Supplementary Information: Predator decline leads to decreased stability in a coastal fish community Gregory Lee Britten 1,2 * Michael Dowd 2 Cóilín Minto 3 Francesco Ferretti 4 Ferdinando Boero 5 Heike K. Lotze 1 1 Department of Biology, Dalhousie University, Halifax, Nova Scotia, Canada 2 Department of Mathematics and Statistics, Dalhousie University, Halifax, Nova Scotia, Canada 3 Marine and Freshwater Research Centre, Galway-Mayo Institute of Technology, Galway, Ireland 4 Hopkins Marine Station, Stanford University, Pacific Grove, California, USA 5 DiSteBA, Universita' del Salento, Lecce, Italy; CNR-ISMAR, Italy * gregbritten@dal.ca Appendix A Methods Formula S1. Formula for 𝐕∞ The MAR(1) model has a stationary distribution which is multivariate normal with mean vector 𝛍∞ and covariance matrix 𝐕∞ given by 𝛍∞ = (𝐈 − 𝐁)−𝟏 , (S1) 𝐕∞ = 𝐁𝐕∞ 𝐁′ + 𝚺, where 𝐈 is the identity matrix, 𝐁 is the autoregressive matrix and 𝚺 is the environmental covariance matrix. An explicit formula for 𝐕∞ is given by Vec(𝐕∞ ) = (𝐈 − 𝐁⨂𝐁) −1 Vec(𝚺), (6) where the Vec operator stacks the columns of a matrix atop one another with the first column on top and last column on bottom, and ⨂ is the Kronecker product. Table S1. All species caught in the Camogli trap along with their trophic group denomination, generation time (in years) and fecundity (# eggs). Generation time and fecundity estimates were extracted from FishBase (Froese & Pauly, 2011) as available (symbol – represents unavailable data). The * symbol represents species with relatively uncertain trophic denomination. Scientific Name Common Name Alosa agone Boops boops Clupea harengus* Diplodus puntazzo* Diplodus vulgaris Hirundichthys rondeletii Luvarus imperialis* Oblada melanura* Pagellus acarne Sardina pilchardus Sarpa salpa Scomber colias Scomber scombrus Scomberesox saurus saurus Sparus aurata Spicara maena Spicara smaris Trachurus trachurus* Auxis rochei rochei Belone belone Cetorhinus maximus* Coryphaena hippurus Dasyatis pastinaca Dentex dentex Euthynnus alletteratus Katsuwonus pelamis Loligo vulgaris Lophius piscatorius Merluccius merluccius Micromesistius poutassou* Mobula mobular Mola mola Mustelus mustelus Pomatomus saltatrix Raja clavata Sepia officinalis Todarodes sagittatus Umbrina cirrosa Agone Bogue Atlantic herring Sharpsnout seabream Common two-banded seabream Black wing flyingfish Luvar Saddled seabream Axillary seabream European pilchard Salema Atlantic chub mackerel Atlantic mackerel Atlantic saury Gilthead seabream Blotched picarel Picarel Atlantic horse mackerel Bullet tuna Garfish Basking shark Common dolphinfish Common stingray Common dentex Little tunny Skipjack tuna European squid Angler European hake Blue whiting Devil fish Ocean sunfish Smooth-hound Bluefish Thornback ray Common cuttlefish European flying squid Shi drum Tropic Group Guild 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 Generation Time 4 4 5.6 2.4 2 0.7 4 3.8 4 3.6 0.8 3.6 1.4 2.4 6.6 6.1 7.7 4.6 3.1 1.5 8.9 5.1 2.1 8.6 6.5 2.4 4.8 8.0 4.8 10.7 3.4 Fecundity 195,000 395,000 213,448 156,525 417,421 104,508 5,916 1 299,684 5 395,221 400,000 1,000,000 3,741,657 30,000 300,000,00 80 1,000,000 94 - Zeus faber Alopias vulpinus Echinorhinus brucus Galeorhinus galeus Isurus oxyrinchus Lichia amia* Prionace glauca Sarda sarda* Sphyrna zygaena Thunnus alalunga* Thunnus thynnus Xiphias gladius John dory Thresher shark Bramble shark Tope shark Shortfin mako shark Leerfish Blue shark Atlantic bonito Smooth hammerhead shark Albacore Atlantic bluefin tuna Swordfish 2 3 3 3 3 3 3 3 3 3 3 3 3.4 12.4 8.6 15 9.0 7.6 4.2 15.6 8.0 13.2 8.6 3 18 8 23 2,449,490 10,000,000 5,385,165 Table S2. R code for calculating the three stability metrics # Code to loop through years, estimate B matrix, stability, and perform bootstrapping # Note ‘data’ is a Nx3 data matrix for(i in 1:length(years)) { y <- log(data[data$year==years[i],][2:4]) T <- nrow(y) X <- as.matrix(y[1:T-1,]) Y <- as.matrix(y[2:T,]) # Estimate AR(1) coefficients B<-solve((t(X)%*%X))%*%(t(X)%*%Y) B1s[[i]] <- B # Calculate the residual Y_hat <- X%*%B e <- Y - Y_hat # Bootstrap for uncertainty for(j in 1:500) { # Resample residuals e_boot <- e[sample(1:nrow(e), replace=TRUE),] T <- nrow(e_boot) x0 <- y[1,] Y_boot <- matrix(NA, T, 3) Y_boot[1,] <- as.numeric(x0) # Add resampled residuals back for(k in 2:T) Y_boot[k,] <- Y_boot[k-1,]%*%B + e_boot[k,] X <- as.matrix(Y_boot[1:(T-1),]) Y <- as.matrix(Y_boot[2:(T),]) B_boot <- pseudoinverse((t(X)%*%X))%*%(t(X)%*%Y) # Use pre-defined stability function to calculate metrics return_time1[i,j] <- return.func(B_boot,T,3) eigen1[i,j] <- dominant.eigen(B_boot) sp_var1[i,j] <- species.var.func(B_boot,T,3) } } # Pre-defined stability functions ################# #-Resistance-#### ################# resistance <- function(B,T,p) { I2 <- diag(p^2) coef <-(1/(T-p-1)) res <- Y-(X%*%B) col <- t(res)%*%res cov <- coef*col varvec <- solve(I2-(kronecker(B,B)))%*%as.vector(cov) var <- matrix(varvec,ncol=3,byrow=F) pro <- det(var-cov)/det(var) return(pro) } #################### # - Resilience - ### #################### resilience <- function(B) { eigen <- eigen(B) maxeigen <- max(abs((eigen$values))) index <- which.max(abs(Re(eigen$values))) maxvec <- eigen$vectors[,index] return(maxeigen) } ##################### # - Reactivity - #### ##################### reactivity <- function(B,T,p) { coef <- (1/(T-p-1)) res <- (Y-(X%*%B)) col <- t(res)%*%res cov <- coef*col I <- diag(p) I <- diag(p^2) varvec <- solve(I2-(kronecker(B,B)))%*%as.vector(cov) var < -matrix(varvec,ncol=3,byrow=F) vartrace <- sum(diag(var)) covtrace <- sum(diag(cov)) return <- -(covtrace/vartrace) return(return) } Figure S1. Kernel density estimates for the distribution of species generation times (in years), according to trophic denomination. Note that the generation times vary several-fold across all trophic levels. Figure S2. Single year example of the raw catch data (Panel A) and the log-transformed, demeaned CPUE data (log(KG)/day; Panel B), as analyzed in the paper. Figure S3. The quantile-quantile (QQ) plot of residuals for the autoregressive fit, separated by trophic level time series. The statistical assumption of normality appears supported. Figure S4. The multivariate auto-correlation function (ACF) for the set of estimated autoregressive model residuals (n=7577, p=3), note HP = high-predators; MP = meso-predators; and LT = low-trophic. Results indicate no significant autocorrelation (diagonal plots) or crosscorrelation (off-diagonal plots). Figure S5. Four additional ecological and environmental variables tested as potential drivers of stability: A) Relative fishing effort series formed by combining two records of # of registered fishing boats in the Ligurian Sea area (1950-1964) and total fishing vessel horsepower (19651972; ISTAT 1972); B) Sea surface temperature (SST) within the Ligurian Sea from 1950-1972; and inset E) SST over 1900-2000 (University of Maryland, 2014); C) Species richness R (# of species caught per year); D) Diversity of the catch, measured by Shannon’s index ∑𝑅𝑖=1 𝑝𝑖 ln 𝑝𝑖 , where 𝑅 is the number of species. Figure S6. Estimated time series of species richness (R) and species diversity (H) by trophic group. Figure S7. Selected species-specific time series of mean annual CPUE. Sarpa salpa, Oblada melanura, Todarodes sagittatus and Loligo vulgaris are among the most frequently caught low trophic species which all increase several fold in abundance as predatory sharks (including Sphyrna zygaena, Prionace glauca, Isurus oxyrinchus, Galeorhinus galeus, Echinorhinus brucu, and Alopias vulpinus) decrease dramatically and are largely absent from the trap by 1974. Figure S8. Results from the leave-one-out stability analysis. Stability metrics were calculated three times by leaving a trophic level out of the CPUE data, estimating the stability of the bivariate community, and correlating it against the excluded trophic level. Panels A-C give the resilience metric; panels D-F the reactivity metric; and G-I the resistance metric (see text for definitions). The sample correlation (r) is given in each plot. HP = high-predators; MP = mespredators; and LT = low trophic. Figure S9: Results from a multiple regression analysis across three individual stability metrics (Resilience, Reactivity, and Resistance). Covariates included species richness (total # of species per year), species diversity (Shannon index value), fishing effort (combined record of total number of regional fishing boats and horsepower), and SST (sea surface temperature of the Ligurian Sea). Results are shown for the full-model including all variables; however, only Effort was retained in AIC model selection for each metric. Figure S10. Estimated time series for the coefficient of variation (CV) of the CPUE data. CV was calculated per year, over the 25 years. Dashed lines give the CV for each individual trophic group and solid black line gives the CV for total CPUE. All trends are statistically insignificant (P > 0.1). Figure S11. Landings (A) and proportional catch (B) data for the Mediterranean Sea over the period 1950-2010 (Source: Sea Around Us database, http://www.seaaroundus.org/lme/26.aspx). The time between dashed lines represents the study period (1950-1974) analyzed in the paper.