Colinus cristatus
NOTES
We elaborated a first habitat suitability model in Maxent for the whole species, the predictions of this model performed fairly well west and north of the Andes, whilst missed extensive areas in the Orinoco and northern Amazon regions. Because of this, we decided to elaborate individual models for each one of the seven subspecies confirmed in the country. Additionally, to improve predictions for the subspecies parvicristatus were added during the elaboration of its model two point localities from Vichada and one from Vaupes (EBIRD, 2015).
There were 123 records not correctly determined to subspecies, which were overridden following their known distribution (Hilty & Brown, 1986; HBW Alive, 2015). Almost half of these were specimens identified as subspecies barnesi, which must be surely badius. Subspecies barnesi is distributed in west central Venezuela in the states of Portuguesa and Barinas (HBW Alive, 2015) and it is not known to occur in Colombia. However, there is one specimen (ICN, Catalogue No. 15184) from the Catatumbo region near Venezuela that well might be barnesi or maybe a non described subspecies; this needs further revision and confirmation.
Most of the original subspecies models elaborated predicted areas where other subspecies are present and therefore they were trimmed to fit better the known distributions. Thus, for the model elaborated for C. cristatus parvicristatus it was trimmed off an area predicted in Norte de Santander, where there is no certainty which subspecies exists, and areas above 2,400 m; for the model elaborated for C. cristatus leucotis it was trimmed off all areas not in the Magdalena valley, except in the low-mid Cauca valley where might be some area of intergradation entering into the Cauca valley; for the model elaborated for C. cristatus littoralis it was trimmed off, with help of the known localities, areas not adjacent to the Sierra Nevada de Santa Marta; for the model elaborated for C. cristatus badius it was trimmed off most areas predicted in the Pacific slope, just leaving areas in the slopes above the Cauca valley below 2,200 m; finally, for the model elaborated for C. cristatus bogotensis it was trimmed off areas north of Sierra Nevada del Cocuy and south of northern Huila, as well as areas above 3,100 m.
Distribution of specimens, according to BioMap, suggests two possible narrow areas of intergradation. The first (≈ 5,870 km2) between subspecies decoratus and leucotis in central Bolivar and south Magdalena to south Cesar. The second (≈ 972 km2) between subspecies badius and leucotis in central Antioquia. Most specimens in both areas have been collected well after the description of both subspecies involved in each pair and we believe it is unlikely they are erroneously identified. Additionally, there is one specimen from Bonda (LSU, Catalogue No. 43405) identified as decoratus, which is in the range of littoralis and might represent individual variation, introgression or intergradation. Similarly, there are nine specimens identified as leucotis from Labateca in south Norte de Santander. Subspecies leucotis is not known in the eastern slope of the Easter Andes and east of it in that area. Therefore, these specimens need revision to clarify their identities.
MODEL METADATA
C. cristatus cristatus
Regularized training gain is 3.929, training AUC is 0.996, unregularized training gain is 4.616.
Algorithm converged after 160 iterations (1 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
responsecurves: true
jackknife: true
maximumiterations: 2000
'Equal Training Sensitivity and Specificity' and 'Equate Entropy of Thresholded and Original Distributions' thresholds and omission rates:
ETSS-EETOD-Description
11.768-8.94-Cumulative threshold
0.161-0.124-Logistic threshold
0.016-0.02-Fractional predicted area
0-0-Training omission rate
C. cristatus parvicristatus
Regularized training gain is 1.149, training AUC is 0.910, unregularized training gain is 1.718.
Algorithm converged after 1300 iterations (39 seconds).
The follow settings were used during the run:
36 presence records used for training.
10036 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: 0.233, categorical: 0.250, threshold: 1.640, hinge: 0.500
Feature types used: linear quadratic hinge
responsecurves: true
jackknife: true
maximumiterations: 2000
'Fixed cumulative value 1', 'Equal Training Sensitivity and Specificity' and 'Equate Entropy of Thresholded and Original Distributions' thresholds and omission rates:
FCV1-ETSS-EETOD-Description
1-31.939-14.313-Cumulative threshold
0.034-0.338-0.201-Logistic threshold
0.671-0.167-0.317-Fractional predicted area
0-0.167-0.111-Training omission rate
C. cristatus leucotis
Regularized training gain is 1.834, training AUC is 0.952, unregularized training gain is 2.212.
Algorithm converged after 1020 iterations (28 seconds).
The follow settings were used during the run:
60 presence records used for training.
10060 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: 0.164, categorical: 0.250, threshold: 1.400, hinge: 0.500
Feature types used: linear quadratic hinge
responsecurves: true
jackknife: true
maximumiterations: 2000
'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
5-17.5-12.172-Cumulative threshold
0.066-0.227-0.164-Logistic threshold
0.25-0.124-0.159-Fractional predicted area
0.017-0.117-0.033-Training omission rate
C. cristatus littoralis
Regularized training gain is 2.466, training AUC is 0.979, unregularized training gain is 2.778.
Algorithm converged after 220 iterations (1 seconds).
The follow settings were used during the run:
13 presence records used for training.
10013 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: 0.671, categorical: 0.393, threshold: 1.870, hinge: 0.500
Feature types used: linear quadratic
responsecurves: true
jackknife: true
maximumiterations: 2000
'Equal Training Sensitivity and Specificity' and 'Equate Entropy of Thresholded and Original Distributions' thresholds and omission rates:
ETSS-EETOD-Description
11.437-7.215-Cumulative threshold
0.213-0.142-Logistic threshold
0.068-0.085-Fractional predicted area
0.077-0-Training omission rate
C. cristatus decoratus
Regularized training gain is 2.483, training AUC is 0.977, unregularized training gain is 2.839.
Algorithm converged after 740 iterations (20 seconds).
The follow settings were used during the run:
24 presence records used for training.
10024 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: 0.365, categorical: 0.250, threshold: 1.760, hinge: 0.500
Feature types used: linear quadratic hinge
responsecurves: true
jackknife: true
maximumiterations: 2000
'Equal Training Sensitivity and Specificity' and 'Equate Entropy of Thresholded and Original Distributions' thresholds and omission rates:
ETSS-EETOD-Description
18.29-9.674-Cumulative threshold
0.283-0.167-Logistic threshold
0.059-0.083-Fractional predicted area
0.042-0.042-Training omission rate
C. cristatus badius
Regularized training gain is 2.748, training AUC is 0.982, unregularized training gain is 3.141.
Algorithm converged after 900 iterations (26 seconds).
The follow settings were used during the run:
37 presence records used for training.
10037 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: 0.230, categorical: 0.250, threshold: 1.630, hinge: 0.500
Feature types used: linear quadratic hinge
responsecurves: true
jackknife: true
maximumiterations: 2000
'Equal Training Sensitivity and Specificity' and 'Equate Entropy of Thresholded and Original Distributions' thresholds and omission rates:
ETSS-EETOD-Description
13.384-11.63-Cumulative threshold
0.152-0.129-Logistic threshold
0.057-0.064-Fractional predicted area
0.054-0.054-Training omission rate
C. cristatus bogotensis
Regularized training gain is 2.743, training AUC is 0.972, unregularized training gain is 3.166.
Algorithm converged after 100 iterations (1 seconds).
The follow settings were used during the run:
13 presence records used for training.
10012 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: 0.671, categorical: 0.393, threshold: 1.870, hinge: 0.500
Feature types used: linear quadratic
responsecurves: true
jackknife: true
maximumiterations: 2000
'Equal Training Sensitivity and Specificity' and 'Equate Entropy of Thresholded and Original Distributions' thresholds and omission rates:
ETSS-EETOD-Description
10.225-12.547-Cumulative threshold
0.095-0.12-Logistic threshold
0.077-0.064-Fractional predicted area
0.077-0.077-Training omission rate