Hylocharis cyanus

NOTES

To elaborate our model we removed uncertain localities such as 'Bogota skins', those localities with no coordinates and all records with coordinates laying down in sites with estimated elevations above 2,193 m.

Initially we elaborated one model in Maxent with all the data. However, predictions east of the Andes were very poor. Following, we elaborated two models, a first model using just data from the northern portion of the distribution and a second model using data east of the Andes. In this case predictions improved substantially. Predictions of the northern model adjusted very well to the data not showing major overprediction, except in the southern slopes of the Sierra Nevada de Santa Marta from were the species is not known and were removed from our potential distribution map. Additionally, our eastern model showed areas suitable in climatic terms for this species west of the Andes. These areas were excluded from the potential distribution related to this model. Finally, we joined predictions from both models to obtain a final potential distribution map.

Distribution of specimens according to BioMap suggest a major extension of the known distribution of subspecies rostrata in Colombia north of the Amazon river. Nonetheless, it is not clear the borders in relation to subspecies viridiventris, and this needs further research.

MODEL METADATA

Northern model

Regularized training gain is 4.169, training AUC is 0.996, unregularized training gain is 4.544.

Algorithm converged after 1080 iterations (39 seconds).

The follow settings were used during the run:

57 presence records used for training.

10055 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: 0.173, categorical: 0.250, threshold: 1.430, hinge: 0.500

Feature types used: hinge linear quadratic

responsecurves: true

jackknife: true

maximumiterations: 2000

Eastern model

Regularized training gain is 1.206, training AUC is 0.922, unregularized training gain is 2.128.

Algorithm converged after 980 iterations (94 seconds).

The follow settings were used during the run:

46 presence records used for training.

10046 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: 0.204, categorical: 0.250, threshold: 1.540, hinge: 0.500

Feature types used: hinge linear quadratic

responsecurves: true

jackknife: true

maximumiterations: 2000

Northern model

'Equal Training Sensitivity and Specificity' and 'Equate Entropy of Thresholded and Original Distributions' thresholds and omission rates:

ETSS-EETOD-Description

13.02-8.893-Cumulative threshold

0.182-0.097-Logistic threshold

0.011-0.015-Fractional predicted area

0.018-0.000-Training omission rate

Eastern model

'Fixed Cumulative Value 5%', 'Equal Training Sensitivity and Specificity' and 'Equate Entropy of Thresholded and Original Distributions' thresholds and omission rates:

FCV5-ETSS-EETOD-Description

5.000-28.001-15.782-Cumulative threshold

0.087-0.281-0.176-Logistic threshold

0.522-0.173-0.300-Fractional predicted area

0.043-0.174-0.130-Training omission rate