Unclear OHC and HYCOM likelihood issue

[marosteg]

One of the spearfish is yielding errors solely for OHC and HYCOM likelihood calculations. I tried removing days with only a surface layer (depth = 0 m) PDT observation but that didn't solve it.

The OHC error reads:
Error in { : task 8 failed - "newsplit: out of vertex space"

The HYCOM error reads:
Error in { :
task 8 failed - "C stack usage 8012680 is too close to the limit"

MWE below....

dir <- '~/Google Drive/Shared drives/MPG_WHOI/data/shortbill_spearfish/MiniPAT/'

#Read in metadata
meta <- read.csv('add_shortbillspearfish_meta.csv',header=T)

## format dates
meta$time_coverage_start <- lubridate::parse_date_time(meta$time_coverage_start, orders='Ymd HMS', tz='UTC')
meta$time_coverage_end <- lubridate::parse_date_time(meta$time_coverage_end, orders='Ymd HMS', tz='UTC')


fish <- list.files(dir)
fish <- fish[-c(8,9)]
i <- 5 #spearfish PTT 227133
fishID <- fish[i] # specify the individual
data_dir <- paste(dir, fishID, sep='')

#### 3.1 Load and prepare the tag data ####
  # SET START/END LOCATIONS iniloc is dataframe containing cols: day, month,
  # year, lat, lon and rows: start, end
  meta.i <- which(meta$ptt == fishID)
  iniloc <- data.frame(matrix(c(day(meta[meta.i,"time_coverage_start"]), month(meta[meta.i,"time_coverage_start"]), year(meta[meta.i,"time_coverage_start"]), meta$geospatial_lat_start[meta.i], meta$geospatial_lon_start[meta.i],   #start
                                day(meta[meta.i,"time_coverage_end"]), month(meta[meta.i,"time_coverage_end"]), year(meta[meta.i,"time_coverage_end"]), meta$geospatial_lat_end[meta.i], meta$geospatial_lon_end[meta.i]),   #end
                              nrow = 2, ncol = 5, byrow = T))
  names(iniloc) <- list("day", "month", "year", "lat", "lon")
  tag <- as.POSIXct(paste(iniloc[1, 1], "/", iniloc[1, 2], "/", iniloc[1, 3],
                          sep = ""), format = "%d/%m/%Y", tz = "UTC")
  pop <- as.POSIXct(paste(iniloc[2, 1], "/", iniloc[2, 2], "/", iniloc[2, 3],
                          sep = ""), format = "%d/%m/%Y", tz = "UTC")
  # VECTOR OF DATES FROM DATA. THIS WILL BE THE TIME STEPS, T, IN THE
  # LIKELIHOODS
  dateVec <- seq.POSIXt(tag, pop, by = "24 hours")


#### 3.1.2 Depth-Temperature Data ####
    pdtFile <- paste(data_dir,paste(fishID,"-PDTs.csv",sep=""),sep="/")
    pdt <- read.wc(pdtFile, type = "pdt", tag = tag, pop = pop, verbose = T)
    pdt <- pdt[, c("Date", "Depth", "MinTemp", "MaxTemp")]


#### 3.3.3 3D Depth-Temperature Likelihood ####
  # OCEAN HEAT CONTENT (INTEGRATED PDTs)
  if (!file.exists(paste0(data_dir, "/L.ohc.rds"))) {
    L.ohc <- calc.ohc.par(pdt, filename = "hycom", ohc.dir = hycom.dir, dateVec = dateVec, 
                          isotherm = "", use.se = T, ncores = 8)
    saveRDS(L.ohc,paste0(data_dir, "/L.ohc.rds"))
  } else {L.ohc <- readRDS(paste0(data_dir, "/L.ohc.rds"))}

  # DEPTH-TEMP LEVELS
  if (!file.exists(paste0(data_dir, "/L.hycom.rds"))) {
    L.hycom <- calc.hycom.par(pdt, filename = "hycom", hycom.dir = hycom.dir, dateVec = dateVec, 
                              use.se = T, ncores = 8)
    saveRDS(L.hycom,paste0(data_dir, "/L.hycom.rds"))
  } else {L.hycom <- readRDS(paste0(data_dir, "/L.hycom.rds"))}

add_shortbillspearfish_meta.csv

[camrinbraun]

I believe I fixed this in the last few months but its possible its a local change that has not yet been pushed to github. Try the following and remove any days that have <=3 bins. This is an issue because we use regression to predict at the appropriate depth levels to match HYCOM, which obviously fails with too few data points.

pdt %>% group_by(Date) %>% summarise(n=n())

[marosteg]

That solved it, thanks!