Calculating areas of different raster classes in R with Superclass

Question

I have a classified image with three categories (specific grass species: 1, ground: 2, other vegetation: 3). I am trying to calculate the area of each category. For the classification process I used the superClass RStoolbox method. For the calculating area method, I am using another code in a similar question, but am having some trouble with my image being in a superClass format type. I tried exporting and importing the file as a new raster and ended up getting a different error. Any suggestions?

img <- brick("C:/Users/name/location/Blung/null.tif")
shp <- st_read("C:/Users/name/location/Blung/Training.shp")
shpS4 <- as_Spatial(shp)

NDVI.Overlay <- function(b1, b4) {
  NDVI.Calc <- (b4 - b1) / (b4 + b4)
  return(NDVI.Calc)
}

NDVI <- overlay(img[[1]], img[[4]], fun = NDVI.Overlay)

img_update <- addLayer(img, NDVI)
names(img_update) <- c('b1', 'b2', 'b3', 'b4', 'NDVI')

SC <- superClass(img, trainData = shpS4, responseCol = "id", model = "rf", tuneLength = 1, trainPartition = 0.8)

SC[] = sample(1:3, ncell(SC), replace=TRUE)
counts <- tapply(area(SC), SC[], sum)

Error in (function (classes, fdef, mtable)  : 
  unable to find an inherited method for function ‘area’ for signature ‘"superClass"’

writeRaster(SC$map, filename = "C:/Users/name/location/Blung_classification_R.tif")
img_processing <- raster("C:/Users/name/location/Blung/Blung_classification_R.tif")

img_processing[] = sample(1:3, ncell(img_processing), replace=TRUE)
counts <- tapply(area(img_processing), img_processing[], sum)

Error in res[i] <- readBin(x@file@con, what = dtype, n = 1, size = dsize,  : 
  replacement has length zero
In addition: Warning messages:
1: In .local(x, ...) :
  This function is only useful for Raster* objects with a longitude/latitude coordinates
2: In .rasterFromRasterFile(grdfile, band = band, objecttype, ...) :
  size of values file does not match the number of cells (given the data type)

Answer 1

You are overwriting the elements in your list in using SC[] <- .... Look at the result of your list object before and after applying that specific line of code.

Let' recreate a semblance of your model.

library(RStoolbox)
library(raster)

data(rlogo)
train <- readRDS(system.file("external/trainingPoints.rds", 
                 package="RStoolbox"))
SC <- superClass(rlogo, trainData = train, responseCol = "class", 
                 model = "rf", tuneLength = 1, 
                 trainPartition = 0.7)

Now, we simply can extract the prediction to a separate object or you can use a double bracket index SC[["map"]] and not create a new object.

pred <- SC[["map"]]

Is your data in a geographic (lat/long) coordinate projection? The raster::area function is intended to return area in a decimal degree geographic projection, which is honestly not good practice for remote sensing analysis. Besides, the function needs an index to know what cells to look at and not an overall count of cells.

If you are working in a projected coordinate system then simply returning the class counts lets you get to area, based on the known cell resolution. This can be done easily by passing a vector of the predicted raster to table and applying the a cell resolution scaling factor, say 30m^2.

table(pred[])*(30^2) # or
raster::freq(pred)*(30^2)

I am curious as to the point of creating a random sample of 1:3 matching the number of cells in the raster.

You are creating a random sample containing the values of [1,2,3].

rs <- sample(1:3, 100, replace=TRUE)

Now, lets create a vector (p) that we would consider the prediction. And get the counts in relation to the random sample.

p <- sample(1:3, 100, replace=TRUE)
  tapply(p, rs, length)

Now, compare those to the actual counts (note; the second line is a brute force approach to verify the tapply results). The random sample aggregation results do not seem to match the actual counts.

tapply(p, p, length)
  length(p[p == 1])