This readme provides a cursory overview of the functionality of spsys. Refer to the manual for deeper descriptions of package functionality and use.
Many environmental surveys use systematic sampling to produce estimates of population parameters. Estimating the precision of estimates from systematic sampling designs has proven a difficult task, with several systematic variance estimators proposed over the past several decades. While not exhaustive, spsys implements several different variance estimators and provides diagnostic and simulation tools to allow analysts to select an appropriate variance estimator for their population. More specifically, spsys provides variance estimation for surveys that rely on point estimates of attributes of interest that use the Horvitz-Thompson estimator in two-dimensional settings.
spsys derives its classes from the popular sp package, thus all sp-related functions such as sp::plot, rgdal::writeOGR, etc. work with no further modifications.
Install this package directly from GitHub using devtools
library(devtools)
devtools::install_github('https://github.com/brycefrank/sys)
spsys operates on a modified version of the now ubiquitious sp package SpatialPointsDataFrame class. The entry point into sys are the HexFrame and RectFrame classes, which represent hexagonal systematic and rectangular systematic sampling configurations respectively.
For example, we can load in a set of points from a hexagonal grid:
hex_points <- readOGR('my_hex_points.shp')
hex_frame <- HexFrame(hex_points, c('vol', 'ba'), N=10000)
HexFrame takes two arguments: a SpatialPointsDataFrame and a vector of column names that indicate attributes we are interested in conducting the analyses on. Here we indicate volume (vol) and basal area (ba) as our attributes of interest, and we indicate that the population size is N=10000. HexFrame, and its sister class RectFrame (for rectangular systematic samples), implement a standard interface upon which variance estimators can be constructed.
Once we have wrapped our sample information within a HexFrame or RectFrame, the next step is to construct a variance estimator. For example, we can construct a variance estimator that assumes simple random sampling was conducted, as is common practice for many environmental sample surveys.
my_srs_estimator <- VarSRS(fpc=TRUE)
my_srs_estimator now represents a function that, when we pass this function a HexFrame or RectFrame, will return the variance estimate.
my_srs_estimator(hex_frame)
spsys allows for the assessment the behavior of variance estimators via simulation. This is done by treating an existing SysFrame as a population, and sampling repeatedly from it. Consider the following example where we treat hex_frame as a population
a <- 3
subsample(hex_frame, c(1,1), a)
where c(1,1) is the starting position in index space and a=3 is the sampling interval. Fans of the dplyr package may want to use pipe operators. That is also possible
hex_frame %>% subsample(c(1,1), a)
In most assessments we will be interested in all possible subsamples. Here we iterate over all possible subsamples and compute the simple random sampling with replacement estimator
all_starts <- subsample_starts(a)
for(i in 1:nrow(all_starts)) {
subsample(hex_frame, all_starts[i,]) %>%
my_srs_estimator()
}
In total, seven unique estimators are implemented, each of which contains the ability for further configuration. The particular formal definitions of these estimators are described in the manual and the publication [].
D'Orazio (2003) - var_dorazio_i, var_dorazio_c
Mátern (1980)* - var_mat
Non-Overlapping Neighborhoods - var_non_overlap
Stevens & Olsen (2003) - var_so
Simple Random Sampling - var_srs
This package is currently in a minimally working condition. Interested collaborators can email the author at [email protected] to contribute to further updates.