@article{fdi:010088389,
  title = {{BFAST}m-{L}2, an unsupervised {LULCC} detection based on seasonal change detection  - an application to large-scale land acquisitions in {S}enegal},
  author = {{S}carpetta, {Y}.{N}. and {L}ebourgeois, {V}. and {L}aques, {A}nne-{E}lisabeth and {D}ieye, {M}. and {B}ourgoin, {J}. and {B}egue, {A}.},
  editor = {},
  language = {{ENG}},
  abstract = {{I}n the context of {G}lobal {C}hange {R}esearch, detection, monitoring and characterization of land use/land cover ({LULC}) changes are of prime importance. {T}he increasing availability of dense satellite image time series ({SITS}) has led to a shift in the change detection paradigm, with algorithms able to exploit the full temporal information laid down in {SITS}. {S}o far, most of these algorithms have focused on the detection of abrupt and gradual changes, and thus developed breakpoint detection based on significant deviations from the mean. {H}owever, {LULC} changes may manifest themselves in other patterns, particularly changes in seasonality (amplitude, number and length of the growing seasons) that are harder to detect. {I}n this paper, we propose a simple method to automatically select the breakpoint linked to the biggest seasonal change in long and dense {SITS} with multiple breakpoints. {T}his approach - {BFAST}m-{L}2 - relies on linking a high-speed algorithm ({BFAST} monitor) with a time series similarity metric ({E}uclidian distance {L}2) sensitive to seasonal changes. {T}he capacity of {BFAST}m-{L}2 to identify the date of change in different situations was tested on two data sets, and compared to the performances of three other algorithms ({BFAST} monitor, {BFAST} lite, and {E}dyn). {T}he data sets are 1. a published benchmark data set composed of 25 200 simulated {SITS} with different change types and change magnitudes, and 2. the 2000-2020 {MODIS} {NDVI} {SITS} over a 200x200 pixels area in {S}enegal including different study sites which have undergone recent {LULC} changes due to agricultural large-scale land acquisitions ({LSLA}s) (as reported in the ground field database used in this study). {T}he results show that {BFAST}m-{L}2 is efficient in accurately detecting in time most  of the changes, and, in contrast with {BFAST} {L}ite and {BFAST}monitor, to spatially highlight {LSLA}s-induced changes without the need of any prior knowledge. {T}he automatic proposed approach, faster than {BFAST} {L}ite and {E}dyn, and with very few tuneable parameters, may thus be easily implemented in unsupervised pipelines to map and analyse generic {LULC} changes at regional scale.},
  keywords = {{SENEGAL} ; {ZONE} {SOUDANOSAHELIENNE}},
  booktitle = {},
  journal = {{I}nternational {J}ournal of {A}pplied {E}arth {O}bservation and {G}eoinformation},
  volume = {121},
  numero = {},
  pages = {103379 [17 ]},
  ISSN = {1569-8432},
  year = {2023},
  DOI = {10.1016/j.jag.2023.103379},
  URL = {https://www.documentation.ird.fr/hor/fdi:010088389},
}