Question explored
In my last blog post, I mapped habitat suitability for Olympia oysters in Yaquina Bay, OR by assessing three environmental parameters: salinity, substrate availability, and elevation. In exercise 2, I brought in oyster location data points collected from field surveys of the intertidal zone to compare against the map of suitable habitat. The question I am examining in this exercise is:

How does the spatial pattern of Olympia oyster location data correspond to the spatial pattern of suitable habitat in Yaquina Bay?

Field surveys of the intertidal zone of Yaquina Bay were conducted on April 19-20 and May 17, 2019 during low tides. Oysters were characterized as ‘present’ if evidence of at least one oyster (living or dead) was detected within the predefined search area.

Name of tool or approach
I first uploaded the data points collected in the field into Google Earth where I could easily verify the locations against my field notes and photos, as well as perform some minor quality control. The points were imported into ArcGIS Pro for spatial analysis. Statistical information was reviewed and plotted in Excel.

Brief description of steps to complete the analysis

1. After validating the data point locations and performing some minor quality control in Google Earth, I saved the points as a KML file. In ArcGIS Pro, I used the geoprocessing tool ‘KML to layer’ to convert them for analysis. Once I added the data points onto the map as a layer, I edited the symbology to display the points as ‘Present’ or ‘Absent’.
2. To assess the neighboring habitat, defined by 4 class types from least suitable to most suitable, surrounding each of the data points, I used the ‘Multiple Ring Buffer’ tool to create 3 buffer rings around each of the points at distances of 75, 150, and 300 meters. For the results in this blog post, only the 300-meter buffer was used. I selected ‘overlapping (disks)’ in the dissolve option to assess the habitat around each data point individually.
3. Once the buffers were created, I used the ‘Zonal Statistics’ tool to overlay the buffered areas onto the raster of habitat suitability. This tool allows the user to select by statistic desired (mean, median, etc.) to generate a spatial output. I chose ‘majority’, which categorizes the buffer zones based on dominant habitat suitability type within the buffer. Majority also represented the median in this output. For example, if the majority of the suitable habitat within the buffer area is class 4 (most suitable) then the buffer display is shown as ‘4’.
4. In addition to the ‘Zonal Statistics’ spatial output, I used the tool ‘Zonal Statistics as Table’ to generate a table of all the statistical information relevant to this analysis. The same input data is used (overlaying the buffered zones on the habitat suitability raster) to create this table.
5. I copied the table generated into Excel where I split up the data on a couple levels: 1) Presence vs. Absence and 2) North shore vs. South shore for comparison. The north and south shores are managed very differently: the north shore is largely composed of rip rap and steeper slopes because Yaquina Bay Road runs right along the edge, whereas the south shore is more natural and less developed. I created box and whisker plots of the majority habitat suitability type, minority, variety, and mean.

Results

The results show some mixed information. When looking ‘Presence’ data points, the majority habitat types surrounding these points on both the north and south shore are 3 or 4, most suitable areas. The minority habitat type is very different between the north and south shore, with the south, more natural shoreline showing a stronger correlation with low suitability habitat being found less often around presence data points. The means for presence data points generally correspond with greater habitat suitability.

However, the absence data points show that the majority habitat type tends to be more closely aligned with predicted most suitable habitat, especially on the south shore. Additionally, the minority habitat types surrounding the absence data points are 1, 2, and 3, indicating that the least suitable habitat does not constitute much of the area. This could be partly due to low coverage by the least suitable habitat type overall (see maps). There appears to be very weak correlation between absence of oysters and location of suitable habitat. Absence of oysters is likely to be recorded in both suitable and unsuitable habitat.

Critique of the method
This method further revealed to me that the importance of the resolution of the baseline data. Based on field observations and conversations with shellfish biologists, the distribution of Olympia oysters is very patchy due to substrate availability. The oysters may be found attached to a pile of rocks in the middle of the mud flat, but will not be found elsewhere in the mud flat. The raster layer I have available for substrate has classified substrate into large generalized categories, which does not reflect the nuanced nature of their opportunistic settling strategy. Dividing habitat suitability into only 4 categories limits the complexity of the analysis which can be helpful, but also means that there’s not a lot of distinction between suitable and unsuitable. Additionally, more data points will help make this analysis more robust.

Using the buffers and the ‘Zonal Statistics’ tools created a generalized output that provides some useful information for analyzing habitat suitability for the oysters. The approach is easily duplicated, which was helpful as I needed to add my field data points in batches as I collected them. What would be more informative for the next iteration is to be able to analyze multiple buffers side-by-side; how does the smallest neighborhood around each point compare to the larger ones?