6

Here I found a very interesting blog:critical threshold in temperature effects and empirical approach is very interesting, so I want to implement its idea in R. However, I have multi-layer raster data of German' historical daily temperatures (15 years' historical daily mean temperature) in large RasterBrick object. According to the empirical approach that discussed in inspired post, I need to construct the distribution of temperature from my multi-layer raster data.

Update 2: reproducible shapefile:

I aware that downloading shapefile from 3rd party website is not practical, so here I come up reproducible shapefile to give it try:

library(sf)
library(maps)
library(rgeos)
library(mapdata)

germany <- st_as_sf(map("Germany", plot = FALSE, fill = TRUE))
write_sf(germany, "germany.shp")

To easily follow up my post, I created reproducible raster data to work with in R. I also provide Germany' shapefile that taken from eurostat website; here is the shapefile on the fly (I can guarantee the link is quite safe and file is very small to use): eurostat' shapefile and here is handy reproducible raster data:

reproducible data

library(raster)
library(lubridate)
library(tidyverse)

r <- raster(xmn=5.75, xmx= 15, ymn = 47.25, ymx =55,res=c(0.25,0.25))
Deu_crop <- do.call(stack,lapply(1:5479,function(i) setValues(r,round(runif(n = ncell(r),min = -10,max = 25)))))
names(Deu_crop) <- paste0('X',gsub('-','.',ymd('1980.01.01') + days(1:5479)))
shp <- shapefile('eurostat_NUTS3/deu_adm_2006.shp')
e <- raster::extract(Deu_crop,shp)
names(e) <- shp$NUTS_ID

so to test the workflow that presented in inspired post, I need to design several global variables which will be used helper functions that presented in here. But I don't understand how to design some critical global variables that used to accomplish its workflow; It is recommended to define global variable like: w - weather data; tempDat: particular aggregated weather data; Trows: span aggregated grid data; and T: vector of integer temperature (details can be found here: details).

I want to estimate the distribution of temperatures over time from the gridded daily weather data. But I have a difficulty to test the empirical steps that presented in this inspired post because it didn't mention the solution for the case of handling multi-layer raster data, so I don't know how to adopt its fantastic idea on my own in R.

Here is my approach to aggregate multi-layer raster data for each polygons from shapefile (eurostat' shapefile) before using helper funcion in inspired post:

initial attempt to manipulate multi-layers raster:

rasterHelper <- function(ix,e){
  gather(data.frame(e[[ix]],stringsAsFactors = F),'colname','temp') %>% 
    group_by(colname) %>% summarise(temp = mean(temp)) %>% ungroup() %>% # spatial mean
    mutate(year = sub('X(\\d{4}).+','\\1',colname)) %>% 
    group_by(year) %>% summarise_all(funs(sum)) %>% mutate(NUTS_ID = names(e)[ix])
}
do.call(rbind,lapply(1:length(e),function(ix) rasterHelper(ix)))

but my above attempt is not working; In my attempt, I intend to aggregate temperature raster data for each polygon. The implementation of inspired post is very useful but it is still hard to follow up for handling multi-layer raster data. I assume I should work on each raster layer and construct temperature distribution over time, but really don't have a solid idea how to do in R. Any idea?

Update:

Here is the paper that I got inspiration from it: nonlinear temperature effect ..., but implementing the proposed method is still challenging for me even I followed the workflow that presented in the respective blog: searching critical threshold in temperature effect

Is there anyone point me out how to adopt its empirical approach on multi-layer raster data in R? How can I estimate the distribution of temperature over time? How can I make this happen in R? Any more thoughts? Thanks

jyson
  • 245
  • 1
  • 8
  • 27
  • Write a function that works with a single cell (time series). Test it for one cell at a time. Then use the function with `raster::calc` – Robert Hijmans Jun 20 '18 at 05:49
  • I do not want to download a file (there are plenty of polygon examples that ship with R). Also, I may have been wrong. Looking at your question again it seems that you want to compute something for each layer, rather than for each cell across layers (time) (which is what calc does). But I find it too difficult to decipher your question. It is up to you to provide a simple clean & clear question. Sorry – Robert Hijmans Jun 20 '18 at 06:05
  • I don't think that this is a minimal example... Think about reducing the computational cost while maintaining the questions core. – loki Jul 04 '18 at 09:51
  • @loki yes, I greatly simplified my question in my new post: [reshape multi-layers raster](https://stackoverflow.com/questions/51171092/any-work-around-to-reshape-and-aggregate-multi-layers-raster-grid-into-plain-tex), could you give your attention and share your thought please? Thank you – jyson Jul 04 '18 at 10:11

1 Answers1

2

I am not quite sure what you want to do.

To set up a smaller example:

library(raster)
lux <- shapefile(system.file("external/lux.shp", package="raster"))
r <- raster(lux)
s <- stack(lapply(1:12, function(i) setValues(r, 1:ncell(r))))

e <- extract(s, lux)

Now you say that you want to aggregate --- that is a bit ambiguous, but perhaps what you want is 

x <- lapply(e, function(i) apply(i,2,mean))

equivalent to

y <- extract(s, lux, fun='mean')
Robert Hijmans
  • 40,301
  • 4
  • 55
  • 63
  • Professor, I restated my problem entirely in my new post: [my new post that new definition of this post](https://stackoverflow.com/questions/51171092/any-workaround-to-reshape-and-aggregate-multi-layers-raster-grid-in-r); Could you elaborate your answer on my new post? For your current answer, I didn't produce the result I want to achieve. Thank you – jyson Jul 05 '18 at 09:38