堆叠栅格并计算每个像素的最大值，然后在 R 中的 rest 个图层中保留该值

Question

I have 181 daily ndvi rasters stacked. I want to find peak for each pixel and then assign peak value to the rest. Original Data Curve:

This is what I want:

This result is for single pixel. I've done it in excel but want to create a code in R for ease. Now every pixel reaches peak at different date (so different Layer in stack).

This is what i tried

fn <- list.files(pn, '*.tif$', full.names = TRUE)
rn <- lapply(fn, raster)
bn <- brick(rn)

for (j in 1:ncell(bn)) {
  x <- bn[j]
  y <- which.max(x)
  x[y:length(x)] <- x[y]
}

Now this code works if I use it without loop. Error is probably because some cells have NA's. Each layer has 24 cells and only 14 have data. Rest are masked.

Answer 1

For processing raster data, I would recommend using the terra package. It is an update to the raster package that is faster and in most cases easier to use (you can install it with install.packages('terra') ). terra has a set of apply functions that allow you to apply functions over each stack of pixels. In the example below, I create a 2x2 pixel raster with some dummy data, and use the terra::app function to carry out the process you outline in your question:

# Create random data for 181 rasters.
x = seq(-0, 10, length.out=181)

# `dnorm` can be used to simulate a bell curve, like the NDVI data. 
vals = rep(dnorm(x, 5, 1), 4)

# I couldn't get the raster to reshape correctly so I am just taking the first
# 50 values so the data are a nice bell curve. 
arr <- array(vals, dim = c(2, 2, 50)) 

# Replace this with your data. From your code, you can just use:
# bn <- terra::rast(fn)
r <- terra::rast(arr)

# Define a custom function that operates on one stack of pixels at a time.
fill_max <- function(timeseries) {
  
  # Find the index with the max value
  idx = which.max(timeseries)
  
  # Set all values past the max to the max value
  timeseries[idx:length(timeseries)] <- timeseries[idx]
  
  return(timeseries)
}

# Apply the function to the raster.
out <- terra::app(r, fun=fill_max)

# Check that process worked. Take the timeseries of the first pixel. 
vals <- terra::values(out)[1,]
# Plot the timeseries. 
plot(1:length(vals), vals)

^{Created on 2023-01-18 with reprex v2.0.2}

Answer 2

If you wanted the curve to never go down, you could use terra::cummax

Example data

library(terra)
r <- rast(ncol=10, nrow=10, nlyr=30)
set.seed(1)
values(r) <- runif(size(r))

Solution

x <- cummax(r)

Illustrated for four cells

par(mfrow=c(2,2), mar=rep(2,4))
for (i in c(2,3,8,9)) {
  plot(unlist(r[i]))
  lines(unlist(x[i]), col="blue", lwd=2)
}