Without -n flag, the complete list of inputs for each cell (including NULLs) is passed to the aggregate function. Individual aggregates can handle data as they choose. Mostly, they just compute the aggregate over the non-NULL values, producing a NULL result only if all inputs are NULL.
The min_raster and max_raster methods generate a map with the number of the raster map that holds the minimum/maximum value of the time-series. The numbering starts at 0 up to n for the first and the last raster listed in input=, respectively.
If the range= option is given, any values which fall outside that range will be treated as if they were NULL. The range parameter can be set to low,high thresholds: values outside of this range are treated as NULL (i.e., they will be ignored by most aggregates, or will cause the result to be NULL if -n is given). The low,high thresholds are floating point, so use -inf or inf for a single threshold (e.g., range=0,inf to ignore negative values, or range=-inf,-200.4 to ignore values above -200.4).
Linear regression (slope, offset, coefficient of determination) assumes equal time intervals. If the data have irregular time intervals, NULL raster maps can be inserted into time series to make time intervals equal (see example).
r.series input="`g.mlist pattern='insitu_data.*' sep=,`" \ output=insitu_data.stddev method=stddev
Note the g.mlist script also supports regular expressions for selecting map names.
Using r.series with NULL raster maps:
r.mapcalc "dummy = null()" r.series in=map2001,map2002,dummy,dummy,map2005,map2006,dummy,map2008 \ out=res_slope,res_offset,res_coeff meth=slope,offset,detcoeff
Example for multiple aggregates to be computed in one run (3 resulting aggregates from two input maps):
r.series in=one,two out=result_avg,res_slope,result_count meth=sum,slope,count
Last changed: $Date$