To elaborate our model we removed uncertain localities such as 'Bogota skins' and all records with coordinates laying down in sites with estimated elevations above 2,016 m. Also three records in BioMap, two from Magdalena and one from Huila, which represent corrupted data, were removed previously to modelling.
The habitat suitability model generated in Maxent showed areas suitable in climatic terms for this species in the slopes above the Magdalena Valley south of Samana (Caldas) and in the east slope of the East Andes. These areas are not known to be occupied by the species and were excluded from the potential distribution map. Furthermore, our model predicted some areas in the interior Cauca River Valley. The presence of the species in this zone is suported by a few data in BioMap and DatAves. There are two records in BioMap from rio Gengue (Cauca) in the high Cauca Valley and two records in DatAves from Jeguadas (Risaralda) and Pueblo Nuevo (Antioquia) in the mid Cauca Valley. Those areas are near low passes of the Western Andes and it may be possible the records are correct. Nevertheless, most of the areas predicted as presence in the Cauca Valley by our model are unlikely to be occupied by the species and were marked as such.
Assuming that the distribution of the species may have filled the complete climatic model generated, its distribution today in remnants of forest is about 56,450 km2, which corresponds to a loss of 43 % of its potential original distribution due to deforestation in Colombia.
Regularized training gain is 2.030, training AUC is 0.965, unregularized training gain is 2.338.
Algorithm converged after 1620 iterations (54 seconds).
The follow settings were used during the run:
88 presence records used for training.
10087 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: 0.136, categorical: 0.250, threshold: 1.120, hinge: 0.500
Feature types used: hinge product linear threshold quadratic
'Equal Training Sensitivity and Specificity' and 'Equate Entropy of Thresholded and Original Distributions' thresholds and omission rates:
0.140-0.266-Fractional predicted area
0.137-0.038-Training omission rate