使用 Earth Mover 距离 (EMD) 计算 .tif 栅格之间空间利用率的相似性

Question

I am analyzing animal tracking data within an acoustic receiver array using dynamic Brownian bridge movement models. The dataset contains an animal identifier and every detection on a receiver has a timestamp and a lat/lon coordinates (in decimal degrees). Further, there is are two additional grouping variables "studyperiod" and "sgroup".

  Elasmo            Datetime      Lat       Lon studyperiod sgroup
1 X10141 2019-02-13 15:29:00 25.72441 -79.30922      IS2019  naive
2 X10141 2019-02-13 15:44:00 25.72441 -79.30922      IS2019  naive
3 X10141 2019-02-13 15:48:00 25.72441 -79.30922      IS2019  naive
4 X10141 2019-02-13 17:17:00 25.72441 -79.30922      IS2019  naive
5 X10141 2019-02-13 17:20:00 25.72441 -79.30922      IS2019  naive
6 X10141 2019-02-13 18:00:00 25.72441 -79.30922      IS2019  naive

I then used the GitHub package 'dBBMMhomeRange' (/https://github.com/SimonDedman/dBBMMhomeRange) to calculate individual-level Utilization distributions first, which are then scaled, weighted, summed to group-level UDs. The resulting group-level UDs are saved as.ascii rasters. As the rasters had different extents, I imported them into ArcMap and specified a shared extent within the program. Rasters with the same shared extent were then exported from Arcmap and imported into R. The rasters have the following properties:

r1 <- raster("~/N_IS2017_sc.tif")
r4 <- raster("~/N_IS2020_sc.tif")

> r1;r4
class      : RasterLayer 
dimensions : 1450, 1264, 1832800  (nrow, ncol, ncell)
resolution : 50, 50  (x, y)
extent     : -31658.67, 31541.33, -36085.92, 36414.08  (xmin, xmax, ymin, ymax)
crs        : NA 
source     : N_IS2017_sc.tif 
names      : N_IS2017_sc 
values     : 0, 0.0003713508  (min, max)

class      : RasterLayer 
dimensions : 1450, 1264, 1832800  (nrow, ncol, ncell)
resolution : 50, 50  (x, y)
extent     : -31658.67, 31541.33, -36085.92, 36414.08  (xmin, xmax, ymin, ymax)
crs        : NA 
source     : N_IS2020_sc.tif 
names      : N_IS2020_sc 
values     : 0, 0.0004588088  (min, max)

Now, I would like to use Earth Mover's Distance as described by Kranstauber et al. (2016) to calculate similarity in space use between the different groups. To do so I used emdDists() function from the move package.

## create a raster stack
allrasters <- stack(r1,r4)

## EMD
emdDists <- emd(allrasters/cellStats(allrasters, sum), threshold = 700)

However, the function starts running but always crashes. Which makes me think that something I did was not correct.

So, this brings me to my two questions:

1. The corresponding paper uses a UDStack as object to supply to the EMDDists(). So I am unsure if it is possible to implement the EMD approach for here described.tif rasters as well?

2. If correct, are there ways to reduce the calculation power demands to R?

Answer 1

Thanks to comments from Chris (see above for background) I found the answer to my question. The important step was to convert the imported rasters to.UD class objects so that they can be used to create a UDStack. The code is pretty straightforward and quick using functions from the "move" package and the "move" package.

library("move")
library("raster")

# Import rasters
r1 <- raster("~/R1.tif")
r2 <- raster("~/R2.tif")
r3 <- raster("~/R3.tif")

# create .UD class rasters
r1ud <- new(".UD", r1)
r2ud <- new(".UD", r2)
r3ud <- new(".UD", r3)

# create a UDStack needed for emd()
stk <- UDStack(as.list(r1ud, r2ud, r3ud))

# 'stk' can now be used within the emd() function as described by Kranstauber et al. 2017