset is to assess the relative proportions of e.g., sightings with respect to the relative proportions of some environmental variable. For example, do 50% of the observations occur in some kind of environmental setting that represents only 20% of the area? Lots of potential here!
Julia