rshift

An R library for paleoecology and regime shift analysis.
These functions assume your data is in tidy format.

A detailed explanation of Rodionov's STARS algorithm, and how to use it for regime shift analysis in R, is available here

Current commands:

Rodionov(): performs STARS analysis (Rodionov, 2004) on a dataset. Takes 7 arguments (4 mandatory):

data - the dataframe that will be used.
col - the column we are measuring change on - variable 'X' in STARS.
time - the column containing time units (e.g. age of a subsample)
l - the cut-off length of a regime; affects sensitivity (see Rodionov, 2004)
prob - the p-value for significance of a regime shift. Defaults to p = 0.05
startrow - the row from which the algorithm starts - use if you want to ignore some rows. Defaults to 1
merge - changes the result to be either a regime-shift only table (if FALSE), or an addition to the original table (if TRUE); see below: Result produced: if merge = FALSE (default), produces a 2-column table of time (the time value for each regime shift) and RSI (the RSI for each regime shift). If merge = TRUE, returns the original dataset with an extra RSI column, giving the RSI for each time unit - 0 for non-shift years.

RSI_graph(): Creates a pair of graphs from a dataset, one of a variable against time, and one of the RSI against time. Good for visualisation of regime shifts analysis. Takes 4 mandatory arguments:

data - the dataframe that will be used.
col - the column we are measuring change on - variable 'X' in STARS.
time - the column containing time units (e.g. age of a subsample)
rsi - the column containing RSI values - for best visualisation (i.e. both graphs on a 1:1 scale), ensure RSI values of 0 are 0's, rather than NA (for example, using the merge functionality of Rodionov()).
Result produced: 2 graphs, one on top of the other, depicting as mentioned above.

Lanzante(): Performs a L-test (Lanzante, 1996) to find regime shifts. Takes 5 arguments (3 mandatory):

data - the dataframe that will be used.
col - the column we are measuring change on - variable 'X' in STARS.
time - the column containing time units (e.g. age of a subsample)
p - the largest p-value you want to check regime shifts for. Defaults to p = 0.05.
merge - changes the result to be either a regime-shift only table (if FALSE), or an addition to the original table (if TRUE); see below: Result produced: if merge = FALSE (default), produces a 2-column table of time (the time value for each regime shift) and p (the p-value for each regime shift). If merge = TRUE, returns the original dataset with an extra p-value column, giving the p-value for each time unit - 0 for non-shift years.

NOTE: it may be useful to visualise this data - you can do so using RSI_graph as seen above, but use the p-value column for the RSI column.

Hellinger_trans(): Hellinger transforms data (Nunerical Ecology, Legendre and Legendre). Mutates the original dataset with a column containing Hellinger transformed values. Takes 3 mandatory arguments:

data - the dataframe that will be used.
col - the column we are measuring change on.
site - the column containing the site of each sample. Result produced: the original dataset, with an added column hellinger_trans_vals, containing hellinger transformed values for each data point.

absolute_to_percentage(): Converts absolute abundance data for each site into percentage of total abundance per site. Takes 3 mandatory arguments:

data - the dataframe that will be used.
col - the column we are measuring change on.
site - the column containing the site of each sample. Result produced: the original dataset, with an added column percentage, containing percentage abundance values for each data point.

rolling_autoc(): Finds rolling autocorrelation over an interval. (Used experimentally for early warning signs: see Liu, Gao & Wang, 2018). Takes 3 mandatory arguments:

data - the dataframe that will be used. col - the column we are measuring change on. l - the time interval (no. of columns) used in the autocorrelation. Result produced: a table of rolling lag-1 autocorrelation values.