Contextual Mann-Kendall Trend Test
tj_contextual_mann_kendall.RdMann-Kendall trend test for 2D rasterStacks, assuming layers represent timepoints.
Usage
tj_contextual_mann_kendall(
x,
...,
neighbourhood = 2,
calc_slope = FALSE,
timevar = "z",
attrvar = "values"
)Arguments
- x
stars-object, with x,y and time dimensions, and at least one attribute.
- ...
ignored
- neighbourhood
2:queen, 1:rook, 0:none (classical M-K test) Only 2 and 0 implemented.
- calc_slope
Calculate Theil-Sen slope estimate for each series? Default:FALSE
- timevar
Name of the time dimension. Default "z"
- attrvar
Name of the value attribute. Default "values
Value
A stars-object with following attributes: 'S' for the Mann-Kendall Statistic (classical or smoothed); 's2' for the variance of the statistic; 'p' for p-value of trend detected; (optional) 'slope' for the Theil-Sen slope estimate.
Details
Assume that the values of each rasterStack location (cell) over time (layers) are a time-series. This function calculates the Mann-Kendall trend test for each series, and returns the relevant statistics as a new rasterStack.
The **Contextual** version (default; see ref.) simply averages the Mann-Kendall-test's S-statistic at each location over neighbours, very much like running 'focal(S, w=matrix(1/9,3,3))' on the non-contextual/cellwise statistic raster. However, this function also calculates the adjusted variances (and hence the p-values).
NOTE assumes equal interval timeseries if 'time' is missing. Matters only for the slope calculation.