Background:
The Tufted Puffin (Fratercula cirrhata), is an iconic seabird that provides a wide range of ecological, economic, and historically important services such as ecotourism for local communities and bringing marine derived nutrients to terrestrial habitats. Tufted Puffin populations on the Oregon Coast have declined dramatically from over 5,000 birds in 1989 to 500 birds in 2021 (Figure 1). In 2018, the Tufted Puffin Species Status Assessment (SSA) determined that factors related to breeding site conditions are one of the most probable causes of puffin decline; however, little is known about the specific characteristics of nesting habitat along the Oregon coast, or how it relates to their population demographics. My research project will directly address this knowledge gap by analyzing how Tufted Puffin breeding habitat has changed on the Oregon Islands National Wildlife Refuge over the past 50 years.
Figure 1. Breeding Tufted Puffin Estimates from 4 Oregon Coast-Wide Surveys in 1979, 1988, 2008, and 2021. Coast-wide surveys obtained from USFWS.
Research Questions:
1) How is the spatial pattern of Tufted Puffin population occupancy related to the spatial pattern of vegetation cover in terms of burrowing habitat availability?
2) How have these spatial patterns at burrowing sites changed over time?
3) Is Tufted Puffin occupancy high when vegetation cover near breeding sites is high, and low when nearby vegetation cover is low?
Datasets:
The temporal extent of my research project is from 1979 to 2021, and the spatial extent is the 62 islands on Oregon Islands National Wildlife Refuge where Tufted Puffins have historically nested. For the purposes of this class and these exercises, I limited my spatial extent to 10 islands, and limited my temporal extent from 2009 to 2018.
To examine percent-cover of vegetation on Oregon Islands refuge over time, I used the National Agriculture Imagery Program (NAIP) imagery because the resolution is 1m, which is fine-scale enough to observe changes in NDVI on my island sites. I obtained NAIP imagery in Oregon for the years 2009 and 2018, because both these years had photography that was available in full color and near infrared. The spatial extent of the NAIP imagery includes the entire state of Oregon, so all my study sites are encapsulated in this dataset.
I also used the data from 4 Tufted Puffin Oregon coast wide surveys to compare Tufted Puffin population occupancy (Figure 2). These coast wide surveys were conducted in 1979, 1988, 2008, and 2021 to estimate the total population of Tufted Puffins in Oregon, and their distribution across islands.
Figure 2. Breeding Tufted Puffin Estimate on the Oregon Coast from 1979, 1988, 2008, and 2021 coast-wide surveys.
Hypotheses:
1) The spatial pattern of Tufted Puffin population occupancy will be clustered in response to the clustered spatial pattern of percent cover vegetation.
2) Suitable Tufted Puffin breeding habitat will have decreased from 2009 to 2018, with more areas being vulnerable to erosion and vegetative degradation.
3)Tufted Puffin occupancy will be high in areas where vegetative cover is high, and low where vegetative cover is low.
I expect percent cover of vegetation to be clustered, because these islands have large areas of sloped rock slabs where there will be no vegetation, close to areas with soil and dense vegetation. I expect the spatial pattern of Tufted Puffin density to be clustered in response to areas with clustered percent cover of vegetation because Tufted Puffin like lush, vegetated areas with grasses, forbs, and shrubs. I expect that vegetation cover will have decreased from 2009 to 2018, with more areas being vulnerable to erosion and vegetative degradation because of more severe and frequent weather conditions related to climate change.
Approaches:
To evaluate the spatial pattern of Tufted Puffin occupancy across islands, I used the average Nearest-Neighbor Distance (NND). To take this analysis one step further and determine if the spatial pattern of Tufted Puffin occupancy varied at different spatial scales, I used the Multi-Distance Spatial Cluster Analysis (Ripley’s K) function. I also used a kernel density estimator to determine if there were Tufted Puffin hot spots and identify where those might be along the Oregon Coast.
To examine the spatial relationship between Tufted Puffin occupancy and vegetation cover, I used multiple tools and steps. First, I downloaded and extracted polygon boundaries of each of the islands. Next, I downloaded 2018 and 2009 NAIP imagery into ArcGIS Pro and clipped the rasters to the extent of the polygon boundaries, so I was only analyzing the data I was interested in. Then, I used NDVI to quantify the vegetation cover on each island. After that, I used focal statistics to identify the mean NDVI value on each island. Finally, I plotted mean NDVI values against the Tufted Puffin occupancy of each island as a logistic regression in R.
Finally, I used a confusion matrix to observe changes in vegetation over time on one island. I began by clipping 2 NAIP rasters from 2009 and 2018 of the island to the previously extracted polygon boundary. Then, I used the iso cluster unsupervised classification tool to classify each raster into 2 classes. Next, I used the combine rasters tool to observe changes between years. The attribute table contained 4 columns that represented: 1) vegetation that stayed the same, 2) vegetation gained, 3) bare earth that stayed the same, 4) vegetation lost. The attribute table displayed the pixel values associated with each of these categories. By dividing each category’s number of pixels by the sum number of pixels in the raster, we could determine what percent each category was.
Results:
From the Ripley’s K and Nearest-Neighbor analysis, I determined that Tufted Puffin occupancy in Oregon is clustered. The kernel density estimator outputs a map displaying the hot spots of breeding Tufted Puffin occupancy on the Oregon Coast (Figure 3). There is a dense hotspot in Northern Oregon, and another dense hot spot in Southern Oregon. The NDVI function outputs a raster with values ranging from –1 to 1, which is used in many scientific applications (Figure 4). The logistic regression analyses output a graph depicting the NDVI values against breeding Tufted Puffin occupancy (Figure 5). I interpreted these results as a positive correlation between breeding bird occupancy and NDVI. The higher the NDVI is at sites, the higher probability there will be more breeding birds.
From the unsupervised classification and combine tools, I obtained a map of Goat Island with 4 color classes, where the colors corresponded to 1) vegetation gained, 2) bare rock that stayed consistent 3) vegetation that stayed the same between years, and 4) vegetation lost (Figure 6). I then used the attribute table to calculate the percentages of the number of pixels in each class relative to the overall number of pixels. From these results, I interpreted the vegetation decreased on Goat Island from 2009 to 2018 by 23%. The results also show an increase in vegetation from 2009 to 2018 by 8% (Table 1).
Figure 3. Hot spots of breeding Tufted Puffins on the Oregon Coast in 2021.
Figure 4. NDVI output raster of Goat Island from 2009 NAIP imagery.
Figure 5. Logistic Regression analyses of NDVI values and breeding Tufted Puffin occupancy.
Figure 6. Supervised classification of 2009 and 2018 NAIP imagery of Goat Island, followed by a raster combination. Red=vegetation lost between years, green=vegetation gained, yellow=vegetation that stayed the same, beige=bare rock that stayed the same.
| Color | Pixel Count | Description | Percent |
| Green | 29469 | Veg gained | 8% |
| Red | 80737 | Veg Lost | 43% |
| Yellow | 90247 | Veg consistent | 26% |
| Beige | 152367 | Bare consistent | 23% |
What I learned from each analysis:
The Ripley’s K and hot spot analyses were very interesting because up until this class, I had mainly been focused on detecting changes on a temporal scale. However, these analyses helped me to reframe some of my initial questions to observe changes on a spatial scale as well. I learned that on a larger spatial scale, Tufted Puffin occupancy was clustered, and there was one Northern hot spot and one southern hot spot. This led me to consider other questions about why Tufted Puffins are not breeding in large quantities along the central coast of Oregon, and what site-specific or climatic factors might be associated with the hot spots. For Exercise 2, I determined the NDVI and plotted it against Tufted Puffin occupancy for high and low sites in a logistic regression analysis. Exercise 2 helped me to learn that Tufted Puffin occupancy is high where vegetation is high, and occupancy is low where vegetation is low. From here, I can further investigate the driving forces behind what might be causing vegetation to differ spatially at different sites. From the supervised classification exercise, I learned that utilizing the “combine” tool provides more information than the raster calculator tool alone, because we get 4 layers to display vegetation change.
Significance.
It is important for land managers to know where the hot spots are along the Oregon coast. Once we understand where high and low occupancy of Tufted Puffin are, we can begin to investigate what characteristics make successful breeding habitat, and why some sites might exhibit higher occupancy than others. For example, we can compare hot spots to “sinks,” and look at factors such as mean sea-surface temperature (SST), mean air temperature, distance to the mainland, amount of human disturbance, etc.
I also learned that there is a positive spatial relationship between Tufted Puffin occupancy and vegetation at breeding sites. Although managers will not be able to directly use the analyses I did in this class, I can build off these results, by incorporating more sites in my analyses, to give the logistic regression more statistical power. This spatial analysis will provide invaluable insight into the habitat requirements of Tufted Puffin- a species of greatest conservation need- in Oregon. Gaining a deeper understanding of the specific vegetative characteristics Tufted Puffin utilize during the breeding season can be used to further address conservation concerns, and guide refuge managers in decision making around habitat restoration.
Finally, the confusion matrix allowed me to quantify the amount of vegetation lost, the amount of vegetation gained, and identify where these processes are occurring. This is significant because it can better inform land managers about where habitat restoration might be effectively implemented. I anticipate that the proposed data collection and analysis will serve as a springboard for future conservation efforts, allowing managers to make comparisons with ecologically similar seabirds that also use the Oregon Islands National Wildlife Refuge for breeding.
What I learned (software):
I did the majority of my analyses in ArcGIS Pro, and after this course, I feel much more comfortable using this software. This was the first time I’ve explored ArcGIS Pro without following the steps of a strict tutorial, and I am very happy with how much I’ve learned and the progress I’ve made on my project. Most importantly, I learned that ArcGIS software can be tricky, but if the original plan does not work, there are lots of adjacent ways to address the same problem. This class helped me get into the habitat of talking my process out with other folks who are working on similar projects and google alternative solutions if I run into problems.
Aside from verbally communicating my results, I also became more skilled at visually communicating my results. One critical component to research I learned in this class was creating figures or graphs in Program R and Excel to better display the results of a spatial analysis.
What I learned (statistics):
One thing I learned about using hot spots was that a few outliers can dramatically skew the distribution of my data. It was very helpful for me to log-transform my data prior to using a kernel density analysis to work with more normally distributed data. I also learned a lot about GWR from talking with other students in this class, and determined I would like to use this approach for my own data further down the line. I might be able to compare Tufted Puffin occupancy with multiple variables, such as slope, elevation, vegetation, precipitation, temperature, and other climate data. The GWR tool in ArcGIS would also allow me to compare the Pearson’s correlation coefficient between my variables, which would help me decide which variables to include in candidate models.
Evolving question.
Initial questions from the beginning of class:
1) How is the spatial pattern of Tufted Puffin population density related to the spatial pattern of percent cover of vegetation, soil depth, elevation, and slope in terms of burrowing habitat availability?
2) How have these spatial patterns at burrowing sites changed over the last 50 years?
Questions I now plan to address:
- How is the spatial and temporal pattern of Tufted Puffin population occupancy related to the spatial pattern of vegetation, elevation, and slope at Tufted Puffin breeding sites?
- How have these spatial patterns at burrowing sites changed over the last 50 years?
- How do these vegetation changes relate to the site-specific, climatic, and environmental variables influencing Tufted Puffin habitat at different spatial scales?
I still plan to identify how spatial patterns at burrowing sites have changed over the last 50 years. However, rather than exploring this question on just a temporal scale, I also plan to look at differences at different spatial scales. I originally planned to compare each island to a previous version of the same island. Now, I am excited to compare each island to other spatially distributed islands within the same year. For example, I might explore how cooler temperatures associated with islands on the Northern Coast affect the Tufted Puffin occupancy differently than warmer temperatures and less precipitation on the Southern Oregon Coast. For this third question, I am planning to look at vegetation cover as my independent variable, and a suite of climate processes as my response varaibles.
Future Research
This project allowed me to gain a deeper understanding of the relationship between Tufted Puffins and vegetative cover. My next step will be to observe this on a longer temporal scale, and a larger spatial scale. I am planning to now incorporate a separate aerial photography dataset I obtained from USFWS to observe changes in vegetation over 4 decades, rather than just 1 decade. I plan to use ArcGIS Pro to georeference this dataset, so the pixels of each island overlap from year to year, to make my results more meaningful. I am also planning to use a supervised classification technique to classify ground cover as vegetation, bare soil, rock, or water. To further compare vegetation cover to the occupancy of breeding Tufted Puffins on each island, I will use a logistic regression.
Moving forward, I am also planning to investigate the relationship between topography and Tufted Puffin occupancy over time. I plan to use the same basic process as I did when analyzing the vegetative cover, but instead of using NAIP rasters, I will use DEM/DTM rasters. I plan to clip DOGAMI DEMs and DTMs to the same polygon boundaries of the islands and subtract rasters from different years. I will also use spatial statistics to find the median elevation and slopes of the islands and use a regression approach to display the results.
I would also like to explore using GWR when I am further along in my research to determine which variables, such as slope, elevation, vegetation, precipitation, and temperature have the greatest impact on Tufted Puffin occupancy. To better understand the climate processes happening at each site, I am planning to download rasters of SST, precipitation, temperature, etc. From NOAA weather service sites, and extract values at each Tufted Puffin colony site.
