#Code is designed to run one species at a time. So focus on one species and run through each section for that species. #Species datasets read in should already have all environmental data assigned to each set setwd("~/BLLOP_Data") library("mgcv") library("doBy") library("glmmTMB") library("nlme") library("HH") library('bbmle') library("rgdal") library("ncf") library("mgcv.helper") library("geosphere") library(tidyverse) library(plyr) library("raster") library(ggplot2) library(sf) library(rnaturalearth) library(rnaturalearthdata) library(ggspatial) library(viridis) library(gstat)#if doesn't load, uninstall gstat and xts. then install xts with depencies, then gstat without dependencies source("http://highstat.com/Books/BGS/GAMM/RCodeP2/HighstatLibV6.R") #for VIF ##################### # Functions ##################### #dispersion disp2=function(mod=NA){ E=residuals(mod,type='pearson') d=sum(E^2)/mod$df.resid return(d) } #Check for correlations in covariates panel.cor <- function(x, y, digits = 2, prefix = "", cex.cor, ...) { usr <- par("usr"); on.exit(par(usr)) par(usr = c(0, 1, 0, 1)) r <- abs(cor(x, y)) txt <- format(c(r, 0.123456789), digits = digits)[1] txt <- paste0(prefix, txt) if(missing(cex.cor)) cex.cor <- 0.8/strwidth(txt) text(0.5, 0.5, txt, cex = cex.cor * r) } ######################## #Read in species dataset ######################## #SB sb_data<-read.csv("sb_bll_Catch_full.csv") sb_data<-subset(sb_data,select=-c(X.2,X.1,X)) sb_data$BINOMIAL<-ifelse(sb_data$Catch>0,1,0) sb_data$BEGIN_SET_DATE_TIME<-as.POSIXct(sb_data$BEGIN_SET_DATE_TIME,format="%Y-%m-%d %H:%M:%S",tz="America/New_York") sb_data<-sb_data[with(sb_data,order(BEGIN_SET_DATE_TIME)),] #remove data west of florida sb_data<-sb_data[which(sb_data$BEGIN_SET_LONGITUDE> -80.5 | (sb_data$BEGIN_SET_LATITUDE>26.7 & sb_data$BEGIN_SET_LONGITUDE> -81.5)),]#need to do this because unlike POP data, some BLL sets occur west of -80.5 but are still in atlantic (e.g. off of jacksonville) sb_data$MONTH<-as.factor(strftime(sb_data$BEGIN_SET_DATE_TIME_UTC,format="%m",tz="America/New_York")) sb_data$YEAR<-as.factor(strftime(sb_data$BEGIN_SET_DATE_TIME_UTC,format="%Y",tz="America/New_York")) #remove 9 sets where soak time is 0 sb_data<-sb_data[-which(sb_data$SOAK_DURATION==0),] sb_data$logEFFORT<-log(sb_data$NUM_HOOKS_SET*sb_data$SOAK_DURATION) sb_data$SPECIES<-"sb" sb_data$Date<-as.Date(sb_data$BEGIN_SET_DATE_TIME,format="%Y-%m-%d",origin="1970-01-01",tz="America/New_York") sb_data$Day<-as.numeric(strftime(sb_data$BEGIN_SET_DATE_TIME,format="%j")) sb_data$Set_Begin_Hour<-as.numeric(strftime(round(as.POSIXct(sb_data$BEGIN_SET_DATE_TIME,format="%m/%d/%Y %H:%M",origin="1970-01-01",tz="America/New_York"), units="hours"),format="%H")) sb_data$BEGIN_SET_TEMPERATURE_C<-(sb_data$BEGIN_SET_TEMPERATURE-32)*5/9 #surface temp taken from bottom of machine sb_data$MEAN_DEPTH<-(sb_data$BOTTOM_DEPTH_MAXIMUM_METER+sb_data$BOTTOM_DEPTH_MINIMUM_METER)/2 #SRF_TRIP is whether set is in shark research fishery or not #Region fished, 22 is south atlantic, 23 is north atlantic, 21 is gulf of mexico sb_datanona<-subset(sb_data, select=c(Catch,BINOMIAL,SRF_TRIP,REGION_FISHED,BEGIN_SET_DATE_TIME,Date,Day,MONTH,YEAR,Set_Begin_Hour,BAITTYPE,HOOK_CONFIG,MEAN_DEPTH,BOTTOM_DEPTH_MAXIMUM_METER,BOTTOM_DEPTH_MINIMUM_METER,BEGIN_SET_TEMPERATURE_C,logEFFORT, BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE,BATHYMETRY,RUGOSITY,BT,BS,SST,SSS,SSH,BTSD,SSTSD,TURB,TURB_ER,CHLA)) #some sets with missing env data (only 14 sets) sb_datanona<-sb_datanona[complete.cases(sb_datanona[c(10:12,17:31)]),] #make sure mean depth (from set) and Bathymetry are similar, remove sets that are not the similar #first convert depths from fathoms to meters (conversion factor is 1.83 m to 1 fathom), also need depth to be negative value sb_datanona$MEAN_DEPTH<-sb_datanona$MEAN_DEPTH*-1 #average between depth at beginning and end of set sb_datanona$BOTTOM_DEPTH_MAXIMUM_METER<-sb_datanona$BOTTOM_DEPTH_MAXIMUM_METER*-1 sb_datanona$BOTTOM_DEPTH_MINIMUM_METER<-sb_datanona$BOTTOM_DEPTH_MINIMUM_METER*-1 par(mfrow=c(2,2)) plot(sb_datanona$BATHYMETRY,sb_datanona$MEAN_DEPTH,ylim=c(-350,0)) abline(a=0,b=1) plot(sb_datanona$BATHYMETRY,sb_datanona$BOTTOM_DEPTH_MAXIMUM_METER,ylim=c(-350,0)) abline(a=0,b=1) plot(sb_datanona$BATHYMETRY,sb_datanona$BOTTOM_DEPTH_MINIMUM_METER,ylim=c(-350,0)) abline(a=0,b=1) #decided to go with mean depth because bathymetry data could have been collected on edge or deeper which means it could be off from min or max, mean should be better which(abs(sb_datanona$BATHYMETRY-sb_datanona$MEAN_DEPTH)>100)#Looks like 4 sets are off by 100 m, understand they could be fishing on ledge but 150 is a large difference sb_datanona<-sb_datanona[-which(abs(sb_datanona$BATHYMETRY-sb_datanona$MEAN_DEPTH)>100),]#this removes 4 sets that off by 100m and 2 sets that have NAs for Mean Depth #1123 rows #DS ds_data<-read.csv("ds_bll_Catch_full.csv") ds_data<-subset(ds_data,select=-c(X.2,X.1,X)) ds_data$BINOMIAL<-ifelse(ds_data$Catch>0,1,0) ds_data$BEGIN_SET_DATE_TIME<-as.POSIXct(ds_data$BEGIN_SET_DATE_TIME,format="%Y-%m-%d %H:%M:%S",tz="America/New_York") ds_data<-ds_data[with(ds_data,order(BEGIN_SET_DATE_TIME)),] #remove data west of florida ds_data<-ds_data[which(ds_data$BEGIN_SET_LONGITUDE> -80.5 | (ds_data$BEGIN_SET_LATITUDE>26.7 & ds_data$BEGIN_SET_LONGITUDE> -81.5)),]#need to do this because unlike POP data, some BLL sets occur west of -80.5 but are still in atlantic (e.g. off of jacksonville) ds_data$MONTH<-as.factor(strftime(ds_data$BEGIN_SET_DATE_TIME_UTC,format="%m",tz="America/New_York")) ds_data$YEAR<-as.factor(strftime(ds_data$BEGIN_SET_DATE_TIME_UTC,format="%Y",tz="America/New_York")) #remove 9 sets where soak time is 0 ds_data<-ds_data[-which(ds_data$SOAK_DURATION==0),] ds_data$logEFFORT<-log(ds_data$NUM_HOOKS_SET*ds_data$SOAK_DURATION) ds_data$SPECIES<-"ds" ds_data$Date<-as.Date(ds_data$BEGIN_SET_DATE_TIME,format="%Y-%m-%d",origin="1970-01-01",tz="America/New_York") ds_data$Day<-as.numeric(strftime(ds_data$BEGIN_SET_DATE_TIME,format="%j")) ds_data$Set_Begin_Hour<-as.numeric(strftime(round(as.POSIXct(ds_data$BEGIN_SET_DATE_TIME,format="%m/%d/%Y %H:%M",origin="1970-01-01",tz="America/New_York"), units="hours"),format="%H")) ds_data$BEGIN_SET_TEMPERATURE_C<-(ds_data$BEGIN_SET_TEMPERATURE-32)*5/9 #surface temp taken from bottom of machine ds_data$MEAN_DEPTH<-(ds_data$BOTTOM_DEPTH_MAXIMUM_METER+ds_data$BOTTOM_DEPTH_MINIMUM_METER)/2 #SRF_TRIP is whether set is in shark research fishery or not #Region fished, 22 is south atlantic, 23 is north atlantic, 21 is gulf of mexico ds_datanona<-subset(ds_data, select=c(Catch,BINOMIAL,SRF_TRIP,REGION_FISHED,BEGIN_SET_DATE_TIME,Date,Day,MONTH,YEAR,Set_Begin_Hour,BAITTYPE,HOOK_CONFIG,MEAN_DEPTH,BOTTOM_DEPTH_MAXIMUM_METER,BOTTOM_DEPTH_MINIMUM_METER,BEGIN_SET_TEMPERATURE_C,logEFFORT, BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE,BATHYMETRY,RUGOSITY,BT,BS,SST,SSS,SSH,BTSD,SSTSD,TURB,TURB_ER,CHLA)) #some sets with missing env data (only 14 sets) ds_datanona<-ds_datanona[complete.cases(ds_datanona[c(10:12,17:31)]),] #make sure mean depth (from set) and Bathymetry are similar, remove sets that are not the similar #first convert depths from fathoms to meters (conversion factor is 1.83 m to 1 fathom), also need depth to be negative value ds_datanona$MEAN_DEPTH<-ds_datanona$MEAN_DEPTH*-1 #average between depth at beginning and end of set ds_datanona$BOTTOM_DEPTH_MAXIMUM_METER<-ds_datanona$BOTTOM_DEPTH_MAXIMUM_METER*-1 ds_datanona$BOTTOM_DEPTH_MINIMUM_METER<-ds_datanona$BOTTOM_DEPTH_MINIMUM_METER*-1 par(mfrow=c(2,2)) plot(ds_datanona$BATHYMETRY,ds_datanona$MEAN_DEPTH,ylim=c(-350,0)) abline(a=0,b=1) plot(ds_datanona$BATHYMETRY,ds_datanona$BOTTOM_DEPTH_MAXIMUM_METER,ylim=c(-350,0)) abline(a=0,b=1) plot(ds_datanona$BATHYMETRY,ds_datanona$BOTTOM_DEPTH_MINIMUM_METER,ylim=c(-350,0)) abline(a=0,b=1) #decided to go with mean depth because bathymetry data could have been collected on edge or deeper which means it could be off from min or max, mean should be better which(abs(ds_datanona$BATHYMETRY-ds_datanona$MEAN_DEPTH)>100)#Looks like 4 sets are off by 100 m, understand they could be fishing on ledge but 150 is a large difference ds_datanona<-ds_datanona[-which(abs(ds_datanona$BATHYMETRY-ds_datanona$MEAN_DEPTH)>100),]#this removes 4 sets that off by 100m and 2 sets that have NAs for Mean Depth #1123 rows #SHH shh_data<-read.csv("shh_bll_Catch_full.csv") shh_data<-subset(shh_data,select=-c(X.2,X.1,X)) shh_data$BINOMIAL<-ifelse(shh_data$Catch>0,1,0) shh_data$BEGIN_SET_DATE_TIME<-as.POSIXct(shh_data$BEGIN_SET_DATE_TIME,format="%Y-%m-%d %H:%M:%S",tz="America/New_York") shh_data<-shh_data[with(shh_data,order(BEGIN_SET_DATE_TIME)),] #remove data west of florida shh_data<-shh_data[which(shh_data$BEGIN_SET_LONGITUDE> -80.5 | (shh_data$BEGIN_SET_LATITUDE>26.7 & shh_data$BEGIN_SET_LONGITUDE> -81.5)),]#need to do this because unlike POP data, some BLL sets occur west of -80.5 but are still in atlantic (e.g. off of jacksonville) shh_data$MONTH<-as.factor(strftime(shh_data$BEGIN_SET_DATE_TIME_UTC,format="%m",tz="America/New_York")) shh_data$YEAR<-as.factor(strftime(shh_data$BEGIN_SET_DATE_TIME_UTC,format="%Y",tz="America/New_York")) #remove 9 sets where soak time is 0 shh_data<-shh_data[-which(shh_data$SOAK_DURATION==0),] shh_data$logEFFORT<-log(shh_data$NUM_HOOKS_SET*shh_data$SOAK_DURATION) shh_data$SPECIES<-"shh" shh_data$Date<-as.Date(shh_data$BEGIN_SET_DATE_TIME,format="%Y-%m-%d",origin="1970-01-01",tz="America/New_York") shh_data$Day<-as.numeric(strftime(shh_data$BEGIN_SET_DATE_TIME,format="%j")) shh_data$Set_Begin_Hour<-as.numeric(strftime(round(as.POSIXct(shh_data$BEGIN_SET_DATE_TIME,format="%m/%d/%Y %H:%M",origin="1970-01-01",tz="America/New_York"), units="hours"),format="%H")) shh_data$BEGIN_SET_TEMPERATURE_C<-(shh_data$BEGIN_SET_TEMPERATURE-32)*5/9 #surface temp taken from bottom of machine shh_data$MEAN_DEPTH<-(shh_data$BOTTOM_DEPTH_MAXIMUM_METER+shh_data$BOTTOM_DEPTH_MINIMUM_METER)/2 #SRF_TRIP is whether set is in shark research fishery or not #Region fished, 22 is south atlantic, 23 is north atlantic, 21 is gulf of mexico shh_datanona<-subset(shh_data, select=c(Catch,BINOMIAL,SRF_TRIP,REGION_FISHED,BEGIN_SET_DATE_TIME,Date,Day,MONTH,YEAR,Set_Begin_Hour,BAITTYPE,HOOK_CONFIG,MEAN_DEPTH,BOTTOM_DEPTH_MAXIMUM_METER,BOTTOM_DEPTH_MINIMUM_METER,BEGIN_SET_TEMPERATURE_C,logEFFORT, BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE,BATHYMETRY,RUGOSITY,BT,BS,SST,SSS,SSH,BTSD,SSTSD,TURB,TURB_ER,CHLA)) #some sets with missing env data (only 14 sets) shh_datanona<-shh_datanona[complete.cases(shh_datanona[c(10:12,17:31)]),] #make sure mean depth (from set) and Bathymetry are similar, remove sets that are not the similar #first convert depths from fathoms to meters (conversion factor is 1.83 m to 1 fathom), also need depth to be negative value shh_datanona$MEAN_DEPTH<-shh_datanona$MEAN_DEPTH*-1 #average between depth at beginning and end of set shh_datanona$BOTTOM_DEPTH_MAXIMUM_METER<-shh_datanona$BOTTOM_DEPTH_MAXIMUM_METER*-1 shh_datanona$BOTTOM_DEPTH_MINIMUM_METER<-shh_datanona$BOTTOM_DEPTH_MINIMUM_METER*-1 par(mfrow=c(2,2)) plot(shh_datanona$BATHYMETRY,shh_datanona$MEAN_DEPTH,ylim=c(-350,0)) abline(a=0,b=1) plot(shh_datanona$BATHYMETRY,shh_datanona$BOTTOM_DEPTH_MAXIMUM_METER,ylim=c(-350,0)) abline(a=0,b=1) plot(shh_datanona$BATHYMETRY,shh_datanona$BOTTOM_DEPTH_MINIMUM_METER,ylim=c(-350,0)) abline(a=0,b=1) #decided to go with mean depth because bathymetry data could have been collected on edge or deeper which means it could be off from min or max, mean should be better which(abs(shh_datanona$BATHYMETRY-shh_datanona$MEAN_DEPTH)>100)#Looks like 4 sets are off by 100 m, understand they could be fishing on ledge but 150 is a large difference shh_datanona<-shh_datanona[-which(abs(shh_datanona$BATHYMETRY-shh_datanona$MEAN_DEPTH)>100),]#this removes 4 sets that off by 100m #1123 rows ####################### #Plot location of Catch ####################### #setwd("~/Enviornmental_Data/USmap") #U.S.<-readOGR(".","us_contig_states") #map looks bad #plot(x=1,y=1,ylim=c(5,51.5),xlim=c(-80.2,-37),ylab="",xlab="",type="n") #rect(xleft= -80.5,ybottom=5,xright= -37,ytop=52,col="lightblue") #plot(U.S.,col="tan",add=T) #points(HAUL_LATITUDE~HAUL_LONGITUDE, data=swo_data[which(swo_data$BINOMIAL==0),],pch=16,col="red",cex=0.5) spatialPlot<-function(species="NA"){ year<-unique(species$YEAR) pdf(paste("Spatial.plot",species$SPECIES[1],"pdf",sep="."),width=10,height=10,onefile=TRUE) par(mfrow=c(3,3)) for(i in year){ s<-which(species$YEAR==i) s1<-species[s,] myCex<-s1$BINOMIAL*1.25 plot(BEGIN_SET_LATITUDE~BEGIN_SET_LONGITUDE,data=s1,pch=21,cex=myCex,bg="blue",col="black",main=paste(species$SPECIES[1],i,sep="_"),ylim=c(20,40),xlim=c(-80.4,-74)) points(BEGIN_SET_LATITUDE~BEGIN_SET_LONGITUDE, data=s1[which(s1$BINOMIAL==0),],pch=16,col="red",cex=0.5) } dev.off() } spatialPlot(species=sb_data) spatialPlot(species=ds_data) spatialPlot(species=shh_data) ########################## #Plot Catch and variables ########################## exporePlot<-function(species_dat=NA){ #Year par(mfrow=c(1,1)) plot(table(species_dat$YEAR[which(species_dat$BINOMIAL==1)]),type="b",main=paste("Number of Sets with",species_dat$SPECIES[1], sep="_"),ylab="Number of Sets",ylim=c(0,100)) lines(table(species_dat$YEAR[which(species_dat$BINOMIAL==0)]),type="b",col="red") #Month par(mfrow=c(1,1)) plot(table(species_dat$MONTH[which(species_dat$BINOMIAL==1)]),type="b",main=paste("Number of Sets with",species_dat$SPECIES[1], sep="_"),ylab="Number of Sets",ylim=c(0,100)) lines(table(species_dat$MONTH[which(species_dat$BINOMIAL==0)]),type="b",col="red") #RUGOSITY par(mfrow=c(2,1)) plot(Catch~RUGOSITY,data=species_dat,pch=".",main=paste(species_dat$SPECIES[1],"rug_Catch_all", sep="_")) plot(BINOMIAL~RUGOSITY,data=species_dat,main=paste(species_dat$SPECIES[1],"rug_pres_abs_all", sep="_")) #BATHYMETRY par(mfrow=c(2,1)) plot(Catch~BATHYMETRY,data=species_dat,pch=".",main=paste(species_dat$SPECIES[1],"bat_Catch_all", sep="_")) plot(BINOMIAL~BATHYMETRY,data=species_dat,main=paste(species_dat$SPECIES[1],"bat_pres_abs_all", sep="_")) #BT par(mfrow=c(2,1)) plot(Catch~BT,data=species_dat,pch=".",main=paste(species_dat$SPECIES[1],"bt_Catch_all", sep="_")) plot(BINOMIAL~BT,data=species_dat,main=paste(species_dat$SPECIES[1],"bt_pres_abs_all", sep="_")) #BS par(mfrow=c(2,1)) plot(Catch~BS,data=species_dat,pch=".",main=paste(species_dat$SPECIES[1],"bs_Catch_all", sep="_")) plot(BINOMIAL~BS,data=species_dat,main=paste(species_dat$SPECIES[1],"bs_pres_abs_all", sep="_")) #SST par(mfrow=c(2,1)) plot(Catch~SST,data=species_dat,pch=".",main=paste(species_dat$SPECIES[1],"sst_Catch_all", sep="_")) plot(BINOMIAL~SST,data=species_dat,main=paste(species_dat$SPECIES[1],"sst_pres_abs_all", sep="_")) #SSH par(mfrow=c(2,1)) plot(Catch~SSH,data=species_dat,pch=".",main=paste(species_dat$SPECIES[1],"ssh_Catch_all", sep="_")) plot(BINOMIAL~SSH,data=species_dat,main=paste(species_dat$SPECIES[1],"ssh_pres_abs_all", sep="_")) #SSS par(mfrow=c(2,1)) plot(Catch~SSS,data=species_dat,pch=".",main=paste(species_dat$SPECIES[1],"sss_Catch_all", sep="_")) plot(BINOMIAL~SSS,data=species_dat,main=paste(species_dat$SPECIES[1],"sss_pres_abs_all", sep="_")) #BTSD par(mfrow=c(2,1)) plot(Catch~BTSD,data=species_dat,pch=".",main=paste(species_dat$SPECIES[1],"sBTSD_Catch_all", sep="_")) plot(BINOMIAL~BTSD,data=species_dat,main=paste(species_dat$SPECIES[1],"BTSD_pres_abs_all", sep="_")) #SSTSD par(mfrow=c(2,1)) plot(Catch~SSTSD,data=species_dat,pch=".",main=paste(species_dat$SPECIES[1],"sstsd_Catch_all", sep="_")) plot(BINOMIAL~SSTSD,data=species_dat,main=paste(species_dat$SPECIES[1],"sstsd_pres_abs_all", sep="_")) #TURB par(mfrow=c(2,1)) plot(Catch~TURB,data=species_dat,pch=".",main=paste(species_dat$SPECIES[1],"turb_Catch_all", sep="_")) plot(BINOMIAL~TURB,data=species_dat,main=paste(species_dat$SPECIES[1],"turb_pres_abs_all", sep="_")) #CHLA par(mfrow=c(2,1)) plot(Catch~CHLA,data=species_dat,pch=".",main=paste(species_dat$SPECIES[1],"chla_Catch_all", sep="_")) plot(BINOMIAL~CHLA,data=species_dat,main=paste(species_dat$SPECIES[1],"chla_pres_abs_all", sep="_")) #BAITTYPE par(mfrow=c(1,1)) plot(table(species_dat$BAITTYPE[which(species_dat$BINOMIAL==1)]),type="b",main=paste("Number of Sets with",species_dat$SPECIES[1], sep="_"),ylab="Number of Sets",ylim=c(0,400)) lines(table(species_dat$BAITTYPE[which(species_dat$BINOMIAL==0)]),type="b",col="red") #HOOK_CONFIG par(mfrow=c(1,1)) plot(table(species_dat$HOOK_CONFIG[which(species_dat$BINOMIAL==1)]),type="b",main=paste("Number of Sets with",species_dat$SPECIES[1], sep="_"),ylab="Number of Sets",ylim=c(0,400)) lines(table(species_dat$HOOK_CONFIG[which(species_dat$BINOMIAL==0)]),type="b",col="red") #MEAN_DEPTH par(mfrow=c(2,1)) plot(Catch~MEAN_DEPTH,data=species_dat,pch=".",main=paste(species_dat$SPECIES[1],"mean_depth_Catch_all", sep="_")) plot(BINOMIAL~MEAN_DEPTH,data=species_dat,main=paste(species_dat$SPECIES[1],"mean_depth_pres_abs_all", sep="_")) #Set_Begin_Hour par(mfrow=c(2,1)) plot(Catch~Set_Begin_Hour,data=species_dat,pch=".",main=paste(species_dat$SPECIES[1],"Set_Begin_Hour_Catch_all", sep="_")) plot(BINOMIAL~Set_Begin_Hour,data=species_dat,main=paste(species_dat$SPECIES[1],"Set_Begin_Hour_pres_abs_all", sep="_")) } exporePlot(species_dat=sb_data) exporePlot(species_dat=ds_data) exporePlot(species_dat=shh_data) ######################## #Check for correlation ######################## pdf("multicolBLL.pdf",width=8,height=6) pairs(cbind(as.numeric(ds_datanona$Day),ds_datanona$MONTH,ds_datanona$YEAR, ds_datanona$BEGIN_SET_LATITUDE,ds_datanona$BEGIN_SET_LONGITUDE,ds_datanona$BAITTYPE,ds_datanona$HOOK_CONFIG, ds_datanona$MEAN_DEPTH,ds_datanona$Set_Begin_Hour,ds_datanona$RUGOSITY, ds_datanona$BATHYMETRY,ds_datanona$BT,ds_datanona$BS,ds_datanona$SST,ds_datanona$SSS,ds_datanona$SSH, ds_datanona$SSTSD,ds_datanona$BTSD,ds_datanona$TURB,ds_datanona$CHLA), lower.panel = panel.cor, labels=c("Dayn","Month","Year","Lat","Lon","Bait","Hook","Hook_D","SetH","Rug","Bat","BT","BS","SST","SSS","SSH","SSTSD","BTSD","TURB","CHLA")) dev.off() #there are a good amount of correlations #dayn and year #lat and lon, lat and BS, lat and SSS #lon and and BS, lon and SSS #hook depth and rug, hook depth and bat #rug and bat, rug and BTSD #BS and SSS #TURB and CHLA #remove hook depth #keep rug and bat and try out different models with each #keep SSS and bs and try out different models with each #keep chla and turb and try out different models with each #SB par(mfrow=c(1,1)) plot(table(sb_datanona$Catch),type='h',ylab='Catch (number/set)') table(sb_datanona$BINOMIAL) #postive Catches make up 78% of sets originally, with added absences its 49% #DS par(mfrow=c(1,1)) plot(table(ds_datanona$Catch),type='h',ylab='Catch (number/set)') table(ds_datanona$BINOMIAL) #postive Catches make up 23% of sets #SHH par(mfrow=c(1,1)) plot(table(shh_datanona$Catch),type='h',ylab='Catch (number/set)') table(shh_datanona$BINOMIAL) #postive Catches make up 29% of sets ################ #GAM (BINOMIAL) ################ ##### #SB ##### #1) GLM or GAM sb_datanona$Daten<-as.numeric(sb_datanona$Date) #start with SSS and Rugosity removed, mess with combinations further down modglm<-glm(BINOMIAL~MONTH+YEAR+BAITTYPE+HOOK_CONFIG+Set_Begin_Hour+BATHYMETRY+BT+SST+BS+SSH+CHLA+BTSD+SSTSD+offset(logEFFORT),data=sb_datanona,family='binomial') modgam<-gam(BINOMIAL~MONTH+YEAR+BAITTYPE+HOOK_CONFIG+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=sb_datanona,family='binomial') AIC(modglm,modgam)#gam is way better, go with GAM for here on out #2)check for temporal autocorrelation (ie. independence vs non independence) par(mfrow=c(1,1)) EL1=residuals(modgam, type="pearson") acf(EL1,main="Residuals") pacf(EL1, main="Residuals") #no temporal autocorrelation #VIF z<-cbind(sb_datanona$YEAR,sb_datanona$MONTH,sb_datanona$Day,sb_datanona$BAITTYPE,sb_datanona$HOOK_CONFIG,sb_datanona$Set_Begin_Hour, sb_datanona$BATHYMETRY,sb_datanona$RUGOSITY,sb_datanona$SST, sb_datanona$SSH,sb_datanona$BT,sb_datanona$SSS,sb_datanona$BS,sb_datanona$CHLA,sb_datanona$BTSD,sb_datanona$SSTSD,sb_datanona$TURB) corvif(z) #correlation among Day of year and Month and BS and SSS z<-cbind(sb_datanona$YEAR,sb_datanona$MONTH,sb_datanona$BAITTYPE,sb_datanona$HOOK_CONFIG,sb_datanona$Set_Begin_Hour,sb_datanona$BATHYMETRY,sb_datanona$RUGOSITY,sb_datanona$SST, sb_datanona$SSH,sb_datanona$BT,sb_datanona$BS,sb_datanona$CHLA,sb_datanona$BTSD,sb_datanona$SSTSD,sb_datanona$TURB) corvif(z) #looks like as long as Dayn and Year are not both in and SSS and BS are not both in we are all good, nothing else is over 5 #3) Check for spatial autocorrelation #meaning of correlog plot not clear go with variogram instead #USE VARIOGRAM #try using variogram, have to convert df to spatial df and use geographic crs sb_datanonasp<-sb_datanona coordinates(sb_datanonasp)<- ~BEGIN_SET_LONGITUDE+BEGIN_SET_LATITUDE proj4string(sb_datanonasp)<-CRS("+init=epsg:4326") modgam.1<-gam(BINOMIAL~MONTH+YEAR+BAITTYPE+HOOK_CONFIG+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=sb_datanonasp,family='binomial') EL.1=residuals(modgam.1, type="pearson") vario500<-variogram(EL.1 ~ BEGIN_SET_LONGITUDE+BEGIN_SET_LATITUDE,data=sb_datanonasp,cutoff=500,width=10,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance sbween intervals plot(vario500,main="500_.1") vario250<-variogram(EL.1 ~ BEGIN_SET_LONGITUDE+BEGIN_SET_LATITUDE,data=sb_datanonasp,cutoff=250,width=5,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance sbween intervals plot(vario250,main="250_.1") vario100<-variogram(EL.1 ~ BEGIN_SET_LONGITUDE+BEGIN_SET_LATITUDE,data=sb_datanonasp,cutoff=100,width=1,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance sbween intervals, # 0 = N; .15 = NE; 90 = E; 135 = SE plot(vario100,main="100_.1") #looks ok to me, but lets see if we can improve it modgam2.6<-gam(BINOMIAL~MONTH+YEAR +BAITTYPE+HOOK_CONFIG+s(Set_Begin_Hour)+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+ s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=sb_datanonasp,family='binomial') EL2.6=residuals(modgam2.6, type="pearson") vario500_2.6<-variogram(EL2.6 ~ BEGIN_SET_LONGITUDE + BEGIN_SET_LATITUDE,data=sb_datanonasp,cutoff=500,width=10,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance sbween intervals plot(vario500_2.6,main="500_2.6") vario250_2.6<-variogram(EL2.6 ~ BEGIN_SET_LONGITUDE + BEGIN_SET_LATITUDE,data=sb_datanonasp,cutoff=250,width=5,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance sbween intervals plot(vario250_2.6,main="250_2.6") AIC(modgam.1,modgam2.6) #2.6 has lowest aic #https://stats.stackexchange.com/questions/264702/concurvity-in-negative-binomial-gam concurvity(modgam2.6,full=TRUE)#concurvity is high between lat/lon and bs and bat df = data.frame(Long=sb_datanona$BEGIN_SET_LONGITUDE, Lat=sb_datanona$BEGIN_SET_LATITUDE,Residuals=EL.1) coordinates(df)<-c("Long", "Lat") bubble(df,zcol="Residuals", col=c("orange","blue"),xlab="long",ylab="lat", maxsize=3) #seems pretty evenly distrubtion of pos and neg resisb so no autocorrelation #because of concurvity issues, don't include spatial term #4) 2 Part Variable selection #Part 1 environmental variables mod2<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=sb_datanona,family='binomial') mod3<-gam(BINOMIAL~MONTH+YEAR+s(RUGOSITY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(SSTSD)+offset(logEFFORT),data=sb_datanona,family='binomial')#put rug in for bat, no btsd (correlates with rug) mod4<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(SST)+s(SSS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=sb_datanona,family='binomial')#put sss in for bs mod5<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=sb_datanona,family='binomial')#put in turb for chla mod6<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+offset(logEFFORT),data=sb_datanona,family='binomial')#no sstsd mod7<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+offset(logEFFORT),data=sb_datanona,family='binomial')#no sstsd and btsd mod8<-gam(BINOMIAL~MONTH+YEAR+s(RUGOSITY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+offset(logEFFORT),data=sb_datanona,family='binomial')#no sstsd, with turb and rug, no btsd (correlates with rug) mod9<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(BTSD)+offset(logEFFORT),data=sb_datanona,family='binomial')#no sstsd and chla mod10<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+offset(logEFFORT),data=sb_datanona,family='binomial')#no sstsd and chla and SSH and btsd mod11<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(BTSD)+offset(logEFFORT),data=sb_datanona,family='binomial')#no sstsd and chla and SSH mod12<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=sb_datanona,family='binomial')#no SST mod13<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+offset(logEFFORT),data=sb_datanona,family='binomial')#no SST and sstsd mod14<-gam(BINOMIAL~MONTH+YEAR+s(RUGOSITY)+s(BT)+s(SST)+s(BS)+offset(logEFFORT),data=sb_datanona,family='binomial')#no sstsd and chla and SSH, with rug, no btsd (correlates with rug) mod15<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=sb_datanona,family='binomial')#no SSH, with turb mod16<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=sb_datanona,family='binomial')#no SSH mod17<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(BS)+s(CHLA)+s(BTSD)+offset(logEFFORT),data=sb_datanona,family='binomial')#no SSH, sst, sstsd AICtab(mod2,mod3,mod4,mod5,mod6,mod7,mod8,mod9,mod10,mod11,mod12,mod13,mod14,mod15,mod16,mod17) #best model is mod4 but mod7 is simpler and within 2 AICs so go with mod7 #Part 2 gear variables with best model from above mod7<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+offset(logEFFORT),data=sb_datanona,family='binomial')#no sstsd and btsd mod18<-gam(BINOMIAL~MONTH+YEAR+BAITTYPE+HOOK_CONFIG+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+offset(logEFFORT),data=sb_datanona,family='binomial') mod19<-gam(BINOMIAL~MONTH+YEAR+BAITTYPE+HOOK_CONFIG+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+offset(logEFFORT),data=sb_datanona,family='binomial') mod20<-gam(BINOMIAL~MONTH+YEAR+HOOK_CONFIG+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+offset(logEFFORT),data=sb_datanona,family='binomial') mod21<-gam(BINOMIAL~MONTH+YEAR+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+offset(logEFFORT),data=sb_datanona,family='binomial') mod22<-gam(BINOMIAL~MONTH+YEAR+BAITTYPE+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+offset(logEFFORT),data=sb_datanona,family='binomial') AICtab(mod7,mod18,mod19,mod20,mod21,mod22) #best model is mod21 is best #5) Tweak k values with REML method #if we can find the model that explains the most variation and the curves make biological sense go with that model mod7<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+offset(logEFFORT),data=sb_datanona,family='binomial')#no sstsd and btsd mod7.1<-gam(BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT,k=7)+s(SST)+s(BS,k=7)+s(SSH)+s(CHLA)+offset(logEFFORT),data=sb_datanona,family='binomial')#no sstsd and btsd par(mfrow=c(1,1)) summary(mod7) gam.check(mod7) #good plot(mod7) disp2(mod7) #good # edf Ref.df F p-value #s(BATHYMETRY) 5.884 6.670 62.771 < 2e-16 *** #s(BT) 3.806 4.741 19.185 0.00174 ** #s(SST) 5.892 6.888 16.182 0.01673 * #s(BS) 5.244 6.331 8.501 0.20932 #s(SSH) 8.723 8.965 24.838 0.00284 ** #s(CHLA) 7.774 8.244 58.904 < 2e-16 *** #R-sq.(adj) = 0.47 Deviance explained = 49% m=mod7 #Plot respose curves #to backtransform binomial model use plogis #to backtransform neg bin use exp #Bathymetry xBat<-seq(min(sb_datanona$BATHYMETRY),max(sb_datanona$BATHYMETRY),length=100) p.data<-expand.grid(YEAR=unique(sb_datanona$YEAR),MONTH=levels(sb_datanona$MONTH),BATHYMETRY=xBat,BT=mean(sb_datanona$BT),SST=mean(sb_datanona$SST),BS=mean(sb_datanona$BS), SSS=mean(sb_datanona$SSS),SSH=mean(sb_datanona$SSH),CHLA=mean(sb_datanona$CHLA),BTSD=mean(sb_datanona$BTSD),SSTSD=mean(sb_datanona$SSTSD),Set_Begin_Hour=mean(sb_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(sb_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(sb_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(sb_datanona$logEFFORT)) p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude=NULL)) outBat<-summaryBy(pred~BATHYMETRY,data=p.data) outBat$pred.Trans<-plogis(outBat$pred.mean) #BT xBT<-seq(min(sb_datanona$BT),max(sb_datanona$BT),length=100) p.data<-expand.grid(YEAR=unique(sb_datanona$YEAR),MONTH=levels(sb_datanona$MONTH),BT=xBT,BATHYMETRY=mean(sb_datanona$BATHYMETRY),SST=mean(sb_datanona$SST),BS=mean(sb_datanona$BS), SSS=mean(sb_datanona$SSS),SSH=mean(sb_datanona$SSH),CHLA=mean(sb_datanona$CHLA),BTSD=mean(sb_datanona$BTSD),SSTSD=mean(sb_datanona$SSTSD),Set_Begin_Hour=mean(sb_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(sb_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(sb_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(sb_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude=NULL)) outBT<-summaryBy(pred~BT,data=p.data) outBT$pred.Trans<-plogis(outBT$pred.mean) #SST xSST<-seq(min(sb_datanona$SST),max(sb_datanona$SST),length=100) p.data<-expand.grid(YEAR=unique(sb_datanona$YEAR),MONTH=levels(sb_datanona$MONTH),SST=xSST,BATHYMETRY=mean(sb_datanona$BATHYMETRY),BT=mean(sb_datanona$BT),BS=mean(sb_datanona$BS), SSS=mean(sb_datanona$SSS),SSH=mean(sb_datanona$SSH),CHLA=mean(sb_datanona$CHLA),BTSD=mean(sb_datanona$BTSD),SSTSD=mean(sb_datanona$SSTSD),Set_Begin_Hour=mean(sb_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(sb_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(sb_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(sb_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude=NULL)) outSST<-summaryBy(pred~SST,data=p.data) outSST$pred.Trans<-plogis(outSST$pred.mean) #BS xBS<-seq(min(sb_datanona$BS),max(sb_datanona$BS),length=100) p.data<-expand.grid(YEAR=unique(sb_datanona$YEAR),MONTH=levels(sb_datanona$MONTH),BS=xBS,BATHYMETRY=mean(sb_datanona$BATHYMETRY),BT=mean(sb_datanona$BT),SST=mean(sb_datanona$SST), SSS=mean(sb_datanona$SSS),SSH=mean(sb_datanona$SSH),CHLA=mean(sb_datanona$CHLA),BTSD=mean(sb_datanona$BTSD),SSTSD=mean(sb_datanona$SSTSD),Set_Begin_Hour=mean(sb_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(sb_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(sb_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(sb_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude=NULL)) outBS<-summaryBy(pred~BS,data=p.data) outBS$pred.Trans<-plogis(outBS$pred.mean) #SSS #xSSS<-seq(min(sb_datanona$SSS),max(sb_datanona$SSS),length=100) #p.data<-expand.grid(YEAR=unique(sb_datanona$YEAR),MONTH=levels(sb_datanona$MONTH),SSS=xSSS,BATHYMETRY=mean(sb_datanona$BATHYMETRY),BT=mean(sb_datanona$BT),SST=mean(sb_datanona$SST), # BS=mean(sb_datanona$BS),SSH=mean(sb_datanona$SSH),CHLA=mean(sb_datanona$CHLA),BTSD=mean(sb_datanona$BTSD),SSTSD=mean(sb_datanona$SSTSD),Set_Begin_Hour=mean(sb_datanona$Set_Begin_Hour), # BEGIN_SET_LATITUDE=mean(sb_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(sb_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(sb_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict #p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude=NULL)) #outSSS<-summaryBy(pred~SSS,data=p.data) #outSSS$pred.Trans<-plogis(outSSS$pred.mean) #SSH xSSH<-seq(min(sb_datanona$SSH),max(sb_datanona$SSH),length=100) p.data<-expand.grid(YEAR=unique(sb_datanona$YEAR),MONTH=levels(sb_datanona$MONTH),BS=mean(sb_datanona$BS),BATHYMETRY=mean(sb_datanona$BATHYMETRY),BT=mean(sb_datanona$BT),SST=mean(sb_datanona$SST), SSS=mean(sb_datanona$SSS),CHLA=mean(sb_datanona$CHLA),SSH=xSSH,BTSD=mean(sb_datanona$BTSD),SSTSD=mean(sb_datanona$SSTSD),Set_Begin_Hour=mean(sb_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(sb_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(sb_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(sb_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude=NULL)) outSSH<-summaryBy(pred~SSH,data=p.data) outSSH$pred.Trans<-plogis(outSSH$pred.mean) #CHLA xCHLA<-seq(min(sb_datanona$CHLA),max(sb_datanona$CHLA),length=100) p.data<-expand.grid(YEAR=unique(sb_datanona$YEAR),MONTH=levels(sb_datanona$MONTH),BS=mean(sb_datanona$BS),BATHYMETRY=mean(sb_datanona$BATHYMETRY),BT=mean(sb_datanona$BT),SST=mean(sb_datanona$SST), SSS=mean(sb_datanona$SSS),SSH=mean(sb_datanona$SSH),CHLA=xCHLA,BTSD=mean(sb_datanona$BTSD),SSTSD=mean(sb_datanona$SSTSD),Set_Begin_Hour=mean(sb_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(sb_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(sb_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(sb_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude=NULL)) outCHLA<-summaryBy(pred~CHLA,data=p.data) outCHLA$pred.Trans<-plogis(outCHLA$pred.mean) #SSTSD #xSSTSD<-seq(min(sb_datanona$SSTSD),max(sb_datanona$SSTSD),length=100) #p.data<-expand.grid(YEAR=unique(sb_datanona$YEAR),MONTH=levels(sb_datanona$MONTH),BS=mean(sb_datanona$BS),BATHYMETRY=mean(sb_datanona$BATHYMETRY),BT=mean(sb_datanona$BT),SST=mean(sb_datanona$SST), # SSS=mean(sb_datanona$SSS),SSH=mean(sb_datanona$SSH),SSTSD=xSSTSD,BTSD=mean(sb_datanona$BTSD),CHLA=mean(sb_datanona$CHLA),Set_Begin_Hour=mean(sb_datanona$Set_Begin_Hour), # BEGIN_SET_LATITUDE=mean(sb_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(sb_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(sb_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict #p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude=NULL)) #outSSTSD<-summaryBy(pred~SSTSD,data=p.data) #outSSTSD$pred.Trans<-plogis(outSSTSD$pred.mean) #BTSD #xBTSD<-seq(min(sb_datanona$BTSD),max(sb_datanona$BTSD),length=100) #p.data<-expand.grid(YEAR=unique(sb_datanona$YEAR),MONTH=levels(sb_datanona$MONTH),BS=mean(sb_datanona$BS),BATHYMETRY=mean(sb_datanona$BATHYMETRY),BT=mean(sb_datanona$BT),SST=mean(sb_datanona$SST), # SSS=mean(sb_datanona$SSS),SSH=mean(sb_datanona$SSH),BTSD=xBTSD,SSTSD=mean(sb_datanona$SSTSD),CHLA=mean(sb_datanona$CHLA),Set_Begin_Hour=mean(sb_datanona$Set_Begin_Hour), # BEGIN_SET_LATITUDE=mean(sb_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(sb_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(sb_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict #p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude=NULL)) #outBTSD<-summaryBy(pred~BTSD,data=p.data) #outBTSD$pred.Trans<-plogis(outBTSD$pred.mean) #Set_Begin_Hour #xSet_Begin_Hour<-seq(min(sb_datanona$Set_Begin_Hour),max(sb_datanona$Set_Begin_Hour),length=100) #p.data<-expand.grid(YEAR=unique(sb_datanona$YEAR),MONTH=levels(sb_datanona$MONTH),BS=mean(sb_datanona$BS),BATHYMETRY=mean(sb_datanona$BATHYMETRY),BT=mean(sb_datanona$BT),SST=mean(sb_datanona$SST), # SSS=mean(sb_datanona$SSS),SSH=mean(sb_datanona$SSH),BTSD=mean(sb_datanona$BTSD),SSTSD=mean(sb_datanona$SSTSD),CHLA=mean(sb_datanona$CHLA),Set_Begin_Hour=xSet_Begin_Hour, # BEGIN_SET_LATITUDE=mean(sb_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(sb_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(sb_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict #p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude=NULL)) #outSet_Begin_Hour<-summaryBy(pred~Set_Begin_Hour,data=p.data) #outSet_Begin_Hour$pred.Trans<-plogis(outSet_Begin_Hour$pred.mean) #MONTH p.data<-expand.grid(YEAR=unique(sb_datanona$YEAR),MONTH=levels(sb_datanona$MONTH),BS=mean(sb_datanona$BS),BATHYMETRY=mean(sb_datanona$BATHYMETRY),BT=mean(sb_datanona$BT),SST=mean(sb_datanona$SST), SSS=mean(sb_datanona$SSS),SSH=mean(sb_datanona$SSH),BTSD=mean(sb_datanona$BTSD),SSTSD=mean(sb_datanona$SSTSD),CHLA=mean(sb_datanona$CHLA),Set_Begin_Hour=mean(sb_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(sb_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(sb_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(sb_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude=NULL)) outMONTH<-summaryBy(pred~MONTH,data=p.data) outMONTH$pred.Trans<-plogis(outMONTH$pred.mean) #YEAR outYEAR<-summaryBy(pred~YEAR,data=p.data) outYEAR$pred.Trans<-plogis(outYEAR$pred.mean) par(mfrow=c(3,3)) plot(pred.Trans~BATHYMETRY, data=outBat,type="l",ylim=c(0,1)) plot(pred.Trans~BT, data=outBT,type="l",ylim=c(0,1)) plot(pred.Trans~SST, data=outSST,type="l",ylim=c(0,1)) plot(pred.Trans~BS, data=outBS,type="l",ylim=c(0,1)) plot(pred.Trans~SSH, data=outSSH,type="l",ylim=c(0,1)) plot(pred.Trans~CHLA, data=outCHLA,type="l",ylim=c(0,1)) #plot(pred.Trans~BTSD, data=outBTSD,type="l",ylim=c(0,0.8)) #plot(pred.Trans~SSTSD, data=outSSTSD,type="l",ylim=c(0,0.8)) #plot(pred.Trans~Set_Begin_Hour, data=outSet_Begin_Hour,type="l",ylim=c(0,0.8)) plot(pred.Trans~MONTH, data=outMONTH,type="l",ylim=c(0,1)) plot(pred.Trans~YEAR, data=outYEAR,type="l",ylim=c(0,1)) ##### #DS ##### #after discussion with tobey, feel that it is appropriate to remove 2 sets where SST was 8C and there was a positive catch #because it is likely outside the temp range of duskies and could be misidentification ds_datanona<-ds_datanona[-which(ds_datanona$SST<10 & ds_datanona$BINOMIAL==1),] #1) GLM or GAM #considered day of year instead, but this correlates with month and we need month ds_datanona$Daten<-as.numeric(ds_datanona$Date) #start with SSS and Rugosity removed, mess with combinations further down modglm<-glm(BINOMIAL~MONTH+YEAR+BAITTYPE+HOOK_CONFIG+Set_Begin_Hour+BATHYMETRY+BT+SST+BS+SSH+CHLA+BTSD+SSTSD+offset(logEFFORT),data=ds_datanona,family='binomial') modgam<-gam(BINOMIAL~MONTH+YEAR+BAITTYPE+HOOK_CONFIG+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial') AIC(modglm,modgam)#gam is way better, go with GAM for here on out #2)check for temporal autocorrelation (ie. independence vs non independence) par(mfrow=c(1,1)) EL1=residuals(modgam, type="pearson") acf(EL1,main="Residuals") pacf(EL1, main="Residuals") #no temporal autocorrelation #VIF z<-cbind(ds_datanona$YEAR,ds_datanona$MONTH,ds_datanona$Day,ds_datanona$BAITTYPE,ds_datanona$HOOK_CONFIG,ds_datanona$Set_Begin_Hour, ds_datanona$BATHYMETRY,ds_datanona$RUGOSITY,ds_datanona$SST, ds_datanona$SSH,ds_datanona$BT,ds_datanona$SSS,ds_datanona$BS,ds_datanona$CHLA,ds_datanona$BTSD,ds_datanona$SSTSD,ds_datanona$TURB) corvif(z) #correlation among Day of year and Month and BS and SSS z<-cbind(ds_datanona$YEAR,ds_datanona$MONTH,ds_datanona$BAITTYPE,ds_datanona$HOOK_CONFIG,ds_datanona$Set_Begin_Hour,ds_datanona$BATHYMETRY,ds_datanona$RUGOSITY,ds_datanona$SST, ds_datanona$SSH,ds_datanona$BT,ds_datanona$BS,ds_datanona$CHLA,ds_datanona$BTSD,ds_datanona$SSTSD,ds_datanona$TURB) corvif(z) #looks like as long as Dayn and Year are not both in and SSS and BS are not both in we are all good, nothing else is over 5 #3) Check for spatial autocorrelation #meaning of correlog plot not clear go with variogram instead #USE VARIOGRAM #try using variogram, have to convert df to spatial df and use geographic crs ds_datanonasp<-ds_datanona coordinates(ds_datanonasp)<- ~BEGIN_SET_LONGITUDE+BEGIN_SET_LATITUDE proj4string(ds_datanonasp)<-CRS("+init=epsg:4326") modgam.1<-gam(BINOMIAL~MONTH+YEAR+BAITTYPE+HOOK_CONFIG+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanonasp,family='binomial') EL.1=residuals(modgam.1, type="pearson") vario500<-variogram(EL.1 ~ BEGIN_SET_LONGITUDE+BEGIN_SET_LATITUDE,data=ds_datanonasp,cutoff=500,width=10,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance dsween intervals plot(vario500,main="500_.1") vario250<-variogram(EL.1 ~ BEGIN_SET_LONGITUDE+BEGIN_SET_LATITUDE,data=ds_datanonasp,cutoff=250,width=5,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance dsween intervals plot(vario250,main="250_.1") vario100<-variogram(EL.1 ~ BEGIN_SET_LONGITUDE+BEGIN_SET_LATITUDE,data=ds_datanonasp,cutoff=100,width=1,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance dsween intervals, # 0 = N; .15 = NE; 90 = E; 135 = SE plot(vario100,main="100_.1") #looks ok to me, but lets see if we can improve it modgam2.6<-gam(BINOMIAL~MONTH+YEAR +BAITTYPE+HOOK_CONFIG+s(Set_Begin_Hour)+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+ s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanonasp,family='binomial') EL2.6=residuals(modgam2.6, type="pearson") vario500_2.6<-variogram(EL2.6 ~ BEGIN_SET_LONGITUDE + BEGIN_SET_LATITUDE,data=ds_datanonasp,cutoff=500,width=10,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance dsween intervals plot(vario500_2.6,main="500_2.6") vario250_2.6<-variogram(EL2.6 ~ BEGIN_SET_LONGITUDE + BEGIN_SET_LATITUDE,data=ds_datanonasp,cutoff=250,width=5,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance dsween intervals plot(vario250_2.6,main="250_2.6") AIC(modgam.1,modgam2.6) #2.6 has lowest aic, go with it concurvity(modgam2.6,full=TRUE)#concurvity is high between lat/lon and bs and bat df = data.frame(Long=ds_datanona$BEGIN_SET_LONGITUDE, Lat=ds_datanona$BEGIN_SET_LATITUDE,Residuals=EL.1) coordinates(df)<-c("Long", "Lat") bubble(df,zcol="Residuals", col=c("orange","blue"),xlab="long",ylab="lat", maxsize=3) #seems pretty evenly distrubtion of pos and neg resids so no autocorrelation #decide to keep spatial term to help make other covariates more realistic #4) 2 Part Variable selection #Part 1 environmental variables mod2<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial') mod3<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(RUGOSITY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial')#put rug in for bat, no btsd (correlates with rug) mod4<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(SSS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial')#put sss in for bs mod5<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial')#put in turb for chla mod6<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+offset(logEFFORT),data=ds_datanona,family='binomial')#no sstsd mod7<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+offset(logEFFORT),data=ds_datanona,family='binomial')#no sstsd and btsd mod8<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(RUGOSITY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+offset(logEFFORT),data=ds_datanona,family='binomial')#no sstsd, with turb and rug, no btsd (correlates with rug) mod9<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(BTSD)+offset(logEFFORT),data=ds_datanona,family='binomial')#no sstsd and chla mod10<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+offset(logEFFORT),data=ds_datanona,family='binomial')#no sstsd and chla and SSH and btsd mod11<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(BTSD)+offset(logEFFORT),data=ds_datanona,family='binomial')#no sstsd and chla and SSH mod12<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial')#no SST mod13<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+offset(logEFFORT),data=ds_datanona,family='binomial')#no SST and sstsd mod14<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(RUGOSITY)+s(BT)+s(SST)+s(BS)+offset(logEFFORT),data=ds_datanona,family='binomial')#no sstsd and chla and SSH, with rug, no btsd (correlates with rug) mod15<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial')#no SSH, with turb mod16<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial')#no SSH mod17<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(BS)+s(CHLA)+s(BTSD)+offset(logEFFORT),data=ds_datanona,family='binomial')#no SSH, sst, sstsd AICtab(mod2,mod3,mod4,mod5,mod6,mod7,mod8,mod9,mod10,mod11,mod12,mod13,mod14,mod15,mod16,mod17) #best models is mod16 #Part 2 gear variables with best model from above mod16<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial')#no SSH mod18<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+BAITTYPE+HOOK_CONFIG+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial') mod19<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+BAITTYPE+HOOK_CONFIG+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial') mod20<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+HOOK_CONFIG+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial') mod21<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial') mod22<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+BAITTYPE+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial') AICtab(mod16,mod18,mod19,mod20,mod21,mod22) #best model is mod21 is best #5) Tweak k values #if we can find the model that explains the most variation and the curves make biological sense go with that model mod21<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='binomial') #model is slightly overdispersed, switch family to quasibinomial because dispersion parameter is not fixed so it can model overdispersion #http://coltekin.net/cagri/R/r-exercisesse11.html#:~:text=Overdispersion%20occurs%20when%20error%20(residuals,distribution%20is%20the%20binomial%20distribution.&text=Underdispersion%2C%20detected%20by%20lower%20residual,also%20possible%20but%20more%20rare. mod21.1<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='quasibinomial') mod21.2<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT,k=7)+s(SST,k=6)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='quasibinomial') #SST curves not going down for cold temps and this is causing presence of duskies in gulf of maine in Jan which is incorrect...will move forward with a model without SST mod21.3<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=ds_datanona,family='quasibinomial') #without SST mod21.4<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(Set_Begin_Hour)+s(BATHYMETRY,k=15)+s(BT,k=9)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD,k=6)+offset(logEFFORT),data=ds_datanona,family='quasibinomial') mod21.5<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(Set_Begin_Hour)+s(BATHYMETRY,k=15)+s(BT,k=9)+s(BS)+s(CHLA)+BTSD+s(SSTSD,k=6)+offset(logEFFORT),data=ds_datanona,family='quasibinomial') par(mfrow=c(1,1)) summary(mod21.4) gam.check(mod21.4) #good plot(mod21.4) disp2(mod21.4) # edf Ref.df F p-value #s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE) 21.433 25.278 1.960 0.00346 ** #s(Set_Begin_Hour) 5.949 7.082 3.372 0.00144 ** #s(BATHYMETRY) 3.373 4.298 0.862 0.51938 #s(BT) 5.048 6.109 5.895 5.08e-06 *** #s(BS) 1.586 1.976 0.288 0.72446 #s(CHLA) 5.266 6.268 1.282 0.23447 #s(BTSD) 2.735 3.492 2.001 0.10038 #s(SSTSD) 1.056 1.108 0.403 0.57694 #R-sq.(adj) = 0.312 Deviance explained = 38.3% m=mod21.4 #Plot respose curves #to backtransform binomial model use plogis #to backtransform neg bin use exp #Bathymetry xBat<-seq(min(ds_datanona$BATHYMETRY),max(ds_datanona$BATHYMETRY),length=100) p.data<-expand.grid(YEAR=unique(ds_datanona$YEAR),MONTH=levels(ds_datanona$MONTH),BATHYMETRY=xBat,BT=mean(ds_datanona$BT),SST=mean(ds_datanona$SST),BS=mean(ds_datanona$BS), SSS=mean(ds_datanona$SSS),SSH=mean(ds_datanona$SSH),CHLA=mean(ds_datanona$CHLA),BTSD=mean(ds_datanona$BTSD),SSTSD=mean(ds_datanona$SSTSD),Set_Begin_Hour=mean(ds_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(ds_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(ds_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(ds_datanona$logEFFORT)) p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outBat<-summaryBy(pred~BATHYMETRY,data=p.data) outBat$pred.Trans<-plogis(outBat$pred.mean) #BT xBT<-seq(min(ds_datanona$BT),max(ds_datanona$BT),length=100) p.data<-expand.grid(YEAR=unique(ds_datanona$YEAR),MONTH=levels(ds_datanona$MONTH),BT=xBT,BATHYMETRY=mean(ds_datanona$BATHYMETRY),SST=mean(ds_datanona$SST),BS=mean(ds_datanona$BS), SSS=mean(ds_datanona$SSS),SSH=mean(ds_datanona$SSH),CHLA=mean(ds_datanona$CHLA),BTSD=mean(ds_datanona$BTSD),SSTSD=mean(ds_datanona$SSTSD),Set_Begin_Hour=mean(ds_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(ds_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(ds_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(ds_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outBT<-summaryBy(pred~BT,data=p.data) outBT$pred.Trans<-plogis(outBT$pred.mean) #SST xSST<-seq(min(ds_datanona$SST),max(ds_datanona$SST),length=100) p.data<-expand.grid(YEAR=unique(ds_datanona$YEAR),MONTH=levels(ds_datanona$MONTH),SST=xSST,BATHYMETRY=mean(ds_datanona$BATHYMETRY),BT=mean(ds_datanona$BT),BS=mean(ds_datanona$BS), SSS=mean(ds_datanona$SSS),SSH=mean(ds_datanona$SSH),CHLA=mean(ds_datanona$CHLA),BTSD=mean(ds_datanona$BTSD),SSTSD=mean(ds_datanona$SSTSD),Set_Begin_Hour=mean(ds_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(ds_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(ds_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(ds_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outSST<-summaryBy(pred~SST,data=p.data) outSST$pred.Trans<-plogis(outSST$pred.mean) #BS xBS<-seq(min(ds_datanona$BS),max(ds_datanona$BS),length=100) p.data<-expand.grid(YEAR=unique(ds_datanona$YEAR),MONTH=levels(ds_datanona$MONTH),BS=xBS,BATHYMETRY=mean(ds_datanona$BATHYMETRY),BT=mean(ds_datanona$BT),SST=mean(ds_datanona$SST), SSS=mean(ds_datanona$SSS),SSH=mean(ds_datanona$SSH),CHLA=mean(ds_datanona$CHLA),BTSD=mean(ds_datanona$BTSD),SSTSD=mean(ds_datanona$SSTSD),Set_Begin_Hour=mean(ds_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(ds_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(ds_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(ds_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outBS<-summaryBy(pred~BS,data=p.data) outBS$pred.Trans<-plogis(outBS$pred.mean) #SSS #xSSS<-seq(min(ds_datanona$SSS),max(ds_datanona$SSS),length=100) #p.data<-expand.grid(YEAR=unique(ds_datanona$YEAR),MONTH=levels(ds_datanona$MONTH),SSS=xSSS,BATHYMETRY=mean(ds_datanona$BATHYMETRY),BT=mean(ds_datanona$BT),SST=mean(ds_datanona$SST), # BS=mean(ds_datanona$BS),SSH=mean(ds_datanona$SSH),CHLA=mean(ds_datanona$CHLA),BTSD=mean(ds_datanona$BTSD),SSTSD=mean(ds_datanona$SSTSD),Set_Begin_Hour=mean(ds_datanona$Set_Begin_Hour), # BEGIN_SET_LATITUDE=mean(ds_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(ds_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(ds_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict #p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) #outSSS<-summaryBy(pred~SSS,data=p.data) #outSSS$pred.Trans<-plogis(outSSS$pred.mean) #SSH #xSSH<-seq(min(ds_datanona$SSH),max(ds_datanona$SSH),length=100) #p.data<-expand.grid(YEAR=unique(ds_datanona$YEAR),MONTH=levels(ds_datanona$MONTH),BS=mean(ds_datanona$BS),BATHYMETRY=mean(ds_datanona$BATHYMETRY),BT=mean(ds_datanona$BT),SST=mean(ds_datanona$SST), # SSS=mean(ds_datanona$SSS),CHLA=mean(ds_datanona$CHLA),SSH=xSSH,BTSD=mean(ds_datanona$BTSD),SSTSD=mean(ds_datanona$SSTSD),Set_Begin_Hour=mean(ds_datanona$Set_Begin_Hour), # BEGIN_SET_LATITUDE=mean(ds_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(ds_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(ds_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict #p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) #outSSH<-summaryBy(pred~SSH,data=p.data) #outSSH$pred.Trans<-plogis(outSSH$pred.mean) #CHLA xCHLA<-seq(min(ds_datanona$CHLA),max(ds_datanona$CHLA),length=100) p.data<-expand.grid(YEAR=unique(ds_datanona$YEAR),MONTH=levels(ds_datanona$MONTH),BS=mean(ds_datanona$BS),BATHYMETRY=mean(ds_datanona$BATHYMETRY),BT=mean(ds_datanona$BT),SST=mean(ds_datanona$SST), SSS=mean(ds_datanona$SSS),SSH=mean(ds_datanona$SSH),CHLA=xCHLA,BTSD=mean(ds_datanona$BTSD),SSTSD=mean(ds_datanona$SSTSD),Set_Begin_Hour=mean(ds_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(ds_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(ds_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(ds_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outCHLA<-summaryBy(pred~CHLA,data=p.data) outCHLA$pred.Trans<-plogis(outCHLA$pred.mean) #SSTSD xSSTSD<-seq(min(ds_datanona$SSTSD),max(ds_datanona$SSTSD),length=100) p.data<-expand.grid(YEAR=unique(ds_datanona$YEAR),MONTH=levels(ds_datanona$MONTH),BS=mean(ds_datanona$BS),BATHYMETRY=mean(ds_datanona$BATHYMETRY),BT=mean(ds_datanona$BT),SST=mean(ds_datanona$SST), SSS=mean(ds_datanona$SSS),SSH=mean(ds_datanona$SSH),SSTSD=xSSTSD,BTSD=mean(ds_datanona$BTSD),CHLA=mean(ds_datanona$CHLA),Set_Begin_Hour=mean(ds_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(ds_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(ds_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(ds_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outSSTSD<-summaryBy(pred~SSTSD,data=p.data) outSSTSD$pred.Trans<-plogis(outSSTSD$pred.mean) #BTSD xBTSD<-seq(min(ds_datanona$BTSD),max(ds_datanona$BTSD),length=100) p.data<-expand.grid(YEAR=unique(ds_datanona$YEAR),MONTH=levels(ds_datanona$MONTH),BS=mean(ds_datanona$BS),BATHYMETRY=mean(ds_datanona$BATHYMETRY),BT=mean(ds_datanona$BT),SST=mean(ds_datanona$SST), SSS=mean(ds_datanona$SSS),SSH=mean(ds_datanona$SSH),BTSD=xBTSD,SSTSD=mean(ds_datanona$SSTSD),CHLA=mean(ds_datanona$CHLA),Set_Begin_Hour=mean(ds_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(ds_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(ds_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(ds_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outBTSD<-summaryBy(pred~BTSD,data=p.data) outBTSD$pred.Trans<-plogis(outBTSD$pred.mean) #Set_Begin_Hour xSet_Begin_Hour<-seq(min(ds_datanona$Set_Begin_Hour),max(ds_datanona$Set_Begin_Hour),length=100) p.data<-expand.grid(YEAR=unique(ds_datanona$YEAR),MONTH=levels(ds_datanona$MONTH),BS=mean(ds_datanona$BS),BATHYMETRY=mean(ds_datanona$BATHYMETRY),BT=mean(ds_datanona$BT),SST=mean(ds_datanona$SST), SSS=mean(ds_datanona$SSS),SSH=mean(ds_datanona$SSH),BTSD=mean(ds_datanona$BTSD),SSTSD=mean(ds_datanona$SSTSD),CHLA=mean(ds_datanona$CHLA),Set_Begin_Hour=xSet_Begin_Hour, BEGIN_SET_LATITUDE=mean(ds_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(ds_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(ds_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outSet_Begin_Hour<-summaryBy(pred~Set_Begin_Hour,data=p.data) outSet_Begin_Hour$pred.Trans<-plogis(outSet_Begin_Hour$pred.mean) #MONTH p.data<-expand.grid(YEAR=unique(ds_datanona$YEAR),MONTH=levels(ds_datanona$MONTH),BS=mean(ds_datanona$BS),BATHYMETRY=mean(ds_datanona$BATHYMETRY),BT=mean(ds_datanona$BT),SST=mean(ds_datanona$SST), SSS=mean(ds_datanona$SSS),SSH=mean(ds_datanona$SSH),BTSD=mean(ds_datanona$BTSD),SSTSD=mean(ds_datanona$SSTSD),CHLA=mean(ds_datanona$CHLA),Set_Begin_Hour=mean(ds_datanona$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(ds_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(ds_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(ds_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outMONTH<-summaryBy(pred~MONTH,data=p.data) outMONTH$pred.Trans<-plogis(outMONTH$pred.mean) #YEAR outYEAR<-summaryBy(pred~YEAR,data=p.data) outYEAR$pred.Trans<-plogis(outYEAR$pred.mean) par(mfrow=c(3,3)) plot(pred.Trans~BATHYMETRY, data=outBat,type="l",ylim=c(0,1)) plot(pred.Trans~BT, data=outBT,type="l",ylim=c(0,1)) #plot(pred.Trans~SST, data=outSST,type="l",ylim=c(0,1)) plot(pred.Trans~BS, data=outBS,type="l",ylim=c(0,1)) #plot(pred.Trans~SSH, data=outSSH,type="l",ylim=c(0,1)) plot(pred.Trans~CHLA, data=outCHLA,type="l",ylim=c(0,1)) plot(pred.Trans~BTSD, data=outBTSD,type="l",ylim=c(0,1)) plot(pred.Trans~SSTSD, data=outSSTSD,type="l",ylim=c(0,1)) plot(pred.Trans~Set_Begin_Hour, data=outSet_Begin_Hour,type="l",ylim=c(0,1)) plot(pred.Trans~MONTH, data=outMONTH,type="l",ylim=c(0,1)) plot(pred.Trans~YEAR, data=outYEAR,type="l",ylim=c(0,1)) ##### #SHH ##### #1) GLM or GAM #after discussion with tobey, feel that it is appropriate to remove set where SST was 8C and there was a positive catch #because it is likely outside the temp range of scalloped hammerheads shh_datanona<-shh_datanona[-which(shh_datanona$SST<10 & shh_datanona$BINOMIAL==1),] #considered day of year instead, but this correlates with month and we need month shh_datanona$Daten<-as.numeric(shh_datanona$Date) #start with SSS and Rugosity removed, mess with combinations further down modglm<-glm(BINOMIAL~MONTH+YEAR+BAITTYPE+HOOK_CONFIG+Set_Begin_Hour+BATHYMETRY+BT+SST+BS+SSH+CHLA+BTSD+SSTSD+offset(logEFFORT),data=shh_datanona,family='binomial') modgam<-gam(BINOMIAL~MONTH+YEAR+BAITTYPE+HOOK_CONFIG+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial') AIC(modglm,modgam)#gam is way better, go with GAM for here on out #2)check for temporal autocorrelation (ie. independence vs non independence) #dont need to worry about temporal autocorrelation, if outside dotted lines means theres correlation par(mfrow=c(1,1)) EL1=residuals(modgam, type="pearson") acf(EL1,main="Residuals") pacf(EL1, main="Residuals") #no temporal autocorrelation #VIF z<-cbind(shh_datanona$YEAR,shh_datanona$MONTH,shh_datanona$Day,shh_datanona$BAITTYPE,shh_datanona$HOOK_CONFIG,shh_datanona$Set_Begin_Hour, shh_datanona$BATHYMETRY,shh_datanona$RUGOSITY,shh_datanona$SST, shh_datanona$SSH,shh_datanona$BT,shh_datanona$SSS,shh_datanona$BS,shh_datanona$CHLA,shh_datanona$BTSD,shh_datanona$SSTSD,shh_datanona$TURB) corvif(z) #correlation among Day of year and Month and BS and SSS z<-cbind(shh_datanona$YEAR,shh_datanona$MONTH,shh_datanona$BAITTYPE,shh_datanona$HOOK_CONFIG,shh_datanona$Set_Begin_Hour,shh_datanona$BATHYMETRY,shh_datanona$RUGOSITY,shh_datanona$SST, shh_datanona$SSH,shh_datanona$BT,shh_datanona$BS,shh_datanona$CHLA,shh_datanona$BTSD,shh_datanona$SSTSD,shh_datanona$TURB) corvif(z) #looks like as long as Dayn and Year are not both in and SSS and BS are not both in we are all good, nothing else is over 5 #3) Check for spatial autocorrelation #meaning of correlog plot not clear go with variogram instead #USE VARIOGRAM #try using variogram, have to convert df to spatial df and use geographic crs shh_datanonasp<-shh_datanona coordinates(shh_datanonasp)<- ~BEGIN_SET_LONGITUDE+BEGIN_SET_LATITUDE proj4string(shh_datanonasp)<-CRS("+init=epsg:4326") modgam.1<-gam(BINOMIAL~MONTH+YEAR+BAITTYPE+HOOK_CONFIG+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanonasp,family='binomial') EL.1=residuals(modgam.1, type="pearson") vario500<-variogram(EL.1 ~ BEGIN_SET_LONGITUDE+BEGIN_SET_LATITUDE,data=shh_datanonasp,cutoff=500,width=10,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance shhween intervals plot(vario500,main="500_.1") vario250<-variogram(EL.1 ~ BEGIN_SET_LONGITUDE+BEGIN_SET_LATITUDE,data=shh_datanonasp,cutoff=250,width=5,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance shhween intervals plot(vario250,main="250_.1") vario100<-variogram(EL.1 ~ BEGIN_SET_LONGITUDE+BEGIN_SET_LATITUDE,data=shh_datanonasp,cutoff=100,width=1,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance shhween intervals, # 0 = N; .15 = NE; 90 = E; 135 = SE plot(vario100,main="100_.1") #looks ok to me, but lets see if we can improve it modgam2.6<-gam(BINOMIAL~MONTH+YEAR +BAITTYPE+HOOK_CONFIG+s(Set_Begin_Hour)+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+ s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanonasp,family='binomial') EL2.6=residuals(modgam2.6, type="pearson") vario500_2.6<-variogram(EL2.6 ~ BEGIN_SET_LONGITUDE + BEGIN_SET_LATITUDE,data=shh_datanonasp,cutoff=500,width=10,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance shhween intervals plot(vario500_2.6,main="500_2.6") vario250_2.6<-variogram(EL2.6 ~ BEGIN_SET_LONGITUDE + BEGIN_SET_LATITUDE,data=shh_datanonasp,cutoff=250,width=5,alpha=c(0,45,90,135))#cutoff is how far away you want to assess and width is distance shhween intervals plot(vario250_2.6,main="250_2.6") AIC(modgam.1,modgam2.6) #2.6 has lowest aic, go with it concurvity(modgam2.6,full=TRUE)#concurvity is high between lat/lon and bs and bat df = data.frame(Long=shh_datanona$BEGIN_SET_LONGITUDE, Lat=shh_datanona$BEGIN_SET_LATITUDE,Residuals=EL.1) coordinates(df)<-c("Long", "Lat") bubble(df,zcol="Residuals", col=c("orange","blue"),xlab="long",ylab="lat", maxsize=3) #seems pretty evenly distrubtion of pos and neg resishh so no autocorrelation #decide to keep spatial term to help make other covariates more realistic #4) 2 Part Variable selection #Part 1 environmental variables mod2<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial') mod3<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(RUGOSITY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial')#put rug in for bat, no btsd (correlates with rug) mod4<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(SSS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial')#put sss in for bs mod5<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial')#put in turb for chla mod6<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+offset(logEFFORT),data=shh_datanona,family='binomial')#no sstsd mod7<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+offset(logEFFORT),data=shh_datanona,family='binomial')#no sstsd and btsd mod8<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(RUGOSITY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+offset(logEFFORT),data=shh_datanona,family='binomial')#no sstsd, with turb and rug, no btsd (correlates with rug) mod9<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(BTSD)+offset(logEFFORT),data=shh_datanona,family='binomial')#no sstsd and chla mod10<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+offset(logEFFORT),data=shh_datanona,family='binomial')#no sstsd and chla and SSH and btsd mod11<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(BTSD)+offset(logEFFORT),data=shh_datanona,family='binomial')#no sstsd and chla and SSH mod12<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial')#no SST mod13<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(BS)+s(SSH)+s(CHLA)+s(BTSD)+offset(logEFFORT),data=shh_datanona,family='binomial')#no SST and sstsd mod14<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(RUGOSITY)+s(BT)+s(SST)+s(BS)+offset(logEFFORT),data=shh_datanona,family='binomial')#no sstsd and chla and SSH, with rug, no btsd (correlates with rug) mod15<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial')#no SSH, with turb mod16<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial')#no SSH mod17<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(BS)+s(CHLA)+s(BTSD)+offset(logEFFORT),data=shh_datanona,family='binomial')#no SSH, sst, sstsd AICtab(mod2,mod3,mod4,mod5,mod6,mod7,mod8,mod9,mod10,mod11,mod12,mod13,mod14,mod15,mod16,mod17) #best models is mod5 #Part 2 gear variables with best model from above mod5<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial')#put in turb for chla mod18<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+BAITTYPE+HOOK_CONFIG+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial') mod19<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+BAITTYPE+HOOK_CONFIG+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial') mod20<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+HOOK_CONFIG+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial') mod21<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial') mod22<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+BAITTYPE+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial') AICtab(mod5,mod18,mod19,mod20,mod21,mod22) #best model is mod22 is best #5) Tweak k values #if we can find the model that explains the most variation and the curves make biological sense go with that model mod22<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+BAITTYPE+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='binomial') #model is slightly overdispersed, switch family to quasibinomial because dispersion parameter is not fixed so it can model overdispersion #http://coltekin.net/cagri/R/r-exercisesse11.html#:~:text=Overdispersion%20occurs%20when%20error%20(residuals,distribution%20is%20the%20binomial%20distribution.&text=Underdispersion%2C%20detected%20by%20lower%20residual,also%20possible%20but%20more%20rare. mod22.1<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+BAITTYPE+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT),data=shh_datanona,family='quasibinomial') mod22.2<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+BAITTYPE+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT,k=12)+s(SST)+s(BS,k=8)+s(SSH,k=8)+s(TURB,k=12)+s(BTSD,k=7)+s(SSTSD,k=8)+offset(logEFFORT),data=shh_datanona,family='quasibinomial') mod22.3<-gam(BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+BAITTYPE+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS,k=8)+s(SSH)+s(TURB)+s(BTSD,k=7)+s(SSTSD,k=12)+offset(logEFFORT),data=shh_datanona,family='quasibinomial') par(mfrow=c(1,1)) summary(mod22.1) gam.check(mod22.1) #good plot(mod22.1) disp2(mod22.1) #close to overdispersion, used quasibinomial distribution to account for it # edf Ref.df F p-value #s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE) 18.638 22.956 1.384 0.109824 #s(Set_Begin_Hour) 8.238 8.817 5.751 6.25e-07 *** #s(BATHYMETRY) 4.107 5.122 4.969 0.000149 *** #s(BT) 6.472 7.595 2.662 0.008827 ** #s(SST) 1.000 1.000 0.015 0.901841 #s(BS) 6.801 7.712 2.504 0.012015 * #s(SSH) 6.672 7.790 3.766 0.000375 *** #s(TURB) 6.740 7.854 1.252 0.296487 #s(BTSD) 5.984 7.127 4.193 0.000133 *** #s(SSTSD) 5.523 6.672 1.540 0.158882 #R-sq.(adj) = 0.299 Deviance explained = 47.2% m=mod22.1 #Plot respose curves #to backtransform binomial model use plogis #to backtransform neg bin use exp #Bathymetry xBat<-seq(min(shh_datanona$BATHYMETRY),max(shh_datanona$BATHYMETRY),length=100) p.data<-expand.grid(YEAR=unique(shh_datanona$YEAR),MONTH=levels(shh_datanona$MONTH),BATHYMETRY=xBat,BT=mean(shh_datanona$BT),SST=mean(shh_datanona$SST),BS=mean(shh_datanona$BS), TURB=mean(shh_datanona$TURB),SSS=mean(shh_datanona$SSS),SSH=mean(shh_datanona$SSH),CHLA=mean(shh_datanona$CHLA),BTSD=mean(shh_datanona$BTSD),SSTSD=mean(shh_datanona$SSTSD),Set_Begin_Hour=mean(shh_datanona$Set_Begin_Hour), BAITTYPE=levels(shh_datanona$BAITTYPE),BEGIN_SET_LATITUDE=mean(shh_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(shh_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(shh_datanona$logEFFORT)) p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outBat<-summaryBy(pred~BATHYMETRY,data=p.data) outBat$pred.Trans<-plogis(outBat$pred.mean) #BT xBT<-seq(min(shh_datanona$BT),max(shh_datanona$BT),length=100) p.data<-expand.grid(YEAR=unique(shh_datanona$YEAR),MONTH=levels(shh_datanona$MONTH),BT=xBT,BATHYMETRY=mean(shh_datanona$BATHYMETRY),SST=mean(shh_datanona$SST),BS=mean(shh_datanona$BS), TURB=mean(shh_datanona$TURB),SSS=mean(shh_datanona$SSS),SSH=mean(shh_datanona$SSH),CHLA=mean(shh_datanona$CHLA),BTSD=mean(shh_datanona$BTSD),SSTSD=mean(shh_datanona$SSTSD),Set_Begin_Hour=mean(shh_datanona$Set_Begin_Hour), BAITTYPE=levels(shh_datanona$BAITTYPE),BEGIN_SET_LATITUDE=mean(shh_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(shh_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(shh_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outBT<-summaryBy(pred~BT,data=p.data) outBT$pred.Trans<-plogis(outBT$pred.mean) #SST xSST<-seq(min(shh_datanona$SST),max(shh_datanona$SST),length=100) p.data<-expand.grid(YEAR=unique(shh_datanona$YEAR),MONTH=levels(shh_datanona$MONTH),SST=xSST,BATHYMETRY=mean(shh_datanona$BATHYMETRY),BT=mean(shh_datanona$BT),BS=mean(shh_datanona$BS), TURB=mean(shh_datanona$TURB),SSS=mean(shh_datanona$SSS),SSH=mean(shh_datanona$SSH),CHLA=mean(shh_datanona$CHLA),BTSD=mean(shh_datanona$BTSD),SSTSD=mean(shh_datanona$SSTSD),Set_Begin_Hour=mean(shh_datanona$Set_Begin_Hour), BAITTYPE=levels(shh_datanona$BAITTYPE),BEGIN_SET_LATITUDE=mean(shh_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(shh_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(shh_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outSST<-summaryBy(pred~SST,data=p.data) outSST$pred.Trans<-plogis(outSST$pred.mean) #BS xBS<-seq(min(shh_datanona$BS),max(shh_datanona$BS),length=100) p.data<-expand.grid(YEAR=unique(shh_datanona$YEAR),MONTH=levels(shh_datanona$MONTH),BS=xBS,BATHYMETRY=mean(shh_datanona$BATHYMETRY),BT=mean(shh_datanona$BT),SST=mean(shh_datanona$SST), TURB=mean(shh_datanona$TURB),SSS=mean(shh_datanona$SSS),SSH=mean(shh_datanona$SSH),CHLA=mean(shh_datanona$CHLA),BTSD=mean(shh_datanona$BTSD),SSTSD=mean(shh_datanona$SSTSD),Set_Begin_Hour=mean(shh_datanona$Set_Begin_Hour), BAITTYPE=levels(shh_datanona$BAITTYPE),BEGIN_SET_LATITUDE=mean(shh_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(shh_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(shh_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outBS<-summaryBy(pred~BS,data=p.data) outBS$pred.Trans<-plogis(outBS$pred.mean) #SSS #xSSS<-seq(min(shh_datanona$SSS),max(shh_datanona$SSS),length=100) #p.data<-expand.grid(YEAR=unique(shh_datanona$YEAR),MONTH=levels(shh_datanona$MONTH),SSS=xSSS,BATHYMETRY=mean(shh_datanona$BATHYMETRY),BT=mean(shh_datanona$BT),SST=mean(shh_datanona$SST), # TURB=mean(shh_datanona$TURB),BS=mean(shh_datanona$BS),SSH=mean(shh_datanona$SSH),CHLA=mean(shh_datanona$CHLA),BTSD=mean(shh_datanona$BTSD),SSTSD=mean(shh_datanona$SSTSD),Set_Begin_Hour=mean(shh_datanona$Set_Begin_Hour), # BAITTYPE=levels(shh_datanona$BAITTYPE),BEGIN_SET_LATITUDE=mean(shh_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(shh_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(shh_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict #p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) #outSSS<-summaryBy(pred~SSS,data=p.data) #outSSS$pred.Trans<-plogis(outSSS$pred.mean) #SSH xSSH<-seq(min(shh_datanona$SSH),max(shh_datanona$SSH),length=100) p.data<-expand.grid(YEAR=unique(shh_datanona$YEAR),MONTH=levels(shh_datanona$MONTH),BS=mean(shh_datanona$BS),BATHYMETRY=mean(shh_datanona$BATHYMETRY),BT=mean(shh_datanona$BT),SST=mean(shh_datanona$SST), TURB=mean(shh_datanona$TURB),SSS=mean(shh_datanona$SSS),CHLA=mean(shh_datanona$CHLA),SSH=xSSH,BTSD=mean(shh_datanona$BTSD),SSTSD=mean(shh_datanona$SSTSD),Set_Begin_Hour=mean(shh_datanona$Set_Begin_Hour), BAITTYPE=levels(shh_datanona$BAITTYPE),BEGIN_SET_LATITUDE=mean(shh_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(shh_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(shh_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outSSH<-summaryBy(pred~SSH,data=p.data) outSSH$pred.Trans<-plogis(outSSH$pred.mean) #CHLA #xCHLA<-seq(min(shh_datanona$CHLA),max(shh_datanona$CHLA),length=100) #p.data<-expand.grid(YEAR=unique(shh_datanona$YEAR),MONTH=levels(shh_datanona$MONTH),BS=mean(shh_datanona$BS),BATHYMETRY=mean(shh_datanona$BATHYMETRY),BT=mean(shh_datanona$BT),SST=mean(shh_datanona$SST), # TURB=mean(shh_datanona$TURB),SSS=mean(shh_datanona$SSS),SSH=mean(shh_datanona$SSH),CHLA=xCHLA,BTSD=mean(shh_datanona$BTSD),SSTSD=mean(shh_datanona$SSTSD),Set_Begin_Hour=mean(shh_datanona$Set_Begin_Hour), # BAITTYPE=levels(shh_datanona$BAITTYPE),BEGIN_SET_LATITUDE=mean(shh_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(shh_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(shh_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict #p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) #outCHLA<-summaryBy(pred~CHLA,data=p.data) #outCHLA$pred.Trans<-plogis(outCHLA$pred.mean) #TURB xTURB<-seq(min(shh_datanona$TURB),max(shh_datanona$TURB),length=100) p.data<-expand.grid(YEAR=unique(shh_datanona$YEAR),MONTH=levels(shh_datanona$MONTH),BS=mean(shh_datanona$BS),BATHYMETRY=mean(shh_datanona$BATHYMETRY),BT=mean(shh_datanona$BT),SST=mean(shh_datanona$SST), TURB=xTURB,SSS=mean(shh_datanona$SSS),SSH=mean(shh_datanona$SSH),CHLA=mean(shh_datanona$CHLA),BTSD=mean(shh_datanona$BTSD),SSTSD=mean(shh_datanona$SSTSD),Set_Begin_Hour=mean(shh_datanona$Set_Begin_Hour), BAITTYPE=levels(shh_datanona$BAITTYPE),BEGIN_SET_LATITUDE=mean(shh_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(shh_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(shh_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outTURB<-summaryBy(pred~TURB,data=p.data) outTURB$pred.Trans<-plogis(outTURB$pred.mean) #SSTSD xSSTSD<-seq(min(shh_datanona$SSTSD),max(shh_datanona$SSTSD),length=100) p.data<-expand.grid(YEAR=unique(shh_datanona$YEAR),MONTH=levels(shh_datanona$MONTH),BS=mean(shh_datanona$BS),BATHYMETRY=mean(shh_datanona$BATHYMETRY),BT=mean(shh_datanona$BT),SST=mean(shh_datanona$SST), TURB=mean(shh_datanona$TURB),SSS=mean(shh_datanona$SSS),SSH=mean(shh_datanona$SSH),SSTSD=xSSTSD,BTSD=mean(shh_datanona$BTSD),CHLA=mean(shh_datanona$CHLA),Set_Begin_Hour=mean(shh_datanona$Set_Begin_Hour), BAITTYPE=levels(shh_datanona$BAITTYPE),BEGIN_SET_LATITUDE=mean(shh_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(shh_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(shh_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outSSTSD<-summaryBy(pred~SSTSD,data=p.data) outSSTSD$pred.Trans<-plogis(outSSTSD$pred.mean) #BTSD xBTSD<-seq(min(shh_datanona$BTSD),max(shh_datanona$BTSD),length=100) p.data<-expand.grid(YEAR=unique(shh_datanona$YEAR),MONTH=levels(shh_datanona$MONTH),BS=mean(shh_datanona$BS),BATHYMETRY=mean(shh_datanona$BATHYMETRY),BT=mean(shh_datanona$BT),SST=mean(shh_datanona$SST), TURB=mean(shh_datanona$TURB),SSS=mean(shh_datanona$SSS),SSH=mean(shh_datanona$SSH),BTSD=xBTSD,SSTSD=mean(shh_datanona$SSTSD),CHLA=mean(shh_datanona$CHLA),Set_Begin_Hour=mean(shh_datanona$Set_Begin_Hour), BAITTYPE=levels(shh_datanona$BAITTYPE),BEGIN_SET_LATITUDE=mean(shh_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(shh_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(shh_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outBTSD<-summaryBy(pred~BTSD,data=p.data) outBTSD$pred.Trans<-plogis(outBTSD$pred.mean) #Set_Begin_Hour xSet_Begin_Hour<-seq(min(shh_datanona$Set_Begin_Hour),max(shh_datanona$Set_Begin_Hour),length=100) p.data<-expand.grid(YEAR=unique(shh_datanona$YEAR),MONTH=levels(shh_datanona$MONTH),BS=mean(shh_datanona$BS),BATHYMETRY=mean(shh_datanona$BATHYMETRY),BT=mean(shh_datanona$BT),SST=mean(shh_datanona$SST), TURB=mean(shh_datanona$TURB),SSS=mean(shh_datanona$SSS),SSH=mean(shh_datanona$SSH),BTSD=mean(shh_datanona$BTSD),SSTSD=mean(shh_datanona$SSTSD),CHLA=mean(shh_datanona$CHLA),Set_Begin_Hour=xSet_Begin_Hour, BAITTYPE=levels(shh_datanona$BAITTYPE),BEGIN_SET_LATITUDE=mean(shh_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(shh_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(shh_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outSet_Begin_Hour<-summaryBy(pred~Set_Begin_Hour,data=p.data) outSet_Begin_Hour$pred.Trans<-plogis(outSet_Begin_Hour$pred.mean) #MONTH p.data<-expand.grid(YEAR=unique(shh_datanona$YEAR),MONTH=levels(shh_datanona$MONTH),BS=mean(shh_datanona$BS),BATHYMETRY=mean(shh_datanona$BATHYMETRY),BT=mean(shh_datanona$BT),SST=mean(shh_datanona$SST), TURB=mean(shh_datanona$TURB),SSS=mean(shh_datanona$SSS),SSH=mean(shh_datanona$SSH),BTSD=mean(shh_datanona$BTSD),SSTSD=mean(shh_datanona$SSTSD),CHLA=mean(shh_datanona$CHLA),Set_Begin_Hour=mean(shh_datanona$Set_Begin_Hour), BAITTYPE=levels(shh_datanona$BAITTYPE),BEGIN_SET_LATITUDE=mean(shh_datanona$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(shh_datanona$BEGIN_SET_LONGITUDE),logEFFORT=mean(shh_datanona$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)")) outMONTH<-summaryBy(pred~MONTH,data=p.data) outMONTH$pred.Trans<-plogis(outMONTH$pred.mean) #YEAR outYEAR<-summaryBy(pred~YEAR,data=p.data) outYEAR$pred.Trans<-plogis(outYEAR$pred.mean) #BAITTYPE outBAIT<-summaryBy(pred~BAITTYPE,data=p.data) outBAIT$pred.Trans<-plogis(outBAIT$pred.mean) par(mfrow=c(3,4)) plot(pred.Trans~BATHYMETRY, data=outBat,type="l",ylim=c(0,1)) plot(pred.Trans~BT, data=outBT,type="l",ylim=c(0,1)) plot(pred.Trans~SST, data=outSST,type="l",ylim=c(0,1)) plot(pred.Trans~BS, data=outBS,type="l",ylim=c(0,1)) plot(pred.Trans~SSH, data=outSSH,type="l",ylim=c(0,1)) plot(pred.Trans~TURB, data=outTURB,type="l",ylim=c(0,1)) plot(pred.Trans~BTSD, data=outBTSD,type="l",ylim=c(0,1)) plot(pred.Trans~SSTSD, data=outSSTSD,type="l",ylim=c(0,1)) plot(pred.Trans~Set_Begin_Hour, data=outSet_Begin_Hour,type="l",ylim=c(0,1)) plot(pred.Trans~BAITTYPE, data=outBAIT,type="l",ylim=c(0,1)) plot(pred.Trans~MONTH, data=outMONTH,type="l",ylim=c(0,1)) plot(pred.Trans~YEAR, data=outYEAR,type="l",ylim=c(0,1)) ################## #Model Validation ################## library(dismo) library(forecast) library(blockCV) ###Random Validation modVal<-function(data=NA,gam.formula=NA,exclude=NULL,method=NULL,drop.unused.levels=TRUE,family=NA){ k <- 10 set.seed(10) group <- kfold(data, k) #assigned all data into 10 groups AUC_Rec<-NULL TSS_Rec<-NULL cor_Rec<-NULL RMSE_test_Rec<-NULL RMSE_train_Rec<-NULL MAPE_test_Rec<-NULL RMSE_testNull_Rec<-NULL RMSE_trainNull_Rec<-NULL nRMSE_train_Rec<-NULL nRMSE_test_Rec<-NULL MASE_Rec<-NULL DMRec<-NULL for (i in 1:k) { train <- data[group != i,] test <- data[group == i,] modbi<-gam(gam.formula,data=train,family=family,method=method,drop.unused.levels=drop.unused.levels) #model run for test pred.test<-predict(modbi,type='link',newdata=test,exclude=exclude) pred.test<-plogis(pred.test) #model run for train pred.train<-predict(modbi,type='link',newdata=train,exclude=exclude) pred.train<-plogis(pred.train) #AUC testpres<-test[which(test$BINOMIAL==1),] testbackg<-test[which(test$BINOMIAL==0),] e<-evaluate(testpres,testbackg,modbi) auc<-e@auc #plot(e, "ROC", cex.lab=1.5, col="blue", type="l", lwd=2) AUC_Rec<-c(AUC_Rec,auc) #TSS #plot(e@t, e@TPR+e@TNR, type="l", lwd=2, cex.lab=1.5) TPR <- e@TPR[which(e@TPR+e@TNR==max(e@TPR+e@TNR))][1] #True Positive Rate at threshold where TPR+TNR was maximized TNR<- e@TNR[which(e@TPR+e@TNR==max(e@TPR+e@TNR))][1] #True Negative Rate at threshold where TPR+TNR was maximized tss<-(TPR+TNR)-1 TSS_Rec<-c(TSS_Rec,tss) #cor cor<-e@cor cor_Rec<-c(cor_Rec,cor) difsqtestRec<-NULL difsqtrainRec<-NULL APE.testRec<-NULL diftestRec<-NULL diftrainRec<-NULL #RMSE and MAE for test for(j in 1:length(pred.test)){ difsqtest<-(pred.test[j]-test$BINOMIAL[j])^2 #first part of numerator (before summation) difsqtestRec<-c(difsqtestRec,difsqtest) diftest<-pred.test[j]-test$BINOMIAL[j] diftestRec<-c(diftestRec,diftest) } RMSE_test<-sqrt(sum(difsqtestRec)/nrow(test)) RMSE_test_Rec<-c(RMSE_test_Rec,RMSE_test) #list of RMSE for testing set nRMSE_test<-RMSE_test/(max(test$BINOMIAL)-min(test$BINOMIAL)) #normalized RMSE by dividing RMSE by range of observed values nRMSE_test_Rec<-c(nRMSE_test_Rec,nRMSE_test) MAE_test<-sum(abs(diftestRec))/nrow(test) #RMSE and MAE for train for(j in 1:length(pred.train)){ difsqtrain<-(pred.train[j]-train$BINOMIAL[j])^2 #first part of numerator (before summation) difsqtrainRec<-c(difsqtrainRec,difsqtrain) diftrain<-pred.train[j]-train$BINOMIAL[j] diftrainRec<-c(diftrainRec,diftrain) } RMSE_train<-sqrt(sum(difsqtrainRec)/nrow(train)) RMSE_train_Rec<-c(RMSE_train_Rec,RMSE_train) #list of RMSE for training set MAE_train<-sum(abs(diftrainRec))/nrow(train) #MAPE for(j in 1:length(pred.test)){ APE.test<-ifelse(test$BINOMIAL[j]>0,abs(test$BINOMIAL[j]-pred.test[j])/test$BINOMIAL[j],0) #first part of numerator (before summation) APE.testRec<-c(APE.testRec,APE.test) } MAPE_test<-(sum(APE.testRec)/nrow(test))*100 MAPE_test_Rec<-c(MAPE_test_Rec,MAPE_test) ###Null model### modbiNull<-gam(BINOMIAL~offset(logEFFORT),data=train,family=family,method=method,drop.unused.levels=drop.unused.levels) #model run for test pred.testNull<-predict(modbiNull,type='link',newdata=test,exclude=exclude) pred.testNull<-plogis(pred.testNull) #model run for train pred.trainNull<-predict(modbiNull,type='link',newdata=train,exclude=exclude) pred.trainNull<-plogis(pred.trainNull) difsqtestNullRec<-NULL difsqtrainNullRec<-NULL APE.testNullRec<-NULL diftestNullRec<-NULL diftrainNullRec<-NULL #RMSE and MAE for test for(j in 1:length(pred.testNull)){ difsqtestNull<-(pred.testNull[j]-test$BINOMIAL[j])^2 #first part of numerator (before summation) difsqtestNullRec<-c(difsqtestNullRec,difsqtestNull) diftestNull<-pred.testNull[j]-test$BINOMIAL[j] diftestNullRec<-c(diftestNullRec,diftestNull) } RMSE_testNull<-sqrt(sum(difsqtestNullRec)/nrow(test)) RMSE_testNull_Rec<-c(RMSE_testNull_Rec,RMSE_testNull) #list of RMSE for testing set MAE_testNull<-sum(abs(diftestNullRec))/nrow(test) #RMSE and MAE for train for(j in 1:length(pred.trainNull)){ difsqtrainNull<-(pred.trainNull[j]-train$BINOMIAL[j])^2 #first part of numerator (before summation) difsqtrainNullRec<-c(difsqtrainNullRec,difsqtrainNull) diftrainNull<-pred.trainNull[j]-train$BINOMIAL[j] diftrainNullRec<-c(diftrainNullRec,diftrainNull) } RMSE_trainNull<-sqrt(sum(difsqtrainNullRec)/nrow(train)) RMSE_trainNull_Rec<-c(RMSE_trainNull_Rec,RMSE_trainNull) #list of RMSE for training set MAE_trainNull<-sum(abs(diftrainNullRec))/nrow(train) MASE<-MAE_test/MAE_testNull MASE_Rec<-c(MASE_Rec,MASE) #Diebold-Marino test statistic: evaluates the loss differential (errors) between two forecast (i.e. null and full GAM), following Kleisner et al. 2017 DM<-dm.test(diftestNullRec,diftestRec,h=1,power=1,alternative = "greater") DM.p<-DM$p.value DMRec<-c(DMRec, DM.p) print(i) } return(list(AUC=AUC_Rec,TSS=TSS_Rec,cor=cor_Rec,RMSE_test=RMSE_test_Rec,RMSE_train=RMSE_train_Rec,nRMSE_test=nRMSE_test_Rec,MASE=MASE_Rec,DM=DMRec)) } ###Spatial Validation #This approach uses spatialBlock to define blocks. This was a little more systematic and less arbitrary than 2.0 method modValSpace3.0<-function(data=NA,gam.formula=NA,exclude=NULL,method=NULL,drop.unused.levels=TRUE,therange=NA,k=NA,family=NA){ datasp<-subset(data,select=c(BEGIN_SET_LONGITUDE,BEGIN_SET_LATITUDE,BINOMIAL)) colnames(datasp)<-c("x","y","Species") #coordinates(datasp)<- ~x+y #proj4string(datasp)<-CRS("+proj=longlat +ellps=WGS84 +datum=WGS84 +no_defs") datasp<-sf::st_as_sf(datasp, coords = c("x", "y"), crs = CRS("+proj=longlat +datum=WGS84 +no_defs +ellps=WGS84 +towgs84=0,0,0")) spBlock<-spatialBlock(speciesData = datasp,species="Species",theRange=therange,k=k,selection="systematic",iteration=10) spBlock$plots + geom_sf(data=datasp %>% dplyr::filter(Species == 1),aes(col=Species)) data$spBlock<-spBlock$foldID AUC_Rec<-NULL TSS_Rec<-NULL cor_Rec<-NULL for (i in 1:k) { train <- data[which(data$spBlock != i),] test <- data[which(data$spBlock == i),] modbi<-gam(gam.formula,data=train,family=family,method=method,drop.unused.levels=drop.unused.levels) #model run for test pred.test<-predict(modbi,type='link',newdata=test,exclude=exclude) pred.test<-plogis(pred.test) #model run for train #pred.train<-predict(modbi,type='link',newdata=train,exclude=exclude) #pred.train<-plogis(pred.train) #AUC testpres<-test[which(test$BINOMIAL==1),] testbackg<-test[which(test$BINOMIAL==0),] e<-tryCatch({ #need to do this in case there are no positive cases in a certain zone (happens for dus) evaluate(testpres,testbackg,modbi) }, warning =function(w){ message("handling warning:",conditionMessage(w)) "Bad" },error=function(e){ message("ignore error:",conditionMessage(e)) "Bad" }) if(nrow(testpres)!=0){ auc<-e@auc #plot(e, "ROC", cex.lab=1.5, col="blue", type="l", lwd=2) #AUC_Rec<-c(AUC_Rec,auc) #TSS #plot(e@t, e@TPR+e@TNR, type="l", lwd=2, cex.lab=1.5) TPR <- e@TPR[which(e@TPR+e@TNR==max(e@TPR+e@TNR))][1] #True Positive Rate at threshold where TPR+TNR was maximized TNR<- e@TNR[which(e@TPR+e@TNR==max(e@TPR+e@TNR))][1] #True Negative Rate at threshold where TPR+TNR was maximized tss<-(TPR+TNR)-1 #TSS_Rec<-c(TSS_Rec,tss) #cor cor<-e@cor #cor_Rec<-c(cor_Rec,cor) }else{ auc<-NA tss<-NA cor<-NA } AUC_Rec<-c(AUC_Rec,auc) TSS_Rec<-c(TSS_Rec,tss) cor_Rec<-c(cor_Rec,cor) print(i) } return(list(AUC=AUC_Rec,TSS=TSS_Rec,cor=cor_Rec)) } #predict over each year with a model trained on data from other years modValTime2.0<-function(data=NA,gam.formula=NA,exclude=NULL,method=NULL,drop.unused.levels=TRUE,family=NA){ Years<-as.factor(2005:2019) AUC_Rec<-NULL TSS_Rec<-NULL cor_Rec<-NULL for (i in Years) { train <- data[which(data$YEAR != i),] test <- data[which(data$YEAR == i),] modbi<-gam(gam.formula,data=train,family=family,method=method,drop.unused.levels=drop.unused.levels) #model run for test pred.test<-predict(modbi,type='link',newdata=test,exclude=exclude) pred.test<-plogis(pred.test) #model run for train #pred.train<-predict(modbi,type='link',newdata=train,exclude=exclude) #pred.train<-plogis(pred.train) #AUC testpres<-test[which(test$BINOMIAL==1),] testbackg<-test[which(test$BINOMIAL==0),] e<-evaluate(testpres,testbackg,modbi) auc<-e@auc #plot(e, "ROC", cex.lab=1.5, col="blue", type="l", lwd=2) AUC_Rec<-c(AUC_Rec,auc) #TSS #plot(e@t, e@TPR+e@TNR, type="l", lwd=2, cex.lab=1.5) TPR <- e@TPR[which(e@TPR+e@TNR==max(e@TPR+e@TNR))][1] #True Positive Rate at threshold where TPR+TNR was maximized TNR<- e@TNR[which(e@TPR+e@TNR==max(e@TPR+e@TNR))][1] #True Negative Rate at threshold where TPR+TNR was maximized tss<-(TPR+TNR)-1 TSS_Rec<-c(TSS_Rec,tss) #cor cor<-e@cor cor_Rec<-c(cor_Rec,cor) print(i) } return(list(AUC=AUC_Rec,TSS=TSS_Rec,cor=cor_Rec)) } ##### #SB ##### formula<-BINOMIAL~MONTH+YEAR+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(CHLA)+offset(logEFFORT) sb_mod_val<-modVal(data=sb_datanona,gam.formula=formula,exclude=NULL,method="GCV.Cp",drop.unused.levels=FALSE,family="binomial") #AUC mean(sb_mod_val$AUC) #good, 0.87 #TSS mean(sb_mod_val$TSS) #okay, 0.65 #Cor mean(sb_mod_val$cor) #okay, 0.50 ###Spatial validation sb_mod_valSpace3.0<-modValSpace3.0(data=sb_datanona,gam.formula=formula,exclude=NULL,method="GCV.Cp",drop.unused.levels=FALSE,therange=65000,k=5,family="binomial") mean(sb_mod_valSpace3.0$AUC)#0.81 mean(sb_mod_valSpace3.0$TSS)#0.55 mean(sb_mod_valSpace3.0$cor)#0.28 ###Temporal Validation sb_mod_valTime2<-modValTime2.0(data=sb_datanona,gam.formula=formula,exclude=NULL,method="GCV.Cp",drop.unused.levels=FALSE,family="binomial") mean(sb_mod_valTime2$AUC[12:14])#0.88 going from 12:14 because that is from 2016-2018 mean(sb_mod_valTime2$TSS[12:14])#0.71 mean(sb_mod_valTime2$cor[12:14])#0.29 #save model output ##### #DS ##### formula<-BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+s(Set_Begin_Hour)+s(BATHYMETRY,k=15)+s(BT,k=9)+s(BS)+s(CHLA)+s(BTSD)+s(SSTSD,k=6)+offset(logEFFORT) ds_mod_val<-modVal(data=ds_datanona,gam.formula=formula,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",method="GCV.Cp",drop.unused.levels=FALSE,family="quasibinomial") #AUC mean(ds_mod_val$AUC) #good, 0.79 #TSS mean(ds_mod_val$TSS) #okay, 0.51 #Cor mean(ds_mod_val$cor) #okay, 0.33 ###Spatial validation ds_mod_valSpace3.0<-modValSpace3.0(data=ds_datanona,gam.formula=formula,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",method="GCV.Cp",drop.unused.levels=FALSE,therange=65000,k=5,family="quasibinomial") mean(ds_mod_valSpace3.0$AUC)#0.72 mean(ds_mod_valSpace3.0$TSS)#0.38 mean(ds_mod_valSpace3.0$cor)#0.22 ###Temporal Validation ds_mod_valTime2<-modValTime2.0(data=ds_datanona,gam.formula=formula,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",method="GCV.Cp",drop.unused.levels=FALSE,family="quasibinomial") mean(ds_mod_valTime2$AUC[12:14])#0.72 going from 12:14 because that is from 2016-2018 mean(ds_mod_valTime2$TSS[12:14])#0.48 mean(ds_mod_valTime2$cor[12:14])#0.29 #save model output ##### #SHH ##### formula<-BINOMIAL~MONTH+YEAR+s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)+BAITTYPE+s(Set_Begin_Hour)+s(BATHYMETRY)+s(BT)+s(SST)+s(BS)+s(SSH)+s(TURB)+s(BTSD)+s(SSTSD)+offset(logEFFORT) shh_mod_val<-modVal(data=shh_datanona,gam.formula=formula,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",method="GCV.Cp",drop.unused.levels=FALSE,family="quasibinomial") #AUC mean(shh_mod_val$AUC) #good, 0.78 #TSS mean(shh_mod_val$TSS) #okay, 0.48 #Cor mean(shh_mod_val$cor) #okay, 0.38 ###Spatial validation shh_mod_valSpace3.0<-modValSpace3.0(data=shh_datanona,gam.formula=formula,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",method="GCV.Cp",drop.unused.levels=FALSE,therange=65000,k=4,family="quasibinomial") mean(shh_mod_valSpace3.0$AUC)#0.75 mean(shh_mod_valSpace3.0$TSS)#0.40 mean(shh_mod_valSpace3.0$cor)#0.36 ###Temporal Validation shh_mod_valTime2<-modValTime2.0(data=shh_datanona,gam.formula=formula,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",method="GCV.Cp",drop.unused.levels=FALSE,family="quasibinomial") mean(shh_mod_valTime2$AUC[12:14])#0.70 going from 12:14 because that is from 2016-2018 mean(shh_mod_valTime2$TSS[12:14])#0.42 mean(shh_mod_valTime2$cor[12:14])#0.25 #save model output ################## #BootStrapping ################## #no hook or bait BinGAMboot<-function(data=NA,gam.formula=NA,exclude=NULL,family=NA,method=NULL){ recbat<-matrix(nrow=100, ncol=1000, NA) #recrug<-matrix(nrow=100, ncol=1000, NA) recsst<-matrix(nrow=100, ncol=1000, NA) recssh<-matrix(nrow=100, ncol=1000, NA) recchl<-matrix(nrow=100, ncol=1000, NA) recbt<-matrix(nrow=100, ncol=1000, NA) recbs<-matrix(nrow=100, ncol=1000, NA) recsstsd<-matrix(nrow=100, ncol=1000, NA) recbtsd<-matrix(nrow=100, ncol=1000, NA) recsethour<-matrix(nrow=100, ncol=1000,NA) recyear<-matrix(nrow=15, ncol=1000, NA) #rechook<-matrix(nrow=4, ncol=1000,NA) #recbait<-matrix(nrow=3,ncol=1000,NA) recmonth<-matrix(nrow=12, ncol=1000, NA) for(i in 1:1000){ boot.sample<-data[sample(nrow(data),nrow(data),replace=TRUE),] mBoot<-tryCatch({ gam(gam.formula,data=boot.sample,family=family,method=method) }, warning=function(w){ message("handling warning:",conditionMessage(w)) "Bad" },error=function(e){ message("ignore error:",conditionMessage(e)) "Bad" }) vars<-attr(m$terms,"term.labels") if(mBoot[1]!="Bad"){ #Bathymetry xBat<-seq(min(data$BATHYMETRY),max(data$BATHYMETRY),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BATHYMETRY=xBat,BT=mean(data$BT),SST=mean(data$SST),BS=mean(data$BS), SSH=mean(data$SSH),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBat<-summaryBy(pred~BATHYMETRY,data=p.data) outBat$pred.Trans<-plogis(outBat$pred.mean) recbat[1:100,i]<-outBat$pred.Trans #BT xBT<-seq(min(data$BT),max(data$BT),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BT=xBT,BATHYMETRY=mean(data$BATHYMETRY),SST=mean(data$SST),BS=mean(data$BS), SSH=mean(data$SSH),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBT<-summaryBy(pred~BT,data=p.data) outBT$pred.Trans<-plogis(outBT$pred.mean) recbt[1:100,i]<-outBT$pred.Trans #SST if(any(vars=="SST")){ xSST<-seq(min(data$SST),max(data$SST),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),SST=xSST,BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),BS=mean(data$BS), SSH=mean(data$SSH),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSST<-summaryBy(pred~SST,data=p.data) outSST$pred.Trans<-plogis(outSST$pred.mean) recsst[1:100,i]<-outSST$pred.Trans }else{ recsst[1:100,i]<-NA} #BS xBS<-seq(min(data$BS),max(data$BS),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=xBS,BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSH=mean(data$SSH),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBS<-summaryBy(pred~BS,data=p.data) outBS$pred.Trans<-plogis(outBS$pred.mean) recbs[1:100,i]<-outBS$pred.Trans #SSH if(any(vars=="SSH")){ xSSH<-seq(min(data$SSH),max(data$SSH),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSH=xSSH,CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSSH<-summaryBy(pred~SSH,data=p.data) outSSH$pred.Trans<-plogis(outSSH$pred.mean) recssh[1:100,i]<-outSSH$pred.Trans }else{ recssh[1:100,i]<-NA} #CHLA xCHLA<-seq(min(data$CHLA),max(data$CHLA),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSH=mean(data$SSH),CHLA=xCHLA,BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outCHLA<-summaryBy(pred~CHLA,data=p.data) outCHLA$pred.Trans<-plogis(outCHLA$pred.mean) recchl[1:100,i]<-outCHLA$pred.Trans #SSTSD if(any(vars=="SSTSD")){ xSSTSD<-seq(min(data$SSTSD),max(data$SSTSD),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSTSD=xSSTSD,BTSD=mean(data$BTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSSTSD<-summaryBy(pred~SSTSD,data=p.data) outSSTSD$pred.Trans<-plogis(outSSTSD$pred.mean) recsstsd[1:100,i]<-outSSTSD$pred.Trans }else{ recsstsd[1:100,i]<-NA} #BTSD if(any(vars=="BTSD")){ xBTSD<-seq(min(data$BTSD),max(data$BTSD),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), BTSD=xBTSD,SSTSD=mean(data$SSTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBTSD<-summaryBy(pred~BTSD,data=p.data) outBTSD$pred.Trans<-plogis(outBTSD$pred.mean) recbtsd[1:100,i]<-outBTSD$pred.Trans }else{ recbtsd[1:100,i]<-NA} #Set_Begin_Hour if(any(vars=="Set_Begin_Hour")){ xSet_Begin_Hour<-seq(min(data$Set_Begin_Hour),max(data$Set_Begin_Hour),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),Set_Begin_Hour=xSet_Begin_Hour, BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSet_Begin_Hour<-summaryBy(pred~Set_Begin_Hour,data=p.data) outSet_Begin_Hour$pred.Trans<-plogis(outSet_Begin_Hour$pred.mean) recsethour[1:100,i]<-outSet_Begin_Hour$pred.Trans }else{ recsethour[1:100,i]<-NA} #MONTH p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outMONTH<-summaryBy(pred~MONTH,data=p.data) outMONTH$pred.Trans<-plogis(outMONTH$pred.mean) recmonth[1:12,i]<-outMONTH$pred.Trans #YEAR outYEAR<-summaryBy(pred~YEAR,data=p.data) outYEAR$pred.Trans<-plogis(outYEAR$pred.mean) recyear[1:15,i]<-outYEAR$pred.Trans #HOOK_CONFIG #outHOOK_CONFIG<-summaryBy(pred~HOOK_CONFIG,data=p.data) #outHOOK_CONFIG$pred.Trans<-plogis(outHOOK_CONFIG$pred.mean) #rechook[1:4,i]<-outHOOK_CONFIG$pred.Trans #BAITTYPE #outBAITTYPE<-summaryBy(pred~BAITTYPE,data=p.data) #outBAITTYPE$pred.Trans<-plogis(outBAITTYPE$pred.mean) #recbait[1:3,i]<-outBAITTYPE$pred.Trans }else{ recbat[1:100,i]<-NA #recrug[1:100,i]<-NA recsst[1:100,i]<-NA recssh[1:100,i]<-NA recchl[1:100,i]<-NA recbt[1:100,i]<-NA recbs[1:100,i]<-NA recbtsd[1:100,i]<-NA recsstsd[1:100,i]<-NA recsethour[1:100,i]<-NA #rechook[1:4,i]<-NA #recbait[1:3,i]<-NA recmonth[1:12,i]<-NA recyear[1:15,i]<-NA } print(i) } return(list(recbat=recbat, recbt=recbt, recsst=recsst, recssh=recssh, recchl=recchl, recbs=recbs, recbtsd=recbtsd, recsstsd=recsstsd, recsethour=recsethour, recmonth=recmonth, recyear=recyear)) } #with hook, no bait BinGAMbootH<-function(data=NA,gam.formula=NA,exclude=NULL,family=NA,method=NULL){ recbat<-matrix(nrow=100, ncol=1000, NA) #recrug<-matrix(nrow=100, ncol=1000, NA) recsst<-matrix(nrow=100, ncol=1000, NA) recssh<-matrix(nrow=100, ncol=1000, NA) recchl<-matrix(nrow=100, ncol=1000, NA) recbt<-matrix(nrow=100, ncol=1000, NA) recbs<-matrix(nrow=100, ncol=1000, NA) recsstsd<-matrix(nrow=100, ncol=1000, NA) recbtsd<-matrix(nrow=100, ncol=1000, NA) recsethour<-matrix(nrow=100, ncol=1000,NA) recyear<-matrix(nrow=15, ncol=1000, NA) rechook<-matrix(nrow=4, ncol=1000,NA) #recbait<-matrix(nrow=3,ncol=1000,NA) recmonth<-matrix(nrow=12, ncol=1000, NA) for(i in 1:1000){ boot.sample<-data[sample(nrow(data),nrow(data),replace=TRUE),] mBoot<-tryCatch({ gam(gam.formula,data=boot.sample,family=family,method=method) }, warning=function(w){ message("handling warning:",conditionMessage(w)) "Bad" },error=function(e){ message("ignore error:",conditionMessage(e)) "Bad" }) if(mBoot[1]!="Bad"){ #Bathymetry xBat<-seq(min(data$BATHYMETRY),max(data$BATHYMETRY),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BATHYMETRY=xBat,BT=mean(data$BT),SST=mean(data$SST),BS=mean(data$BS), SSH=mean(data$SSH),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBat<-summaryBy(pred~BATHYMETRY,data=p.data) outBat$pred.Trans<-plogis(outBat$pred.mean) recbat[1:100,i]<-outBat$pred.Trans #BT xBT<-seq(min(data$BT),max(data$BT),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BT=xBT,BATHYMETRY=mean(data$BATHYMETRY),SST=mean(data$SST),BS=mean(data$BS), SSH=mean(data$SSH),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBT<-summaryBy(pred~BT,data=p.data) outBT$pred.Trans<-plogis(outBT$pred.mean) recbt[1:100,i]<-outBT$pred.Trans #SST xSST<-seq(min(data$SST),max(data$SST),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),SST=xSST,BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),BS=mean(data$BS), SSH=mean(data$SSH),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSST<-summaryBy(pred~SST,data=p.data) outSST$pred.Trans<-plogis(outSST$pred.mean) recsst[1:100,i]<-outSST$pred.Trans #BS xBS<-seq(min(data$BS),max(data$BS),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=xBS,BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSH=mean(data$SSH),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBS<-summaryBy(pred~BS,data=p.data) outBS$pred.Trans<-plogis(outBS$pred.mean) recbs[1:100,i]<-outBS$pred.Trans #SSH xSSH<-seq(min(data$SSH),max(data$SSH),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSH=xSSH,CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSSH<-summaryBy(pred~SSH,data=p.data) outSSH$pred.Trans<-plogis(outSSH$pred.mean) recssh[1:100,i]<-outSSH$pred.Trans #CHLA xCHLA<-seq(min(data$CHLA),max(data$CHLA),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSH=mean(data$SSH),CHLA=xCHLA,BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outCHLA<-summaryBy(pred~CHLA,data=p.data) outCHLA$pred.Trans<-plogis(outCHLA$pred.mean) recchl[1:100,i]<-outCHLA$pred.Trans #SSTSD xSSTSD<-seq(min(data$SSTSD),max(data$SSTSD),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSTSD=xSSTSD,BTSD=mean(data$BTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSSTSD<-summaryBy(pred~SSTSD,data=p.data) outSSTSD$pred.Trans<-plogis(outSSTSD$pred.mean) recsstsd[1:100,i]<-outSSTSD$pred.Trans #BTSD xBTSD<-seq(min(data$BTSD),max(data$BTSD),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), BTSD=xBTSD,SSTSD=mean(data$SSTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBTSD<-summaryBy(pred~BTSD,data=p.data) outBTSD$pred.Trans<-plogis(outBTSD$pred.mean) recbtsd[1:100,i]<-outBTSD$pred.Trans #Set_Begin_Hour xSet_Begin_Hour<-seq(min(data$Set_Begin_Hour),max(data$Set_Begin_Hour),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=xSet_Begin_Hour, Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSet_Begin_Hour<-summaryBy(pred~Set_Begin_Hour,data=p.data) outSet_Begin_Hour$pred.Trans<-plogis(outSet_Begin_Hour$pred.mean) recsethour[1:100,i]<-outSet_Begin_Hour$pred.Trans #MONTH p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outMONTH<-summaryBy(pred~MONTH,data=p.data) outMONTH$pred.Trans<-plogis(outMONTH$pred.mean) recmonth[1:12,i]<-outMONTH$pred.Trans #YEAR outYEAR<-summaryBy(pred~YEAR,data=p.data) outYEAR$pred.Trans<-plogis(outYEAR$pred.mean) recyear[1:15,i]<-outYEAR$pred.Trans #HOOK_CONFIG outHOOK_CONFIG<-summaryBy(pred~HOOK_CONFIG,data=p.data) outHOOK_CONFIG$pred.Trans<-plogis(outHOOK_CONFIG$pred.mean) rechook[1:4,i]<-outHOOK_CONFIG$pred.Trans #BAITTYPE #outBAITTYPE<-summaryBy(pred~BAITTYPE,data=p.data) #outBAITTYPE$pred.Trans<-plogis(outBAITTYPE$pred.mean) #recbait[1:3,i]<-outBAITTYPE$pred.Trans }else{ recbat[1:100,i]<-NA #recrug[1:100,i]<-NA recsst[1:100,i]<-NA recssh[1:100,i]<-NA recchl[1:100,i]<-NA recbt[1:100,i]<-NA recbs[1:100,i]<-NA recbtsd[1:100,i]<-NA recsstsd[1:100,i]<-NA recsethour[1:100,i]<-NA rechook[1:4,i]<-NA #recbait[1:3,i]<-NA recmonth[1:12,i]<-NA recyear[1:15,i]<-NA } print(i) } return(list(recbat=recbat, recbt=recbt, recsst=recsst, recssh=recssh, recchl=recchl, recbs=recbs, recbtsd=recbtsd, recsstsd=recsstsd, recsethour=recsethour, recmonth=recmonth, recyear=recyear, rechook=rechook)) } #with bait, no hook BinGAMbootB<-function(data=NA,gam.formula=NA,exclude=NULL,family=NA,method=NULL){ recbat<-matrix(nrow=100, ncol=1000, NA) #recrug<-matrix(nrow=100, ncol=1000, NA) recsst<-matrix(nrow=100, ncol=1000, NA) recssh<-matrix(nrow=100, ncol=1000, NA) recchl<-matrix(nrow=100, ncol=1000, NA) recturb<-matrix(nrow=100, ncol=1000, NA) recbt<-matrix(nrow=100, ncol=1000, NA) recbs<-matrix(nrow=100, ncol=1000, NA) recsstsd<-matrix(nrow=100, ncol=1000, NA) recbtsd<-matrix(nrow=100, ncol=1000, NA) recsethour<-matrix(nrow=100, ncol=1000,NA) recyear<-matrix(nrow=15, ncol=1000, NA) #rechook<-matrix(nrow=4, ncol=1000,NA) recbait<-matrix(nrow=3,ncol=1000,NA) recmonth<-matrix(nrow=12, ncol=1000, NA) for(i in 1:1000){ boot.sample<-data[sample(nrow(data),nrow(data),replace=TRUE),] mBoot<-tryCatch({ gam(gam.formula,data=boot.sample,family=family,method=method) }, warning=function(w){ message("handling warning:",conditionMessage(w)) "Bad" },error=function(e){ message("ignore error:",conditionMessage(e)) "Bad" }) vars<-attr(m$terms,"term.labels") if(mBoot[1]!="Bad"){ #Bathymetry xBat<-seq(min(data$BATHYMETRY),max(data$BATHYMETRY),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BATHYMETRY=xBat,BT=mean(data$BT),SST=mean(data$SST),BS=mean(data$BS), SSH=mean(data$SSH),TURB=mean(data$TURB),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),BAITTYPE=levels(data$BAITTYPE),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBat<-summaryBy(pred~BATHYMETRY,data=p.data) outBat$pred.Trans<-plogis(outBat$pred.mean) recbat[1:100,i]<-outBat$pred.Trans #BT xBT<-seq(min(data$BT),max(data$BT),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BT=xBT,BATHYMETRY=mean(data$BATHYMETRY),SST=mean(data$SST),BS=mean(data$BS), SSH=mean(data$SSH),TURB=mean(data$TURB),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),BAITTYPE=levels(data$BAITTYPE),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBT<-summaryBy(pred~BT,data=p.data) outBT$pred.Trans<-plogis(outBT$pred.mean) recbt[1:100,i]<-outBT$pred.Trans #SST if(any(vars=="SST")){ xSST<-seq(min(data$SST),max(data$SST),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),SST=xSST,BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),BS=mean(data$BS), SSH=mean(data$SSH),TURB=mean(data$TURB),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),BAITTYPE=levels(data$BAITTYPE),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSST<-summaryBy(pred~SST,data=p.data) outSST$pred.Trans<-plogis(outSST$pred.mean) recsst[1:100,i]<-outSST$pred.Trans }else{ recsst[1:100,i]<-NA} #BS xBS<-seq(min(data$BS),max(data$BS),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=xBS,BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSH=mean(data$SSH),TURB=mean(data$TURB),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),BAITTYPE=levels(data$BAITTYPE),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBS<-summaryBy(pred~BS,data=p.data) outBS$pred.Trans<-plogis(outBS$pred.mean) recbs[1:100,i]<-outBS$pred.Trans #SSH if(any(vars=="SSH")){ xSSH<-seq(min(data$SSH),max(data$SSH),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSH=xSSH,TURB=mean(data$TURB),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),BAITTYPE=levels(data$BAITTYPE),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSSH<-summaryBy(pred~SSH,data=p.data) outSSH$pred.Trans<-plogis(outSSH$pred.mean) recssh[1:100,i]<-outSSH$pred.Trans }else{ recssh[1:100,i]<-NA} #CHLA if(any(vars=="CHLA")){ xCHLA<-seq(min(data$CHLA),max(data$CHLA),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSH=mean(data$SSH),TURB=mean(data$TURB),CHLA=xCHLA,BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),BAITTYPE=levels(data$BAITTYPE),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outCHLA<-summaryBy(pred~CHLA,data=p.data) outCHLA$pred.Trans<-plogis(outCHLA$pred.mean) recchl[1:100,i]<-outCHLA$pred.Trans }else{ recchl[1:100,i]<-NA} #TURB if(any(vars=="TURB")){ xTURB<-seq(min(data$TURB),max(data$TURB),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSH=mean(data$SSH),TURB=xTURB,CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),BAITTYPE=levels(data$BAITTYPE),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outTURB<-summaryBy(pred~TURB,data=p.data) outTURB$pred.Trans<-plogis(outTURB$pred.mean) recturb[1:100,i]<-outTURB$pred.Trans }else{ recturb[1:100,i]<-NA} #SSTSD xSSTSD<-seq(min(data$SSTSD),max(data$SSTSD),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSTSD=xSSTSD,TURB=mean(data$TURB),BTSD=mean(data$BTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),BAITTYPE=levels(data$BAITTYPE),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSSTSD<-summaryBy(pred~SSTSD,data=p.data) outSSTSD$pred.Trans<-plogis(outSSTSD$pred.mean) recsstsd[1:100,i]<-outSSTSD$pred.Trans #BTSD xBTSD<-seq(min(data$BTSD),max(data$BTSD),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), BTSD=xBTSD,TURB=mean(data$TURB),SSTSD=mean(data$SSTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),BAITTYPE=levels(data$BAITTYPE),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBTSD<-summaryBy(pred~BTSD,data=p.data) outBTSD$pred.Trans<-plogis(outBTSD$pred.mean) recbtsd[1:100,i]<-outBTSD$pred.Trans #Set_Begin_Hour xSet_Begin_Hour<-seq(min(data$Set_Begin_Hour),max(data$Set_Begin_Hour),length=100) p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), BTSD=mean(data$BTSD),TURB=mean(data$TURB),SSTSD=mean(data$SSTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),BAITTYPE=levels(data$BAITTYPE),Set_Begin_Hour=xSet_Begin_Hour, BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSet_Begin_Hour<-summaryBy(pred~Set_Begin_Hour,data=p.data) outSet_Begin_Hour$pred.Trans<-plogis(outSet_Begin_Hour$pred.mean) recsethour[1:100,i]<-outSet_Begin_Hour$pred.Trans #MONTH p.data<-expand.grid(YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), BTSD=mean(data$BTSD),TURB=mean(data$TURB),SSTSD=mean(data$SSTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),BAITTYPE=levels(data$BAITTYPE),Set_Begin_Hour=mean(data$Set_Begin_Hour), BEGIN_SET_LATITUDE=mean(data$BEGIN_SET_LATITUDE),BEGIN_SET_LONGITUDE=mean(data$BEGIN_SET_LONGITUDE),logEFFORT=mean(data$logEFFORT)) p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outMONTH<-summaryBy(pred~MONTH,data=p.data) outMONTH$pred.Trans<-plogis(outMONTH$pred.mean) recmonth[1:12,i]<-outMONTH$pred.Trans #YEAR outYEAR<-summaryBy(pred~YEAR,data=p.data) outYEAR$pred.Trans<-plogis(outYEAR$pred.mean) recyear[1:15,i]<-outYEAR$pred.Trans #HOOK_CONFIG #outHOOK_CONFIG<-summaryBy(pred~HOOK_CONFIG,data=p.data) #outHOOK_CONFIG$pred.Trans<-plogis(outHOOK_CONFIG$pred.mean) #rechook[1:4,i]<-outHOOK_CONFIG$pred.Trans #BAITTYPE outBAITTYPE<-summaryBy(pred~BAITTYPE,data=p.data) outBAITTYPE$pred.Trans<-plogis(outBAITTYPE$pred.mean) recbait[1:3,i]<-outBAITTYPE$pred.Trans }else{ recbat[1:100,i]<-NA #recrug[1:100,i]<-NA recsst[1:100,i]<-NA recssh[1:100,i]<-NA recchl[1:100,i]<-NA recturb[1:100,i]<-NA recbt[1:100,i]<-NA recbs[1:100,i]<-NA recbtsd[1:100,i]<-NA recsstsd[1:100,i]<-NA recsethour[1:100,i]<-NA #rechook[1:4,i]<-NA recbait[1:3,i]<-NA recmonth[1:12,i]<-NA recyear[1:15,i]<-NA } print(i) } return(list(recbat=recbat, recbt=recbt, recsst=recsst, recssh=recssh, recchl=recchl, recturb=recturb, recbs=recbs, recbtsd=recbtsd, recsstsd=recsstsd, recsethour=recsethour, recmonth=recmonth, recyear=recyear, recbait=recbait)) } #with hook and bait BinGAMbootHB<-function(data=NA,gam.formula=NA,exclude=NULL,family=NA,method=NULL){ recbat<-matrix(nrow=100, ncol=1000, NA) #recrug<-matrix(nrow=100, ncol=1000, NA) recsst<-matrix(nrow=100, ncol=1000, NA) recssh<-matrix(nrow=100, ncol=1000, NA) recchl<-matrix(nrow=100, ncol=1000, NA) recbt<-matrix(nrow=100, ncol=1000, NA) recbs<-matrix(nrow=100, ncol=1000, NA) recsstsd<-matrix(nrow=100, ncol=1000, NA) recbtsd<-matrix(nrow=100, ncol=1000, NA) recsethour<-matrix(nrow=100, ncol=1000,NA) recyear<-matrix(nrow=15, ncol=1000, NA) rechook<-matrix(nrow=4, ncol=1000,NA) recbait<-matrix(nrow=3,ncol=1000,NA) recmonth<-matrix(nrow=12, ncol=1000, NA) for(i in 1:1000){ boot.sample<-data[sample(nrow(data),nrow(data),replace=TRUE),] mBoot<-tryCatch({ gam(gam.formula,data=boot.sample,family=family,method=method) }, warning=function(w){ message("handling warning:",conditionMessage(w)) "Bad" },error=function(e){ message("ignore error:",conditionMessage(e)) "Bad" }) if(mBoot[1]!="Bad"){ #Bathymetry xBat<-seq(min(data$BATHYMETRY),max(data$BATHYMETRY),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BATHYMETRY=xBat,BT=mean(data$BT),SST=mean(data$SST),BS=mean(data$BS), SSH=mean(data$SSH),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),BAITTYPE=levels(data$BAITTYPE),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBat<-summaryBy(pred~BATHYMETRY,data=p.data) outBat$pred.Trans<-plogis(outBat$pred.mean) recbat[1:100,i]<-outBat$pred.Trans #BT xBT<-seq(min(data$BT),max(data$BT),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BT=xBT,BATHYMETRY=mean(data$BATHYMETRY),SST=mean(data$SST),BS=mean(data$BS), SSH=mean(data$SSH),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),BAITTYPE=levels(data$BAITTYPE),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBT<-summaryBy(pred~BT,data=p.data) outBT$pred.Trans<-plogis(outBT$pred.mean) recbt[1:100,i]<-outBT$pred.Trans #SST xSST<-seq(min(data$SST),max(data$SST),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),SST=xSST,BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),BS=mean(data$BS), SSH=mean(data$SSH),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),BAITTYPE=levels(data$BAITTYPE),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSST<-summaryBy(pred~SST,data=p.data) outSST$pred.Trans<-plogis(outSST$pred.mean) recsst[1:100,i]<-outSST$pred.Trans #BS xBS<-seq(min(data$BS),max(data$BS),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=xBS,BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSH=mean(data$SSH),CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),BAITTYPE=levels(data$BAITTYPE),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBS<-summaryBy(pred~BS,data=p.data) outBS$pred.Trans<-plogis(outBS$pred.mean) recbs[1:100,i]<-outBS$pred.Trans #SSH xSSH<-seq(min(data$SSH),max(data$SSH),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSH=xSSH,CHLA=mean(data$CHLA),BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),BAITTYPE=levels(data$BAITTYPE),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSSH<-summaryBy(pred~SSH,data=p.data) outSSH$pred.Trans<-plogis(outSSH$pred.mean) recssh[1:100,i]<-outSSH$pred.Trans #CHLA xCHLA<-seq(min(data$CHLA),max(data$CHLA),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSH=mean(data$SSH),CHLA=xCHLA,BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),BAITTYPE=levels(data$BAITTYPE),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outCHLA<-summaryBy(pred~CHLA,data=p.data) outCHLA$pred.Trans<-plogis(outCHLA$pred.mean) recchl[1:100,i]<-outCHLA$pred.Trans #SSTSD xSSTSD<-seq(min(data$SSTSD),max(data$SSTSD),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), SSTSD=xSSTSD,BTSD=mean(data$BTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),BAITTYPE=levels(data$BAITTYPE),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSSTSD<-summaryBy(pred~SSTSD,data=p.data) outSSTSD$pred.Trans<-plogis(outSSTSD$pred.mean) recsstsd[1:100,i]<-outSSTSD$pred.Trans #BTSD xBTSD<-seq(min(data$BTSD),max(data$BTSD),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), BTSD=xBTSD,SSTSD=mean(data$SSTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),BAITTYPE=levels(data$BAITTYPE),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outBTSD<-summaryBy(pred~BTSD,data=p.data) outBTSD$pred.Trans<-plogis(outBTSD$pred.mean) recbtsd[1:100,i]<-outBTSD$pred.Trans #Set_Begin_Hour xSet_Begin_Hour<-seq(min(data$Set_Begin_Hour),max(data$Set_Begin_Hour),length=100) p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),BAITTYPE=levels(data$BAITTYPE),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=xSet_Begin_Hour, Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outSet_Begin_Hour<-summaryBy(pred~Set_Begin_Hour,data=p.data) outSet_Begin_Hour$pred.Trans<-plogis(outSet_Begin_Hour$pred.mean) recsethour[1:100,i]<-outSet_Begin_Hour$pred.Trans #MONTH p.data<-expand.grid(Dayn=mean(data$Dayn),YEAR=unique(data$YEAR),MONTH=levels(data$MONTH),BS=mean(data$BS),BATHYMETRY=mean(data$BATHYMETRY),BT=mean(data$BT),SST=mean(data$SST), BTSD=mean(data$BTSD),SSTSD=mean(data$SSTSD),SSH=mean(data$SSH),CHLA=mean(data$CHLA),BAITTYPE=levels(data$BAITTYPE),HOOK_CONFIG=levels(data$HOOK_CONFIG),Set_Begin_Hour=mean(data$Set_Begin_Hour), Lat_Lon1="30_-80",logEFFORT=mean(data$logEFFORT))#the Lat_Lon1 value I choose doesn't matter because it is excluded in predict p.data<-cbind(p.data,pred=predict(mBoot,type="link",newdata=p.data,exclude=exclude)) outMONTH<-summaryBy(pred~MONTH,data=p.data) outMONTH$pred.Trans<-plogis(outMONTH$pred.mean) recmonth[1:12,i]<-outMONTH$pred.Trans #YEAR outYEAR<-summaryBy(pred~YEAR,data=p.data) outYEAR$pred.Trans<-plogis(outYEAR$pred.mean) recyear[1:15,i]<-outYEAR$pred.Trans #HOOK_CONFIG outHOOK_CONFIG<-summaryBy(pred~HOOK_CONFIG,data=p.data) outHOOK_CONFIG$pred.Trans<-plogis(outHOOK_CONFIG$pred.mean) rechook[1:4,i]<-outHOOK_CONFIG$pred.Trans #BAITTYPE outBAITTYPE<-summaryBy(pred~BAITTYPE,data=p.data) outBAITTYPE$pred.Trans<-plogis(outBAITTYPE$pred.mean) recbait[1:3,i]<-outBAITTYPE$pred.Trans }else{ recbat[1:100,i]<-NA #recrug[1:100,i]<-NA recsst[1:100,i]<-NA recssh[1:100,i]<-NA recchl[1:100,i]<-NA recbt[1:100,i]<-NA recbs[1:100,i]<-NA recbtsd[1:100,i]<-NA recsstsd[1:100,i]<-NA recsethour[1:100,i]<-NA rechook[1:4,i]<-NA recbait[1:3,i]<-NA recmonth[1:12,i]<-NA recyear[1:15,i]<-NA } print(i) } return(list(recbat=recbat, recbt=recbt, recsst=recsst, recssh=recssh, recchl=recchl, recbs=recbs, recbtsd=recbtsd, recsstsd=recsstsd, recsethour=recsethour, recmonth=recmonth, recyear=recyear, rechook=rechook, recbait=recbait)) } CIfuncFix<-function(data=NA){ CIrec<-NULL for(i in 1:nrow(data)){ datrow<-data[i,] CI<-1.96*sd(datrow) CIrec<-c(CIrec,CI) } return(CIrec) } ####### #SB ####### #read in model data output #no hook or bait is in model so use BinGAMboot BinGAMbootout<-BinGAMboot(data=sb_datanona,gam.formula = formula,exclude=NULL,family="binomial",method="GCV.Cp") #save bootstrap output as "speciesname_BinGAMbootout.RData" batbootmat<-BinGAMbootout$recbat[, colSums(is.na(BinGAMbootout$recbat)) != nrow(BinGAMbootout$recbat)] btbootmat<-BinGAMbootout$recbt[, colSums(is.na(BinGAMbootout$recbt)) != nrow(BinGAMbootout$recbt)] sstbootmat<-BinGAMbootout$recsst[, colSums(is.na(BinGAMbootout$recsst)) != nrow(BinGAMbootout$recsst)] bsbootmat<-BinGAMbootout$recbs[, colSums(is.na(BinGAMbootout$recbs)) != nrow(BinGAMbootout$recbs)] chlbootmat<-BinGAMbootout$recchl[, colSums(is.na(BinGAMbootout$recchl)) != nrow(BinGAMbootout$recchl)] sshbootmat<-BinGAMbootout$recssh[, colSums(is.na(BinGAMbootout$recssh)) != nrow(BinGAMbootout$recssh)] #btsdbootmat<-BinGAMbootout$recbtsd[, colSums(is.na(BinGAMbootout$recbtsd)) != nrow(BinGAMbootout$recbtsd)] #sstsdbootmat<-BinGAMbootout$recsstsd[, colSums(is.na(BinGAMbootout$recsstsd)) != nrow(BinGAMbootout$recsstsd)] monthbootmat<-BinGAMbootout$recmonth[, colSums(is.na(BinGAMbootout$recmonth)) != nrow(BinGAMbootout$recmonth)] yearbootmat<-BinGAMbootout$recyear[, colSums(is.na(BinGAMbootout$recyear)) != nrow(BinGAMbootout$recyear)] #sethourbootmat<-BinGAMbootout$recsethour[, colSums(is.na(BinGAMbootout$recsethour)) != nrow(BinGAMbootout$recsethour)] #hookbootmat<-BinGAMbootout$rechook[, colSums(is.na(BinGAMbootout$rechook)) != nrow(BinGAMbootout$rechook)] #baitbootmat<-BinGAMbootout$recbait[, colSums(is.na(BinGAMbootout$recbait)) != nrow(BinGAMbootout$recbait)] outbootCIbat<-CIfuncFix(batbootmat) outbootCIbt<-CIfuncFix(btbootmat) outbootCIsst<-CIfuncFix(sstbootmat) outbootCIbs<-CIfuncFix(bsbootmat) outbootCIchl<-CIfuncFix(chlbootmat) outbootCIssh<-CIfuncFix(sshbootmat) #outbootCIbtsd<-CIfuncFix(btsdbootmat) #outbootCIsstsd<-CIfuncFix(sstsdbootmat) outbootCImonth<-CIfuncFix(monthbootmat) outbootCIyear<-CIfuncFix(yearbootmat) #outbootCIsethour<-CIfuncFix(sethourbootmat) #outbootCIhook<-CIfuncFix(hookbootmat) #outbootCIbait<-CIfuncFix(baitbootmat) #add and subtract CI from mean outbatupperCI<-outBat$pred.Trans+outbootCIbat outbatlowerCI<-outBat$pred.Trans-outbootCIbat outbtupperCI<-outBT$pred.Trans+outbootCIbt outbtlowerCI<-outBT$pred.Trans-outbootCIbt outsstupperCI<-outSST$pred.Trans+outbootCIsst outsstlowerCI<-outSST$pred.Trans-outbootCIsst outbsupperCI<-outBS$pred.Trans+outbootCIbs outbslowerCI<-outBS$pred.Trans-outbootCIbs outchlupperCI<-outCHLA$pred.Trans+outbootCIchl outchllowerCI<-outCHLA$pred.Trans-outbootCIchl outsshupperCI<-outSSH$pred.Trans+outbootCIssh outsshlowerCI<-outSSH$pred.Trans-outbootCIssh #outbtsdupperCI<-outBTSD$pred.Trans+outbootCIbtsd #outbtsdlowerCI<-outBTSD$pred.Trans-outbootCIbtsd #outsstsdupperCI<-outSSTSD$pred.Trans+outbootCIsstsd #outsstsdlowerCI<-outSSTSD$pred.Trans-outbootCIsstsd outmonthupperCI<-outMONTH$pred.Trans+outbootCImonth outmonthlowerCI<-outMONTH$pred.Trans-outbootCImonth outyearupperCI<-outYEAR$pred.Trans+outbootCIyear outyearlowerCI<-outYEAR$pred.Trans-outbootCIyear #outsethourupperCI<-outSet_Begin_Hour$pred.Trans+outbootCIsethour #outsethourlowerCI<-outSet_Begin_Hour$pred.Trans-outbootCIsethour #outhookupperCI<-outHOOK_CONFIG$pred.Trans+outbootCIhook #outhooklowerCI<-outHOOK_CONFIG$pred.Trans-outbootCIhook #outbaitupperCI<-outBAITTYPE$pred.Trans+outbootCIbait #outbaitlowerCI<-outBAITTYPE$pred.Trans-outbootCIbait #marginal mean plots with errors tiff("sb_response_curvestiff.tif",width=8.5,height=6,units="in",res=300) par(mfrow=c(3,3)) par(mar=c(5,1,1,2),oma=c(0,3.5,0,0)) plot(pred.Trans~BATHYMETRY, data=outBat,type="l",ylim=c(0,1),ylab="",xlab="Bathymetry (m)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xBat,rev(xBat)),y=c(outbatlowerCI,rev(outbatupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~BT, data=outBT,type="l",ylim=c(0,1),ylab="",xlab="Bottom Temperature (ºC)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xBT,rev(xBT)),y=c(outbtlowerCI,rev(outbtupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~SST, data=outSST,type="l",ylim=c(0,1),ylab="",xlab="SST (ºC)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xSST,rev(xSST)),y=c(outsstlowerCI,rev(outsstupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~BS, data=outBS,type="l",ylim=c(0,1),ylab="",xlab="Bottom Salinity (ppt)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xBS,rev(xBS)),y=c(outbslowerCI,rev(outbsupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) mtext(side=2,line=3,"Probability of Occurrence") plot(pred.Trans~CHLA, data=outCHLA,type="l",ylim=c(0,1),ylab="",xlab=expression(paste("Chlorophyll a (mg m"^"-3",")")),cex.lab=1.25,cex.axis=1.15) polygon(x=c(xCHLA,rev(xCHLA)),y=c(outchllowerCI,rev(outchlupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~SSH, data=outSSH,type="l",ylim=c(0,1),ylab="",xlab="SSH (m)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xSSH,rev(xSSH)),y=c(outsshlowerCI,rev(outsshupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) #plot(pred.Trans~BTSD, data=outBTSD,type="l",ylim=c(0,1),ylab="",xlab="Bottom Temperature SD (ºC)",cex.lab=1.25,cex.axis=1.15) #polygon(x=c(xBTSD,rev(xBTSD)),y=c(outbtsdlowerCI,rev(outbtsdupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) #plot(pred.Trans~SSTSD, data=outSSTSD,type="l",ylim=c(0,1),ylab="",xlab="SST SD (ºC)",cex.lab=1.25,cex.axis=1.15) #polygon(x=c(xSSTSD,rev(xSSTSD)),y=c(outsstsdlowerCI,rev(outsstsdupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) #plot(pred.Trans~Set_Begin_Hour, data=outSet_Begin_Hour,type="l",ylim=c(0,1),ylab="",xlab="Begin Set Time (hr)",cex.lab=1.25,cex.axis=1.15) #polygon(x=c(xSet_Begin_Hour,rev(xSet_Begin_Hour)),y=c(outsethourlowerCI,rev(outsethourupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) #plot(pred.Trans~c(1:4), data=outHOOK_CONFIG,type="b",ylim=c(0,1),xlab="Hook Configuration",ylab="",xaxt="n",cex.lab=1.25,cex.axis=1.15) #axis(side=1,at=c(1:4),c(">16/0C","<=16/0C",">=12/0J","<12/0J"),cex=0.8) #polygon(x=c(1:4,rev(1:4)),y=c(outhooklowerCI,rev(outhookupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) #plot(pred.Trans~c(1:3), data=outBAITTYPE,type="b",ylim=c(0,1),xlab="Bait Type",ylab="",xaxt="n",cex.lab=1.25,cex.axis=1.15) #axis(side=1,at=c(1:3),c("Elasmo","Teleost","Unknown"),cex=0.8) #polygon(x=c(1:3,rev(1:3)),y=c(outbaitlowerCI,rev(outbaitupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~c(1:12), data=outMONTH,type="b",ylim=c(0,1),xlab="Month",ylab="",cex.lab=1.25,cex.axis=1.15,xaxt="n") axis(side=1,at=c(1,4,7,10),c("Jan","Apr","Jul","Oct"),cex.axis=1.15) polygon(x=c(1:12,rev(1:12)),y=c(outmonthlowerCI,rev(outmonthupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~c(1:15), data=outYEAR,type="b",ylim=c(0,1),xlab="Year",ylab="",xaxt="n",cex.lab=1.25,cex.axis=1.15) axis(side=1,at=c(1,6,11),c("2005","2010","2015"),cex.axis=1.15) polygon(x=c(1:15,rev(1:15)),y=c(outyearlowerCI,rev(outyearupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) dev.off() ####### #DS ####### #read in model data output #hook and bait are not in model so use BinGAMboot BinGAMbootout<-BinGAMboot(data=ds_datanona,gam.formula = formula,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",family="quasibinomial",method="GCV.Cp") #save bootstrap output as "speciesname_BinGAMbootout.RData" batbootmat<-BinGAMbootout$recbat[, colSums(is.na(BinGAMbootout$recbat)) != nrow(BinGAMbootout$recbat)] btbootmat<-BinGAMbootout$recbt[, colSums(is.na(BinGAMbootout$recbt)) != nrow(BinGAMbootout$recbt)] bsbootmat<-BinGAMbootout$recbs[, colSums(is.na(BinGAMbootout$recbs)) != nrow(BinGAMbootout$recbs)] chlbootmat<-BinGAMbootout$recchl[, colSums(is.na(BinGAMbootout$recchl)) != nrow(BinGAMbootout$recchl)] btsdbootmat<-BinGAMbootout$recbtsd[, colSums(is.na(BinGAMbootout$recbtsd)) != nrow(BinGAMbootout$recbtsd)] sstsdbootmat<-BinGAMbootout$recsstsd[, colSums(is.na(BinGAMbootout$recsstsd)) != nrow(BinGAMbootout$recsstsd)] monthbootmat<-BinGAMbootout$recmonth[, colSums(is.na(BinGAMbootout$recmonth)) != nrow(BinGAMbootout$recmonth)] yearbootmat<-BinGAMbootout$recyear[, colSums(is.na(BinGAMbootout$recyear)) != nrow(BinGAMbootout$recyear)] sethourbootmat<-BinGAMbootout$recsethour[, colSums(is.na(BinGAMbootout$recsethour)) != nrow(BinGAMbootout$recsethour)] outbootCIbat<-CIfuncFix(batbootmat) outbootCIbt<-CIfuncFix(btbootmat) outbootCIbs<-CIfuncFix(bsbootmat) outbootCIchl<-CIfuncFix(chlbootmat) outbootCIbtsd<-CIfuncFix(btsdbootmat) outbootCIsstsd<-CIfuncFix(sstsdbootmat) outbootCImonth<-CIfuncFix(monthbootmat) outbootCIyear<-CIfuncFix(yearbootmat) outbootCIsethour<-CIfuncFix(sethourbootmat) #add and subtract CI from mean outbatupperCI<-outBat$pred.Trans+outbootCIbat outbatlowerCI<-outBat$pred.Trans-outbootCIbat outbtupperCI<-outBT$pred.Trans+outbootCIbt outbtlowerCI<-outBT$pred.Trans-outbootCIbt outbsupperCI<-outBS$pred.Trans+outbootCIbs outbslowerCI<-outBS$pred.Trans-outbootCIbs outchlupperCI<-outCHLA$pred.Trans+outbootCIchl outchllowerCI<-outCHLA$pred.Trans-outbootCIchl outbtsdupperCI<-outBTSD$pred.Trans+outbootCIbtsd outbtsdlowerCI<-outBTSD$pred.Trans-outbootCIbtsd outsstsdupperCI<-outSSTSD$pred.Trans+outbootCIsstsd outsstsdlowerCI<-outSSTSD$pred.Trans-outbootCIsstsd outmonthupperCI<-outMONTH$pred.Trans+outbootCImonth outmonthlowerCI<-outMONTH$pred.Trans-outbootCImonth outyearupperCI<-outYEAR$pred.Trans+outbootCIyear outyearlowerCI<-outYEAR$pred.Trans-outbootCIyear outsethourupperCI<-outSet_Begin_Hour$pred.Trans+outbootCIsethour outsethourlowerCI<-outSet_Begin_Hour$pred.Trans-outbootCIsethour #marginal means plot with errors tiff("ds_response_curvestiff.tif",width=8,height=6,units="in",res=300) par(mfrow=c(3,3)) par(mar=c(5,1,1,2),oma=c(0,3.5,0,0)) plot(pred.Trans~BATHYMETRY, data=outBat,type="l",ylim=c(0,1),ylab="",xlab="Bathymetry (m)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xBat,rev(xBat)),y=c(outbatlowerCI,rev(outbatupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~BT, data=outBT,type="l",ylim=c(0,1),ylab="",xlab="Bottom Temperature (ºC)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xBT,rev(xBT)),y=c(outbtlowerCI,rev(outbtupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~BS, data=outBS,type="l",ylim=c(0,1),ylab="",xlab="Bottom Salinity (ppt)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xBS,rev(xBS)),y=c(outbslowerCI,rev(outbsupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~CHLA, data=outCHLA,type="l",ylim=c(0,1),ylab="",xlab=expression(paste("Chlorophyll a (mg m"^"-3",")")),cex.lab=1.25,cex.axis=1.15) polygon(x=c(xCHLA,rev(xCHLA)),y=c(outchllowerCI,rev(outchlupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) mtext(side=2,line=3,"Probability of Occurrence") plot(pred.Trans~BTSD, data=outBTSD,type="l",ylim=c(0,1),ylab="",xlab="Bottom Temperature SD (ºC)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xBTSD,rev(xBTSD)),y=c(outbtsdlowerCI,rev(outbtsdupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~SSTSD, data=outSSTSD,type="l",ylim=c(0,1),ylab="",xlab="SST SD (ºC)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xSSTSD,rev(xSSTSD)),y=c(outsstsdlowerCI,rev(outsstsdupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~Set_Begin_Hour, data=outSet_Begin_Hour,type="l",ylim=c(0,1),ylab="",xlab="Begin Set Time (hr)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xSet_Begin_Hour,rev(xSet_Begin_Hour)),y=c(outsethourlowerCI,rev(outsethourupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) #plot(pred.Trans~c(1:4), data=outHOOK_CONFIG,type="b",ylim=c(0,1),xlab="Hook Configuration",ylab="",xaxt="n",cex.lab=1.25,cex.axis=1.15) #axis(side=1,at=c(1:4),c(">16/0C","<=16/0C",">=12/0J","<12/0J")) #polygon(x=c(1:4,rev(1:4)),y=c(outhooklowerCI,rev(outhookupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~c(1:12), data=outMONTH,type="b",ylim=c(0,1),xlab="Month",ylab="",cex.lab=1.25,cex.axis=1.15,xaxt="n") axis(side=1,at=c(1,4,7,10),c("Jan","Apr","Jul","Oct"),cex.axis=1.15) polygon(x=c(1:12,rev(1:12)),y=c(outmonthlowerCI,rev(outmonthupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~c(1:15), data=outYEAR,type="b",ylim=c(0,1),xlab="Year",ylab="",xaxt="n",cex.lab=1.25,cex.axis=1.15) axis(side=1,at=c(1,6,11),c("2005","2010","2015"),cex.axis=1.15) polygon(x=c(1:15,rev(1:15)),y=c(outyearlowerCI,rev(outyearupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) dev.off() ####### #SHH ####### #read in model data output #hook is in model so use BinGAMbootB BinGAMbootout<-BinGAMbootB(data=shh_datanona,gam.formula = formula,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",family="quasibinomial",method="GCV.Cp") #save bootstrap output as "speciesname_BinGAMbootout.RData" batbootmat<-BinGAMbootout$recbat[, colSums(is.na(BinGAMbootout$recbat)) != nrow(BinGAMbootout$recbat)] btbootmat<-BinGAMbootout$recbt[, colSums(is.na(BinGAMbootout$recbt)) != nrow(BinGAMbootout$recbt)] sstbootmat<-BinGAMbootout$recsst[, colSums(is.na(BinGAMbootout$recsst)) != nrow(BinGAMbootout$recsst)] bsbootmat<-BinGAMbootout$recbs[, colSums(is.na(BinGAMbootout$recbs)) != nrow(BinGAMbootout$recbs)] #chlbootmat<-BinGAMbootout$recchl[, colSums(is.na(BinGAMbootout$recchl)) != nrow(BinGAMbootout$recchl)] turbbootmat<-BinGAMbootout$recturb[, colSums(is.na(BinGAMbootout$recturb)) != nrow(BinGAMbootout$recturb)] sshbootmat<-BinGAMbootout$recssh[, colSums(is.na(BinGAMbootout$recssh)) != nrow(BinGAMbootout$recssh)] btsdbootmat<-BinGAMbootout$recbtsd[, colSums(is.na(BinGAMbootout$recbtsd)) != nrow(BinGAMbootout$recbtsd)] sstsdbootmat<-BinGAMbootout$recsstsd[, colSums(is.na(BinGAMbootout$recsstsd)) != nrow(BinGAMbootout$recsstsd)] monthbootmat<-BinGAMbootout$recmonth[, colSums(is.na(BinGAMbootout$recmonth)) != nrow(BinGAMbootout$recmonth)] sethourbootmat<-BinGAMbootout$recsethour[, colSums(is.na(BinGAMbootout$recsethour)) != nrow(BinGAMbootout$recsethour)] #hookbootmat<-BinGAMbootout$rechook[, colSums(is.na(BinGAMbootout$rechook)) != nrow(BinGAMbootout$rechook)] baitbootmat<-BinGAMbootout$recbait[, colSums(is.na(BinGAMbootout$recbait)) != nrow(BinGAMbootout$recbait)] yearbootmat<-BinGAMbootout$recyear[, colSums(is.na(BinGAMbootout$recyear)) != nrow(BinGAMbootout$recyear)] outbootCIbat<-CIfuncFix(batbootmat) outbootCIbt<-CIfuncFix(btbootmat) outbootCIsst<-CIfuncFix(sstbootmat) outbootCIbs<-CIfuncFix(bsbootmat) #outbootCIchl<-CIfuncFix(chlbootmat) outbootCIturb<-CIfuncFix(turbbootmat) outbootCIssh<-CIfuncFix(sshbootmat) outbootCIbtsd<-CIfuncFix(btsdbootmat) outbootCIsstsd<-CIfuncFix(sstsdbootmat) outbootCImonth<-CIfuncFix(monthbootmat) outbootCIsethour<-CIfuncFix(sethourbootmat) #outbootCIhook<-CIfuncFix(hookbootmat) outbootCIbait<-CIfuncFix(baitbootmat) outbootCIyear<-CIfuncFix(yearbootmat) #add and subtract CI from mean outbatupperCI<-outBat$pred.Trans+outbootCIbat outbatlowerCI<-outBat$pred.Trans-outbootCIbat outbtupperCI<-outBT$pred.Trans+outbootCIbt outbtlowerCI<-outBT$pred.Trans-outbootCIbt outsstupperCI<-outSST$pred.Trans+outbootCIsst outsstlowerCI<-outSST$pred.Trans-outbootCIsst outbsupperCI<-outBS$pred.Trans+outbootCIbs outbslowerCI<-outBS$pred.Trans-outbootCIbs #outchlupperCI<-outCHLA$pred.Trans+outbootCIchl #outchllowerCI<-outCHLA$pred.Trans-outbootCIchl outturbupperCI<-outTURB$pred.Trans+outbootCIturb outturblowerCI<-outTURB$pred.Trans-outbootCIturb outsshupperCI<-outSSH$pred.Trans+outbootCIssh outsshlowerCI<-outSSH$pred.Trans-outbootCIssh outbtsdupperCI<-outBTSD$pred.Trans+outbootCIbtsd outbtsdlowerCI<-outBTSD$pred.Trans-outbootCIbtsd outsstsdupperCI<-outSSTSD$pred.Trans+outbootCIsstsd outsstsdlowerCI<-outSSTSD$pred.Trans-outbootCIsstsd outmonthupperCI<-outMONTH$pred.Trans+outbootCImonth outmonthlowerCI<-outMONTH$pred.Trans-outbootCImonth outsethourupperCI<-outSet_Begin_Hour$pred.Trans+outbootCIsethour outsethourlowerCI<-outSet_Begin_Hour$pred.Trans-outbootCIsethour #outhookupperCI<-outHOOK_CONFIG$pred.Trans+outbootCIhook #outhooklowerCI<-outHOOK_CONFIG$pred.Trans-outbootCIhook outbaitupperCI<-outBAIT$pred.Trans+outbootCIbait outbaitlowerCI<-outBAIT$pred.Trans-outbootCIbait outyearupperCI<-outYEAR$pred.Trans+outbootCIyear outyearlowerCI<-outYEAR$pred.Trans-outbootCIyear #marginal means plots with errors tiff("shh_response_curvestiff.tif",width=8.5,height=6,units="in",res=300) par(mfrow=c(3,4)) par(mar=c(5,1,1,1.5),oma=c(0,3.5,0,0)) plot(pred.Trans~BATHYMETRY, data=outBat,type="l",ylim=c(0,1),ylab="",xlab="Bathymetry (m)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xBat,rev(xBat)),y=c(outbatlowerCI,rev(outbatupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~BT, data=outBT,type="l",ylim=c(0,1),ylab="",xlab="Bottom Temperature (ºC)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xBT,rev(xBT)),y=c(outbtlowerCI,rev(outbtupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~SST, data=outSST,type="l",ylim=c(0,1),ylab="",xlab="SST (ºC)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xSST,rev(xSST)),y=c(outsstlowerCI,rev(outsstupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~BS, data=outBS,type="l",ylim=c(0,1),ylab="",xlab="Bottom Salinity (ppt)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xBS,rev(xBS)),y=c(outbslowerCI,rev(outbsupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~TURB, data=outTURB,type="l",ylim=c(0,1),ylab="",xlab="Turbidity (m)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xTURB,rev(xTURB)),y=c(outturblowerCI,rev(outturbupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) mtext(side=2,line=3,"Probability of Occurrence") plot(pred.Trans~SSH, data=outSSH,type="l",ylim=c(0,1),ylab="",xlab="SSH (m)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xSSH,rev(xSSH)),y=c(outsshlowerCI,rev(outsshupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~BTSD, data=outBTSD,type="l",ylim=c(0,1),ylab="",xlab="Bottom Temperature SD (ºC)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xBTSD,rev(xBTSD)),y=c(outbtsdlowerCI,rev(outbtsdupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~SSTSD, data=outSSTSD,type="l",ylim=c(0,1),ylab="",xlab="SST SD (ºC)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xSSTSD,rev(xSSTSD)),y=c(outsstsdlowerCI,rev(outsstsdupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~Set_Begin_Hour, data=outSet_Begin_Hour,type="l",ylim=c(0,1),ylab="",xlab="Begin Set Time (hr)",cex.lab=1.25,cex.axis=1.15) polygon(x=c(xSet_Begin_Hour,rev(xSet_Begin_Hour)),y=c(outsethourlowerCI,rev(outsethourupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(c(NA,outBAIT$pred.Trans)~c(1:4),type="p",ylim=c(-0.05,1),xlab="Bait Type",ylab="",xaxt="n",cex.lab=1.25,cex.axis=1.15,xlim=c(1.75,4.25)) axis(side=1,at=c(2:4),c("Elasmo","Teleost","Unknown"),cex.axis=1.15,gap.axis=0.3) arrows(y0=outbaitlowerCI,y1=outbaitupperCI, x0=c(2:4),x1=c(2:4),code=3, angle=90,length=.1) plot(pred.Trans~c(1:12), data=outMONTH,type="b",ylim=c(0,1),xlab="Month",ylab="",cex.lab=1.25,cex.axis=1.15,xaxt="n") axis(side=1,at=c(1,4,7,10),c("Jan","Apr","Jul","Oct"),cex.axis=1.15) polygon(x=c(1:12,rev(1:12)),y=c(outmonthlowerCI,rev(outmonthupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) plot(pred.Trans~c(1:15), data=outYEAR,type="b",ylim=c(0,1),xlab="Year",ylab="",xaxt="n",cex.lab=1.25,cex.axis=1.15) axis(side=1,at=c(1,6,11),c("2005","2010","2015"),cex.axis=1.15) polygon(x=c(1:15,rev(1:15)),y=c(outyearlowerCI,rev(outyearupperCI)),col=rgb(0,0,0,alpha=0.4),border=NA) dev.off() ############### #Predicting ############### setwd("~/Enviornmental_Data/USmap") useez<-readOGR(".","US_Fed_Waters") eastcoastmask<-readOGR(".","east_coast_shp_to_mask_eez") useez<-spTransform(useez,"+init=epsg:4326") eastcoastmask<-spTransform(eastcoastmask,"+init=epsg:4326") #crop out other parts of eez, only keep east coast eez (crop takes areas outside extent of an object [eastcoastmask] and ignores them) useez<-crop(useez,eastcoastmask) noestuaries<-readOGR(".","no_estuaries_wholeeastcoast")#need to use this one because it doesn't have the section that dips further south which gets into finished map and it covers entire map noestuaries<-spTransform(noestuaries,"+init=epsg:4326") useezsimple<-readOGR(".","US_State_Fed_waters_simple_dissolve")#us eez but also includes areas inland of state waters line useezsimple<-spTransform(useezsimple,"+init=epsg:4326") #remove useez too far west (gulf of mexico) useezsimple<-crop(useezsimple,eastcoastmask) setwd("~/Enviornmental_Data/USmap") fisherykud95<-readOGR(".","fisherykud95_bllop")#use dissolved shp because it is smaller size setwd("~/Enviornmental_Data/Mid_Atlantic_closure") midclose<-readOGR(".","MIDClosure") midclose<-spTransform(midclose,"+init=epsg:4326") setwd("~/Enviornmental_Data/Prediction_Rasters/BLLOP_rasters/raster2016_2018") depthraster16_18<-raster("depthraster16_18_hycom.tif") rugosityraster16_18<-raster("rugosityraster16_18_hycom.tif") theme_set(theme_bw()) #https://www.r-spatial.org/r/2018/10/25/ggplot2-sf.html midclosedata<- st_as_sf(midclose) midclosedata$FMC<-as.factor("South Atlantic") useezdata<-st_as_sf(useez) fisherykuddata<-st_as_sf(fisherykud95) fisherykuddata$ID<-"map" makeMap<-function(habrast=NA,type=NA,monthabr=NA,species=NA,maxval=NA){ habrast<-mask(habrast,mask=fisherykud95) r.df <- as.data.frame(rasterToPoints(habrast)) names(r.df) <- c('Longitude', 'Latitude', 'Occurrence Probability') world<-ne_countries(scale="medium",returnclass = "sf") mp<-ggplot(data=world) + geom_sf(color="black",fill="tan") + coord_sf(xlim=c(-81.5, -67),ylim=c(23,39),expand=FALSE) + geom_raster(data=r.df, aes(x=Longitude, y=Latitude, fill=`Occurrence Probability`)) + theme(axis.title.x = element_blank(), panel.grid = element_blank())+ theme(axis.title.y = element_blank())+ scale_fill_viridis(option="C",limits=c(0,maxval)) + geom_sf(data = midclosedata, aes(colour = "FMC"), fill = NA,size=1) + geom_sf(data = fisherykuddata, aes(colour = "ID"), fill = NA,size=1) + geom_sf(data = useezdata, aes(colour = "Jurisdicti"), fill = NA) + #scale_colour_manual(values = c("FMC" = "green","ID"="cyan3","Jurisdicti" = "black"), name = NULL, # labels = c("FMC" ="Mid Atlantic","ID"="Fishery Domain","Jurisdicti" ="U.S. EEZ"), # guide = guide_legend(override.aes = list(shape = NA),reverse = F)) + #for some reason PLLOP one this is needed, but when use it for BLLOP it puts eez and closure above scale bar, so leave this line out scale_colour_manual(values = c("Jurisdicti"="black","ID"="cyan3","FMC"="green"), name = NULL, labels = c("Jurisdicti"="U.S. EEZ","ID"="Fishery Domain","FMC"="Mid Atlantic"), guide = guide_legend(override.aes = list(shape = NA),reverse = T)) + coord_sf(xlim=c(-81.5, -67),ylim=c(23,39),expand=FALSE) + #annotate("text",x=-80,y=38.5,label="b",size=6) + #if I want to add letter in map theme(text = element_text(size = 13)) + #if I want to change size of text in map annotation_scale(location="br" ,width_hint=0.25,style="bar",unit_category="metric",text_cex=0.85) + annotation_north_arrow(location = "tl", which_north = "true", height=unit(0.75,"cm"),width=unit(0.6,"cm"),style = north_arrow_orienteering) ggsave(filename = paste(species,monthabr,type,".pdf",sep=""),plot=mp,width=8,height=6,units="in",dpi=300) ggsave(filename = paste("png",species,monthabr,type,".png",sep=""),plot=mp,width=8,height=6,units="in",dpi=300) #ggsave(filename = paste(species,".pdf",sep=""),plot=mp) #ggsave(filename = "blah.pdf",plot=mp) } #PPP (Prediction Probability of Presence) PPPraster<-function(month=NA,monthabr=NA,data=NA,exclude=NULL,species=NULL,maxval=NA,hooks=NA,bait=NA,excludeYear=NA){ newraster<-setValues(depthraster16_18,NA) #create blank raster with correct parameters newrasterupper<-setValues(depthraster16_18,NA) #create blank raster with correct parameters newrasterlower<-setValues(depthraster16_18,NA) #create blank raster with correct parameters #Creates raster of lat and raster of lon to put into p.data latraster<-init(depthraster16_18,'y') lonraster<-init(depthraster16_18,'x') #no need to read in depth and rugosity rasters everytime (no time component) setwd("~/Enviornmental_Data/Prediction_Rasters/BLLOP_rasters/raster2016_2018") monthbt<-raster(paste(monthabr,"BT16_18_hycom",".tif",sep="")) monthbs<-raster(paste(monthabr,"BS16_18_hycom",".tif",sep="")) monthsst<-raster(paste(monthabr,"SST16_18_hycom",".tif",sep="")) monthssh<-raster(paste(monthabr,"SSH16_18_hycom",".tif",sep="")) monthchla<-raster(paste(monthabr,"CHLA16_18_hycom",".tif",sep="")) monthbtsd<-raster(paste(monthabr,"BTSD16_18_hycom",".tif",sep="")) monthsstsd<-raster(paste(monthabr,"SSTSD16_18_hycom",".tif",sep="")) monthturb<-raster(paste(monthabr,"Turb16_18_hycom",".tif",sep="")) dayval<-mean(data$Day[which(data$MONTH==month)]) p.data<-data.frame(Day=mean(data$Day),MONTH=month,BEGIN_SET_LATITUDE=values(latraster),BEGIN_SET_LONGITUDE=values(lonraster),Set_Begin_Hour=mean(data$Set_Begin_Hour),BATHYMETRY=values(depthraster16_18),RUGOSITY=values(rugosityraster16_18),CHLA=values(monthchla),SST=values(monthsst),SSH=values(monthssh), TURB=values(monthturb),BT=values(monthbt),BS=values(monthbs),BTSD=values(monthbtsd),SSTSD=values(monthsstsd),MONTH=month,YEAR="2016",logEFFORT=mean(data$logEFFORT),vals=1:length(values(monthbt)))#vals used as a dummy variable to summaryBy over if(excludeYear=="YES"){ p.data<-p.data #no need to make three datasets one for each year because Year is excluded in the prediction (when Year is a random effect), meaning the prediction will be the same among years for a given grid cell }else{ #but when no a random effect, need to predict over each of three prediciton years cause prediction will be different p.data16<-p.data p.data17<-p.data16 p.data17$Year<-"2017" p.data18<-p.data16 p.data18$Year<-"2018" p.data<-rbind(p.data16,p.data17,p.data18) } if(hooks=="YES"){ p.data_sc<-p.data p.data_lc<-p.data p.data_sj<-p.data p.data_lj<-p.data p.data_sc$HOOK_CONFIG<-"Circle_small" p.data_lc$HOOK_CONFIG<-"Circle_large" p.data_sj$HOOK_CONFIG<-"J_small" p.data_lj$HOOK_CONFIG<-"J_large" p.data<-rbind(p.data_sc,p.data_lc,p.data_sj,p.data_lj) p.data$HOOK_CONFIG<-as.factor(p.data$HOOK_CONFIG) }else{ p.data=p.data } if(bait=="YES"){ p.data_e<-p.data p.data_t<-p.data p.data_u<-p.data p.data_e$BAITTYPE<-"ELASMO" p.data_t$BAITTYPE<-"TELEOST" p.data_u$BAITTYPE<-"UNKNOWN" p.data<-rbind(p.data_e,p.data_t,p.data_u) p.data$BAITTYPE<-as.factor(p.data$BAITTYPE) }else{ p.data=p.data } p.data<-cbind(p.data,pred=predict(m,type="link",newdata=p.data,na.action=na.pass,exclude=exclude,se.fit=T)) preds<-summaryBy(pred.fit~vals,data=p.data) sepreds<-summaryBy(pred.se.fit~vals,data=p.data) #suppose to add and subtract SEs prior to back transform from link scale to response scale preds$pred.fit.mean.upper<-preds$pred.fit.mean+sepreds$pred.se.fit.mean preds$pred.fit.mean.lower<-preds$pred.fit.mean-sepreds$pred.se.fit.mean preds$predtrans<-plogis(preds$pred.fit.mean) #can also use VGAMs logitlink funcition and select inverse =T preds$preduppertrans<-plogis(preds$pred.fit.mean.upper) preds$predlowertrans<-plogis(preds$pred.fit.mean.lower) newraster[1:553938]<-preds$predtrans newrasterupper[1:553938]<-preds$preduppertrans newrasterlower[1:553938]<-preds$predlowertrans #EVENTUALLY REMOVE AREAS NORTH OF GULF OF MAINE #turn all projections west of -81.5 and areas south of Panama to NA newraster[428:726,]<-NA #removes area south of~23 degrees, not concerns with areas south newraster[,1:235]<-NA #removes everything west of -81.5 newrasterupper[428:726,]<-NA #removes area south of~23 degrees, not concerns with areas south newrasterupper[,1:235]<-NA #removes everything west of -81.5 newrasterlower[428:726,]<-NA #removes area south of~23 degrees, not concerns with areas south newrasterlower[,1:235]<-NA #removes everything west of -81.5 setwd(paste("~/BLLOP_Data/Results/",species,"/R_raster_outputs",sep="")) writeRaster(newraster, file=paste(species,monthabr,"PPP.tif",sep=""),overwrite=T) writeRaster(newrasterupper, file=paste(species,monthabr,"PPPupper.tif",sep=""),overwrite=T) writeRaster(newrasterlower, file=paste(species,monthabr,"PPPlower.tif",sep=""),overwrite=T) #IGNORE, gets removed when masking to fishery domain in makeMap function #removes estuaries for the most part #newraster<-mask(newraster,mask=noestuaries,inverse=TRUE) #newrasterupper<-mask(newrasterupper,mask=noestuaries,inverse=TRUE) #newrasterlower<-mask(newrasterlower,mask=noestuaries,inverse=TRUE) #remove areas east of eez #newraster<-mask(newraster,mask=useezsimple) #newrasterupper<-mask(newrasterupper,mask=useezsimple) #newrasterlower<-mask(newrasterlower,mask=useezsimple) setwd(paste("~/BLLOP_Data/Results/",species,"/GIS_maps",sep="")) makeMap(habrast=newraster,type="pred",monthabr=monthabr,species=species,maxval=maxval) makeMap(habrast=newrasterupper,type="upper",monthabr=monthabr,species=species,maxval=maxval) makeMap(habrast=newrasterlower,type="lower",monthabr=monthabr,species=species,maxval=maxval) #ignore warning about geom_tile, geom_raster does better (shifts less compared to geom_tile) return(list(newraster=newraster,newrasterupper=newrasterupper,newrasterlower=newrasterlower)) } ##### #SB ##### #read in model data output sbJanPPP<-PPPraster(month="01",monthabr="Jan",data=sb_datanona,exclude=NULL,species="sb",maxval=1,hooks="NO",bait="NO",excludeYear="YES") sbFebPPP<-PPPraster(month="02",monthabr="Feb",data=sb_datanona,exclude=NULL,species="sb",maxval=1,hooks="NO",bait="NO",excludeYear="YES") sbMarPPP<-PPPraster(month="03",monthabr="Mar",data=sb_datanona,exclude=NULL,species="sb",maxval=1,hooks="NO",bait="NO",excludeYear="YES") sbAprPPP<-PPPraster(month="04",monthabr="Apr",data=sb_datanona,exclude=NULL,species="sb",maxval=1,hooks="NO",bait="NO",excludeYear="YES") sbMayPPP<-PPPraster(month="05",monthabr="May",data=sb_datanona,exclude=NULL,species="sb",maxval=1,hooks="NO",bait="NO",excludeYear="YES") sbJunPPP<-PPPraster(month="06",monthabr="Jun",data=sb_datanona,exclude=NULL,species="sb",maxval=1,hooks="NO",bait="NO",excludeYear="YES") sbJulPPP<-PPPraster(month="07",monthabr="Jul",data=sb_datanona,exclude=NULL,species="sb",maxval=1,hooks="NO",bait="NO",excludeYear="YES") sbAugPPP<-PPPraster(month="08",monthabr="Aug",data=sb_datanona,exclude=NULL,species="sb",maxval=1,hooks="NO",bait="NO",excludeYear="YES") sbSepPPP<-PPPraster(month="09",monthabr="Sep",data=sb_datanona,exclude=NULL,species="sb",maxval=1,hooks="NO",bait="NO",excludeYear="YES") sbOctPPP<-PPPraster(month="10",monthabr="Oct",data=sb_datanona,exclude=NULL,species="sb",maxval=1,hooks="NO",bait="NO",excludeYear="YES") sbNovPPP<-PPPraster(month="11",monthabr="Nov",data=sb_datanona,exclude=NULL,species="sb",maxval=1,hooks="NO",bait="NO",excludeYear="YES") sbDecPPP<-PPPraster(month="12",monthabr="Dec",data=sb_datanona,exclude=NULL,species="sb",maxval=1,hooks="NO",bait="NO",excludeYear="YES") ##### #DS ##### #read in model data output dsJanPPP<-PPPraster(month="01",monthabr="Jan",data=ds_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="ds",maxval=1,hooks="NO",bait="NO",excludeYear="YES") dsFebPPP<-PPPraster(month="02",monthabr="Feb",data=ds_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="ds",maxval=1,hooks="NO",bait="NO",excludeYear="YES") dsMarPPP<-PPPraster(month="03",monthabr="Mar",data=ds_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="ds",maxval=1,hooks="NO",bait="NO",excludeYear="YES") dsAprPPP<-PPPraster(month="04",monthabr="Apr",data=ds_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="ds",maxval=1,hooks="NO",bait="NO",excludeYear="YES") dsMayPPP<-PPPraster(month="05",monthabr="May",data=ds_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="ds",maxval=1,hooks="NO",bait="NO",excludeYear="YES") dsJunPPP<-PPPraster(month="06",monthabr="Jun",data=ds_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="ds",maxval=1,hooks="NO",bait="NO",excludeYear="YES") dsJulPPP<-PPPraster(month="07",monthabr="Jul",data=ds_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="ds",maxval=1,hooks="NO",bait="NO",excludeYear="YES") dsAugPPP<-PPPraster(month="08",monthabr="Aug",data=ds_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="ds",maxval=1,hooks="NO",bait="NO",excludeYear="YES") dsSepPPP<-PPPraster(month="09",monthabr="Sep",data=ds_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="ds",maxval=1,hooks="NO",bait="NO",excludeYear="YES") dsOctPPP<-PPPraster(month="10",monthabr="Oct",data=ds_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="ds",maxval=1,hooks="NO",bait="NO",excludeYear="YES") dsNovPPP<-PPPraster(month="11",monthabr="Nov",data=ds_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="ds",maxval=1,hooks="NO",bait="NO",excludeYear="YES") dsDecPPP<-PPPraster(month="12",monthabr="Dec",data=ds_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="ds",maxval=1,hooks="NO",bait="NO",excludeYear="YES") ##### #SHH ##### #read in model data output shhJanPPP<-PPPraster(month="01",monthabr="Jan",data=shh_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="shh",maxval=1,hooks="NO",bait="YES",excludeYear="YES") shhFebPPP<-PPPraster(month="02",monthabr="Feb",data=shh_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="shh",maxval=1,hooks="NO",bait="YES",excludeYear="YES") shhMarPPP<-PPPraster(month="03",monthabr="Mar",data=shh_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="shh",maxval=1,hooks="NO",bait="YES",excludeYear="YES") shhAprPPP<-PPPraster(month="04",monthabr="Apr",data=shh_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="shh",maxval=1,hooks="NO",bait="YES",excludeYear="YES") shhMayPPP<-PPPraster(month="05",monthabr="May",data=shh_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="shh",maxval=1,hooks="NO",bait="YES",excludeYear="YES") shhJunPPP<-PPPraster(month="06",monthabr="Jun",data=shh_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="shh",maxval=1,hooks="NO",bait="YES",excludeYear="YES") shhJulPPP<-PPPraster(month="07",monthabr="Jul",data=shh_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="shh",maxval=1,hooks="NO",bait="YES",excludeYear="YES") shhAugPPP<-PPPraster(month="08",monthabr="Aug",data=shh_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="shh",maxval=1,hooks="NO",bait="YES",excludeYear="YES") shhSepPPP<-PPPraster(month="09",monthabr="Sep",data=shh_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="shh",maxval=1,hooks="NO",bait="YES",excludeYear="YES") shhOctPPP<-PPPraster(month="10",monthabr="Oct",data=shh_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="shh",maxval=1,hooks="NO",bait="YES",excludeYear="YES") shhNovPPP<-PPPraster(month="11",monthabr="Nov",data=shh_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="shh",maxval=1,hooks="NO",bait="YES",excludeYear="YES") shhDecPPP<-PPPraster(month="12",monthabr="Dec",data=shh_datanona,exclude="s(BEGIN_SET_LATITUDE,BEGIN_SET_LONGITUDE)",species="shh",maxval=1,hooks="NO",bait="YES",excludeYear="YES")