Mapping Prosopis spp. with Landsat 8 data in arid environments:Evaluating effectiveness of different methods and temporal imageryselection for Hargeisa, Somaliland

Prosopis spp. is a fast and aggressive invader threatening many arid and semi-arid areas globally. Thespecies is native to the American dry zones and was introduced in Somaliland for dune stabilizationand fuel wood production in the 1970’s and 1980’s. Its deep rooting system is capable of tapping into thegroundwater table thereby reducing its reliance on infrequent rainfalls and near-surface water. The com-petitive advantage of Prosopis is further fuelled by the hybridization of the many introduced subspeciesthat made the plant capable of adapting to the new environment and replacing endemic species. Thisstudy aimed to test the mapping accuracy achievable with Landsat 8 data acquired during the wet andthe dry seasons within a Random Forest (RF) classifier, using both pixel- and object-based approaches.Maps are produced for the Hargeisa area (Somaliland), where reference data was collected during thedry season of 2015. Results were assessed through a 10-fold cross-validation procedure. In our study, thehighest overall accuracy (74%) was achieved when applying a pixel-based classification using a combi-nation of the wet and dry season Earth observation data. Object-based mapping were less reliable due tothe limitations in spatial resolution of the Landsat data (15–30 m) and problems in finding an appropriatesegmentation scale.

NG Wai-Tim;  MERONI Michele;  IMMITZER Markus;  BOCK Sebastian;  LEONARDI Ugo;  REMBOLD Felix;  GADAIN Hussein;  ATZBERGER Clement; 

2016-12-20

ELSEVIER SCIENCE BV

JRC104055

0303-2434,   

http://www.sciencedirect.com/science/article/pii/S0303243416301271,    https://publications.jrc.ec.europa.eu/repository/handle/JRC104055,   

10.1016/j.jag.2016.07.019,