ElevDistr

Getting started

The package ElevDistr can be installed via CRAN or from GitHub using the package devtools.
Stable version from CRAN:

install.packages("ElevDistr")
library("ElevDistr")

Developmental from GitHub:

install.packages("devtools", repos = "http://cran.us.r-project.org")
devtools::install_github("LivioBaetscher/ElevDistr")
library("ElevDistr")

Process a single data point

Computing the distance to the treeline for a single point is the easiest task. Except for climatic layers and the elevation model, ElevDistr contains everything needed for this task. First of all, it is essential that two climatic raster are imported: one for growing season length (gsl) and one for growing season temperature (gst). I suggest to use the layers from CHELSA. The user is free to use different climatic layers, but it is highly recommended to customize the object pointsAboveTreeline when using layers other than the CHELSA raster (see below). The algorithm does not fail if this adjustment is not made, but it is very likely to have a negative impact on computation time.

gstURL <- paste0("https://os.zhdk.cloud.switch.ch/chelsav2/GLOBAL/",
                 "climatologies/1981-2010/bio/CHELSA_gst_1981-2010_V.2.1.tif")
gslURL <- paste0("https://os.zhdk.cloud.switch.ch/chelsav2/GLOBAL/",
                 "climatologies/1981-2010/bio/CHELSA_gsl_1981-2010_V.2.1.tif")

gst <- terra::rast(gstURL, vsi = TRUE)
gsl <- terra::rast(gslURL, vsi = TRUE)

Furthermore, a digital elevation model (DEM) needs to be imported. I selected a part of the GMTED2010 model, which is provided by the U.S. Geological Survey. Importing the layer works the same way as with the climatic raster above.

gmted2010URL <- paste0("https://edcintl.cr.usgs.gov/downloads/sciweb1/shared/topo/downloads/GMTED/",
                      "Global_tiles_GMTED/300darcsec/med/E000/30N000E_20101117_gmted_med300.tif")

gmted2010Part <- terra::rast(gmted2010URL, vsi = TRUE)

The only thing that needs to be done now is to call the wrapper and specify the longitude, latitude, and elevation of the point of interest.
Note: All coordinates need to be in World Geodetic System 1984 (‘WGS-84’).

distance_to_treeline(lon = 8.65, lat = 46.87,  gstRaster = gst,  gslRaster = gsl, 
                     elevationRaster = gmted2010Part, elevation = 504, pointDf = pointsAboveTreeline, 
                     plot = FALSE, plotHist = FALSE, gstMin = 6.4, gslMin = 94)

If using only a few points, the user should plot a map and a histogram of the local treeline height. This can be achieved by changing the default parameters: plot = TRUE for the map and plotHist = TRUE for the histogram.
Please note that you need to register a Google API in order to produce a map. A detailed explanation can be found here: Register a Google API. You can check if you have an already registered key with ggmap::has_google_key(), and use the command ggmap::ggmap_show_api_key() to allow the wrapper to download the map of interest.

Example of a sampled area
Example of a sampled area
Example of treeline distribution
Example of treeline distribution

Furthermore, although it is possible to change the treeline definition by adjusting the thresholds for minimal growing season length and growing season temperature, this option should only be modified if you know what you are doing. Changed thresholds significantly impact the computational output and are carefully selected based on Paulsen and Körner (2014) https://doi.org/10.1007/s00035-014-0124-0.

Process GBIF data

After explaining the basics of the wrapper function distance_to_treeline, I will now show an example of how to calculate entire species distribution based on GBIF data. This chapter gives the user an overview of how this R package could be used to process bigger data frames.
Remember that just because this tool is specially designed to handle the uncertainty of spatial data, it does not mean that this approach is a one-size-fits-all solution. Depending on your question you might want to change the approach.

Download data

First, we need to access GBIF data for processing. As an example, I will download records of two species from GBIF: Ranunculus pygmaeus and Ranunculus thora. The data can be imported directly from a CSV file or using the package rgbif. Here the fully automatically approach with rgbif is demonstrated. However, if you are looking for many data points, you may hit the hard limit of 100,000 occurrences. In such a case downloading the csv file from the GBIF homepage becomes necessary.

#install.packages("rgbif") #Remove hashtag if you have not installed this package
myspecies <- c("Ranunculus pygmaeus", "Ranunculus thora")
gbifData <- rgbif::occ_data(scientificName = myspecies, hasCoordinate = TRUE, limit = 20000)

To keep the example simple, I will only store the variables of interest: species name, longitude, latitude, elevation and the taxon rank.

temp1 <- gbifData$`Ranunculus pygmaeus`$data[, c("scientificName", "decimalLatitude", "decimalLongitude",
                                                 "elevation", "taxonRank")]
temp2 <- gbifData$`Ranunculus thora`$data[, c("scientificName", "decimalLatitude", "decimalLongitude",
                                              "elevation", "taxonRank")]

ranunculus <- rbind(temp1, temp2)

Filter data

Now we are ready to process the downloaded data frame. Here it is only presented how to exclude species with no meaningful elevation. However, depending on your data and your question this very basic filtering might not be sufficient. I recommend to check out the package CoordinateCleaner and think properly of what kind of potential bias you must exclude from your data set.
Here I exclude all data points that are below 0 meters or above 8,850 meters above the sea level or contain a NA. This is achieved using the Tidyverse package.

#install.packages("tidyverse") #Remove hashtag if you have not installed this package
library("tidyverse")
ranunculusFiltered <- ranunculus %>% filter(!is.na(elevation) & 0 < elevation & elevation < 8850)

Furthermore, I will only keep entries that have an assigned taxon of “species”.

ranunculusFiltered <- ranunculusFiltered %>% filter(taxonRank == "SPECIES")

Down sampling

To speed up the computation, only 100 random samples are retained. The user might choose a different approach depending on the performed filtering and the desired output accuracy.

set.seed(42) #Set a seed for reproducible
ranunculusSampled <- ranunculusFiltered %>% group_by(scientificName) %>% slice_head(n = 100)

Compute distance to the treeline

Now the distance to the closest local treeline can be computed. The computation works similarly to what was explained above with a single data point, with the difference that vectors for longitude, latitude, and elevation are used as input.

#Import climatic layers
gstURL <- paste0("https://os.zhdk.cloud.switch.ch/chelsav2/GLOBAL/",
                 "climatologies/1981-2010/bio/CHELSA_gst_1981-2010_V.2.1.tif")
gslURL <- paste0("https://os.zhdk.cloud.switch.ch/chelsav2/GLOBAL/",
                 "climatologies/1981-2010/bio/CHELSA_gsl_1981-2010_V.2.1.tif")

gst <- terra::rast(gstURL, vsi = TRUE)
gsl <- terra::rast(gslURL, vsi = TRUE)

#Import the DEM
gmted2010URL2 <- paste0("https://edcintl.cr.usgs.gov/downloads/sciweb1/shared/topo/downloads/GMTED/",
                      "Global_tiles_GMTED/300darcsec/med/E000/50N000E_20101117_gmted_med300.tif")

gmted2010Part2 <- terra::rast(gmted2010URL2, vsi = TRUE)

#Run classification for the first five elements
elev <- distance_to_treeline(lon = ranunculusSampled$decimalLongitude[1:5], 
                             lat = ranunculusSampled$decimalLatitude[1:5], gstRaster = gst,
                             gslRaster = gsl, elevationRaster = gmted2010Part2, pointDf = pointsAboveTreeline,
                             elevation = ranunculusSampled$elevation[1:5], plot = FALSE, 
                             plotHist = FALSE, gstMin = 6.4, gslMin = 94)
#>   |                                                                    |                                                            |   0%
#> Warning in distance_to_treeline(lon = ranunculusSampled$decimalLongitude[1:5],
#> : The "gridSize" parameter was to small. The parameter was automatically
#> increased by 15 km.
#>   |                                                                    |============                                                |  20%
#> Warning in distance_to_treeline(lon = ranunculusSampled$decimalLongitude[1:5],
#> : The "gridSize" was increased 5 times, to avoid an endlose loop we stop here.
#> Consider to exclude this point it might not be representative.
#> Warning in distance_to_treeline(lon = ranunculusSampled$decimalLongitude[1:5],
#> : The "gridSize" parameter was to small. The parameter was automatically
#> increased by 30 km.
#> Warning in distance_to_treeline(lon = ranunculusSampled$decimalLongitude[1:5],
#> : No treeline was found within the defined gridd. Therfore, the result is "NA".
#> Consider retying with a larger gird size.
#>   |                                                                    |========================                                    |  40%  |                                                                    |====================================                        |  60%  |                                                                    |================================================            |  80%
#> Warning in distance_to_treeline(lon = ranunculusSampled$decimalLongitude[1:5],
#> : The "gridSize" parameter was to small. The parameter was automatically
#> increased by 10 km.
#>   |                                                                    |============================================================| 100%

Be aware that the computation time for each point is roughly three seconds, the time increases linear with the amount of input points; \(O(n)\). Filtering and parallelization are suitable methods for reducing the computation time.
The result stored in elev are the computed distances to the closest local treeline.

elev
#> [1] 705  NA 250 172 457

The function distance_to_treeline returns an NA if the required number of samples (specified by the input value treelineSamplingSize; default value of 10) is not reached, even with an increased search radius, or if the DEM does not cover the area of interest (as is the case in elev[1]).

Increase the speed

The way the three raster are loaded in this tutorial, is not optimized for speed. If you would like to increase the speed it is recommended to download the layers from CHELSA:

and the U.S. Geological Survey:

If a digital elevation model of the whole globe is needed the GTOPO30.tif file can be downloaded from my Google Drive. GTOPO30 is also courtesy of the U.S. Geological Survey. All the raster can be imported by customizing:

temp <- terra::rast("your_raster_layer.tif")

How to customize “pointsAboveTreeline”

As stated in the first paragraph, users can use different climate raster than those presented in the tutorial. However, to keep computation time low, it is recommended to recalculate the pointsAboveTreeline data frame.
After importing the climate raster of your choice, you can use the generate_point_df function to identify all points above the treeline.

#Import raster of your choice
gst <- terra::rast(file = "your_gst_layer.tif")
gsl <- terra::rast(file = "your_gsl_layer.tif")

#Compute new data frame
newPointsAboveTreeline <- generate_point_df(gstRaster = gst, gslRaster = gsl, stepSize = 0.0416666, 
                                            gstTreshold = 6.4, gslTreshold = 94)

#Save the output
save(newPointsAboveTreeline, file = "newPointsAboveTreeline.Rdata")

The computation takes some time, but it only needs to be executed once.

It is probably relevant to exclude all the points that are that are outside of the mountain polygons GMBA. This was done for the pointsAboveTreeline data frame. To do so, download the folder from the link above and follow these steps:

#Load the polygons of all the mountains (from the GMBA project)
mountain_polygons <- terra::vect("GMBA mountain inventory V1.2(entire world)/
                                 GMBA Mountain Inventory_v1.2-World.shp")

#Keep only the points that are in a alpine polygon
#"keep" will be a vector containing all row numbers that contain coordinate, which lie in the Alps
keep <- terra::is.related(terra::vect(newPointsAboveTreeline, geom = c("longitude", "latitude")),
                          mountain_polygons, "intersects") |> which() 

newPointsAboveTreeline2 <- newPointsAboveTreeline [keep,] #Pick the lines of interest

How the wrapper works

In case users are interested in how the algorithm works, this paragraph explains what happens behind the scenes. For simplicity, it is explained using the same data as in the first paragraph.
First, the wrapper searches for the closest point above the treeline in the pointsAboveTreeline data frame (or what ever the user feeds into the wrapper). For the search, a k-dimensional tree and a nearest neighbour approach are used by calling the get_nearest_point function from the package RANN. It is important to find a point close to the treeline because both alpine and non-alpine points are needed in the following steps.

pointAbove <- get_nearest_point(lon = 8.65, lat = 46.87, pointDf = pointsAboveTreeline)
pointAbove
#> $lon
#> [1] 8.728898
#> 
#> $lat
#> [1] 46.93756

After getting a point that lies above the treeline, a grid is generated with the extracted point as grid center. The width (in km) and the distance between the points (in degrees) are defined by the input arguments.

grid <- generate_grid(lon = pointAbove$lon, lat = pointAbove$lat, squareSize = 10, stepSize = 0.0025)
head(grid$df)
#>   longitude latitude
#> 1  8.663898 46.89251
#> 2  8.663898 46.89501
#> 3  8.663898 46.89751
#> 4  8.663898 46.90001
#> 5  8.663898 46.90251
#> 6  8.663898 46.90501

Now for each grid point the growing season length and growing season temperature are extracted. Based on the selected input thresholds, it is determined whether the point is above or below the treeline. The extracted information is added to the data frame from the generate_grid function output.

grid$df <- classify_above_treeline(coords = grid$df, gstRaster = gst, gslRaster = gsl,
                                   gstTreshold = 6.4, gslTreshold = 94)
head(grid$df)
#>   longitude latitude growingSeasonTemperature growingSeasonLength aboveTreeline
#> 1  8.663898 46.89251                     8.95                 133         FALSE
#> 2  8.663898 46.89501                     8.95                 133         FALSE
#> 3  8.663898 46.89751                     8.95                 133         FALSE
#> 4  8.663898 46.90001                     8.95                 135         FALSE
#> 5  8.663898 46.90251                     8.95                 135         FALSE
#> 6  8.663898 46.90501                     8.95                 135         FALSE

Based on this classification, the treeline is drawn, if two points with a different classification lie next to each other, a vertical or horizontal line is drawn between the points and NA values are ignored. The lines are stored in a data frame with the following information: an identifier, a start latitude and longitude, and an end latitude and longitude.

treelineDf <- sample_treeline(df = grid$df, lonLength = grid$lonLength, latLength = grid$latLength)
head(treelineDf)
#>              id     lat1     lon1     lat2     lon2
#> 2 horizontal195 46.91626 8.675148 46.91626 8.677648
#> 3 horizontal198 46.92376 8.675148 46.92376 8.677648
#> 4 horizontal232 46.91626 8.677648 46.91626 8.680148
#> 5 horizontal235 46.92376 8.677648 46.92376 8.680148
#> 6 horizontal269 46.91626 8.680148 46.91626 8.682648
#> 7 horizontal272 46.92376 8.680148 46.92376 8.682648

After this step, the wrapper contains a safety feature: whenever there are fewer than ten treeline elements, the last three functions are recalculated with a bigger grid (+5 km). It is implemented in case only a few trellises are found, which indicates potential problems and might increase uncertainty.
After this safety loop the plot argument is evaluated. If it is = TRUE a map is generated by calling the function plot_distr.

plot_distr(nearestCorner = pointAbove, grid = grid$df, treelineDf = treelineDf, size = 12)

In the final step, the treeline is sampled with a step size defined in the input argument. From all the drawn points, the elevation is extracted from the elevationRaster (containing a digital elevation model). Of all extracted elevations, the median is calculated and subtracted from the pointElevation input to estimate the relative distance to the local treeline. This function is capable of plotting a histogram of the extracted elevation points, including the median.

calculate_distance(treeline = treelineDf, elevationRaster = gmted2010Part, pointElevation = 504,
                   treelineSampling = 10, plot = FALSE)
#> [1] -1491

If desired, all the functions can be called independently of the wrapper.