R/to_closest.R

In manubio13/ku.enm: Detailed development of ecological niche models using Maxent

#' Move occurrences to closest pixel with environmental data
#'
#' @description kuenm_toclosest helps in changing the longitude and latitude values
#' of occurrences with no environmental data, so they move to the closest pixel
#' of a raster layer that contains relevant information. This process prevents
#' NAs in future analyses.
#'
#' @param data data.frame or matrix of occurrence records. Columns must include
#' longitude and latitude. Other columns are optional and wont be changed.
#' @param longitude (character) name of the column with longitude data.
#' @param latitude (character) name of the column with latitude data.
#' @param raster.layer RasterLayer to be used as a reference.
#' @param limit.distance (numeric) maximun distance in km at which an occurrence
#' could be to be moved. Records farther than this distance wont be moved.
#'
#' @return
#' A data.frame with the corrected coordinates and four additional columns.
#' The first of the new columns indicates the condition of the coordinates:
#' Correct, if it was not moved because it was on a pixel with data; Moved, if
#' it was moved to the nearest pixel; and Not_moved, if it was not moved because
#' the occurrence was farther than the \code{limit_distance} to the closest pixel.
#' The second new column indicates the distance to the closest pixel with data.
#' The other two additional columns will contain the initial longitudes and
#' latitudes.
#'
#' @export
#'
#' @examples
#' data("sp_test", package = "kuenm")
#'
#' var <- raster::raster(list.files(system.file("extdata", package = "kuenm"),
#'                                  pattern = "Mbio_", full.names = TRUE)[1])
#'
#' raster::plot(var)
#'
#' out <- rbind(c(-103, 27), c(-90, 26.5), c(-109, 40), c(-70, 41))
#' colnames(out) <- colnames(sp_test)
#'
#' data <- rbind.data.frame(sp_test, out)
#'
#' points(data)
#'
#' data1 <- kuenm_toclosest(data, longitude = "Longitude", latitude = "Latitude",
#'                          raster.layer = var, limit.distance = 200)
#'
#' points(data1[, 1:2], col = "red")

kuenm_toclosest <- function(data, longitude, latitude, raster.layer, limit.distance) {
  # detecting potential errors
  if (missing(data)) {
    stop("Argument data is necessary to perform the analysis")
  }
  if (missing(longitude)) {
    stop("Argument longitude is not defined.")
  }
  if (missing(latitude)) {
    stop("Argument latitude is not defined.")
  }
  if (missing(raster.layer)) {
    stop("Argument raster_layer is not defined.")
  }
  if (missing(limit.distance)) {
    stop("Argument limit.distance is not defined.")
  }

  # preparing data
  xy <- data[, c(longitude, latitude)]
  vals <- raster::extract(raster.layer, xy)

  tomove <- which(is.na(vals))
  xyout <- xy[tomove, ]

  if (nrow(xyout) > 0) {
    xyras <- raster::rasterToPoints(raster.layer)[, 1:2]

    dists <- raster::pointDistance(xyout, xyras, lonlat = TRUE)

    condition <- rep("Correct", nrow(data))
    distss <- rep(0, nrow(data))

    limdist <- limit.distance * 1000

    # running process
    cat("\nMoving occurrences to closest pixels:\n")
    if (class(xyout)[1] %in% c("matrix", "data.frame")) {
      no <- nrow(xyout)
    } else {
      no <- length(xyout)
    }

    for (i in 1:no) {
      if (class(xyout)[1]  %in% c("matrix", "data.frame")) {
        mindis <- min(dists[i, ])
      } else {
        mindis <- min(dists[i])
      }

      if (mindis <= limdist) {
        if (class(xyout)[1] %in% c("matrix", "data.frame")) {
          xyin <- xyras[dists[i, ] == mindis, ]
        } else {
          xyin <- xyras[dists[i] == mindis, ]
        }

        if (class(xyin)[1] %in% c("matrix", "data.frame")) {
          xyin <- xyin[1, ]
        }
        data[tomove[i], longitude] <- xyin[1]
        data[tomove[i], latitude] <- xyin[2]
        condition[tomove[i]] <- "Moved"
        distss[tomove[i]] <- mindis / 1000
        cat("\tOccurrence", i, "of", nrow(xyout), "moved\n")
      } else {
        condition[tomove[i]] <- "Not_moved"
        distss[tomove[i]] <- mindis / 1000
        cat(paste0("\tOccurrence ", i," of ", nrow(xyout), " was not moved because it is more than\n\t",
                   limit.distance, " km apart from the closest pixel with environmental values\n"))
      }
    }
    data <- data.frame(data, condition = condition, distance_km = distss,
                       initial_lon = xy[, 1], initial_lat = xy[, 2],
                       stringsAsFactors = FALSE)
  } else {
    cat("No occurrence is out of the area of the raster layer of reference with values.")
  }



  return(data)
}