The goal of SimPoly is to simulate a EU-PoMS like datasets with known parameters and appropriate built in stochasticity. The only goal is to explore how different types of variability affect model performance under strong simplified assumptions and should be interpreted as such.
DISCLAIMER: This is work in progress and the assumptions used to build the simulations can be enhanced.
# install.packages("devtools")
devtools::install_github("ibartomeus/SimPoly")library(SimPoly)
set.seed(32468)This is a basic example to show how we build one dataset:
First we define the number of species, sites and years to simulate. We
do this using the define_sites_years() functions. The pool argument
is a call to sp_pool(), which simply defines the number of species
(here pool = 100), and the occupancy of the rarest and commonest
species.
site_years <- define_sites_years_rich(pool = sp_pool(pool = 1000,
mean_occ = 0.25),
n_years = 7, n_sites = 100)Let’s take a closer look at the data.
str(site_years)
head(site_years)
length(unique(site_years$siteID))
length(unique(site_years$species))Note that the number of species in the dataset is lower than the species pool. This reflects stochasticity in how the species are assigned to sites.
Second, we specify species attributes such as phenology, abundance and detectability
pars <- sp_responses(site_years = site_years,
pheno_peak_mean = 120, pheno_peak_sd = 50,
pheno_range_mean = 25, pheno_range_sd = 5,
trend_max = 1, trend_min = 0.8)
str(pars)
hist(pars$h)
head(pars)There is one row per species, with the following columns:
opt = day since 1 january of maximum abundance.
tol = spread -> range. h = expected species abundance along a
dominant-rare log normal distribution.
slope = trend in abundance per year on the abundance scale.
detect = detectability in probability of a species being detected in a
transect (independent of its abundance).
detect_pan = detectability in probability of a specimen falling in a
pantrap.
Third, we sample the true abundance values expected at each sampling point.
dat <- true_abundance(n_round = 8,
site_years = site_years,
sp_responses = pars, white_noiseCV = 0.5)
str(dat)
head(dat)There as many observations per species per site.
year: an integer from 1-3 (as defined by n_years, above). siteID:
which site is it. round: month of the year of this survey. species:
which species.
abundance: abundance during that visit.
From this table we can calculate species richness per site
library(reshape2)
(rich <- dcast(dat, siteID ~ "richness", value.var = 'species', function(x) length(unique(x))))
#and persistance (% of years where the species is present)
hist((persistence <- dcast(dat, siteID + species ~ "persistence", value.var = 'abundance', function(x) length(which(x == 0))/length(x)))$persistence, las = 1)
#average persistence over rounds, not over years.
yearly <- dcast(dat, year + siteID + species ~ "abundance", value.var = 'abundance', sum)
hist(yearly$abundance) #up to 400 indiv per year seems realistic
hist((persistence <- dcast(yearly, siteID + species ~ "persistence", value.var = 'abundance', function(x) length(which(x == 0))/length(x)))$persistence, las = 1)Finally, we sample with detection error from the true values.
dat_obs <- obs_abundance(true_abundance = dat, sp_responses = pars, fraction_observed = 0.1)
head(dat_obs)
plot(dat_obs$obs, dat_obs$abundance) #nice expected correlation, but more noisy.
plot(dat_obs$obs, dat_obs$presences_pan) #nice expected correlation, but more noisy.We can simulate a second transect if desired:
dat_obs$obs2 <- obs_abundance(true_abundance = dat, sp_responses = pars, fraction_observed = 0.1)$obs #note order is preserved
head(dat_obs)
plot(dat_obs$obs, dat_obs$obs2) #nice expected correlation.This is the final output:
head(dat_obs)plot(dat_obs$abundance ~ dat_obs$obs, las = 1, xlim = c(0,max(dat_obs$abundance)),
ylim = c(0,max(dat_obs$abundance)))
#you can summarize observed variables per species, site and year
s_dat <- summary_poms(dat_obs, var_name = "obs")
head(s_dat)
scatter.smooth(s_dat$obs_abund ~ s_dat$year, las = 1)
#This can be easily loop to obtain several simulationsIf using this package, please cite it:
citation("SimPoly")
To cite SimPoly in publications use:
Bartomeus I, Isaac N and Schweiger O. 2023. SimPoly: Generating
EU-PoMS like datasets https://github.com/ibartomeus/SimPoly
A BibTeX entry for LaTeX users is
@Manual{,
title = {SimPoly: Generating EU-PoMS like datasets},
author = {Ignasi Bartomeus and Nick Isaac and Oliver Schweiger},
year = {2023},
url = {https://github.com/ibartomeus/SimPoly},
}STING, JRC and specially Nick Isaac and Oliver Schweiger and the coenocliner package.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent