NEWS

---

bfast 1.7.0 (2024-10-22)

• Added the option to run a structural change test before running bfastlite() with two parameters: 'level' and 'type', equivalent to bfast() except that anything equal to or below 0 in 'level' will skip the sctest.
• Fixed a bug where bfastpp() would throw an error when annual data is input.
• Rework of the 'breaks' argument in bfast(), it now works correctly with a numeric input and gives the opportunity to specify the breaks/statistic for computing the optimal number of breaks separately per component.
• Added a reference to the paper describing the BFAST Lite algorithm, and updated author names to use the proper UTF-8 symbols.
• Updated the examples, migrating from the 'raster' to the 'terra' package, and adding a raster procesing example into '?bfastlite'.
• The 'history_roc()' generic is now consistent in parameters with the 'history_roc.matrix()' and 'history_roc.formula()' methods.

bfast 1.6.1 (2021-05-10)

• Fixed a heap overflow issue in the C++ version of bfastts().

bfast 1.6.0 (2021-05-03)

• Integrated work by Marius Appel on speeding up BFAST algorithms using Rcpp. Whether the speedups are enabled is controlled by set_fast_options() vs set_fallback_options(). The fast options are now default.
• Integrated work by Martin Jung on allowing BFAST to operate on time series that have NA values by making use of the stlplus package. This is controlled by the "decomp" parameter in bfast(). 'stl' is still the default for backwards compatibility and for use in sparse time series.
• New bfastlite() function for running BFAST Lite (BFAST without STL).
• bfast01classify() now has an alternative typology type optimised for dryland change classification (thanks to Paulo Negri Bernardino).
• bfastpp() and bfastmonitor() now have a parameter 'sbins' that controls the number of seasonal dummies (bins) when using 'season' in the formula.
• bfast() now defaults to max.iter = 10, making the argument optional. It also no longer has the argument 'reg', as it was previously unused. Any additional arguments to the function now are either sent to the stlplus::stlplus() function, or ignored if decomp = 'stl'.
• create16dayts() is now deprecated in favour of bfastts().
• Switched dependencies from strucchange to strucchangeRcpp for better performance.
• The raster package is no longer a dependency, but is suggested.
• bfastmonitor() now allows to specify the significance 'level' separately for the ROC and the monitoring algorithms. If only one specified, both are set to the same level.
• bfastts() now officially handles matrix inputs to make use of external regressors easier.

bfast 1.5-9

• minor bug x_coor <- out$bp.Wt$datatsp[[1]] with x_coor <- tsp(out$Wt)[1] in seasonal()

bfast 1.5-7

• all required packages are now in imports so you have to load the package e.g. zoo yourself now.

bfast 1.5-5

• Bfast01 classification function added

bfast 1.5

• Bfast01 function added

bfast 1.4-4

• Bfastmonitor function added

bfast 1.4-3

• Preparing helper functions for processing of different types of time series data
• Preparing structure and plan for raster brick processing (satellite image time series processing)

bfast 1.4-1

• Plotting functionality is improved for bfastmonitor() output (i.e. when dealing with daily data and lot's of missing data points)

bfast 1.4-0

• Added bfastmonitor() for near real-time detection of breaks in BFAST-type model. Data pre-processing is handled by a new function bfastpp() whose results can easily be plugged into strucchange (or other modeling/testing functions).
• New data set "som" with NDVI series from Somalia.