To elaborate the habitat suitability model we removed five corrupted records placed west of the Andes; three in BioMap from the WFVZ, a dataset known to have problems Verhelst et al. (non published data) and two in ProAves, coincidentally all from Santander.
The habitat suitability model generated in Maxent showed extensively areas that are suitable in climatic terms for this species in most of tropical Colombia. Areas west of the Andes are not known to be occupied by this curassow and were excluded from its potential distribution map. Also the model showed suitable areas south of the rio Caqueta; there are no records from this zone and it is believed the distribution of this curassow does not reach that far south. It is interesting to note that when applied a lower threshold (≈ FCV5) than the usual we have used (EETOD), the marginally suitable areas increased extending to the Amazon Trapezium. Additionally, it is important to highlight that in more recent years there have been several sightings from central south to central north Casanare (EBIRD, 2015), confirming those records areas predicted by out model in that department.
Assuming that the distribution of the species may have filled the areas predicted as suitable (i.e. marginally suitable, suitable and highly suitable), its potential distribution today in remnants of forest is about 358,062 km2, which corresponds to a loss of 10 % of its potential original distribution due to deforestation in that region. It is interesting to note that as a whole this figure must be higher, since we do not account for areas around Macarena and in the rio Guayabero where there is a very strong colonisation front that has reduced forested areas severely in the southwestern Orinoco region.
Regularized training gain is 0.358, training AUC is 0.802, unregularized training gain is 0.659.
Algorithm converged after 100 iterations (0 seconds).
The follow settings were used during the run:
7 presence records used for training.
10007 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: 0.545, threshold: 1.930, hinge: 0.500
Feature types used: linear
'Fixed Cumulative Value 5', 'Equal Training Sensitivity and Specificity' and 'Equate Entropy of Thresholded and Original Distributions' thresholds and omission rates:
0.73-0.299-0.699-Fractional predicted area
0-0.286-0-Training omission rate