--- title: "enerscape" output: rmarkdown::html_vignette vignette: > %\VignetteIndexEntry{enerscape} %\VignetteEngine{knitr::rmarkdown} %\VignetteEncoding{UTF-8} --- ```{r, include = FALSE} knitr::opts_chunk$set( collapse = TRUE, comment = "#>" ) ``` # Energy Landscapes With Enerscape Energy landscapes are spatially explicit projections of the energy cost of travel (kcal or J) for an animal. In enerscape, energy landscapes depend on the body mass of the animal and the incline between locations, following the ARC model from Pontzer (2016): . For a general overview of energy landscapes, refer to Berti et al. (2021): . ```{r setup} library(terra) library(enerscape) ``` Enerscape comes with the default dataset _sirente_; this is a matrix with values the digital elevation model (DEM) for Monte Sirente in Central Italy: ```{r sirente, fig.width=6, fig.height=3.65} data("sirente") dem <- rast( sirente, crs = "+proj=utm +zone=32 +datum=WGS84 +units=m +no_defs", extent = ext(879340, 885280, 4672880, 4676810) ) plot(dem) ``` I provided the correct coordinate reference system (CRS) and the extent of the DEM; in most cases, this will already be present in downloaded raster layers. Enerscape assumes that the coordinate system is in meters: ```{r crs} crs(dem, proj = TRUE) ``` You **have** to check this by yourself to make sure _+units=m_ or enerscape results will not be meaningful. Enerscape checks if this is the case and returns a warning if _+units_ does not match _m_, but it will still return its output. The energy landscapes can be obtained with the _enerscape()_ function. There are two mandatory inputs: _dem_, which is the SpatRaster of the DEM, and _mass_, which is the body mass of the animal (kg). ```{r enerscape} en <- enerscape(dem, 10, neigh = 4, unit = "kcal") ``` Before visualizing it, it is worth nothing that, from my experience, energy landscape values are usually right-skewed, i.e. with most cells having low values and few cells having high values. In such cases, it is useful to log-transform energy landscape for visualization: ```{r visualize, fig.width=6, fig.height=3.65} en_log <- log(en) plot(en_log) terra::contour(dem, add = TRUE, nlevels = 5) ``` ```{r zoom, fig.width=6, fig.height=3.65} en_log <- crop(en_log, ext(881306, 882853, 4674928, 4675984)) plot(en_log) terra::contour(dem, add = TRUE, nlevels = 20) ``` # In or Out movement By default, enerscape calculates the cost of moving _into_ a cell. However, it is possible to calculate the cost of moving _from_ the cell since enerscape v1.1.1. This is achieved by using the optional argument `direction = "out"`. ```{r out} en_out_log <- log(enerscape(dem, 10, neigh = 4, unit = "kcal", direction = "out")) en_out_log <- crop(en_out_log, ext(881306, 882853, 4674928, 4675984)) plot(c(en_log, en_out_log)) ``` # Initialization files for Circuitscape/Omniscape Circuitscape and Omniscape are circuit theory approaches to calculate connectivity in landscapes. The main idea of these two algorithm is that resistances in the landscape (e.g. due to energy costs) behave like resistances in electric circuit. For instance, parallel resistances will give a lower overall resistance than each of the individual resistances (). The major advantage of a circuit theory approach is that it consider all possible paths of travel, instead of focusing only on one, e.g. a least cost path. Circuity theory has, thus, more realistic assumptions on how animals behave and I preferred Circuitscape/Omniscape over least-cost path analysis. Following _rgeos_ and _rgdal_, I rewrote the core of the package in C++, removing the dependency also on _gdistance_. In doing so, I also removed the function to calculate least-cost paths using the transition matrix approach from _gdistance_. If you want to use these functions, check older releases on enerscape (; until 0.1.3, included). Currently, there are no implementations of Circuitscape/Omniscape in R. There are, however, two great Julia libraries that are extremely easy to use: - Circuitscape.jl: . - Omniscape.jl: Both are one-liners in Julia, e.g. `run_omniscape("omniscape.ini")`. The major obstacle in using these libraries is to create the initialization file (_.ini_). ## Crcuitscape To create the initialization file for Circuitscape, enerscape has the dedicated function _circuitscape_skeleton()_: ```{r circuitscape, eval=FALSE} # two random points p <- spatSample(en, 2, xy = TRUE)[, c("x", "y")] circuitscape_skeleton(en, path = tempdir(), points = p) ``` Circuitscape requires two locations (_points_) to calculate the connectivity among them. A _circuitscape.ini_ file will be write to the _path_ directory. From within this directory, start Julia and run: ```{r compute, eval=FALSE} julia> using Circuitscape julia> compute("circuitscape.ini") [ Info: 2022-12-16 10:05:18 : Precision used: Double [ Info: 2022-12-16 10:05:18 : Reading maps [ Info: 2022-12-16 10:05:18 : Resistance/Conductance map has 231472 nodes [ Info: 2022-12-16 10:05:21 : Solver used: AMG accelerated by CG [ Info: 2022-12-16 10:05:22 : Graph has 231472 nodes, 2 focal points and 1 connected components [ Info: 2022-12-16 10:05:22 : Total number of pair solves = 1 [ Info: 2022-12-16 10:05:24 : Time taken to construct preconditioner = 2.510170134 seconds [ Info: 2022-12-16 10:05:24 : Time taken to construct local nodemap = 0.021110616 seconds [ Info: 2022-12-16 10:05:27 : Solving pair 1 of 1 [ Info: 2022-12-16 10:05:27 : Time taken to solve linear system = 0.512068804 seconds [ Info: 2022-12-16 10:05:28 : Time taken to calculate current maps = 0.82549585 seconds [ Info: 2022-12-16 10:05:29 : Time taken to complete job = 11.5802219 3×3 Matrix{Float64}: 0.0 1.0 2.0 1.0 0.0 0.559379 2.0 0.559379 0.0 ``` Output is in the same folder as the _circuitscape.ini_ file. ## Omniscape Omniscape basically runs omniscape between all location within a window radius, repeating this step for all raster cells. Omniscape does not require input points, but it needs the radius of the moving window: ```{r omniscape, eval=FALSE} omniscape_skeleton(en, path = tempdir(), radius = 10) ``` A _omniscape.ini_ file will be write to the _path_ directory. From within this directory, start Julia and run: ```{r run, eval=FALSE} julia> using Oircuitscape julia> run_omniscape("circuitscape.ini") [ Info: Starting up Omniscape with 1 workers and double precision [ Info: Using Circuitscape with the cg+amg solver... [ Info: Solving moving window targets... days,Progress: 100%| | Time: 0:08:08 [ Info: Time taken to complete job: 490.2733 seconds [ Info: Outputs written to /tmp/RtmpXZb2Hv//tmp/RtmpXZb2Hv/omniscape (Union{Missing, Float64}[missing missing … missing missing; missing 3.4477693535486376 … 4.462051188132685 missing; … ; missing 10.365746449782868 … 4.186837622531591 missing; missing missing … missing missing], Union{Missing, Float64}[missing missing … missing missing; missing 0.652059842432581 … 0.743080786816455 missing; … ; missing 0.7753872880414857 … 0.7333620170587554 missing; missing missing … missing missing]) ``` Output is in the folder _omniscape_ within the folder specified in _path_. # Changelog ## 1.1.0 This follows a correction on the Pontzer model: https://doi.org/10.1098/rsbl.2023.0492. ## 1.0.0 This follows the transfer of functionalists of old GIS packages to _terra_, with _rgdal_, _rgeos_ scheduled to be retired in 2023: . I removed dependencies on _gDistance_, _raster_, _sp_, _rgeos_, and _rgdal_. Energy landscapes are now calculated using a zonal (kernel) based method, implemented in C++ functions. This is faster that previous versions, but it does not return the transition matrix or the conductance matrix. Because of this, least cost paths functions are not supported any more. Instead, use circuitscape or omniscape; see *circuitscape_skeleton()* and *omniscape_skeleton()* to generate the initialization files to run in Julia. The cyclist model is also not supported any more; custom models must be written in the C++ functions. I plan to add a customizable function soon to do that. # References Pontzer, H. (2016). A unified theory for the energy cost of legged locomotion. Biology letters, 12(2), 20150935. Berti, E., Davoli, M., Buitenwerf, R., Dyer, A., Hansen, O. L., Hirt, M., ... & Vollrath, F. (2022). The r package enerscape: A general energy landscape framework for terrestrial movement ecology. Methods in Ecology and Evolution, 13(1), 60-67.