The habitat suitability model generated in Maxent showed areas that might be possibly occupied by this species southwest of Sierra Nevada de Santa Marta, the southern Central Andes and the Colombian Massif and south. These areas, were removed from the potential distribution map of the species; with the exception of the areas in the Colombian massif and south (east slopes).
It is necessary to bear in mind that this model was elaborated using just two point localities, which is a extremely small sample and that there are just three collected 'Bogota' skins for the subspecies castaneus. The model is useful in the sense that can give an indication of the potential areas where the species may be looked for in Colombia. The distribution of this rare and local species is extremely unknown in Colombia.
Assuming that the distribution of the species may have filled the complete climatic model generated in the Eastern Andes, its distribution today in remnants of forest is about 13,942 km2, which corresponds to a loss of 76 % of its potential original distribution due to deforestation.
Some authors (MacMullan & Donegan, 2014) have pointed to records of this species in the extreme northern portion of central northeast Colombia (i.e. north Norte de Santander) and have suggested the presence of the subspecies knoxi in this zone of the Eastern Andes just south of serrania del Perija. There are no collected specimens that confirm this observation and the presence of this subspecies in Colombia needs confirmation.
Regularized training gain is 2.189, training AUC is 0.987, unregularized training gain is 2.877.
Algorithm converged after 80 iterations (0 seconds).
The follow settings were used during the run:
2 presence records used for training.
10002 points used to determine the Maxent distribution (background points and presence points).
Environmental layers used (all continuous): bio10co bio11co bio12co bio13co bio14co bio15co bio16co bio17co bio18co bio19co bio1co bio2co bio3co bio4co bio5co bio6co bio7co bio8co bio9co
Regularization values: linear/quadratic/product: 1.000, categorical: 0.620, threshold: 1.980, hinge: 0.500
Feature types used: linear
'Equal Training Sensitivity and Specificity' and 'Equate Entropy of Thresholded and Original Distributions' thresholds and omission rates:
0.023-0.112-Fractional predicted area
0-0-Training omission rate