#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#load functions ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

library(probGLS)
library(GeoLight)
library(readxl)

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#load meta data ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

meta <- read_excel('meta.xlsx')

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#load light data ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

setwd('raw data location')
ll <- list.files(pattern="*driftadj.lux$")

for(i in c(1:length(ll))){
  lig           <- read.table(ll[i],sep="\t",skip=19,header=T)
  colnames(lig) <- c('dtime','lig')
  lig$dtime     <- as.POSIXct(strptime(lig$dtime, format="%d/%m/%Y %H:%M:%S"), tz="UTC")
  trn           <- twilightCalc(lig$dtime,lig$lig,ask=F,LightThreshold = 2)
  trn           <- trn[trn$type>0,]
  trn$id        <- substr(ll[i],1,4)
  if(i==1) trn2 <- trn else trn2 <- rbind(trn2,trn)
}
trn2$daylength  <- abs(as.numeric(difftime(trn2$tFirst, trn2$tSecond, units="hours")))
trn2            <- trn2[trn2$daylength < 24,]

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#load sensor data ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

# mode 9
# wet/dry every 6 sec and length recorded at every switch
# full light range every 5 min


setwd('raw data location')
aa <- list.files(pattern="*driftadj.deg$")
tt <- list.files(pattern="*.sst$")

for( i in c(1:length(aa))){
  ad             <- read.table(aa[i],sep="\t",skip=19,header=T)
  ad$dtime       <- as.POSIXct(strptime(ad[,1], format = "%d/%m/%Y %H:%M:%S"),tz='UTC')
  ad$date        <- as.Date(ad$dtime)
  ad$id          <- substr(aa[i],1,4)
  ad             <- ad[ad$dtime >= meta$Deployed[meta$GLS_id==substr(aa[i],2,4)] & ad$dtime <= meta$Retrieved[meta$GLS_id==substr(aa[i],2,4)],]
  ad$WetDryState <- ad[,2]/60/60
  ad             <- ad[ad$wet.dry=="wet",]
  
  td             <- read.csv(tt[i],sep="\t",skip=19,header=T)
  td$dtime       <- as.POSIXct(strptime(td[,1], format = "%d/%m/%Y %H:%M:%S"),tz='UTC')
  td$date        <- as.Date(td$dtime)
  
  sensor         <- sst_deduction(datetime = td$dtime,temp = td[,4])
  sensor$doy     <- as.numeric(strftime(sensor$date, format = "%j"))
  sensor$month   <- as.numeric(strftime(sensor$date, format = "%m"))
  sensor$year    <- as.numeric(strftime(sensor$date, format = "%Y"))
  sensor$jday    <- as.numeric(julian(sensor$date))
  sensor$id      <- substr(aa[i],1,4)
  
  if(i==1) act   <- ad else act <- rbind(act,ad)
  if(i==1) sen   <- sensor else sen <- rbind(sen,sensor)
}

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#load GPS data ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

setwd('gps data location')
gps       <- read.csv('gps.csv')
gps$dtime <- as.POSIXct(strptime(paste(gps$Date,gps$Time,sep=" "),'%d/%m/%Y %H:%M:%S'),tz="UTC")
gps$lon   <- gps$Longitude
gps$lat   <- gps$Latitude

for( i in c(1:length(ll))){
  start            <- meta$Deployed[i]
  end              <- meta$Retrieved[i]
  id               <- paste('P',meta$GLS_id[i],sep="")
  
  gps$id  [gps$dtime >= start & gps$dtime <= end & gps$Id == meta$Darvic[i]]      <- id
  gps$i   [gps$dtime >= start & gps$dtime <= end & gps$Id == meta$Darvic[i]]      <- i
  gps$i.id[gps$dtime >= start & gps$dtime <= end & gps$Id == meta$Darvic[i]]      <- paste(gps$i[gps$dtime >= start & gps$dtime <= end & gps$Id == meta$Darvic[i]],gps$id[gps$dtime >= start & gps$dtime <= end & gps$Id == meta$Darvic[i]],sep='.')
}
gps <- gps[!is.na(gps$i.id),]

rm(co,ad,sensor,td,temp,trn,lig)
rm(end,i,id,ll,start,tt,aa)

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#run model with 200 iterations and 1 to 10 000 particles ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

tw            <- twilight_error_estimation()

for(part in c(1,5,10,20,40,60,80,100,200,300,400,500,600,800,1000,1250,1500,2000,3000,4000,5000,10000)){
  
  for(i in c(1:nrow(meta))){
    
    start            <- meta$Last_fix_colony[i]
    end              <- meta$first_fix_colony[i]
    id               <- paste('P',meta$GLS_id[i],sep="")
    trn              <- trn2[trn2$id==id,]
    sensor           <- sen[sen$id==id,]
    trn              <- trn[trn$tFirst > (start) & trn$tSecond < (end),]
    
    if(nrow(trn)>1){
      trn$id           <- id
      trn$i            <- i
      trn$i.id         <- paste(trn$i,trn$id,sep='.')
      sensor           <- sensor[sensor$date >= as.Date(start) & sensor$date <= as.Date(end),]
      sensor$id        <- id
      sensor$i         <- i
      sensor$i.id      <- paste(sensor$i,sensor$id,sep='.')
      act2             <- act[act$id==id,]
      act2$i           <- i
      act2$i.id        <- paste(act2$i,act2$id,sep='.')
      
      
      mm.list <- prob_algorithm(particle.number    = part, 
                       iteration.number            = 200,
                       trn                         = trn, 
                       sensor                      = sensor, 
                       act                         = act2, 
                       loess.quartile              = NULL, 
                       sunrise.sd                  = tw,
                       sunset.sd                   = tw,
                       tagging.location            = c(-36.816,-54.316), 
                       tagging.date                = start, 
                       retrieval.date              = end, 
                       speed.wet                   = c(1,1.3,5), 
                       speed.dry                   = c(12,6,45),
                       boundary.box                = c(-120,40,-90,0),
                       days.around.spring.equinox  = c(20,20), 
                       days.around.fall.equinox    = c(20,20),
                       ice.conc.cutoff             = 1.1, 
                       land.mask                   = T,
                       med.sea                     = F,
                       black.sea                   = F,
                       baltic.sea                  = F,
                       caspian.sea                 = F,
                       wetdry.resolution           = 1)
      
      # all computed tracks
      newt2      <- mm.list[[1]]
      newt2$id   <- id
      newt2$i    <- i
      newt2$i.id <- paste(newt2$i,newt2$id,sep='.')
      
      if(i==1) newt3 <- newt2 else newt3 <- spRbind(newt3,newt2)
     }
  }
  
  newt3$available.particles <- part
  if(part==1) newt4 <- newt3 else newt4 <- spRbind(newt4,newt3)
}

# dataframe containing all computed tracks for each individual using 200 iterations and 1 to 10 000 particles
newt4          <- data.frame(newt4)
newt4$identity <- paste(newt4$available.particles, newt4$step, newt4$i.id, sep=".")


#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#calculate distance for each set of tracks using 1 to 200 
#iterations to compute the geographic median track ----
#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

for(vb in c(1,5,10,15,seq(20,200,20))){
  nat <- newt4[newt4$bootstrap %in% c(1:vb),]
  print(vb)
  
  nag2 <- NULL
  
  for(g in c(1:length(unique(nat$identity)))){
    print(paste(g,length(unique(nat$identity)),sep=" / "))
    
    identity <- unique(nat$identity)[g]
    nat2     <- nat[nat$identity == identity,]
    
    # create new median location with available bootstraps
    
    newt2               <- nat2
    sf                  <- data.frame(spDists(as.matrix(newt2[,1:2],ncol=2),longlat=T),ncol=length(newt2$step))
    sa                  <- data.frame(sum.dist=rowMeans(sf),bot=seq(1,length(newt2$step),1))
    gmp                 <- newt2 [sa$sum.dist==min(sa$sum.dist),] [1,]
    gmp$median.sat.sst  <- median(newt2$sat.sst)
    gmp$median.sun.elev <- median(newt2$sun.elev)
    gmp$median.wrel     <- median(newt2$wrel)
    
    nag <- gmp
    
    nag$gps.lon.mean  <- NA
    nag$gps.lat.mean  <- NA
    
    nag$mean.median.distance   <- NA
    nag$mean.distance          <- NA
    
    nag$wrel.median <- median(nat$wrel[nat$identity==identity],na.rm=T)
    nag$wrel.mean   <- mean  (nat$wrel[nat$identity==identity],na.rm=T)
    nag$wrel.sd     <- sd    (nat$wrel[nat$identity==identity],na.rm=T)
    nag$wrel.chosen <- nag$wrel
    
    if(g==1){
      nag.empty <- nag
      nag.empty$lon <- NA
    }
    
    # check if any GPS position for this time and id are available
    gps3b   <- gps[gps$dtime >= nag$tFirst-1800 & gps$dtime <= nag$tSecond+1800 & gps$i.id==nag$i.id,]
    gps3b   <- gps3b[!is.na(gps3b$lon),]
    
    gps3c   <- gps3b
    gps3c   <- gps3c[!is.na(gps3c$lon),]
    
    if(nrow(gps3b)>0){
      nag$gps.lon.mean  <- mean(gps3c$lon)
      nag$gps.lat.mean  <- mean(gps3c$lat)
      
      nag$mean.median.distance   <- spDistsN1(matrix(c(nag$gps.lon.mean,nag$gps.lat.mean),ncol=2),matrix(c(nag$lon,nag$lat),ncol=2),longlat=T)
      nag$mean.distance          <- min(spDistsN1(matrix(c(nat2$lon,nat2$lat),ncol=2),matrix(c(nag$gps.lon.mean,nag$gps.lat.mean),ncol=2),longlat=T))[1]
      
      if(is.null(nag2)==T) nag2 <- nag else nag2 <- rbind(nag2,nag)
      
    } else { nag <- nag.empty}
    
    
  }
  
  nag2$daylength  <- abs(as.numeric(difftime(nag2$tFirst, nag2$tSecond, units="hours")))
  
  nag2$sst.used <- 0
  nag2$sst.used[!is.na(nag2$sst.diff)] <- 1
  
  nag2$available.bootstraps <- vb
  
  if(vb == 1) nag3 <- nag2 else nag3 <- rbind(nag3,nag2)
}

### nag3 is a dataframe containing all distance measurements for the geopgraphic 
### median track as well as the closest location to each average GPS location for all 
### tracks computed with a range of particles and iterations