J.A. Laney – Final Project Blog Post GEOG 566
Background & Research Questions
One chapter of my dissertation research is focused on exploring patterns of functional diversity and heterogeneity–diversity relationships in communities of disparate small-bodied vertebrate taxa. I am interested in (1) understanding how functional diversity (estimated by metrics such as functional richness, functional divergence, functional redundancy, functional dispersion, etc.) vary within and across communities of songbirds and small mammals along the elevational gradient of my study system, and (2) relating these patterns to changes in habitat heterogeneity that occur across elevation. In this course I focussed on the first of these two objectives by exploring autocorrelation of functional metrics of communities and also by describing the dissimilarity of communities by distance.
In Exercise 1, I set out to assess the spatial autocorrelation of various functional metrics across the 16 localities for which I have both bird and mammal survey data. In Exercise 2, I asked how taxonomic and functional Sørensen dissimilarity of passerine bird and small mammal communities in my dataset vary as a function of geographic distance. In Exercise 3., I explored the sensitive of my analysis dissimilarity of bird-mammal communities to changes in the underlying data, specifically when filtering bird observations by distance from observation points?
Description of the Dataset
The dataset I analyzed is comprised of records of small mammal and bird occurrences, as well as associated habitat information, from a comprehensive biological survey project I am leading on a desert-montane gradient in the Northern Great Basin: The Steens Mountain Resurvey Project. To date, I have surveyed 21 discrete localities across the elevational gradient of this region with sites ranging in elevation from the mountain’s summit to the basin floor (Figure 1.). Each locality consists of a circular survey area with an approximate area of 0.8km. Localities were selected based on the availability of historical small mammal survey data and to maximize sampling of the elevations and habitats in the Steens Mountain-Alvord Desert system. Of the total localities in this project, I selected 16 sites where both bird and small mammal data were collected for the years 2019 and 2021.
Within each survey locality (Figure 1.), I conducted avian point count surveys for breeding songbirds and small mammal trapline surveys in the summers of 2019 and 2021. Point-counts were conducted within 4 hours of sunrise between early June and early July to coincide with hours of peak bird activity during the breeding season in this region. I detected birds by sight or sound and recorded each bird’s distance from the observer to the nearest meter using a digital rangefinder. Small mammal trapline surveys consist of removal trapping along multiple traplines (usually 3-6) arrayed at each survey locality to capture the heterogeneity of habitat types and plant communities and to detect the highest diversity in the shortest amount of time. Using a combination of baited Sherman live-traps, Havahart traps, museum snap traps, Victor rat traps, and pitfall traps along traplines, I targeted rodent and shrew species (Orders Rodentia and Eulipotypla) (<500g). In most scenarios, mammal surveys were conducted for a minimum of four nights to maximize sampling. Small mammal surveys are conducted between early July and mid-August, tailored to align with periods of historical sampling while accounting for moon phase.
Analytical Approach
Methods – Exercise 1.
To characterize functional diversity of the passerine-rodent-shrew communities, I calculated functional metrics, such as functional richness, functional redundancy, functional divergence, and functional dispersion of communities using unique trait combinations of the birds and mammals detected in each survey locality in this project. These metrics were derived using the “mFD” packing in program r for each locality using the abundance data of the birds and mammals detected during field surveys. To assess spatial autocorrelation of the various functional metrics across the 16 localities for which I had both bird and mammal survey data. To do so, I calculated the global Moran’s I correlation coefficient for each functional metric.
I began by sorting my bird point-count records by taxonomic order (Passeriformes) and distance (< 100 meter from observer). I then joined bird data with small mammal data by locality. I selected unique mensural traits from the ecological literature that were both biologically relevant and shared both the small-bodied songbirds and small mammals in this system. These traits included body mass, litter/clutch size, % diet invertebrate, % diet vertebrate, %diet scavenged, % diet seed, % diet fruit, and % diet plant other. Traits came from two different databases, the “Amniote” database, and “Elton” database.
I calculated standard functional metrics for each locality using the “mFD” r package and matrices of species by site, and traits by species. To calculate functional redundancy, I first binned traits using the Sturgis algorithm to derive “functional entities, or species with unique trait combinations (UTCs). In r, I generated a distance matrix of localities using their associated geographic coordinate information. I then took the inverse of the matrix values and replace the diagonal entries with zero to complete a distance matrix that I could use to assess spatial autocorrelation. Once I had functional metrics for each locality, I computed Moran’s I in the r programming language using the ‘Moran.I’ function in the ‘Ape’ package. This was a relatively straight forward process using minimal lines of code, though I fist had to create a custom function to compute the coefficient across multiple columns representing individual functional metrics.
Methods – Exercise 2.
For this exercise 2, I was interested in exploring how the taxonomic and functional dissimilarity of the passerine bird and small mammal communities in my dataset vary as a function of geographic distance. Specifically, I modeled the Sørensen dissimilarity between localities, as well as its turnover and nestedness components, as both a function of geographic distance in kilometers and xyz distance (latitude, longitude, and elevation) to see how dissimilarity is related to distance.
I began by calculating the pairwise taxonomic dissimilarity of all pairs of localities using the Sørensen index in the ‘betapart’ package in R. The output of this package provides overall Sørensen distance, as well as the turnover and nestedness components of the dissimilarity. I then calculated the pairwise functional dissimilarity of all pairs of localities using the functional beta diversity function in the ‘mFD’ package (which utilizes the ’betapart’ framework). The output provides the functional Sørensen distance between all pairs of localities, as well as the functional turnover and nestedness components of the dissimilarity. Next, using the ‘sf’ package, I created geographic centroids for all my localities and calculated the pairwise geographic distance in km for all localities using the ‘st_distance’ function in that same package. Considering that these localities are distributed along an elevation gradient in a mountain-desert system, elevation is “built into” the geographic distances between localities. However, I also wanted to explicitly incorporate elevation into the distances. Thus, I used the ‘scatterplot3d’ package to create a 3-dimensional space composed of the xy (latitude and longitude) and z (elevation in meters) coordinates of the localities (Figure 2.). I then used the ‘dist’ function to calculate the xyz distance between all pairs of localities in this space. Finally, I modeled distance decay of taxonomic and functional pairwise dissimilarity, turnover, and nestedness for all pairs of localities against geographic distance in xy space and xyz space and plotted the results. These analyses were done using the ‘decay.model’ and ‘plot.decay’ functions in the ‘betapart’ package.
Methods – Exercise 3.
For this exercise 3., I was interested in assessing parameter sensitivity of the spatial pattern I described in Exercise 2. I tested the sensitivity of the dissimilarity distance decay analyses by filtering the dataset so that it only contained passerine bird observation that were within 50 meters or less of the points during surveys. I use distance-detection methods in my avian point count surveys while in the field, whereby I estimate the distance from me to all birds detected during a count to the nearest meter (calibrated by a laser rangefinder). Thus, I have distance estimates for every data point. In Exercise 2, I filtered bird observation to 150 m from observer. This is large distance and probably more than is reasonable for the dataset as it could introduce error, such as misidentified birds and potentially double counting of birds within localities due to overlapping count radii. As I wanted to test the method while ensuring I had enough species points to calculate the functional diversity, I decided to leave in those observation for the the first pass of this analysis in Exercise 2. In this exercise, however, I filtered the bird observations to passerines only detected within 50m or less of the observer during point counts. This is a much more reasonable distance and is in line with other approaches in passerine data collection methods using point counts. This filtering step reduced the number of observations in the bird-mammal input dataset to 1211 from the original 1668 species records and from 56 bird species to 51. I did not modify the small mammal data.
Figure 2. A simple 3-dimensional representation of the space containing the xy (latitude and longitude) and z (elevation in meters) coordinates of the 16 localities used in this analyses created using the ‘scatterplot3d’ package in R.
Results
Using the results of the Moran’s I global test (Table 1.) in Exercise 1., we can reject the null hypothesis that there is no spatial autocorrelation present for a given functional metric across these localities if p-value is < 0.05. Based on these results, functional richness, functional originality, and the number of functional entities appear to be spatially autocorrelated. All other functional metrics are not spatially autocorrelated correlated across the localities within the extent of this study.
Table 1. Output from Moran’s I analysis of all functional metrics across all localities.
The results from Exercise 2. show that turnover is the primary driver of taxonomic Sørensen dissimilarity between all localities (Figure 3.). Unlike taxonomic dissimilarity, functional dissimilarity does not seem to be driven exclusively by either turnover nor nestedness components and the relationship is less clear. I did not see a major difference in distance decay when dissimilarity is plotted against either xy or xyz distance, thus I have only presented the results from the xyz distance analysis here. The results after adjusting the dataset were strikingly similar to the results I obtained in the previous analysis (Figure 4.). As in Exercise 2, we do see a triangle plot emerge. This shows that taxonomic dissimilarity can be both extreme and low in near localities, but only extreme in localities that are spatially separated to greater degrees. This may indicate some lower bound of taxonomic dissimilarity as distance increases.
Figure 3. Taxonomic dissimilarity and functional dissimilarity as a function of three-dimensional distance (latitude, longitude and elevation in meters) for communities of rodents, shrews, and passerines modeled using a power function. Triangle symbols denote pairwise comparisons of localities. The y-axis indicates dissimilarity, and its turnover and nestedness components, in species composition (A – C) and functional composition (D – F) between localities (measured using the Sorensen dissimilarity index), with higher values indicating more dissimilar communities. Also shown are the slope (b) and coefficient of determination (R2) for the fitted models.
Figure 4. Taxonomic dissimilarity and functional dissimilarity as a function of three-dimensional distance (latitude, longitude and elevation in meters) for communities of rodents, shrews, and passerines modeled using a power function after filtering the data to only include bird observations < 50 m from observer. Triangle symbols denote pairwise comparisons of localities. The y-axis indicates dissimilarity, and its turnover and nestedness components, in species composition (A – C) and functional composition (D – F) between localities (measured using the Sorensen dissimilarity index), with higher values indicating more dissimilar communities. Also shown are the slope (b) and coefficient of determination (R2) for the fitted models.
Interpretation & Significance
Functional richness (FRic), functional originality (FOri), and the number of functional entities appear to be spatially autocorrelated based on the results of the Moran’s I global analysis. FRic indicates reflects the amount of niche space filled by species in the community. In this analysis I primarily chose functional traits that correspond to diet, thus changes in species composition may modify the functional richness if those species consume drastically different resources. It makes sense that communities that have similar FRic would be geographically autocorrelated due to the environmental filtering along the elevational gradient, as communities closer to each other are most likely comprised of similar species that use similar dietary resources. FOri quantifies how changes in species abundances modify the functional redundancy between species (i.e., minimal functional distances among species pairs). Species tend to be functionally less original in the pool if they tend to share their traits more closely with other species. The interpretation here is a bit trickier, but it seems that localities that have similar abundances of particular species are autocorrelated.
Turnover appears to be the primary driver of taxonomic dissimilarity in the bird-mammal communities along this gradient. As turnover between communities is though to indicate environmental filtering processes structuring species composition, this pattern makes sense considering these localities are distributed along an elevation gradient with major differences in environmental conditions and habitat between localities at different elevations. Interestingly, we do see a triangle plot emerge (panels A and B in Figure 3.). This shows that taxonomic dissimilarity can be both extreme and low in near localities, but only extreme in localities that are spatially separated to greater degrees. This may indicate some lower bound of taxonomic dissimilarity as distance increases.
The fact that the same dissimilarity pattern emerges when I adjusted the dataset to only include bird observations within 50 m of the observer is interesting (Figure 4.) and somewhat of a relief. For one, it suggests that that my field sampling was thorough enough to capture similar species both near and far from observation points. More useful though, this sensitivity analysis shows that the dissimilarity patterns observed are robust to slight variations in the underlying data set. The numbers of bird species was reduced in the dataset by 5, and the total number of observations was reduced by 457. This did have an effect on functional richness, as the functional metric calculations used in the ‘mFD’ package take both species identity and the abundances of each species into account. However the overall pattern of dissimilarity remained the same between pairwise comparisons of sites across the elevation gradient despite the reduction of species data.
The findings I have produced in this course are an important part of describing the pattern in functional diversity of these bird-mammal communities along the gradient of the Steens-Alvord system. These results tell part of the story and will be valuable as I connect functional diversity of these disparate taxa to underlying environmental habitat characteristics.
Learning Reflection
Over the course of these three exercises I increased my learning of using program r to perform several analyses and wrangle data. Specifically, I used the ‘mFD’ and ‘betapart’ packages to perform multivariate analyses in order to calculate estimates of both taxonomic and functional diversity across communities, as well as dissimilarity of these communities as a function of geographic distance in three dimensions. I did a great deal of data sorting, which was time consuming but helpful in the long run—both for my dissertation chapter and my increased understand of working in r. I learned that certain spatial analyses were not appropriate for my data given the structure of the dataset and the aggregate metrics I was interested in. These techniques included kernel density and autocorrelation function, for example. I spent a great deal of time at the end of the course attempting to perform interpolation techniques on habitat data associate with my species observation data. Specifically I wanted to assess origins techniques to interpolate a habitat surface across locality landscapes from discrete habitat point data I have. Ultimately, I was unable to accomplish this in r, though I think this speaks more to my limitations that the program. I also failed at attempting to do this in QGIS before running out of time at the end of the course. I plan to continue attempting to achieve this goal.
Future Directions & Techniques
The work I have conducted in this course has been beneficial in that I have described spatial patterns of taxonomic and functional diversity in bird-mammal communities across the Steens-Alvord elevation gradient. This is a necessary first step for the work I am planning to do for this chapter of my dissertation. Ultimately I also plan to investigate the relationship between these patterns and covarying changes in habitat heterogeneity across elevation. To do this I will utilize habitat data I have calculated along small mammal trapline (i.e., structural complexity indices derived from desecrate habitat quadrate data collected in the field), as well as remotely sensed data I have pulled in using ArcGIS. I am interested in trying regression kriging and Empirical Bayesian kriging to develop habitat complexity surfaces that could be used as predictors in models that describe the influence of habitat heterogeneity an available area by elevation to functional metrics of these communities. Thus, my work continues.