FOR ONLINE PUBLICATION
Appendix 1
COMPARISON OF THE TRADITIONAL, VISUAL ASSESSMENT OF THE BAIT LAMINAE WITH AN IMAGE BASED METHOD
In addition to the visual assessment for the feeding activity on the bait laminae, where a binary classification was used and each aperture was classified as perforated or not by holding the stick against bright light, we also tested an image based method. In the image based method, each stick was photographed against a constant light source and the area of the perforations was analysed with ImageJ software (Schneider and others 2012).
The results were compared using a Passing-Bablok non-parametric regression (Passing and Bablok 1983; Bilic Zulle 2011) in the mcr package for the R software (Manuilova and others 2012). The two different methods to quantify the perforations of the bait apertures showed a fairly consistent relationship overall (Figure A1.1), with a Pearson’s correlation coefficient of 0.88. The conclusions of the study would have been the same regardless of the method
There was, however, a proportional bias between the two methods: the difference between the methods became progressively larger at higher feeding rates, the visual method showing higher feeding activities than the image based method. This highlights the difference between the two methods: when the majority of the apertures are perforated this will result in a high visual score, however, if the bait has not been completely consumed in those apertures, then the feeding activity estimate using the image software will be lower.
The visual method is faster, if the perforation status each individual aperture is of interest (for example, for analysing feeding depth patterns). However, if the main interest is the proportion of bait eaten per each stick, or per a cluster of sticks (for example, all sticks within one plot), the image based method may be advantageous.
References (Appendix 1)
Bilic Zulle L. 2011. Comparison of methods: Passing and Bablok regression. Biochemia medica 21: 49-52.
Manuilova E, Schuetzenmeister A, Model F. 2012. mcr: Method Comparison Regression. R package version 1.1.
Passing H, Bablok W. 1983. A new biometrical procedure for testing the equality of measurements from 2 different analytical methods - application of linear-regression procedures for method comparison studies in clinical-chemistry.1. Journal of Clinical Chemistry and Clinical Biochemistry 21: 709-720.
Schneider CA, Rasband WS, Eliceiri KW. 2012. NIH Image to ImageJ: 25 years of image analysis. Nat Meth 9: 671-675.
Figure A1.1
Figure A1.1. Comparison of two methods to quantify the percentage of bait eaten in the bait lamina sticks. In the visual analysis a binary variable was used, where each aperture was recorded as perforated or not based on whether the perforation was visible to a naked eye when holding the stick against bright light. The measure of the feeding activity was the percentage of perforated apertures out of the total number of apertures. In the image analysis, the sticks were photographed against a constant light source and the area of the perforations was calculated using image software. The measure of the feeding activity was the percentage of the perforated area out of the total surface area of the apertures. In the figure, each data point represents the mean feeding activity in one plot.
Appendix 2
UPSCALING RESULTS FROM ONE STUDY SITE TO THE FOREST LANDSCAPE IN ENGLAND – UNCERTAINTY ANALYSIS
To assess the edge effect on forest soil fauna feeding activity at a landscape scale, we extrapolated the results from our study site to all broadleaved and mixed forest patches in England. The upscaling shows that edge effects are important for this ecosystem process: the mean feeding activity across the fragmented forest landscape may be reduced by around 17%. This estimate has to be treated with caution, however, as the results are based only on a single study site. Therefore, rather than an exact value, the estimate should be seen as indicative of the magnitude of the edge effect.
We do not have empirical data to evaluate how well our study site represents the whole landscape in terms of soil fauna feeding activity. Therefore, we carried out an uncertainty analysis that considers two main sources of uncertainty associated with our study: 1) the moisture gradient from the edge towards the forest core may differ from that observed in our study site, and 2) the overall level of soil fauna feeding activity and the sensitivity of feeding activity to moisture and to distance from the edge may differ from that observed in our study site. The aim was to create a parameter space that covers all plausible scenarios of soil moisture conditions and soil fauna feeding activities in forest patches and to illustrate how this variation changes the estimate of the edge effect on soil fauna feeding activity.
As forest edges lose more water than forest core areas, due to higher evapotranspiration at the edge (Didham and Lawton 1999; Herbst and others 2007), soil moisture typically decreases towards the forest edge (Kapos 1989; Didham and Lawton 1999). However, there are likely to be concomitant, site specific changes, such as changes in soil type and forest stand structure and composition from the core towards the edge, which may change the expected soil moisture gradient. We described the soil moisture gradient along the distance from the forest edge with a logarithmic model:
Moist = a+b*LOG(Dist+1)(Eq. 1)
In Eq. 1, Moist is the volumetric soil moisture content (%), a and b are parameters, and Dist is the distance (m) to the nearest edge from within the forest. To simulate the site-specific variation in moisture, we generated a range of possible soil moisture gradients by drawing the values for parameters a and b from random normal distributions. The mean of a (2.96) and b (26.4) were from the model fitted to our study site. The standard deviations of a and b were derived by fitting the model for each individual edge to core transect within the site (n=6), and by calculating the standard deviation among the transects. The results were multiplied by 3, to take into account that the variation across the landscape is likely to be larger than the variation observed within one study site. This resulted in standard deviations of 1.43 and 7.32 for a and b, respectively. In the simulations, the decreasing soil moisture level towards the edge is the most probable outcome, but the moisture and distance to the edge relationship may vary from positive to negative and from flat to steep (Figure A2.1a). The random parameter set was constrained so that the predicted moisture content (%) was plausible, that is, between 0 and 100.
Soil processes can be assumed to have a unimodal response to moisture (Martin and Bolstad 2005; Schrama and others 2013). The process rate, in this case soil fauna feeding activity, decreases when moving away from the moisture optimum towards drier or wetter conditions. We described this unimodal response with a Gaussian function (Eq. 2)
FA = FAMAX*EXP(-0.5*((Moist-MOPT)/MTOL)^2),(Eq. 2)
In the equation, FA is soil fauna feeding activity, Moist is the volumetric soil moisture content (%), FAmax is a parameter denoting the maximum feeding activity, which is reached at optimum moisture level, MOPT is a parameter denoting the optimum moisture level and MTOL is a parameter denoting the tolerance of the soil fauna to moisture variation.
For the uncertainty analysis, we used the parameter estimates based on the data from the study site as mean values for the landscape: FAMAX = 46.2, MOPT=69.3 and MTOL=30.0. We constructed random normal distributions around these mean values to create alternative parameter spaces.
We made the assumption that the maximum soil fauna feeding activity, FAMAX, was likely to have large variation across the landscape, from very high to very low activity, due to variability associated with tree species composition, edaphic conditions, and the abundance and community composition of the soil fauna. Therefore, for the parameter FAMAX we used a random normal distribution with mean of 46.2 and standard deviation of 30.0, truncated between 0 and 100 (% of bait lamina stick apertures perforated).
We assumed the moisture optimum, MOPT, to have a slightly narrower distribution, with a mean of 69.3 and standard deviation of 15. Values were truncated between 15 and 85, as neither extremely dry nor extremely wet conditions are likely to be optimal for the soil fauna activity.
We assumed that tolerance of soil fauna to variation in moisture was likely to be fairly large, as the natural variation in moisture level throughout the growing season can be considerable, although soil fauna has been shown to become inactive in dry conditions (Hassall and Tuck 2007; Collison and others 2013). Therefore, for MTOL we assumed a mean=30.0 and standard deviation=30.0, but truncated between 20 to 100, to take into account that minimum tolerance is likely to be considerably higher than zero (zero tolerance indicating that the soil fauna community would became inactive at moisture levels even slightly outside the moisture optimum). The maximum tolerance, on the other hand, can be very large, representing a situation where soil fauna is not sensitive to variation in moisture.
We assumed the soil fauna community to be locally adapted to the prevailing moisture conditions, soil fauna in dry sites having lower moisture optimum than soil fauna in wet sites. Therefore, we further constrained the acceptable MOPT values so that the moisture optimum was within one MTOL of the simulated moisture level at the core (>450 m from the edge). We further checked that the predicted feeding activity values were plausible, that is, between 0 and 100 (% of bait lamina stick apertures perforated). The resulting soil fauna feeding activity ranged from high to low and the response to variation in moisture ranged from flat to steep (Figure A2.2b)
The combination of Eq.1 and Eq. 2 describes the response of soil fauna feeding activity to the distance to the edge (as moisture co-varies with distance). These simulations ranged from positive to negative edge effects and from flat to steep responses (Figure A2.2c).
We then generated 5000 alternative parameter sets for Eq. 1 and Eq.2 to calculate 5000 alternative estimates for the mean soil fauna feeding activity across the landscape, assuming (i) uncertainty in the soil moisture gradient only (Eq.1), keeping the soil fauna response (Eq.2) parameters constant, (ii) uncertainty in the soil fauna response only (Eq. 2), keeping the moisture gradient as a function of distance from the edge (Eq.1) parameters constant, and (iii) assuming uncertainty in both moisture and soil fauna response, drawing all the parameters of Eq.1 and Eq. 2 from the simulated parameter sets.
The simulations showed that the variation in the soil fauna response resulted in a much broader distribution of the landscape scale feeding activity than the variation in moisture conditions (Figure A2.2, left panels). However, the moisture and soil fauna response contributed more or less equally to the variation in the reduction in the feeding activity due to edge effect (that is, the landscape-level feeding activity compared to the feeding activity in the forest core in each simulation) (Figure A2.2, right panels). A small proportion of the simulations also showed a positive edge effect, but such cases were not within 90% confidence intervals.
References (Appendix 2)
Collison EJ, Riutta T, Slade EM. 2013. Macrofauna assemblage composition and soil moisture interact to affect soil ecosystem functions. Acta Oecologica in press.
Didham RK, Lawton JH. 1999. Edge structure determines the magnitude of changes in microclimate and vegetation structure in tropical forest fragments. Biotropica 31: 17-30.
Hassall M, Tuck JM. 2007. Sheltering behavior of terrestrial isopods in grasslands. Invertebrate Biology 126: 46-56.
Herbst M, Roberts JM, Rosier PTW, Taylor ME, Gowing DJ. 2007. Edge effects and forest water use: A field study in a mixed deciduous woodland. Forest Ecology and Management 250: 176-186.
Kapos V. 1989. Effects of isolation on the water status of forest patches in the Brazilian Amazon. Journal of Tropical Ecology 5: 173-185.
Martin J, Bolstad P. 2005. Annual soil respiration in broadleaf forests of northern Wisconsin: influence of moisture and site biological, chemical, and physical characteristics. Biogeochemistry 73: 149-182.
Schrama M, Veen GF, Bakker ES, Ruifrok JL, Bakker JP, Olff H. 2013. An integrated perspective to explain nitrogen mineralization in grazed ecosystems. Perspectives in Plant Ecology, Evolution and Systematics 15: 32-44.
Figure A2.1
Figure A2.2
Figure A2.1. Simulated relationships between a) soil moisture and distance from the forest edge (Eq. 1), b) soil fauna feeding activity and soil moisture (Eq. 2) and c) soil fauna feeding activity and distance from the edge, taking into account that moisture co-varies with distance. Each curve represents one simulation with random parameters. For visual clarity, 50 simulations out of the total number of 5000 are shown in each panel, including those simulations that resulted in the highest and lowest feeding activity and largest positive and negative edge effect on feeding activity. The black lines are the models using the parameters from the empirical study site.
Figure A2.2. Frequency histograms of 5000 simulations of soil fauna feeding activity across the landscape (left panels) and the reduction of soil fauna feeding activity due to edge effect, i.e. landscape feeding activity compared with the feeding activity in the forest core (right panels). In the simulations, parameters of the a) moisture vs. distance to the edge model (Eq. 1), b) soil fauna feeding activity model (Eq. 2) and c) both models were allowed to vary. The solid horizontal line denotes the value observed in the empirical study and the dashed lines are upper and lower 90% confidence intervals of the simulations.
1