The following supplements accompany the article Effects of temperature and humidity on activity and microhabitat selection by Littorina subrotundata Karen J. C. Rickards 1, Elizabeth Grace Boulding* *Corresponding author: boulding@uoguelph.ca Marine Ecology Progress Series 537: 163 173 (2015) CONTENTS Supplement 1. Additional figures and tables Fig. S1. Study sites. Fig. S2. Measuring air temperature, relative humidity and dewpoint. Table S1. Definitions of the five microhabitat categories used in this study. Table S2. Definitions of the three levels of activity used in this study. Table S3. Nine sets of independent variables used in the logistic regression models. Table S4. % cover of barnacles and snail density inside the surveyed quadrats. Table S5. Environmental variables for 2011 and 2012 for the study sites. Table S6. Pearson correlations among independent environmental variables. Table S7. Rankings for the set of models used to assess microhabitat selection. Table S8. Components and fit of the accepted best model evaluating microhabitat selection. Table S9. Rankings for the set of models used to assess activity level. Table S10. Components and fit of the accepted best model evaluating activity level. Table S11. Rankings for the set of models used to assess aggregation propensity. Table S12. Components and fit of the accepted best model evaluating aggregation propensity. Supplement S2. Back-calculation of air temperature from RH and t dp in 2011. Fig. S3. VPD c obtained using air temperature 1
Supplement 1. Additional figures and tables A B Fig. S1. Study sites (A) Bamfield Marine Sciences Centre, BMSC, (48.836N, 125.136W), located on Vancouver Island, British Columbia. (B) From left to right in photo: Cape Beale, CB (48.786058N 125.217142W), Nudibranch Point, NP, (48.815N, 125.1758W), Prasiola Point, PP, (48.81667N, 125.16944W), and Bamfield Marine Sciences Centre, BMSC. Image credit; Google Earth; https://maps.google.com/maps?ll=48.822478,-125.15077&z=14&t=h& hl=en June 24 2012. 2
A B Fig. S2. (A) Measuring substrate temperature of a littorinid snail found in the microhabitat bare rock next to barnacle(s). (B) Bottom; measuring air temperature, relative humidity and dewpoint of the same snail. 3
Table S1. Definitions of the five microhabitat categories used in this study (See Fig. 1). Microhabitat Bare rock/crevice: rock Bare rock next to barnacle: next Live barnacle/barnacle test: on Between barnacles: btwn other Description Snail is not in physical contact with any biotic structure; may be in small crevice in the rock. Snail is in physical contact with a barnacle test, and shell aperture is facing/in contact with underlying rock substrate. Snail is either on top of a live barnacle within the crevice formed by shell plates, or else fully within the test of a deceased barnacle. Snail is in physical contact with at least two barnacles; it is not in contact with the underlying rock substrate. Snail is in physical contact with a structural element of the habitat that is not rock or barnacle; in almost all cases algae or a limpet. Table S2. Definitions of the three levels of activity used in this study. Activity Level Category Withdrawn Foot out Moving Description Snail is completely withdrawn into its shell; operculum sealed over tissue and no part of the body is visible. A portion of the snail s body (the foot) is visible at the shell aperture; operculum is not sealed. A large portion of the snail s body is visible, and the snail is displaying noticeable movement across the substrate. 4
Table S3. Description of each of the nine multiple working hypotheses and the corresponding independent variables used in the logistic regression models. Model Hypothesis Independent Variables used in Model 1 None of the variables examined influence behaviour Constant only 2 Temperature is a cue for behaviour Temperature 3 VPD is a cue for behaviour Vapour Pressure Deficit 4 Temperature and VPD are cues for behaviour Temperature and Vapour Pressure Deficit 5 Temperature and VPD interact as a cue for behaviour Temperature, Vapour Pressure Deficit, and Interaction 6 Site influences behaviour Site 7 Size influences behaviour Size 8 Site and snail size-class both influence behaviour Site and Size 9 Time since last immersion influences behaviour Emersion Time (2011 only) Table S4. Mean (± se) % cover of barnacle and snail density per 100 cm 2 inside the N quadrats used to measure microhabitat selection at three sites in 2011 and 2012. Site and Period Year N Snail density 1 % cover barnacles 2 Nudibranch Point (NP_2011) 2011 29 48.4 (± 8.45) 35.5 (±4.59) Prasiola Point (PP_2011) 2011 26 32.0 (± 3.87) 56.3 (±4.79) Nudibranch Point (NP_2012) 2012 18 43.4 (±3.84) 36.1 (±4.64) Prasiola Point (PP_2012) 2012 21 34.4 (± 3.42) 64.0 (± 4.86) Cape Beale (CB_2012) 2012 12 30.2 (± 8.86) 57.1 (± 5.56) 1 Significant difference in snail density between sites but not between years (Two way ANOVA df=1, 90: Year p = 0.830, Site p=0.042, Interaction p=0.552; (NP > PP). In 2012 there was no difference in snail density among the three sites (ANOVA df=2, 48: p=0.188). 2 Highly significant difference in % cover of barnacles between the two sites used in both years (Two way ANOVA df=1,90: Year p = 0.393, Site p< 0.000), Interaction p = 0.463). NP had a significantly higher % cover of barnacles than the other two sites in 2012. (ANOVA df=2, 48: p <0.001, Tukey tests: p= 0.024 (NP < CB), p< 0.001 NP < PP), p= 0.625 (PP = CB)) 5
Table S5. Environmental variables for 2011 and 2012 for study sites. Site and Period Tair 1 VPD 2 T s 3 2011 May NP N of Cases 294 294 294 294 Minimum 8.96 0.160 9.00 0.106 Maximum 28.9 2.190 23.9 0.353 Arithmetic Mean 17.7 0.858 16.0 0.217 Standard Error of Arithmetic Mean 0.259 0.025 Emersion Time 4 2011 May PP N of Cases 219 219 219 219 Minimum 8.68 0.030 9.70 0.122 Maximum 23.0 1.620 23.5 0.403 Arithmetic Mean 14.8 0.552 14.3 0.238 Standard Error of Arithmetic Mean 0.250 0.025 2011 August NP N of Cases 460 460 460 460 Minimum 18.1 0.460 14.7 16.3 Maximum 37.2 3.370 26.3 36.0 Arithmetic Mean 27.5 1.641 20.2 26.8 Standard Error of Arithmetic Mean 0.239 0.037 2011 August PP N of Cases 510 510 510 510 Minimum 16.4 0.290 14.3 0.252 Maximum 34.7 3.000 28.2 0.857 Arithmetic Mean 25.0 1.226 20.5 0.680 Standard Error of Arithmetic Mean 0.198 0.026 2012 July/August NP N of Cases 626 626 626 Minimum 14.720 0.235 14.200 Maximum 24.480 1.194 26.900 Arithmetic Mean 19.883 0.709 20.835 Standard Error of Arithmetic Mean 0.101 0.010 0.126 6
Site and Period Tair 1 VPD 2 T s 3 2012 July/August PP N of Cases 610 610 610 Minimum 14.730 0.220 15.500 Maximum 25.440 1.500 26.000 Arithmetic Mean 19.721 0.626 19.990 Standard Error of Arithmetic Mean 0.093 0.013 0.101 Emersion Time 4 2012 July/August CB N of Cases 605 605 605 Minimum 15.270 0.307 15.100 Maximum 23.820 1.252 29.500 Arithmetic Mean 19.653 0.706 20.731 Standard Error of Arithmetic Mean 0.081 0.009 0.126 1. In 2012 this was measured air temperature which did not differ among the three sites (ANOVA df=1, 1838: p=0.19). In 2011 we forgot to record air temperature ( C) so it was calculated from the humidity probe s measurements of RH and DP (see text of Methods). 2. VPD The calculated Vapour Pressure Deficit (kpa). In 2012 VPD differed among sites (ANOVA df=1, 1838: p<0.001, Tukey HSD p<0.001: PP <NP = CB). In 2011 VPD was estimated using calculated air temperature (see text of Methods). 3. T s Measured substrate temperature ( C). In 2011 the interaction between site and month for the two-way ANOVA for T s was highly significant (ANOVA df =1,1479: p<0.001). Tukey pairwise posthoc tests showed that: 1) Both sites were cooler in May than in August, and 2) PP was significantly cooler than NP in May (p<0.001) but was not significantly different in August (p=0.310). In 2012 PP was significantly cooler than the other two sites (ANOVA df=1, 1848: p<0.001, Tukey HSD p<0.001: PP < CB =NP). The interaction between site and year was significant for an ANOVA of the summer data for PP and NP from both years (ANOVA df=1, 1848: p<0.001); substrate temperatures were higher at NP in 2012 than in 2011 but were not higher at PP (see above)). 4. Emersion Time the time since a snail was last inundated by the tide was calculated from the Hobologger measurements. Models containing this variable were not ranked highly so it was only used for 2011 models (see text of Methods). 7
Table S6. Pearson correlations among independent environmental variables. 2011 (N=1,483 snails) Tair 1 VPD 2 3 T s Tair 1 1.000 VPD 2 0.928 1.000 T s 3 0.801 0.673 1.000 2012 (N= 1,736 snails) Tair 1 1.000 Tair 1 VPD 2 T s 3 VPD 2 0.898 1.000 T s 3 0.729 0.743 1.000 1. In 2012 this was measured air temperature. In 2011 the air temperature ( C) was calculated based upon measurements of RH and DP (see text of Methods). 2. Calculated Vapour Pressure Deficit (kpa). In 2011 calculated Tair was used to calculate VPD (kpa). 3. Substrate temperature, T s ( C). 8
Table S7. 2011 and 2012 rankings for the models representing multiple working hypotheses used to assess microhabitat selection. All values are calculated as described in Anderson (2008). For 2011 model set, n = 1483 and ĉ = 5.461 (ĉ = estimated overdispersion parameter). QAIC c is the corrected value for AIC c, ΔQAIC c indicates the difference between the QAIC c value of that model and the smallest QAIC c value within the model set being used for comparisons (i.e. the best model will have a ΔQAIC c of zero). w represents the Akaike weight of each model. Rank 2011 Model QAIC c ΔQAIC c w 1 Site + Size (mm) 717.087 0.000 0.968 2 Site 724.951 7.864 0.019 3 Temperature ( C) + VPD (kpa) + Temperature*VPD 727.131 10.045 0.006 4 Size (mm) 727.958 10.872 0.004 5 Temperature ( C)+ VPD (kpa) 730.935 13.848 0.001 6 Temperature ( C) 732.420 15.333 0.000 7 Emersion Time (hh:mm) 732.835 15.748 0.000 8 VPD (kpa) 733.626 16.540 0.000 9 Constant Only 739.099 22.012 0.000 Rank 2012 1 Site + Size (mm) 1288.781 0.000 0.993 2 Site 1298.601 9.820 0.007 3 Size (mm) 1323.373 34.592 0.000 4 Temperature ( C) + VPD (kpa) + Temperature*VPD 1327.354 38.573 0.000 5 Temperature ( C) + VPD (kpa) 1333.160 44.379 0.000 6 VPD (kpa) 1337.013 48.232 0.000 7 Temperature ( C) 1338.644 49.863 0.000 8 Constant Only 1339.326 50.545 0.000 9
Table S8. Components and fit of the accepted best model from the set evaluating microhabitat selection. In 2011 the calculated odds ratio for site uses NP as the reference and PP as the response, whereas in 2012 it uses CB as the reference and both NP and PP as the responses. The Odds q1 Ratio for size is in response to a one millimeter increase in size. ROC represents the area found under an ROC curve for a given model. Parameter Model Component: Coefficient Odds Ratio Model Fit: ROC Classification Success 2011 Next to Barnacle 0.748 0.755 constant 3.453 Site 1.438 4.210 Size -1.268 0.281 Barnacle Shell 0.701 0.672 constant 2.399 Site 1.176 3.241 Size -1.050 0.350 Between Barnacles 0.756 0.844 constant 4.174 Site 1.554 4.731 Size -1.232 0.292 Other 0.665 0.650 constant 2.138 Site 0.931 2.536 constant 2.138 2012 Next to Barnacle 0.586 0.921 constant 3.706 Site (PP) 0.355 1.427 Site (NP) -0.103 0.902 Size -0.52 0.594 Barnacle Shell 0.673 0.701 constant 2.534 Site (PP) 1.020 2.772 Site (NP) -0.037 0.963 Size -0.810 0.445 Between Barnacles 0.731 0.845 constant 3.836 Site (PP) 1.455 4.285 Site (NP) -0.063 0.939 Size -1.094 0.335 Other 0.588 0.637 constant 0.367 Site (PP) 0.071 1.074 Site (NP) 0.574 1.776 Size -0.441 0.643 10
Table S9. 2011 and 2012 rankings for the set of models representing multiple working hypotheses used to assess activity level. All values are calculated as described in Anderson (2008). For 2011 model set, n = 1368 and ĉ = 34.282 (ĉ = estimated overdispersion parameter). QAIC c is the corrected value for AIC c, ΔQAIC c indicates the difference between the QAIC c value of that model and the smallest QAIC c value within the model set being used for comparisons (i.e. the best model will have a ΔQAIC c of zero). w represents the Akaike weight of each model. Rank 2011 Model QAIC c ΔQAIC c w 1 Temperature ( C) + VPD (kpa) + Temperature*VPD 38.992 0.000 0.372 2 Temperature ( C)+ VPD (kpa) 39.346 0.354 0.312 3 Emersion Time (hh:mm) 41.558 2.566 0.103 4 Temperature ( C) 41.750 2.758 0.094 5 Site + Size (mm) 43.979 4.987 0.031 6 Size (mm) 44.153 5.161 0.028 7 Site 44.639 5.647 0.022 8 VPD (kpa) 44.921 5.929 0.019 9 Constant Only 44.957 5.965 0.019 2012 1 Temperature ( C) + VPD (kpa) + Temperature*VPD 42.213 0.000 0.303 2 Temperature ( C) + VPD (kpa) 42.301 0.088 0.290 3 Temperature ( C) 42.665 0.452 0.241 4 VPD (kpa) 45.072 2.859 0.072 5 Site + Size (mm) 46.576 4.363 0.034 6 Site 47.078 4.866 0.027 7 Size (mm) 47.838 5.625 0.018 8 Constant Only 48.182 5.969 0.015 11
Table S10. Components and fit of the accepted best model (incorporating temperature and VPD as independent variables) from the set evaluating activity level. The calculated odds ratio for temperature is for a 10 C increase in temperature. The odds ratio for Vapour Pressure Deficit is for a one kpa increase in pressure deficit. ROC represents the area found under an ROC curve for a given model. Model Component: Model Fit: Parameter Coefficient Odds Ratio ROC Classification Success 2011 constant 3.651 Foot Out 0.761 0.860 Temperature ( C) -0.395 0.674 Vapour Pressure 1.269 3.931 Deficit (kpa) constant 2.921 Moving 0.795 0.957 Temperature ( C) -0.447 0.640 Vapour Pressure 1.548 4.704 Deficit (kpa) 2012 Foot Out 0.749 0.790 constant 5.349 Temperature ( C) -0.344 0.709 Moving 0.818 0.969 constant 6.355 Temperature ( C) -0.507 0.602 12
Table S11. 2011 and 2012 rankings for the set of models representing multiple working hypotheses used to assess aggregation propensity. All values are calculated as described in Anderson (2008). For this model set, n = 1483 and ĉ = 5.121 (ĉ = estimated overdispersion parameter). QAIC c is the corrected value for AIC c, ΔQAIC c indicates the difference between the QAIC c value of that model and the smallest QAIC c value within the model set being used for comparisons (i.e. the best model will have a ΔQAIC c of zero). w represents the Akaike weight of each model. Rank Model QAICc ΔQAICc w 2011 1 Site + Size (mm) 371.320 0.000 0.501 2 Site 371.767 0.446 0.400 3 Emersion Time (hh:mm) 377.600 6.279 0.022 4 Temperature ( C) + VPD (kpa) + Temperature*VPD 377.705 6.385 0.021 5 Temperature ( C) + VPD (kpa) 378.126 6.806 0.017 6 Temperature ( C) 378.165 6.845 0.016 7 VPD (kpa) 379.177 7.856 0.010 8 Size (mm) 379.319 7.999 0.009 9 Constant only 380.673 9.353 0.005 2012 1 Site + Size (mm) 204.879 0.000 0.381 2 Site 205.485 0.605 0.281 3 Temperature ( C) + VPD (kpa) + Temperature*VPD 207.940 3.060 0.082 4 Temperature ( C) + VPD (kpa) 208.096 3.217 0.076 5 Temperature ( C) 208.248 3.369 0.071 6 VPD (kpa) 208.731 3.851 0.056 7 Size (mm) 209.673 4.793 0.035 8 Constant Only 210.912 6.033 0.019 13
Table S12. Components and fit of the accepted best model (incorporating Site as the only independent variable) from the set of models evaluating aggregation propensity. ROC represents the area found under an ROC curve for a given model. Model Components: Model Fit: Parameter Coefficient Odds Ratio ROC Classification Success 2011 constant -0.202 0.591 0.633 Site -0.736 0.479 2012 constant -0.202 0.591 0.633 Site -0.736 0.479 14
Supplement S2. Back-calculation of air temperature from RH and t dp in 2011 Estimation of Air Temperature T c in 2011 In 2011 it was necessary to estimate calculated air temperature (T c ; C) because we had neglected to record it directly. Although we did not record air temperature during field observations in 2011; we did record relative humidity (RH; %) and dewpoint (t dp ; C) that had been measured with the same probe. T c was estimated using equations taken from Lawrence (2005) that had been rearranged so that air temperature could be back-calculated from RH and t dp : where: Tc = Bz ( dp ) A B+ t z (S1) RH RH z = Bln tdp ln + Atdp (S2) 100 100 The values for the constants A=17.625 and B=243.04 C are recommended by Alduchov & Eskridge (1996; found in Lawrence 2005). Several of very large values of VPD obtained using back-calculated air temperatures were suspected of being outliers. To identify outliers back-calculated air temperature values from Eq. S1 were plotted against the air temperatures recorded at the geographically closest weather station at BMSC: http://www.islandweather.ca/dataavgs.php?field=temperature&interval=60&id=161&year=2011 &month=5 Points with a residual (estimated from a least squares linear regression; p<0.001) greater than ten were excluded from the subsequent logistic regression analyses. 15
Validation of VPD c from Calculated Air Temperature using the 2012 dataset Introduction The increase in activity level of Littorina subrotundata in 2011 in response to increased desiccation stress - represented by vapour pressure deficit (VPD) - was counter to a large body of previous studies on other species. There was concern that this discrepancy was due to the back-calculation method used to calculate T c that was used in the VPD calculations. Fortunately we were able to use the 2012 dataset to test whether this method of VPD calculation could have contributed to erroneous results. In 2012 both direct air temperature and back-calculated air temperature could be used to calculate VPD. Methods Measurements of environmental variables were taken between July 23, 2012 and August 13, 2012 at NP, PP and CB (see text). Measurements of RH and t dp were taken using a Panther brand hand held meter and recorded (see Methods text). At the same time and using the same probe, measurements of air temperature were also recorded. RH, DP, and measured air temperature were input into a spreadsheet and used to calculate VPD. The method of calculating VPD using back-calculated air temperature from Eqs. A1 and A2 above (referred to as VPD c ). These values were compared to a standard method of calculating VPD (referred to as VPD m ) that makes use of measured air temperature and RH. First, saturation vapour pressure (e s ) in Pascals was calculated using calculated air temperature input into equation 1 (Methods text of main manuscript). Second, the vapour pressure of the air (e a ) is calculated using equation 2 (Methods text). Finally VPD m was then calculated using equations 1-3 (Methods text). After VPD m and VPD c had been calculated, an ordinary least squares regression was performed in SYSTAT (Version 13.00.05) using VPD m as the predictor variable and VPD c as the response variable. Results It was found that values of VPD c calculated using RH and t dp as described in this study were very similar to VPD m values calculated using measured air temperature and RH. VPD m was found to be highly correlated with VPD c (r=0.998, df =1, 467, p<0.001; Fig. S3). Values of VPD m were consistently slightly lower than VPD c (slope β=0.909, constant=11.3, Fig. S3). 16
Discussion The high correlation between VPD m and VPD c suggests that the back-calculation method used to calculate VPD in 2011 in this study did not influence the analysis of L. subrotundata behaviour in response to this environmental variable. Abnormally high values of VPD c from the 2011 dataset used for logistic regression in this study therefore cannot be attributed to the calculations used to obtain these values. These results suggest that the response of activity level in L. subrotundata to VPD was not being influenced by the method used to calculate VPD. Figure S3. VPD c obtained using air temperature that was back-calculated from RH and t dp versus VPD m obtained using measured air temperature using the summer 2012 dataset. Note the presence of several points with larger than expected deviations from the regression line. LITERATURE CITED Anderson DR (2008) Model based inference in the life sciences: a primer on evidence. Springer Science+Business Media, LLC, New York, NY. Lawrence MG (2005) The relationship between relative humidity and the dewpoint temperature in moist air: a simple conversion and applications. Bull Am Meteorol Soc 86: 225-233. 17