To elaborate our model we removed uncertain localities such as those localities with no coordinates and all records with coordinates laying down in sites with estimated elevations above 311 m. Also three records in BioMap, three in DatAves and one in EBIRD from the interior Caribbean region and the noreastern Andean region, which constitute corrupted information in the database were deleted previous to modelling.
The habitat suitability model generated in Maxent showed a very few areas suitable in climatic terms for this species in the interior of the Caribbean region, these areas are not known to be occupied by the species and were excluded from the potential distribution map of this hummingbird. South of the Buenaventura area our maps point areas not predicted but that are likely suitable for this species. This areas where catalogued as such since our model showed an area in the farthest southwest. Nonetheless, there are no known records so far south and those areas are most certainly not part of the actual distribution of this hummingbird.
Assuming that the distribution of the species may have filled the complete climatic model generated, its distribution today in remnants of forest is about 17,395 km2, which corresponds to a loss of 70 % of its potential original distribution due to deforestation. This has particularly noticeable in the Caribbean where most coastal and mangrove forests have been cleared.
Distribution of specimens according to BioMap shows a few specimens in the zone of the golfo de Uraba identified as the nominate subspecies, which has not been recorded for Colombia. This may indicate a zone of intergradation between subspecies coelina and confinis or an identification error. Otherwise, it might be due to the fact that sometimes confinis and the nominate are considered synonyms, in which case the distribution of subspecies confinis may extend slightly further east than what actually appears in our maps covering the whole golfo de Uraba. This needs further revision to be clarified.
Regularized training gain is 3.045, training AUC is 0.988, unregularized training gain is 3.497.
Algorithm terminated after 2000 iterations (66 seconds).
The follow settings were used during the run:
63 presence records used for training.
10063 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: 0.156, categorical: 0.250, threshold: 1.370, hinge: 0.500
Feature types used: hinge linear quadratic
'Equal Training Sensitivity and Specificity' and 'Equate Entropy of Thresholded and Original Distributions' thresholds and omission rates:
0.038-0.047-Fractional predicted area
0.032-0.016-Training omission rate