1

I have downloaded a land surface temperature (LST) raster from ECOSTRESS as well as it's associate cloud mask (you can sign-up for free here and download your own data).

NASA's tutorial (minute 31:15) on YouTube uses Python to remove cloudy pixels from the LST image. I'd like to work with R for the same task. So, I did:

library(terra)

# working dir
setwd <- "path/"

# read the raw LST
lst_raw <- rast(paste0(wd, "ECO2LSTE.001_SDS_LST_doy2022152172444_aid0001.tif"))

# read the cloud mask
cloud_mask <- rast(paste0(wd, "ECO2CLD.001_SDS_CloudMask_doy2022152172444_aid0001.tif"))

# convert the cloud mask to binary rast
cloud_mask_binary <- ?????????

The part of converting the cloud mask to binary raster in Python is:

cloud.data = (cloud.data >> 2) & 1

I quote from the YouTube video:

To process this quality flag image into a cloud mask we are going to access the data array of this dataset with .data and we are going to bit shift this array two bits to the right and then and with one and this will produce a binary cloud mask.

What is the equivalent Python's code for converting the cloud mask to binary raster in R?

R 4.4.0, RStudio 2024.04.2 Build 764, Windows 11.

2 Answers 2

2

@Spacedman I found another solution, which seems to work okay and there is no need to install other libraries.

library(terra)

# working dir
wd <- "path/"

# read the raw LST
lst_raw <- rast(paste0(wd, "ECO2LSTE.001_SDS_LST_doy2022152172444_aid0001.tif"))
plot(lst_raw)

raw lst

# read the cloud mask
cloud_mask <- rast(paste0(wd, "ECO2CLD.001_SDS_CloudMask_doy2022152172444_aid0001.tif"))
plot(cloud_mask)

cloud mask

 # convert the cloud mask to binary raster
    cloud_mask_binary <- (cloud_mask %/% 4) %% 2
    plot(cloud_mask_binary)

cloud mask binary

lst_masked <- ifel(cloud_mask_binary == 0, lst_raw, NA)
plot(lst_masked)

masked lst

I haven't converted the LST image to Celsius nor I applied the scale factor yet, that's why the "strange" pixel values.

2
  • Yes, integer divide by 4 is the same as two binary shifts right, then mod 2 tests the 0/1 status of the rightmost bit! Probably not much difference in speed to using bitops.
    – Spacedman
    Commented Jun 30 at 10:06
  • I haven't tested the bitops package yet. I will do it later today.
    – Nikos
    Commented Jun 30 at 10:10
1

Using the bitops package you can work on values at the bit level:

> library(bitops)

First test a matrix with all 8-bit values from 0 to 255 in a 16x16 raster:

> r = rast(matrix(0:255,16,16))

We can't operate directly on this so let's copy it and work on the values of a copy:

> rmask = r
> rmask[] = rmask[] %&% 2^4
> plot(rmask)

Here's I've used the bitwise "AND" operator with binary pattern 00010000 to see if the 5th bit from the right (4th counting from zero) is a one.

enter image description here

Then you can compare rmask with 0 to get a 0/1 mask raster.

That python code looks like its testing against bit 3 (0-indexed) so you want to do cloudmask[] = cloudmask[] %&% 2^2 but check this (I've not downloaded the data to test). Ooh, and you can use terra::app to save the initial making a copy step:

> rmask = app(r, \(x){x %&% 4})
> plot(rmask)
1
  • as you said, bitops can't work directly on spatraster objects (it either works with matrices or vectors) Here is what I tried: library(bitops) # raster to matrix cloud_mask_matrix <- values(cloud_mask) cloud_mask_binary_values <- bitAnd(bitShiftR(cloud_mask_matrix, 2), 1) # create an empty raster with the same dim, ext and crs as the original cloud mask raster cloud_mask_binary <- rast(cloud_mask) values(cloud_mask_binary) <- cloud_mask_binary_values plot(cloud_mask_binary) Identical results to my answer. The speed is almost the same as the terra lib.
    – Nikos
    Commented Jun 30 at 13:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.