Apply temporal filter
This option replaces each data point with the dfference between the
value of this year and the previous year. It acts as a (primitive)
high-pass filter and is often used to eliminate to first order slow
variations or trends. Note that higher-order effects, such as a
slowly varying variance, are not removed.
Subtract mean of previous years
This is an extension of the year-on-year differences in which the mean
of the previous N
years is subtracted. (For N
coincide). Again, these high-pass filters are used to remove
low-frequency signals or trends. For N
>1 the filter is more
gentle than the year-on-year difference method and retains
more low-frequency signal.
Average with previous years
The converse is to average with N
previous years, making a
+1 year running mean. This primitive low-pass filter can be
used to improve the signal/noise ratio of slowly-varying signals.
Note that the degrees of freedom in the significance calculations are
adjusted accordingly, so the signal may not be more significant after
this low-pass filtering.
If there is an option "no overlap" (currently only on the
verification pages), this means that instead of the default sliding
window, non-overlapping windows are taken. An example: if you have 100
years of data with a 10-yr running window ("average with  previous
years"), you will normally get 90 or more data points that are not
independent: years 1-10, 2-11, 3-12, ..., 91-100. With this option
you will get 10 independent data points: years 1-10, 11-20, ..., 91-100.
One day I will implement more sophisticated low-pass, high-pass and
See also the detrend option.