Land cover classification of the Southern Yucatan & Improvement of secondary successional forest mapping
Rebecca
P
Dickson, Clark University, Graduate School of Geography, rdickson@clarku.edu
(Presenting)
J
Ronald
Eastman, Clark University, Graduate School of Geography & Clark Labs, George Perkins Marsh Institute, Clark University, reastman@clarku.edu
Birgit
Schmook, Clark University, Graduate School of Geography & ECOSUR (EL Colegio de la Frontera SUR), Chetumal, Q. Roo, Mexico, bschmook@ecosur-qroo.mx
This research explores land cover classification in the dry tropical forests of the Southern Yucatan Peninsular Region (SYPR), Mexico, with additional emphasis on secondary successional forests. Detailed land-cover characterizations facilitate analysis of both the human and ecological landscape, but classifications of medium spatial resolution remotely sensed data are challenging. The SYPR presents both typical and distinct challenges. Typical challenges include subtle vegetation transitions and within-class variability across ecological gradients making strict delineation between classes difficult, as well as frequent cloud cover and atmospheric disturbances. Tropical dry forests are defined by a distinct amount (250-2000 mm) and seasonality (received in 4-9 months) of precipitation, creating unique classification challenges related to variable annual deciduousness (0-100%) and the resulting spectral variability in land covers. A hybrid supervised and unsupervised methodology (IPCA: In Process Classification Approach) was used to both overcome the challenges of the landscape and to explore the level of information that could be uncovered from the imagery. This approach required an iterative process including extensive field and lab work to differentiate the maximum number of land cover classes which were both spectrally separable and ecologically meaningful. Despite high overall map accuracy of 87.6%, secondary successional forest classes remain difficult to distinguish. Further research on the properties of secondary forests in the region includes extensive field data collection on species composition and structural properties over a 15 year chronosequence. Analysis of the changing spectral properties through succession and methods which best differentiate multiple age classes and late secondary stands is ongoing.