A SPATIAL ANALYSIS OF SEA TURTLE AND HUMAN INTERACTION IN KAHALU U BAY, HI By Nathan D. Stewart USC/SSCI 586 Spring 2015
1. INTRODUCTION Currently, sea turtles are an endangered species. This project looks at the work that Rachel Silverman has been conducting on site at Kahalu u Bay in Hawaii. With the assistance of Dr. Karen Kemp and Dr. Jennifer Swift, this project is designed to help Rachel determine where in the bay there is the most human and sea turtle interaction so that she may design ways to protect the turtles in those areas better in the future. The spatial problem is determining where the humans come across the turtles the most and what sites would benefit the most from her help. Her data is compiled from the last six months from October 2014 through April of 2015. With her data, this project is designed to create a visual map of all of the turtle sightings and a spatial analysis of possible locations that need her help. Figure 1 shows the study area of the project.
Figure 1: Kahalu u Bay Grid reference given to volunteers to ascertain location of sea turtles. Figure shows the full extent of the study area. 2. DATA AND METHODS This project is based solely on the data given to me from Rachel Silverman and her work on the Kahalu u Bay Education Center Turtle Project. Rachel has uploaded all of the data that I am using to the Kohala Center Portal. From there I am able to retrieve it in a spreadsheet format and use it in ArcGIS. The first dataset that I am using is the Kahalu u Bay Turtle Survey data. This data details all of the volunteer s visual sightings of sea turtles in an area of the bay. This data is broken down by grid location and further by quadrant location. Each piece of data is sorted by date and time of the sighting and each line of data
represents a single sighting. Each of the turtles is represented by a discrete point vector system. There are 347 entities located in this dataset than span a period of six months. The scale of this dataset is set around the big island of Hawaii s Kahalu u Bay and all of the points that are placed on the map are located at the center of the grid and quadrant they are in. This is due to the lack of spatial information in the data. The collection of this data is done by visual interpretation by volunteers. The collection and original map of the area was not done with geographic coordinates and the grid system originally designed gave no way to distinguish where the exact location of the turtles was. Therefore, each turtle has been placed inside of the center of the quadrant within the grid specified. Due to this, there are many points that overlap in quadrants. An example of this is shown in Figure 1. The key attributes taken from this dataset are the grid, quadrant, and date attributes. Each of these is used as ordinal scales of measurement and there are no measurement units for these attributes. There is a good amount of fuzzy data, as the predictive models will leave some room for error. Overall, the idea is not necessarily to focus on the exactness of the model, but rather to demonstrate what areas the turtles are sighted in the most. Lastly, when it comes to the error of this data, as previously stated, it relies on volunteers and visual sightings rather than exact data points. Moreover, the grid locations that are given in the data are roughly 60m x 60m squares and the quadrants 15m x 15m squares, which leaves a vast amount of room for error in the visual sightings. The final problem that was come across is that none of the quadrants or grids had coordinates associated with them. Therefore, I had to georeference the map from a picture of a grid overlay on top of a Google
Earth map of the bay. This is the map that was given to all of the volunteers to determine where the turtles were (Figure 1). The next dataset that is used is a compilation of the grid and quadrant references from the turtle data in the last dataset mixed with my own georeferencing and total counts of sea turtle sightings. This dataset allows me to establish a system in which to count the number of sightings of the turtles in a particular quadrant within each grid. This allows me to do the majority of my geostatistical analysis (Figure 2 and 3). Each of the points again represents a number of sea turtles that were documented to have been spotted in the same grid quadrant. Once again this data is vector point data, but may contain more than one sighting in a single point. As far as the size goes, there are 45 entities in this table. The scale of this dataset is the same as the last; we are still looking at Kahalu u Bay on the big island of Hawaii. The measure of this dataset will be in meters. The key attributes of this dataset are the grid, quadrant, latitude, longitude, and number of turtles spotted. Each of these will be measured in ordinal except for the number of turtles spotted which will be nominal. The measurement units for the number of turtles spotted are simply a count in integers; the remaining attributes units are not measured. As for the fuzziness of the attributes, like the last dataset, that is some fuzziness and room for error as the dataset is built by myself and based on eyewitness accounts rather than hard latitudes and longitudes. To this however, I feel as though the level of accuracy needed for this project is not a precise level, but more of a relative location. Therefore, the level of fuzziness does not impede the
process or outcome of the final data. Lastly, there is still plenty of error built into this dataset as I have stated. The data is based solely on volunteer findings with no actual equipment and then I interpreted that information and georeferenced my own map of the area and data from those findings. Figure 2: Graduated symbols of the number of turtle sightings
Figure 3: Kriging of the number of turtle sightings When it comes to the coordinate systems for these datasets I decided to use NAD 1983 StatePlane Hawaii 1 FIPS 5101 (Meters) as the projected coordinate system. All datasets use this coordinate system and I am using a world imagery basemap that I projected into this system so that everything aligns properly. I have done all of my analysis in this coordinate system. Using these datasets I was able to create a model to help build the spatial analysis maps. Figure 4 demonstrates what the model called Turtle Spatial Analysis, looks like and the outputs that came from it. Among the outputs are analysis maps of IDW, KDE, Hot Spot analysis, and Kriging. This model was built using model builder and PyScripter. Figure 5 is a flow chart that shows the process by which I built this
project. This model fulfills the goals of the project because it is able to create a number of analysis maps that show where the levels of turtle sightings are the greatest and then when overlaid with the human interaction zones, you are able to identify the areas with the highest levels of interaction. Figure 4: Model of inputs and outputs of spatial analysis.
Figure 5: Flow chart representing steps taken to complete project. 3. RESULTS Identifying the areas of Kahalu u Bay in which the humans and turtles interacted the most was the goal of this project. As a result of that goal, various analytical processes were used to obtain an understanding of where those interaction points were. The results came back quite promising as you can see in Figure 6 with the Hot Spot Analysis, I was able to locate the area that had the most turtles and then when you look at Figure 7 you can see the human interaction zones with the number of turtles.
Figure 6: Hot Spot Analysis of the Number of Turtle Sightings from October 2014 to April 2015.
Figure 7: A look at the human and turtle interaction zones. The polygons are the human occupied zones. For the spatial analysis of the project I focused on four types of analysis. The first was the Hot Spot analysis that you see above in Figure 6, the second was the Kriging prediction model you see in Figure 3. The third form of analysis was the IDW prediction model (Figure 8) to determine the unknown points and by using the weighted averages determine what areas are more likely to have turtle population. Lastly, I used the idea of Kernel Density to show the areas turtle population and where the highest chances of turtle sightings were (Figure 9).
Figure 8: Depicts the Inverse Distance Weighted number of turtle sightings Figure 9: Depicts the Kernel Density of the number of turtle sightings within the barrier of data
After completing these analysis methods I was able to see that the area with the most interaction between the number of turtle sightings and the human zones was located in grid D3, quadrant 2 (Figure 1). Although this location did not have the most turtle sightings, it was the location that had the most interaction between the sightings and the human zones. This information showed that the edge of the swimming and snorkeling zone was where the humans were more likely to interact with the turtles. With this information Rachel will now be able to plan accordingly and use this to help her educated people in that area about their interaction with the turtles. Moreover, the results showed that the swimming, snorkeling, and people on the beach were more likely to encounter the turtles than the surfers. I was able to verify the models and analysis information by way of QQ plots and assessing the error in the prediction models. Below in Figure 10 are examples of the QQ plots and prediction error of the statistical analysis. Using this data I was able to see that each of them are fairly accurate to what I am looking for in the data. Considering that I had to make the coordinate data myself and place multiple sightings in the same area, I assumed the error would be relatively small and the QQ plots would be similar to where they are.
A. B. C. Figure 10: A) Prediction Error from IDW Model. B) QQ Plot from the Kernel Density Model. C) QQ Plot from the Kriging Model. 4. CONCLUSIONS AND FUTURE WORK Although the analysis was able to identify the area in which the majority of the human and turtle interaction occurs, It barely touched the surface of what could be done with this project. With the proper data and time, there are many things that could be done to this project. To start, one could do a proper temporal analysis if the data matched up between the human data and the turtle data. One could also turn this into a web browser to allow for proper placement of geographical positions by simply having the user click where the turtle was sighted.
As for recommendations for improvement on the data itself; to start, I would first make sure that the area that the study was done in was georeferenced and that proper coordinates were put on a map. Next, I would use a map of the area to have the volunteers that are collecting the data mark where the turtle is sighted, but also mark where they are when they see the turtles. I would also have a survey that would ask questions such as, how many people are near the sight of the turtle and what are they doing? Is it just a single turtle or is there more than one in that area? What is the weather like at the time of the sighting? What were you doing when you saw the turtle? These are just a few of the questions that could help in further analysis and the weather question could help with outside affects that could produce correlation to when turtles appear. Furthermore, I would add a second map that has a zoomed in area of Grid C2, C3, D2, and D3, which have the most turtles spotted, in order for the location data to be more precise for the user. In conclusion, this project has barely touched all that it could be with the proper data; however, it has allowed us to see the areas that we were looking for in terms of interaction. With the proper updates to the data this project could be even more accurate and narrow the zone from 60m grids to a possible 1m error. I think this could help tremendously in the protection of these endangered turtles and help Rachel Silverman teach proper precautions to tourists and visitors.
5. REFERENCES Ahmed, M., Dutton, P., Howard, P., & Squires, D. (2011). Conservation of Pacific sea turtles. Honolulu, Hawaii: University of Hawaiʻi Press. Gardner, B., Sullivan, P., Morreale, S., & Epperly, S. (2008). Spatial and temporal statistical analysis of bycatch data: Patterns of sea turtle bycatch in the North Atlan. Canadian Journal of Fisheries and Aquatic Sciences, 65(11), 2461-2471. Silverman, R. (n.d.). Kahalu'u Bay Turtle Survey. Retrieved April 1, 2015, from http://hbmpweb.pbrc.hawaii.edu/tkc/turtle_survey_data 6. ACKNOWLEDGEMENTS Special thanks to Rachel Silverman for the use of her data and Dr. Karen Kemp and Dr. Jennifer Swift for their help and guidance throughout this project. For more information, contact Nathan Stewart at ndstewar@usc.edu