+-Upon application of the calibration models, it was discovered that the use of log transformed models interacted with patterns in the data to create unreliable biomass estimates. The log transformed models were therefore displaced by non transformed models. Additionally, the models were modified to force the intercept through the origin. The reasons for both of these decisions are explained in the discussion.
There was not sufficiently large sample sizes to create valid models for all present species. In the subsequent analysis of species composition and biodiversity parameters, the best available model was therefore used. Genus models were prioritized, followed by morphological type models and finally the general model.
A non-metric multidimensional scaling (NMDS) was performed in R, in order to investigate the importance of environmental variables related to geographical location and management for the composition of species in treatment plots. NMDS was chosen to deal with skewed species distributions and categorical nominal variables (Shipley, 2021). A RxC matrix of species occurences was created, where the rows were constituted of the samples and the columns were the number of intercepts per registered species. Using the function “metaMDS” from the “vegan” package, a two-dimensional NMDS ordination was conducted. metaMDS is a wrapper function, meaning that it performs several functions in one command (Oksanen, 2021b). In this case, the default setting was used, which resulted in the function performing a Wisconsin double standardization on the data whereby each element is divided by its column maximum and then by the row total. Subsequently, the function applied a square root transformation. The Bray-Curtis dissimilarity measure was used for its ability to deal with count data and general popularity when analyzing data on species occurrences (Shipley, 2021). Two dimensions yielded a stress value of 0.146, where increasing the number of dimensions would yield only small additional reductions in stress.
A second RxC matrix of environmental variables was uploaded to R, containing the same samples as rows and with location, block and treatment as factor variables. The environmental factors were fitted to the ordination using the “envfit” function from the “vegan” package and the output gave the R2 and p values associated with the variables. The centroid value coordinates of the environmental variables were extracted and plotted along with the sample coordinates using the “ggplot” function from the “ggplot2” package.
The above process was repeated for another dataset where the intercept values were transformed into biomass estimations using the calibration models previously constructed.
A canonical correspondence analysis (CCA) was carried out using the same dataframes of species occurences and environmental values as previously mentioned in order to test whether a constrained ordination would show a significant effect of treatment on species composition (Legendre & Legendre, 2012). In order to deal with potential errors associated with very rare species, the data was first log+1 transformed. The analysis was then carried out using the function “cca” from the “vegan” package. The function uses chi-square transformation of the data matrix and performs a weighted linear regression on the constraining variables, the fitted values of which are then subjected to correspondence analysis via singular value decomposition (Oksanen, 2021a). Variance inflation factors were calculated using the “vif.cca”  function from “vegan”, with no values >10, indicating no collinearity between the environmental variables (Oksanen, 2021c). An analysis of variance was then perfomed for the constraining variables using the function “anova.cca” from “vegan”. The same function was also used to test the significance of the constraining axes, of which the first three were significant. The centroid value coordinates of the environmental variables were then extracted and plotted along with the sample coordinates using the “ggplot” function from the “ggplot2” package.
The process was then repeated for the biomass estimates.
In order to test the importance of treatment on the selected biodiversity parameters, a data frame was constructed in which the parameter values were listed for each OU, as well as the variables location, block and treatment. All parameter variables are continuous, except for species richness, withered intercepts and sensitive species, which only consist of integers. The values for species richness, unicity and Ellenberg L were normally distributed, whereas the values for sensitive species, Simpsons diversity and evenness, the ratio of forbs to graminoids, withered intercepts and Ellenberg N had skewed distributions.
Levene’s test was carried out, using the function “leveneTest” from the “car” package, to test whether variances were significantly different between variable groups, which is a requirement for e.g. analysis of variance. Only the variable “location” showed significant between-group differences and only for the parameters Simpsons diversity and evenness and Ellenberg N. There were no instances of significant differences in variance for treatment.
Linear regression models on the parameter values revealed that the errors were in many cases not normally distributed.
A generalized linear model (GLM) was therefore constructed for each parameter, using the “glm” function from basic r, with the treatment, block and location as independent variables. GLM’s were chosen for their ability to deal with count data, and irregular error distributions (Crawley, 2007). Poisson’s error was used when modelling the species richness, withered intercepts and sensitive species, as it is able to deal with count data (Crawley, 2007).
The process was repeated for a second dataset where intercept values were transformed into biomass estimates, using the calibration models previously constructed. Since the parameters of species richness, sensitive species and unicity were not based on the number intercepts, but rather the presence of species in the plot, these were not repeated.
Block location in either the hills or on the old seabed beneath the Littorina slope was identified in the field and in the subsequent analyses as a significant factor for several parameters.