我需要使用R计算由多边形覆盖的高分辨率栅格（具有8层栅格堆叠）中每个栅格像元的比例。
我的标准方法是使用raster :: rasterize（...，getCover = TRUE），但是，这种方法会花费很长的时间，尤其是当多边形的大小增加时。
作为替代方案，我尝试将栅格堆栈裁剪到多边形的范围，将栅格堆栈转换为多边形，然后根据所得形状与原始多边形的交点计算比例。这对于较小的多边形效果很好，但是会随着多边形的增加而分解，这是因为R用完了内存（我的内存限制为16GB）或相交的计算时间过长。
这是一个使用我当前解决方案的可重现示例，其中的形状文件非常小。
library(raster)
library(spex)
library(dplyr)
library(sf)
library(data.table)
# setup a dummy example
r <- raster(nrow = 21600, ncol = 43200)
r[] <- 1:933120000
r_stack <- stack(r,r,r,r,r,r,r,r)
# get a small dummy shapefile
shp_small <- raster::getData(name = "GADM", country = "CHE", level = 2, download = TRUE)
shp_small <- st_as_sf(shp_small)[1, ]
# for comparison, use a big dummy shapefile
# shp_big <- raster::getData(name = "GADM", country = "BRA", level = 0, download = TRUE)
## Approach for a small shape file
stack_small <- raster::crop(r_stack, shp_small, snap = "out")
## transform to polygon
stack_small_poly <- spex::polygonize(stack_small)
stack_small_poly$ID <- 1:nrow(stack_small_poly)
## I can then perform the necessary calculations on the polygons to obtain
## the proportional overlay
# transform to mollweide for area calculation
mollw <- "+proj=moll +lon_0=0 +x_0=0 +y_0=0 +datum=WGS84 +units=m +no_defs"
stack_small_crs <- st_crs(stack_small_poly)
stack_small <- st_transform(stack_small_poly, mollw)
stack_small_poly <- st_transform(stack_small_poly, mollw)
# calculate area for each cell
stack_small_poly$area_org <- st_area(stack_small_poly)
# transform to world equidistant cylindrical
stack_small_poly<- st_transform(stack_small_poly, 4087)
shp_small <- st_transform(shp_small, 4087)
# get call the cells that intersect with the shape (this might take a while)
i <- st_intersects(stack_small_poly, shp_small, sparse = FALSE)
stack_small_poly <- dplyr::filter(stack_small_poly, i)
# now calculate the extact intersection (this might take a while)
st_agr(stack_small_poly) <- "constant"
stack_small_poly <- st_intersection(stack_small_poly, st_geometry(shp_small))
# calculate the new areas and backtransform
stack_small_poly <- st_transform(stack_small_poly, mollw)
stack_small_poly$new_area <- st_area(stack_small_poly)
stack_small_poly <- st_transform(stack_small_poly, stack_small_crs)
# calculate proportion
stack_small_poly$proportion <- as.numeric(stack_small_poly$new_area/stack_small_poly$area_org)
# finally transform to data.table for subsequent analysis
st_geometry(stack_small_poly) <- NULL
setDT(stack_small_poly)
我正在寻找R中的解决方案，该解决方案能够在10-15分钟内（最好更快）执行任务，并且表示巴西的shapefile的内存限制为16 GB RAM（请参见上面的代码shp_big）。
我很清楚，这种最佳效果可能无法实现，并且导致执行时间和/或内存使用量减少的每条建议都超过了我们的要求。