nlcd functions refactoring

- Missing geometry is fixed in process_nlcd - calc_nlcd does not take exceptions into account - Missing class in inst/extdata/nlcd_classes.csv is added
NIEHS · May 30, 2024 · 21658ab · 21658ab
1 parent ee50712
commit 21658ab
Show file tree

Hide file tree

Showing 7 changed files with 133 additions and 68 deletions.
diff --git a/NAMESPACE b/NAMESPACE
@@ -178,7 +178,6 @@ importFrom(terra,buffer)
 importFrom(terra,coltab)
 importFrom(terra,crop)
 importFrom(terra,crs)
-importFrom(terra,deepcopy)
 importFrom(terra,describe)
 importFrom(terra,distance)
 importFrom(terra,expanse)

diff --git a/R/calculate_covariates.R b/R/calculate_covariates.R
@@ -241,11 +241,15 @@ calc_koppen_geiger <-
 #' @param from SpatRaster(1). Output of \code{process_nlcd()}.
 #' @param locs terra::SpatVector of points geometry
 #' @param locs_id character(1). Unique identifier of locations
+#' @param mode character(1). One of `"exact"` 
+#'   (using [`exactextractr::exact_extract()`])
+#'   or `"terra"` (using [`terra::freq()`]).
 #' @param radius numeric (non-negative) giving the
 #' radius of buffer around points
 #' @param max_cells integer(1). Maximum number of cells to be read at once.
 #' Higher values may expedite processing, but will increase memory usage.
-#' Maximum possible value is `2^31 - 1`.
+#' Maximum possible value is `2^31 - 1`. Only valid when
+#' `mode = "exact"`.
 #' See [`exactextractr::exact_extract`] for details.
 #' @param geom logical(1). Should the geometry of `locs` be returned in the
 #' `data.frame`? Default is `FALSE`. If `geom = TRUE` and `locs` contain
@@ -254,8 +258,13 @@ calc_koppen_geiger <-
 #' coordinate reference system of the `$geometry` is the coordinate
 #' reference system of `from`.
 #' @param ... Placeholders.
-#' @note NLCD is available in U.S. only. Users should be cautious
-#' the spatial extent of the data.
+#' @note NLCD is available in U.S. only. Users should be aware of
+#' the spatial extent of the data. The results are different depending
+#' on `mode` argument. The `"terra"` mode is less memory intensive
+#' but less accurate because it counts the number of cells
+#' intersecting with the buffer. The `"exact"` may be more accurate
+#' but uses more memory as it will account for the partial overlap
+#' with the buffer.
 #' @seealso [`process_nlcd`]
 #' @returns a data.frame object
 #' @importFrom utils read.csv
@@ -264,7 +273,6 @@ calc_koppen_geiger <-
 #' @importFrom terra project
 #' @importFrom terra vect
 #' @importFrom terra crs
-#' @importFrom terra deepcopy
 #' @importFrom terra set.crs
 #' @importFrom terra buffer
 #' @importFrom sf st_union
@@ -276,11 +284,14 @@ calc_koppen_geiger <-
 calc_nlcd <- function(from,
                       locs,
                       locs_id = "site_id",
+                      mode = c("exact", "terra"),
                       radius = 1000,
                       max_cells = 5e7,
                       geom = FALSE,
+                      nthreads = 1L,
                       ...) {
   # check inputs
+  mode <- match.arg(mode)
   if (!is.numeric(radius)) {
     stop("radius is not a numeric.")
   }
@@ -307,43 +318,61 @@ calc_nlcd <- function(from,
   # select points within mainland US and reproject on nlcd crs if necessary
   data_vect_b <-
     terra::project(locs_vector, y = terra::crs(from))
-
   # create circle buffers with buf_radius
-  bufs_pol <- terra::buffer(data_vect_b, width = radius) |>
-    sf::st_as_sf() |>
-    sf::st_geometry()
-  # ratio of each nlcd class per buffer
-  nlcd_at_bufs <-
-    exactextractr::exact_extract(
-      from,
-      bufs_pol,
-      fun = "frac",
-      force_df = TRUE,
-      progress = FALSE,
-      max_cells_in_memory = max_cells
-    )
-
-  # select only the columns of interest
+  bufs_pol <- terra::buffer(data_vect_b, width = radius)
   cfpath <- system.file("extdata", "nlcd_classes.csv", package = "amadeus")
   nlcd_classes <- utils::read.csv(cfpath)
-  nlcd_at_bufs <-
-    nlcd_at_bufs[
-      sort(names(nlcd_at_bufs)[
-        grepl(paste0("frac_(", paste(nlcd_classes$value, collapse = "|"), ")"),
-              names(nlcd_at_bufs))
-      ])
-    ]
+
+  if (mode == "fast") {
+    # fast mode
+    class_query <- "names"
+    # extract land cover class in each buffer
+    nlcd_at_bufs <-
+      terra::freq(
+        from,
+        zones = bufs_pol,
+        wide = TRUE
+      )
+    nlcd_at_bufs <- nlcd_at_bufs[, -seq(1, 2)]
+    nlcd_cellcnt <- nlcd_at_bufs[, seq(1, ncol(nlcd_at_bufs), 1)]
+    nlcd_cellcnt <- nlcd_cellcnt / rowSums(nlcd_cellcnt)
+    nlcd_at_bufs[, seq(1, ncol(nlcd_at_bufs), 1)] <- nlcd_cellcnt
+  } else {
+    class_query <- "value"
+    # ratio of each nlcd class per buffer
+    bufs_pol <- bufs_pol |>
+      sf::st_as_sf() |>
+      sf::st_geometry()
+    nlcd_at_bufs <-
+      exactextractr::exact_extract(
+        from,
+        bufs_pol,
+        fun = "frac",
+        force_df = TRUE,
+        progress = FALSE,
+        max_cells_in_memory = max_cells
+      )
+
+    # select only the columns of interest
+    nlcd_at_buf_names <- names(nlcd_at_bufs)
+    nlcd_val_cols <-
+      grep("^frac_", nlcd_at_buf_names)
+    nlcd_at_bufs <- nlcd_at_bufs[, nlcd_val_cols]
+  }
+
   # change column names
   nlcd_names <- names(nlcd_at_bufs)
   nlcd_names <- sub(pattern = "frac_", replacement = "", x = nlcd_names)
-  nlcd_names <- sort(as.numeric(nlcd_names))
-  nlcd_names <- nlcd_classes[nlcd_classes$value %in% nlcd_names, c("class")]
-  new_names <- sapply(
-    nlcd_names,
-    function(x) {
-      sprintf("LDU_%s_0_%05d", x, radius)
-    }
-  )
+  nlcd_names <-
+    switch(
+      mode,
+      exact = sort(as.numeric(nlcd_names)),
+      fast = nlcd_names
+    )
+  nlcd_names <-
+    nlcd_classes$class[match(nlcd_names, nlcd_classes[[class_query]])]
+  #nlcd_classes[nlcd_classes[[class_query]] %in% nlcd_names, c("class")]
+  new_names <- sprintf("LDU_%s_0_%05d", nlcd_names, radius)
   names(nlcd_at_bufs) <- new_names
   # merge locs_df with nlcd class fractions
   new_data_vect <- cbind(locs_df, as.integer(year), nlcd_at_bufs)
@@ -404,7 +433,7 @@ calc_ecoregion <-
     locs_df <- locs_prepared[[2]]
 
     extracted <- terra::intersect(locsp, from)
-    return(extracted)
+
     # Generate field names from extracted ecoregion keys
     # TODO: if we keep all-zero fields, the initial reference
     # should be the ecoregion polygon, not the extracted data

diff --git a/R/process.R b/R/process.R
@@ -683,6 +683,8 @@ process_koppen_geiger <-
 #' returning a single `SpatRaster` object.
 #' @param path character giving nlcd data path
 #' @param year numeric giving the year of NLCD data used
+#' @param extent numeric(4) or SpatExtent giving the extent of the raster
+#'   if `NULL` (default), the entire raster is loaded
 #' @param ... Placeholders.
 #' @description Reads NLCD file of selected `year`.
 #' @returns a `SpatRaster` object
@@ -695,6 +697,7 @@ process_nlcd <-
   function(
     path = NULL,
     year = 2021,
+    extent = NULL,
     ...
   ) {
     # check inputs
@@ -717,7 +720,7 @@ process_nlcd <-
     if (length(nlcd_file) == 0) {
       stop("NLCD data not available for this year.")
     }
-    nlcd <- terra::rast(nlcd_file)
+    nlcd <- terra::rast(nlcd_file, win = extent)
     terra::metags(nlcd) <- c(year = year)
     return(nlcd)
   }
@@ -752,9 +755,11 @@ process_ecoregion <-
     ecoreg <- ecoreg[, grepl("^(L2_KEY|L3_KEY)", names(ecoreg))]
     ecoreg_edit_idx <- sf::st_intersects(ecoreg, poly_tukey, sparse = FALSE)
     ecoreg_edit_idx <- vapply(ecoreg_edit_idx, function(x) any(x), logical(1))
-    ecoreg_else <- ecoreg[!ecoreg_edit_idx, ]
-    ecoreg_edit <- sf::st_union(ecoreg[ecoreg_edit_idx, ], poly_tukey)
-    ecoreg <- rbind(ecoreg_else, ecoreg_edit)
+    if (!all(ecoreg_edit_idx == 0)) {
+      ecoreg_else <- ecoreg[!ecoreg_edit_idx, ]
+      ecoreg_edit <- sf::st_union(ecoreg[ecoreg_edit_idx, ], poly_tukey)
+      ecoreg <- rbind(ecoreg_else, ecoreg_edit)
+    }
     ecoreg$time <- paste0(
       "1997 - ", data.table::year(Sys.time())
     )

diff --git a/inst/extdata/nlcd_classes.csv b/inst/extdata/nlcd_classes.csv
@@ -1,17 +1,18 @@
 "","value","class","names","col"
-"1",0,"TUNCL","Unclassified","white"
+"1",0,"TUNCL","Unclassified","#ffffff00"
 "2",11,"TWATR","Open Water","#476ba1"
-"3",21,"TDVOS","Developed, Open Space","#decaca"
-"4",22,"TDVLO","Developed, Low Intensity","#d99482"
-"5",23,"TDVMI","Developed, Medium Intensity","#ee0000"
-"6",24,"TDVHI","Developed, High Intensity","#ab0000"
-"7",31,"TBARN","Barren Land","#b3aea3"
-"8",41,"TDFOR","Deciduous Forest","#68ab63"
-"9",42,"TEFOR","Evergreen Forest","#1c6330"
-"10",43,"TMFOR","Mixed Forest","#b5ca8f"
-"11",52,"TSHRB","Shrub/Scrub","#ccba7d"
-"12",71,"THERB","Herbaceous","#e3e3c2"
-"13",81,"TPAST","Hay/Pasture","#dcd93d"
-"14",82,"TPLNT","Cultivated Crops","#ab7028"
-"15",90,"TWDWT","Woody Wetlands","#bad9eb"
-"16",95,"THWEM","Emergent Herbaceous Wetlands","#70a3ba"
+"3",12,"TSNOW","Perennial Snow/Ice","#F5F5F5"
+"4",21,"TDVOS","Developed, Open Space","#decaca"
+"5",22,"TDVLO","Developed, Low Intensity","#d99482"
+"6",23,"TDVMI","Developed, Medium Intensity","#ee0000"
+"7",24,"TDVHI","Developed, High Intensity","#ab0000"
+"8",31,"TBARN","Barren Land","#b3aea3"
+"9",41,"TDFOR","Deciduous Forest","#68ab63"
+"10",42,"TEFOR","Evergreen Forest","#1c6330"
+"11",43,"TMFOR","Mixed Forest","#b5ca8f"
+"12",52,"TSHRB","Shrub/Scrub","#ccba7d"
+"13",71,"THERB","Herbaceous","#e3e3c2"
+"14",81,"TPAST","Hay/Pasture","#dcd93d"
+"15",82,"TPLNT","Cultivated Crops","#ab7028"
+"16",90,"TWDWT","Woody Wetlands","#bad9eb"
+"17",95,"THWEM","Emergent Herbaceous Wetlands","#70a3ba"
diff --git a/man/calc_nlcd.Rd b/man/calc_nlcd.Rd
diff --git a/man/process_nlcd.Rd b/man/process_nlcd.Rd
diff --git a/tests/testthat/test-calculate_covariates.R b/tests/testthat/test-calculate_covariates.R
@@ -441,10 +441,10 @@ testthat::test_that("Check calc_nlcd works", {
   withr::local_package("sf")
   withr::local_options(list(sf_use_s2 = FALSE))
 
-  point_us1 <- cbind(lon = -114.7, lat = 38.9, id = 1)
-  point_us2 <- cbind(lon = -114, lat = 39, id = 2)
-  point_ak <- cbind(lon = -155.997, lat = 69.3884, id = 3) # alaska
-  point_fr <- cbind(lon = 2.957, lat = 43.976, id = 4) # france
+  point_us1 <- cbind(lon = -114.7, lat = 38.9, site_id = 1)
+  point_us2 <- cbind(lon = -114, lat = 39, site_id = 2)
+  point_ak <- cbind(lon = -155.997, lat = 69.3884, site_id = 3) # alaska
+  point_fr <- cbind(lon = 2.957, lat = 43.976, site_id = 4) # france
   eg_data <- rbind(point_us1, point_us2, point_ak, point_fr) |>
     as.data.frame() |>
     terra::vect(crs = "EPSG:4326")
@@ -474,6 +474,22 @@ testthat::test_that("Check calc_nlcd works", {
               radius = -3),
     "radius has not a likely value."
   )
+
+  # -- buf_radius has likely value
+  testthat::expect_no_error(
+    calc_nlcd(locs = eg_data,
+              from = nlcdras,
+              mode = "exact",
+              radius = 300)
+  )
+  testthat::expect_no_error(
+    calc_nlcd(locs = eg_data,
+              from = nlcdras,
+              mode = "terra",
+              radius = 300)
+  )
+
+
   # -- year is numeric
   testthat::expect_error(
     process_nlcd(path = path_testdata, year = "2021"),
@@ -492,7 +508,7 @@ testthat::test_that("Check calc_nlcd works", {
   )
   testthat::expect_error(
     calc_nlcd(locs = 12,
-              locs_id = "id",
+              locs_id = "site_id",
               from = nlcdras)
   )
   testthat::expect_error(
@@ -519,14 +535,14 @@ testthat::test_that("Check calc_nlcd works", {
   testthat::expect_no_error(
     calc_nlcd(
       locs = eg_data,
-      locs_id = "id",
+      locs_id = "site_id",
       from = nlcdras,
       radius = buf_radius
     )
   )
   output <- calc_nlcd(
     locs = eg_data,
-    locs_id = "id",
+    locs_id = "site_id",
     radius = buf_radius,
     from = nlcdras
   )
@@ -556,7 +572,7 @@ testthat::test_that("Check calc_nlcd works", {
   )
   output_geom <- calc_nlcd(
     locs = eg_data,
-    locs_id = "id",
+    locs_id = "site_id",
     radius = buf_radius,
     from = nlcdras,
     geom = TRUE