ShapeToMask.R

#'Convert Shapefile to Mask Array
#' 
#'This function reads a shapefile (.shp) containing information about polygonal 
#'regions. It then transfers the shapefile data into an array and subsets the 
#'output based on requested region names or IDs. The accepted shapefile 
#'databases are 'NUTS', 'LAU', and 'GADM', each with its own unique format. 
#'However, the function can use other shapefiles databases with specifying the 
#'categories names with the parameter 'shp_col_name_ids'. 
#'
#'To ensure accurate comparison with the shapefile, the function loads a 
#'reference dataset that provides longitude and latitude information. By 
#'intersecting each subset of the shapefile with the reference coordinates, the 
#'function selects only the desired regions. The final step involves creating a 
#'mask array. Depending on the chosen option, the mask array is either returned 
#'as the function's output or saved into a NetCDF format in the specified 
#'directory.
#'
#'Note: Modules GDAL, PROJ and GEOS are required.
#' 
#'@param shp_file A character string indicating the shp file path.
#'@param ref_grid A character string indicating the path to the reference 
#'  data. Either (1) a netCDF file or (2) a list of lon and lat to provide the 
#'  reference grid points. It is NULL by default.
#'@param shp_system A character string containing the Shapefile System Database 
#'  Name used to subset the shapefile into regions by using parameters 'reg_ids' 
#'  or 'reg_names'. The accepted systems are: 'NUTS', 'LAU', and 'GADM'. When it 
#'  is used, you must specify either 'reg_ids' or 'reg_names'; if you don't need 
#'  to subset different regions, set it to NULL. It is set to "NUTS" by default
#'  (optional). 
#'@param shp_col_name_ids A character string indicating the column name of the 
#'  column in where the specified 'reg_ids' will be taken (optional).
#'@param reg_ids A character string indicating the unique ID in shapefile. 
#'  It is NULL by default (optional).
#'@param reg_names A named list of character string vectors indicating the 
#'  country and the region name. The name of the list stands for the country 
#'  name code and the vector character strings indicate the region name for 
#'  each country. It is NULL by default (optional).
#'@param reg_level An integer number from 1 to 3 indicating the 'NUTS' dataset 
#'  level. For other datasets this parameter is not used. One mask can only have 
#'  a unique level. It is set to 3 by default (optional).
#'@param lat_dim A character string indicating the latitudinal dimension. If it 
#'  is NULL, the latitudinal name will be searched using an internal function 
#'  with the following possible names: 'lat', 'latitude', 'y', 'j' and 
#'  'nav_lat'. It is set to NULL by default.
#'@param lon_dim A character string indicating the longitudinal dimension. If it 
#'  is NULL, the longitudinal name will be searched using an internal function 
#'  with the following possible names: 'lon', 'longitude', 'x', 'i' and 
#'  'nav_lon'. It is set to NULL by default.
#'@param target_crs A character string indicating the target 'Coordinate 
#'  Reference System'.
#'@param region A logical value indicating if we want a dimension for the 
#'  regions in the resulting mask array. It is FALSE by default.
#'@param check_valid A logical value that when it is TRUE it uses the function 
#'  'sf::st_make_valid' applied to the shapefile and to the coordinates.
#'@param find_min_dist A logical value indicating if we want to look for the 
#'  nearest coordinate between the shapefile region and the reference grid when 
#'  there is no intersection between the shapefile and the reference grid. It is 
#'  FALSE by default.
#'@param max_dist A numeric value indicating the maximum distance is accepted 
#'  to the closest gridpoint when there is no intersection between the shapefile 
#'  and the reference grid. 
#'@param ncores The number of parallel processes to spawn for the use for 
#'  parallel computation in multiple cores.
#'@param ... Arguments passed on to 's2_options' in function 'st_intersection'.  
#'  See 's2 package'.
#' 
#'@return A multidimensional array containing a mask array with longitude and 
#'latitude dimensions. If 'region' is TRUE, there will be a dimension for 
#'the region. 
#' 
#'@examples 
#'# Exmple (1): NUTS
#'shp_file <- paste0('/esarchive/shapefiles/NUTS3/NUTS_RG_60M_2021_4326.shp/', 
#'                   'NUTS_RG_60M_2021_4326.shp')
#'ref_grid <- list(lon = seq(10, 40, 0.5), lat = seq(40, 85, 0.5))
#'NUTS_name <- list(FI = c('Lappi', 'Kainuu'), SI = c('Pomurska', 'Podravska'))
#'mask2 <- ShapeToMask(shp_file = shp_file, ref_grid = ref_grid, 
#'                     reg_names = NUTS_name)
#'@import easyNCDF
#'@import sf
#'@import foreach
#'@importFrom doParallel registerDoParallel
#'@export
ShapeToMask <- function(shp_file, ref_grid, return_area = FALSE, 
                        shp_system = "NUTS", reg_names = NULL, reg_ids = NULL, 
                        shp_col_name_ids = NULL, reg_level = 3, 
                        lat_dim = NULL, lon_dim = NULL, 
                        region = FALSE, target_crs = NULL, 
                        check_valid = FALSE, find_min_dist = FALSE, 
                        max_dist = 50, ncores = NULL, ...) {

  # TODO: Suppress warnings?
  # TODO: Add saving option?

  # Initial checks
  # shp_file
  if (is.null(shp_file)) {
    stop("Parameter 'shp_file' cannot be NULL.")
  }
  if (!is.character(shp_file)) {
    stop("Parameter 'shp_file' must be a character string.")
  }
  # lon_dim, lat_dim
  if (!is.null(lon_dim)) {
    if (!is.character(lon_dim)) {
      stop("Parameter 'lon_dim' must be a character string.")
    }
  }
  if (!is.null(lat_dim)) {
    if (!is.character(lat_dim)) {
      stop("Parameter 'lat_dim' must be a character string.")
    }
  }
  # ref_grid
  if (is.null(ref_grid)) {
    stop("Parameter 'ref_grid' cannot be NULL.")
  }
  if (!any(is.character(ref_grid) | is.list(ref_grid))) {
    stop("Parameter 'ref_grid' must be either a netCDF file or a list of lon and lat.")
  }
  if (is.character(ref_grid)) {
    if (!file.exists(ref_grid)) {
      stop("ref_grid file does not exist.")
    } 
  }
  if (is.list(ref_grid)) {
    if (length(ref_grid) != 2) {
      stop("If 'ref_grid' is a list, it must have two items for longitude and latitude.")
    }
    if (is.null(lat_dim) | is.null(lon_dim)) {
      lon_known_names <- c(.KnownLonNames(), 'lons')
      lat_known_names <- c(.KnownLatNames(), 'lats')
      lon_dim <- lon_known_names[which(lon_known_names %in% names(ref_grid))]
      lat_dim <- lat_known_names[which(lat_known_names %in% names(ref_grid))]

      if (identical(lon_dim, character(0)) | identical(lat_dim, character(0))) {
        stop("longitude and latitude names are not recognized in 'ref_grid'. ",  
             "Please specify 'lon_dim' and 'lat_dim'.")
      }
    }
  }

  # shp_system, reg_ids, reg_names, shp_col_name_ids
  if (!is.null(shp_system)) {
    if (!is.character(shp_system)) {
      stop("Parameter 'shp_system' must be a character strinig.")
    }
    if (all(is.null(reg_ids), is.null(reg_names))) {
      stop("If 'shp_system' is used, you must provide either parameter ",
           "'reg_ids' or 'reg_names'.")
    } 
  } else if (!is.null(shp_col_name_ids)) {
    if (is.null(reg_ids)) {
      stop("If 'shp_col_name_ids' is used, parameter 'reg_ids' must be provided.")
    }
    if (!is.character(shp_col_name_ids)) {
      stop("Parameter 'shp_col_name_ids' must be a character strinig.")
    }
  }
  if (all(!is.null(reg_ids), !is.null(reg_names))) {
    warning("Only use 'reg_ids' to get the shape region. 'reg_names' is not used.")
  }
  # reg_level
  if (!is.null(reg_level)) {
    if (!is.numeric(reg_level)) {
      stop("Parameter 'reg_level' must be numeric.")
    }
  }
  # region
  if (!is.logical(region)) {
    stop("Parameter 'region' must be a logical value.")
  }
  # target_crs
  if (!is.null(target_crs)) {
    if (!is.character(target_crs)) {
      stop("Parameter 'target_crs' must be a character string.")
    }
    transform_shp_crs <- TRUE
  } else {
    transform_shp_crs <- FALSE
    target_crs <- sf::st_crs(shp)
  }
  # check_valid
  if (!is.logical(check_valid)) {
    stop("Parameter 'check_valid' must be a logical value.")
  }
  # find_min_dist
  if (!is.logical(find_min_dist)) {
    stop("Parameter 'find_min_dist' must be a logical value.")
  }
  # max_dist
  if (!is.null(max_dist)) {
    if (!is.numeric(max_dist)) {
      stop("Parameter 'max_dist' must be a numeric.")
    }
  }
  # ncores
  if (!is.null(ncores)) {
    if (!is.numeric(ncores)) {
      stop("Parameter 'ncores' must be numeric.")
    }
    ncores <- round(ncores)
    if (ncores < 2) {
      ncores <- NULL
    }
  }

  # Step 1: Load the shapefile
  shp <- sf::st_read(shp_file)  # class sf
  if (transform_shp_crs) {
    transformed_shapefile <- st_transform(shp, crs = target_crs)
    shp <- transformed_shapefile
  }

  NUTS_ID <- ADM1_PCODE <- ISO <- NULL
  CNTR_CODE <- NUTS_NAME <- ADM0_EN <- ADM1_EN <- Name <- LEVL_CODE <- NULL

  if (!is.null(reg_ids)) {
    ## Method 1: Directly use IDs
    if (!is.null(shp_col_name_ids)) {
      if (shp_col_name_ids %in% names(shp)) {
        shp <- subset(shp, get(shp_col_name_ids) %in% reg_ids)
      } else {
        stop("Shape system name not found in shapefile names.")
      }
    } else if (shp_system == "NUTS") {
      shp <- subset(shp, NUTS_ID %in% reg_ids)
      shp_col_name_ids <- "NUTS_ID"
    } else if (shp_system == "ADM") {
      shp <- subset(shp, ADM1_PCODE %in% reg_ids)
      shp_col_name_ids <- "ADM1_PCODE"
    } else if (shp_system == "GADM") {
      shp <- subset(shp, ISO %in% reg_ids)
      shp_col_name_ids <- "ISO"
    } else {
      stop("shp_system ", shp_system, " is not defined yet.")
    }
  
  } else if (!is.null(reg_names)) {
    shp_col_name_ids <- NULL
    ## Method 2: Use country code & region name
    for (cntr_i in 1:length(reg_names)) {
      if (shp_system == "NUTS") {
        tmp <- subset(shp, CNTR_CODE == names(reg_names)[cntr_i])
        tmp <- subset(tmp, NUTS_NAME %in% reg_names[[cntr_i]])
        shp_col_name_ids <- 'NUTS_NAME'
      } else if (shp_system == "ADM") {
        tmp <- subset(shp, ADM0_EN == names(reg_names)[cntr_i])
        tmp <- subset(tmp, ADM1_EN %in% reg_names[[cntr_i]])
        shp_col_name_ids <- 'ADM1_EN'
      } else if (shp_system == "GADM") {
        tmp <- subset(shp, Name %in% reg_names)
        shp_col_name_ids <- 'Name'
      }
      if (cntr_i == 1) {
        shp_tmp <- tmp
      } else {
        shp_tmp <- rbind(shp_tmp, tmp)
      }
    }
    if (shp_system == "NUTS") {
      shp <- subset(shp_tmp, LEVL_CODE == reg_level)
    } else if (shp_system == "ADM" | shp_system == "GADM") {
      shp <- shp_tmp
    }
  }
  
  # Step (2.1): Use the reference file to get lat and lon
  if (all(tools::file_ext(ref_grid) == 'nc')) {
    ## Method 1: ref_grid is a netCDF file
    if (is.null(lat_dim) | is.null(lon_dim)) {
      var_names <- easyNCDF::NcReadVarNames(ref_grid)
      lat_dim <- var_names[which(var_names %in% .KnownLatNames())]
      lon_dim <- var_names[which(var_names %in% .KnownLonNames())]
    }
    latlon <- easyNCDF::NcToArray(ref_grid, vars_to_read = c(lat_dim, lon_dim))
    lat <- easyNCDF::NcToArray(ref_grid, vars_to_read = lat_dim)[1, ]
    lon <- easyNCDF::NcToArray(ref_grid, vars_to_read = lon_dim)[1, ]
  } else if (is.list(ref_grid)) {
    ## Method 2: ref_grid is a list of lon and lat
    lat <- ref_grid[[lat_dim]]
    lon <- ref_grid[[lon_dim]]
  }

  # Step (2.2): Create the grid
  lonlat_df <- data.frame(lon = rep(as.vector(lon), length(lat)),
                          lat = sort(rep(as.vector(lat), length(lon)), decreasing = TRUE))

  if (return_area) {
    locations <- array(1:(length(lon)*length(lat)), c(lon = length(lon), 
                                                      lat = length(lat)))
    xy.sfc <- areagrid(data = locations, lon = lon, lat = lat, 
                       target_proj = target_proj)
    region <- TRUE
  } else {
    ref.df <- data.frame(data = 0,
                         lon = rep(lon, times = length(lat)),
                         lat = rep(lat, each = length(lon)))
    coord <- as.matrix(data.frame(x = ref.df$lon, y = ref.df$lat))
    
    xy.sfg <- sf::st_multipoint(coord)
    xy.sfc <- sf::st_sfc(xy.sfg)
    # Assign crs of original shapefile
    st_crs(xy.sfc) <- sf::st_crs(target_crs) #initial_crs # asign crs of original shapefile
    xy.sfc <- sf::st_transform(xy.sfc, st_crs(shp)) 
    # Check valid
    if (check_valid) {
      xy.sfc <- st_make_valid(xy.sfc)
      shp <- st_make_valid(shp)
    }
  }

  ## Step (3): Loop through each shp region to create the mask
  if (region) {
    cfun <- function(a, b) {
      if (length(dim(a)) == 2) {
        dims <- c(dim(a), 2)
      } else {
        dims <- c(dim(a)[1:2], dim(a)[length(dim(a))]+1)
      }
      return(array(c(a,b), dim = dims))
    }
  } else {
    cfun <- function(a, b) {
      a[which(b != 0)] <- b[which(b != 0)]
      return(a)
    }
  }
  shp_i <- NULL
  if (is.null(ncores)) {
    mask <- foreach(shp_i = 1:nrow(shp), .combine = 'cfun') %do%
              .shapetomask(shp = shp, n = shp_i, lon = lon, lat = lat, 
                           xy.sfc = xy.sfc, return_area = return_area, 
                           find_min_dist = find_min_dist, 
                           shp_col_name_ids = shp_col_name_ids, 
                           max_dist = max_dist, region = region, ...)
  } else {
    registerDoParallel(ncores)
    mask <- foreach(shp_i = 1:nrow(shp), .combine = 'cfun', .packages='sf') %dopar%
              .shapetomask(shp = shp, n = shp_i, lon = lon, lat = lat, 
                           xy.sfc = xy.sfc, return_area = return_area, 
                           find_min_dist = find_min_dist, 
                           shp_col_name_ids = shp_col_name_ids, 
                           max_dist = max_dist, region = region, ...)
    registerDoSEQ()
  }
  if (region) {
    if (length(dim(mask)) == 2) dim(mask) <- c(dim(mask), 1)
    names(dim(mask)) <- c(lon_dim, lat_dim, 'region')
  } else {
    names(dim(mask)) <- c(lon_dim, lat_dim)
  }
  
  # Step 4: Add attributes
  # attr(mask, lon_dim) <- lon
  # attr(mask, lat_dim) <- lat
  # if (shp_system == "NUTS") {
  #   attr(mask, "index") <- as.list(shp$NUTS_ID)
  # } else if (shp_system == "ADM") {
  #   attr(mask, "index") <- as.list(shp$ADM1_PCODE)
  # } else if (shp_system == "GADM") {
  #   attr(mask, "index") <- as.list(shp$ISO)
  # }
  # names(attr(mask, "index")) <- 1:nrow(shp)

  # ## Return all the info from shp
  # attr(mask, "shapefile") <- attributes(shp)
  
  return(mask)
}

.shapetomask <- function(shp, n, lon, lat, xy.sfc, return_area = FALSE, 
                         find_min_dist = FALSE, 
                         shp_col_name_ids = NULL, max_dist = 50, 
                         region = FALSE, ...) {
  mask <- array(0, dim = c(length(lon), length(lat)))
  shpi <- shp[n, ]
  if (return_area) {
    datapoly_int <- xy.sfc %>% dplyr::mutate(int = area_cov(xy.sfc$geometry, shpi))
    vals <- datapoly_int %>% dplyr::select(int, value)
    mask <- vals[order(vals$value),] %>% dplyr::pull(int)
    dim(mask) <- c(length(lon), length(lat)) 
  } else {
    # NOTE: Don't know it's a problem in st_intersection or st_coordinates, tmp_coords may 
    #       not be identical as lon/lat. E.g., (29, 65) may be (29 - -3.552714e-15, 65).
    shp_pol <- sf::st_intersection(x = xy.sfc, y = shpi, ...)
    tmp_coords <- sf::st_coordinates(shp_pol)[, 1:2]
    if (length(tmp_coords) == 0) {
      dim(tmp_coords) <- NULL
    } else if (is.null(dim(tmp_coords))) {
      tmp_coords <- array(tmp_coords, dim = c(1, length(tmp_coords)))
    }
    if (!is.null(dim(tmp_coords))) {

      # polygon_instersection
      for (ii in 1:nrow(tmp_coords)) {
        # pt_x <- which(lon == tmp_coords[ii, 1])
        # pt_y <- which.min(abs(lat - tmp_coords[ii, 2]))
        if (!region) {
          # min(abs(lon - tmp_coords[ii, 1]))
          # min(abs(lat - tmp_coords[ii, 2]))
          mask[which.min(abs(lon - tmp_coords[ii, 1])),
              which.min(abs(lat - tmp_coords[ii, 2]))] <- n
        } else {
          mask[which.min(abs(lon - tmp_coords[ii, 1])),
              which.min(abs(lat - tmp_coords[ii, 2]))] <- 1
        }
      }
    } else if (find_min_dist) {
      x.centroid.shpi <- sf::st_coordinates(sf::st_centroid(shpi))[,1]
      y.centroid.shpi <- sf::st_coordinates(sf::st_centroid(shpi))[,2]
      dist <- sqrt((xy.sfc[,1] - x.centroid.shpi)**2 + (xy.sfc[,2] - y.centroid.shpi)**2)
      tmp_coords <- array(xy.sfc[which(dist == min(dist, na.rm = TRUE)),], dim = c(1,2))
      colnames(tmp_coords) <- c('X', 'Y')
      if (max(dist) <= max_dist & (any(round(lat,2) == round(tmp_coords[1,2],2)) & 
                                  any(round(lon,2) == round(tmp_coords[1,1],2))) ) {
        if (length(dim(mask)) == 2) {
          mask[which.min(abs(lon - tmp_coords[, 1])),
              which.min(abs(lat - tmp_coords[, 2]))] <- n
        } else {
          mask[which.min(abs(lon - tmp_coords[, 1])),
              which.min(abs(lat - tmp_coords[, 2]))] <- 1
        }
        # warning(paste0('The reference grid has no intersection with region ', 
        #         ifelse(is.character(shp_col_name_ids), shpi[[shp_col_name_ids]], paste0('n° ', n)), 
        #         ' from the shapefile; the provided grid cell is at a distance of ', dist[which(dist == min(dist, na.rm = TRUE))], 
        #         ' to the centroid of the region (units are: ° or meters depending on the crs of the shapefile).'))
      } else {
        # warning(paste0('The reference grid has no intersection with region ', 
        #         ifelse(is.character(shp_col_name_ids), shpi[[shp_col_name_ids]], paste0('n° ', n))))
      }
    }
  }
  return(mask)
}


areagrid <- function(data, lon, lat, target_proj) {

  # Build data dataframe
  lonlat_df <- data.frame(lon = rep(as.vector(lon), length(lat)),
                          lat = sort(rep(as.vector(lat), length(lon)), decreasing = TRUE))
                          
  data_df <- lonlat_df %>%
    dplyr::mutate(dat = as.vector(data))

  lonlat_df_ori <- NULL

  #NOTE: if target_proj = "ESRI:54030", Nord3v2 has different behavior from hub and ws!!
  data_df <- st_as_sf(data_df, coords = c("lon", "lat"), crs = target_proj)
  # data_df <- st_transform(data_df, crs = target_proj)
  data_df <- data_df %>%
    dplyr::mutate(long = st_coordinates(data_df)[, 1],
                  lat = st_coordinates(data_df)[, 2])

    # Calculate polygon points from regular lat/lon
    #NOTE: The original grid must be regular grid with same space
    d_lon <- abs(lon[2] - lon[1]) / 2
    d_lat <- abs(lat[2] - lat[1]) / 2
    lon_poly <- lat_poly <- rep(NA, 4 * dim(lonlat_df)[1])
    for (ii in 1:dim(lonlat_df)[1]) {
      lon_poly[(ii*4-3):(ii*4)] <- c(lonlat_df$lon[ii] - d_lon, lonlat_df$lon[ii] + d_lon,
                                     lonlat_df$lon[ii] + d_lon, lonlat_df$lon[ii] - d_lon)
      lat_poly[(ii*4-3):(ii*4)] <- c(lonlat_df$lat[ii] - d_lat, lonlat_df$lat[ii] - d_lat,
                                     lonlat_df$lat[ii] + d_lat, lonlat_df$lat[ii] + d_lat)
    }
    #  # To prevent out-of-global lon
    #  lon_poly[which(lon_poly >= 180)] <- 179.9
    #  lon_poly[which(lon_poly < -180)] <- -180
    # To prevent out-of-global lat
    lat_poly[which(lat_poly > 90)] <- 90
    lat_poly[which(lat_poly < -90)] <- -90

    lonlat_df <- data.frame(lon = lon_poly, lat = lat_poly)
    # Transfer lon/lat to projection
    proj_lonlat <- st_as_sf(lonlat_df, coords = c("lon", "lat"), crs = target_proj)
    #NOTE: if target_proj = "ESRI:54030", on Nord3v2, st_transform has lon and lat swapped!
    # proj_lonlat <- st_transform(proj_lonlat, crs = target_proj)
    lonlat_df_proj <- st_coordinates(proj_lonlat)

    # Use id to create groups for each polygon
    ids <- factor(paste0("id_", 1:dim(data_df)[1]))
    values <- data.frame(id = ids,
                         value = data_df$dat)
    positions <- data.frame(id = rep(ids, each = 4),
                            x = lonlat_df_proj[, 1],
                            y = lonlat_df_proj[, 2])
    datapoly <- merge(values, positions, by = "id")
    datapoly <- st_as_sf(datapoly, coords = c("x", "y"), crs = target_proj)

    datapoly <- datapoly %>%
                dplyr::group_by(.data$id) %>%
                dplyr::summarise() %>%  #NOTE: VERY SLOW if plot global
                dplyr::mutate(value = values[order(values$id), ]$value) %>%
                st_cast("POLYGON") %>%
                st_convex_hull()  # maintain outer polygen (no bowtie shape)

    return(datapoly)

}