Skip to content

Latest commit

 

History

History
624 lines (477 loc) · 26.5 KB

README.md

File metadata and controls

624 lines (477 loc) · 26.5 KB
Raw

sdmLit 📚

Author: João Frederico Berner

R-CMD-check

Have any suggestions? Contact me on Twitter :)

Outline

Introduction

The goal of sdmLit is to allow easy implementation of methods not implemented in the sdm package, using the standard outputs of the sdm:: functions. The methods are all described in the literature, and are usually not compiled in any single package, which leads to very long coding and implementing sessions. The author felt ENM analyses and mapping currently sparsed in the literature could be centralized in a single package.

There are four currently working “branches/tools” in the package. Namely,

  1. Boyce Index (Hirzel, Le Lay, Helfer, Randin, & Guisan, 2006) , with the function sdm_to_boyceIndex();
  2. Consistency/Congruence Maps (Morales-Barbero & Vega-Álvarez, 2019) with the function sdm_to_consistency(). Original source code here. I intend to allow new congruence maps to be created from a set of environmental predictors with a new function in the near future, but for now only the original congruence map is available, which was built using WorldClim v.2, MERRAclim (C. Vega, Pertierra, & Olalla-Tárraga, 2017), CHELSA, ocean databases MARSPEC and Bio-ORACLE v. 2.0. See Morales-Barbero & Vega-Álvarez (2019) for details.
  3. Accumulation of Occurrences Curve (Jiménez & Soberón, 2020) with the functions sdm_to_occ.pnts() and sdm_to_output.mods(), to be passed onto the original functions accum.occ(), and later comp.accplot(), originally found at: DOI.
  4. Frequency Ensemble (Sobral-Souza, Francini, & Lima-Ribeiro, 2015) with functions sdm_to_freqEnsemble() and frequency_ensemble_plot(). The approach is the same as in the paper, but the code I got from a class I had with one of the authors.

Below, I do my best to explain when/why use each of these and give examples of usage with the standard outputs from the sdm package, mainly from sdm() and predict().

Installation

You can install the development version of sdmLit from GitHub with:

# install.packages("devtools")
devtools::install_github("jfberner/sdmLit")

Examples

Base Objects

Disclaimer

  • The following are the basic objects that are going to be used in all the examples. They are not meant to be fully reproducible, as species data, variable and algorithm selection vary considerably and should be picked according to each dataset and objectives. Instead, I intend to briefly take the reader through the process of creating an ENM with the sdm package solely for clarifying what each object is (or is supposed to be).

  • Details on the nature and origin of the datasets should not be important for the use of this package’s functions.

  • Mind that some lines of the code are a MUST if one’s to implement my functions, and they are described as such.

The ‘species’ I’m using here is going to be called ‘charinus’. It is not a species, but is a model from another project I have easy at hand. I will not be sharing the dataset. I have two datasets: one called occ_train (with model training data) and another called occ_test (with model test data, which includes absence records). These objects are SpatialPointsDataFrame(s).

# The following are shapefiles (.shp) being loaded:
occ_train <- rgdal::readOGR('data/processed/shapefiles/train.shp')
occ_test <- rgdal::readOGR('data/processed/shapefiles/test.shp')
# As SpatialPointsDataFrame
class(occ_train)

Climate Data (predictors) are four uncorrelated bioclimatic variables from BioClim v2.1. In the code they’ll be called envpres, refferring to ‘present’ climate (average from 1970-2000, more on this on worldclim’s website linked above).

envpres <- dir(path = 'data/processed/envcropped/Present/', pattern = ".tif$", full.names = T)

envpres <- raster::stack(envpres)
class(envpres)

I’ll be using four algorithms in my all of the examples: SVM, MaxEnt, BioClim and Domain. References for those can be found in e.g. Sobral-Souza, Francini, et al. (2015), I will not go into details here.

# Saving algorithm list can make life easier, for at least a couple reasons: 
# it retains the order in which they are modeled inside the sdm::sdm() function to be used in sdmLit functions, 
# and it saves you from re-typing them everytime.

algorithms <- c('svm', 'maxent', 'bioclim', 'domain') # order will be important

The Model will (naturally) be constructed using the sdm package. Details on sdm package usage can be found here in video format and here is the github of the package. I won’t go into further, unnecessary details.

First, we prepare the data:

d_occ <- sdm::sdmData(formula = charinus~., train=occ_train,
                 predictors = envpres, bg = 18, # same number as occurrences
                 method = 'eRandom') # Create sdmData object
# eRandom = random in Environmental Space

Then we run the models:

m_occ <- sdm::sdm(formula = charinus~.,data = d_occ,
              methods = algorithms,
                           replication = c('bootstrapping'), n=5)

This m_occ object is our sdmModel object, and it’ll be often used.

Finally, we predict using built models:

p_occ <- sdm::predict(m_occ, newdata = envpres, # new data is the environment in which we want to predict
                 filename = 'data/processed/model-build/predictions/predictions.present-rasterStd.tif',
                 prj = T, overwrite = T, nc = 7)

# This creates a file in the specified folder BUT the file doesn't retain the layer names, which will be super important for us to distinguish among methods. To do so:

This will save the predictions as a raster file. However, layer names WILL NOT BE RETAINED. Not only is retaining layer names extremely important to implement the functions of sdmLit package, it will make it easier for the user/researcher to distinguish them later. To retain them, simply:

# The next line grabs the 'fullname' inside the p_occ object
names(p_occ) <- p_occ@z$fullname # This is SUPER important. It ensure both SPECIES and ALGORITHM names are retained in layer names

p_occ_stack <- raster::stack(p_occ) #convert rasterBrick to rasterStack
p_occ_spatraster <- terra::rast(p_occ_stack) #convert rasterStack to SpatRaster
names(p_occ_spatraster) <- names(p_occ_stack) #grab the names again, because in converting the terra::rast() function grabs the names from the .tif file that p_occ refers to.

terra::writeRaster(x = p_occ_spatraster, 
                    filename = 'data/processed/model-build/predictions/predictions.present-terraStd.tif',
                    overwrite = TRUE) # Using terra package because it retains layer names

When you read your predictions raster file, make sure to so using the terra package and then converting it to a RasterStack to retain sdm package standards (not required, but I haven’t tested).

pocc <- terra::rast(x = 'data/processed/model-build/predictions/predictions.present-terraStd.tif')
pocc <- raster::stack()

Note: there are now two different prediction objects, and they are significantly different. The object p_occ is the one originally obtained from the sdm::predict() function, and pocc is the one read from the terra package. Since is always best practice to have separate scripts for different steps in modelling I’ll be using pocc for all functions from here on.

And these are all the objects we need to implement the functions of the sdmLit package: occ_train, occ_test, envpres, m_occ and pocc. The following sections will be examples of how to use the functions and a brief description, in my words, of when to use them.

Boyce Index

The Continuous Boyce Index Hirzel et al. (2006) is a threshold-independent evaluator for models. It provides predicted-to-expected ratio curves that offer further insights into the model quality: robustness, habitat suitability resolution and deviation from randomness. This information helps reclassifying predicted maps into meaningful habitat suitability classes.

When absence data are unreliable or unavailable, the model evaluation should be assessed for presences only. Two simple evaluators are the absolute validation index (AVI) and contrast validation index (CVI), but both rely on fixed thresholds. Boyce, Vernier, Nielsen, & Schmiegelow (2002) proposed a way to relieve somewhat the threshold constraint. Their method consists in partitioning the habitat suitability range into b classes (or bins), instead of only two. The main shortcoming of the Boyce index is its sensitivity to the number of suitability classes b and to their boundaries.

To overcome this problem, Hirzel et al. (2006) derived a new evaluator based on a “moving window” of width W (say W = 0.1) instead of fixed classes.

The continuous Boyce index is thus both a complement to usual evaluation of presence/absence models and a reliable measure of presence-only based predictions.

Usage is as follows:

library(sdmLit)
boyce.output <- sdm_to_boyceIndex(preds = pocc, layer = mean(1:5), occurrence = occ_test)
# preds = pocc NOTE that it is NOT p_occ, refer to the text above for the difference.
# layer = mean(1:5) will take the mean of the first five layers, in this case the five SVM models

Which outputs a list with F.ratio, cor, and HS, and plots the CBI which should look something like this:

The sdm_to_boyceIndex relies on ecospat::ecospat.boyce() and simply passes the arguments onto this function. Any additional arguments can be passed onto it and reading its help is strongly suggested (run ?ecospat::ecospat.boyce() in R Console).

Congruence Maps

The lack of bioclimatic congruence (degree of agreement) between different databases is a main concern in distribution modelling and it is critical in single-source models, for which the database choice is decisive. In order to prevent unreliable predictions derived from distorted input data, SDMs accuracy can be assessed by mapping model predictions according to a bioclimatic congruence measure derived from the comparison of multiple databases, which can be achieved with the bioclimatic consistency maps proposed by Morales-Barbero & Vega-Álvarez (2019).

In this package, we provide the congruence map generated by the original authors (image below). If you only need the dataset for plotting or any other reason, you can access it using data(congruence) or sdmLit::congruence, both will load the data to your R environment. This is not needed to plot your results using the method however, as explained below.

Global-scale bioclimatic congruence map to analyse environmental mismatches between recently updated bioclimatic databases WorldClim v.2 (Fick & Hijmans, 2017), MERRAclim (C. Vega et al., 2017), CHELSA (Karger et al., 2017), and ocean databases MARSPEC (Sbrocco & Barber, 2013) and Bio-ORACLE v.2.0 (Tyberghein et al., 2012)

The method consists in plotting your model predictions on top of this congruence map. The colors are coded in two axes, Habitat Suitability and Climatic Congruence, both ranging from zero to one. Note that if your model predictions do not range from 0 to 1, you should resample them to range between these values before running the sdmLit::sdm_to_consistency() function. Function usage is as follows, using the objects presented in the introduction of this document:

# There are only two arguments to this function, the prediction object and the layer you wish to plot onto the consistency map. 
sdmLit::sdm_to_consistency(preds = pocc,
                            layer = 1)
# Here I'm using the first layer, which corresponds to an SVM model that ranges from 0.15 to 0.92 in its predictions.
Congruence Color-Codes for Map Legend Congruence/Suitability Bivariate Map

As you can see, there are two outputs from running this function. The first the congruence color-coded to be used by users as legend or nested-plot. The second is the final bivariate congruence/suitability map.

Regarding this second plot: (a) it is cropped using the limits of your original envpres (and consequently of your pocc object). Also, it is plotted using baseR, just as in the original code (links in the introduction of this document), so you can add points and whatever else you need using this syntax.

There is no need to assign this function to an object, as the output would only be a RasterLayer object, but you are welcome to do so if desired.

I have plans to implement two more functions regarding this method. The one is to allow the user to create their own congruence maps from any set of rasters. The other is to allow any kind of bivariate map using pocc and user-generated congruence maps. No predictions on when this will happen.

Accumulation of Occurrences Curve

The area under the curve (AUC) of the receiving-operating characteristic (or certain modifications of it) is almost universally used to assess the performance of species distribution models (SDMs), despite the well-recognized problems encountered with this approach, mainly present when dealing with presence-only data.

Jiménez & Soberón (2020) present a probabilistic treatment of the presence-only problem and derive a method to assess the performance of SDMs based on the analysis of an area-presence plot and the SDM outputs represented in both geographic and environmental spaces.

This allows assignment of the performance of an algorithm that classifies presence-only data by two factors: (a) the degree of non-randomness of the classification at every step in the accumulation curve of presences, and (b) the amount of uninformative niche space used for the classification.

To implement the original functions accum.occ() and comp.accplot() DOI, we need first to manipulate the sdm output objects to create the inputs required by the functions: output.mods and occ.pnts. To do so, simply use the functions:

# If you need help in the definition of the objects below, please refer back to the Introduction of this document.
occ.pnts <- sdmLit::sdm_to_occ.pnts(env = envpres, occ = occ_train, algorithms = algorithms, predict_object = pocc)

output.mod <- sdmLit::sdm_to_output.mods(env = envpres, algorithms = algorithms, predict_object = pocc, long = 'x', lat = 'y') 

# Longitude is also commonly typed as 'lon', and Latitude as 'lat'.
# To check how yours is called, I suggest running:
# head(as_tibble(rasterToPoints(pocc))[,1:2]) # which will output longitude and latitude respectively.

Now that the data has been prepared, simply apply the original functions.

The first one to be run is the accum.occ function, which returns three plots, and a matrix to the assigned object. This matrix will be used later in the comp.accplot function.

# The accum.occ function is a bit annoying and creates three different plots in three different devices.
# To save them we'll use dev.print() in the reverse order in which they appear. I could automatize
# this part of the script but I don't have time for that, I'm sorry dear reader.

# First, enable device control
dev.new() ; dev.control('enable') ; dev.off()

# WARNING: Before running this next section of the script, make sure nothing
# is plotted and no devices are open with dev.list()

aocc_svm <- sdmLit::accum.occ(sp.name = 'charinus',
                     output.mod = output.mod[[1]], # The index 1 here refers to the algorithm index, if you followed the exact same syntax so far, it should be the same index as in the 'algorithms' object
                     occ.pnts = occ.pnts[[1]], # Same observation as above
                     null.mod = "hypergeom",
                     conlev = 0.05, bios = 0)

# To save the plots, we have to do it in the reverse order as they were plotted:
dev.print(png, file = "figs/aocc/svm_aocc_aocc.png", # Accumulation of Occurrences Plot
          width=8, height=8, units="in", res=300);dev.off()

dev.print(png, file = "figs/aocc/svm_aocc_map.png", # Map Predictions of this Algorithm
          width=8, height=8, units="in", res=300);dev.off()

dev.print(png, file = "figs/aocc/svm_aocc_env.png", # Environmental Space predictions by this algorithm
          width=8, height=8, units="in", res=300);dev.off()

The three plots should look something like this:

Environmental Space Map of Predictions Accumulation of Occurrences

After repeating the chunk above for each algorithm (i.e. each of the items in the occ.pnts object), you have to list them in a single object, as follows:

aocc_list <- list(aocc_svm, aocc_maxent, aocc_bioclim, aocc_domain)

And then, you can compare the performance of each algorithm using the comp.aocc function:

model_comp <- sdmLit::comp.accplot(mods = aocc_list,
                          nocc = length(occ_train),
                          ncells = raster::ncell(envpres),
                          sp.name = 'Charinus',
                          mods.names = c('svm', 'maxent', 'bioclim', 'domain'), alpha = 0.05)

Which should return a plot that looks something like:

For further details in the arguments of the functions above, I strongly suggest reading Jiménez & Soberón (2020) and referring to the links in the beggining of this section and the Introduction.

Ensemble by Frequency

This method of ensembling is reasonably straightforward. A threshold is applied to suitability maps (predictions, here pocc), and then the frequency that each grid cell is predicted as presence is computed to obtain a consensual map from all algorithms. This can also be done for different GCMs, as long as they’re all stacked as layers of the same object. See examples of this approach in Sobral-Souza, Lima-Ribeiro, & Solferini (2015), Da Silveira, Vancine, Jahn, Pizo, & Sobral-Souza (2021) and Sobral-Souza, Francini, et al. (2015).

Thresholds available in this package are the ones computed by the sdm() function, and these values are obtained directly by the sdm::getEvaluation function.

In this package, I provide two functions for frequency ensembling. One computes the frequency map (sdm_to_freqEnsemble()) and the other one plots it in a reproducible style (frequency_ensemble_plot()). Of course, the plotting can be done by other means, as the output of the first function is a RasterStack.

For this particular functions it is extremely important that both the name of the species and the algorithm name are present in each layer name. I explain thoroughly how to make sure this is true in the Base Objects part of this document.

Usage is as follows:

freqE <- sdmLit::sdm_to_freqEnsemble(model = m_occ, preds =  pocc, species =  c('charinus'), threshold = 2)

# Argument threshold: which parameter to use as threshold for binarizing the predictions. The possible value can be between 1 to 10 for "sp=se", "max(se+sp)", "min(cost)", "minROCdist", "max(kappa)", "max(ppv+npv)", "ppv=npv", "max(NMI)", "max(ccr)", "prevalence" criteria, respectively.

Which returns a RasterStack that can be plotted with this package using:

freqPlot <- sdmLit::frequency_ensemble_plot(freqE)

freqPlot

Which should return something like:

The good thing about using this second function is that it returns a ggplot object, and any further tinkering with the map can be done using ggplot syntax.

References

Boyce, M. S., Vernier, P. R., Nielsen, S. E., & Schmiegelow, F. K. A. (2002). Evaluating resource selection functions. Ecological Modelling, 157(2), 281–300. doi: 10.1016/S0304-3800(02)00200-4

C. Vega, G., Pertierra, L. R., & Olalla-Tárraga, M. Á. (2017). MERRAclim, a high-resolution global dataset of remotely sensed bioclimatic variables for ecological modelling. Scientific Data, 4(1), 170078. doi: 10.1038/sdata.2017.78

Da Silveira, N. S., Vancine, M. H., Jahn, A. E., Pizo, M. A., & Sobral-Souza, T. (2021). Future climate change will impact the size and location of breeding and wintering areas of migratory thrushes in South America. Ornithological Applications, 123(2), duab006. doi: 10.1093/ornithapp/duab006

Fick, S. E., & Hijmans, R. J. (2017). WorldClim 2: New 1-km spatial resolution climate surfaces for global land areas. International Journal of Climatology, 37(12), 4302–4315. doi: 10.1002/joc.5086

Hirzel, A. H., Le Lay, G., Helfer, V., Randin, C., & Guisan, A. (2006). Evaluating the ability of habitat suitability models to predict species presences. Ecological Modelling, 199(2), 142–152. doi: 10.1016/j.ecolmodel.2006.05.017

Jiménez, L., & Soberón, J. (2020). Leaving the area under the receiving operating characteristic curve behind: An evaluation method for species distribution modelling applications based on presence-only data. Methods in Ecology and Evolution, 11(12), 1571–1586. doi: 10.1111/2041-210X.13479

Karger, D. N., Conrad, O., Böhner, J., Kawohl, T., Kreft, H., Soria-Auza, R. W., … Kessler, M. (2017). Climatologies at high resolution for the earth’s land surface areas. Scientific Data, 4, 170122. doi: 10.1038/sdata.2017.122

Morales-Barbero, J., & Vega-Álvarez, J. (2019). Input matters matter: Bioclimatic consistency to map more reliable species distribution models. Methods in Ecology and Evolution, 10(2), 212–224. doi: 10.1111/2041-210X.13124

Sbrocco, E. J., & Barber, P. H. (2013). MARSPEC: Ocean climate layers for marine spatial ecology. Ecology, 94(4), 979–979. doi: 10.1890/12-1358.1

Sobral-Souza, T., Francini, R. B., & Lima-Ribeiro, M. S. (2015). Species extinction risk might increase out of reserves: Allowances for conservation of threatened butterfly Actinote quadra (Lepidoptera: Nymphalidae) under global warming. Natureza & Conservação, 13(2), 159–165. doi: 10.1016/j.ncon.2015.11.009

Sobral-Souza, T., Lima-Ribeiro, M. S., & Solferini, V. N. (2015). Biogeography of Neotropical Rainforests: Past connections between Amazon and Atlantic Forest detected by ecological niche modeling. Evolutionary Ecology, 29(5), 643–655. doi: 10.1007/s10682-015-9780-9

Tyberghein, L., Verbruggen, H., Pauly, K., Troupin, C., Mineur, F., & De Clerck, O. (2012). Bio-ORACLE: A global environmental dataset for marine species distribution modelling. Global Ecology and Biogeography, 21(2), 272–281. doi: 10.1111/j.1466-8238.2011.00656.x