In richardjtelford/bio250: What the Package Does (One Line, Title Case)

knitr::opts_chunk$set(echo = TRUE, eval = FALSE, message = FALSE)

Setup

Transfer functions for palaeoenvironmental reconstructions can be fitted with the rioja or analogue packages in R. Other packages such as vegan and palaeoSig can help interpretation.

library("rioja")
library("palaeoSig")
library("tidyverse")
library("ggpalaeo")

Packages rioja and analogue have functions with the same names - don't load both at the same time.

set.seed(42)#ensures results are same every time

Data

For convenience, we are going to use some of the datasets available within the rioja package.

#NB any text after a '#' is a comment and ignored by R 
#get help on any function with '?function_name' 
data("SWAP")#SWAP calibration set
data("RLGH")#fossil data from the Round Loch of Glenhead

This has imported the data in lists. We can break these up into their component parts to simplify the code later.

# '<-' assigns the result on the right to the name on the left
# '$' extracts a named component of a list
spp <- SWAP$spec
pH <- SWAP$pH
spNames <- SWAP$names

fos <- RLGH$spec
ageDepth <- RLGH$depths

Species-environement relationships

Plot calibration set species abundances against pH. Add a smoother to highlight the main trends.

ggplot(spp/100, aes(x = pH, y = TA003A)) + 
  geom_point() +
  geom_smooth(method = "gam", method.args = list(family = "quasibinomial"))

The relationship between all species and the environment can be seen with an ordination.

ord <- cca(sqrt(spp) ~ pH)
ord
autoplot(ord)
#learn more about ordinations in Bio303

Fit a transfer function

mod <- WA(y = sqrt(spp), x = pH)
#crossvalidate
mod <- crossval(mod)
performance(mod)
autoplot(mod)

mod <- WA(y = sqrt(spp), x = pH)
#crossvalidate
mod <- crossval(mod)

performance(mod)

Question: do any of the other transfer function methods perform better - try WAPLS, MAT, MLRC? Look only at the cross-validated performance.

By default crossval uses leave-one-out cross-validation. Bootstrap cross-validation is also useful as it gives sample specific errors. Other forms of cross-validation are appropriate if there is pseudoreplication or autocorrelation.

Plot a stratigraphic plot of the fossil data

strat.plot(d = fos, yvar = ageDepth$Age, scale.percent = TRUE, y.rev = TRUE)

Apply the transfer function

recon <- predict(mod, sqrt(fos))
tibble(age = ageDepth$Age, pH = recon$fit[, "WA.inv"]) %>% 
  ggplot(aes(x = age, y = pH)) + 
  geom_line()

Diagnostics & checks

Several diagnostics available - goodness of fit, distance to closest analogue - in analogue package.

Timetrack of fossil samples in calibration set ordination

#use 'analogue::' to access functions without loading library - prevents name conflicts with rioja
tt <- analogue::timetrack(X = sqrt(spp), passive = sqrt(fos), env = pH) 
plot(tt)

Residual length to constraining ordination axis

rlen <- analogue::residLen(X = sqrt(spp), env = pH, passive = sqrt(fos))

plot(rlen)

autoplot(rlen, df = ageDepth, x_axis = "Age") +
labs(x = "Age", y = "Squared residual distance", fill = "Goodness of fit")

Colours show how good the fit is.

Analogue distances

AD <- analogue_distances(spp, fos)
autoplot(AD, df = ageDepth, x_axis = "Age") +
 labs(y = "Squared-chord distance", x = "Age")

Significance tests in palaeoSig.

tst1 <- randomTF(spp = sqrt(spp), env = pH, fos = sqrt(fos), n = 99, col = 1, fun = WA)
tst1$sig
autoplot(tst1)

Further reading

Juggins and Birks (2012) Quantitative Environmental Reconstructions from Biological Data. In Tracking Environmental Change Using Lake Sediments, Data Handling and Numerical Techniques https://www.springer.com/gp/book/9789400727441

richardjtelford/bio250 documentation built on Oct. 12, 2022, 2:16 a.m.