Test for seasonality in a time series Using AIC-based model selection

Fits a simple linear regression model and segmented regression models (with 1 to 3 breakpoints) to a numeric time series. Compares their Akaike Information Criterion (AIC) values, with a penalty parameter k, to determine if the time series is best described as flat (no seasonality) or as having significant seasonal changes

Usage

determine_seasonality(ts, doy = 1:length(ts), k = 50)

Arguments

ts

Numeric vector. The time series signal to evaluate (e.g., EVI, NDVI, or other index).

doy

Numeric vector. The day-of-year indices corresponding to the time series. Default is 1:length(ts). If your ts is not daily, you need to provide the day-of-year indices.

k

Numeric. Penalty parameter for the AIC calculation. Higher values make the function more conservative in detecting seasonality. Default is 50.

Value

Logical. Returns TRUE if the time series is seasonal (segmented model preferred), or FALSE if the time series is flat (linear model preferred).

Examples

if (FALSE) { # \dontrun{
# Simulate a seasonal time series with noise and missing values
t <- 1:365
simulate_ts <- sin(2 * pi * t / 365) + rnorm(365, sd = 0.1)
simulate_ts[sample(1:365, 30)] <- NA # introduce some missing data

# Test for seasonality
determine_seasonality(ts = simulate_ts, k = 50)

# Example with a flat (non-seasonal) time series
flat_ts <- rnorm(365, mean = 0.5, sd = 0.05)
determine_seasonality(ts = flat_ts)
} # }