0

I have run MaxEnt using the regular software user interface and I have got a nice model which predicts very accurately the distribution of a species. Then I ran MaxEnt in R using the same variables, same spatial extent and same occurrences and got a quite different model. The model from R predicts a similar distribution but the probability values for each pixel are considerably lower. Does anyone have any idea of what I may be doing to get such differences? I would expect some little differences for each run but these are too large. Below I give details of the features used in MaxEnt and the code use in R.

The follow settings were used during the run in MaxEnt interface:

300 presence records used for training, 100 for testing.
10298 points used to determine the Maxent distribution (background points and presence points).
Environmental layers used: Bio1 Bio10 Bio11 Bio12 Bio13 Bio14 Bio15 Bio16 Bio17 Bio18 Bio19 Bio2 Bio3 Bio4 Bio5 Bio6 Bio7 Bio8 Bio9 litologia_Peninsula2_ASC(categorical)
Regularization values: linear/quadratic/product: 0.050, categorical: 0.250, threshold: 1.000, hinge: 0.500
Feature types used: hinge product linear quadratic
responsecurves: true
jackknife: true
samplesfile: C:...Random_points\\1+2.csv
environmentallayers: C:...Variable_clips\\ASC
randomtestpoints: 25

The follow settings were used during the run in R:

300 presence records used for training.
4339 points used to determine the Maxent distribution (background points and presence points).
Environmental layers used: Bio_10clip_Present Bio_11clip_Present Bio_12clip_Present Bio_13clip_Present Bio_14clip_Present Bio_15clip_Present Bio_16clip_Present Bio_17clip_Present Bio_18clip_Present Bio_19clip_Present Bio_1clip_Present Bio_2clip_Present Bio_3clip_Present Bio_4clip_Present Bio_5clip_Present Bio_6clip_Present Bio_7clip_Present Bio_8clip_Present Bio_9clip_Present litologia_Peninsula2_ASC_Present(categorical)
Regularization values: linear/quadratic/product: 0.050, categorical: 0.250, threshold: 1.000, hinge: 0.500
Feature types used: hinge product linear threshold quadratic
samplesfile: C:...maxent/8126817298/presence
environmentallayers: C:...maxent/8126817298/absence
autorun: true
visible: false
prefixes: false

I can see some differences between both runs but I don't know how to make the run in R with exactly the same setting. Any help would be much appreciated.

Gregor Thomas
  • 136,190
  • 20
  • 167
  • 294
M.G.
  • 1
  • 1
  • I have now managed to make the run reports looking the same. So the points used to determine the Maxent distribution are now the same, but still the probability values in the predictions from the regular Maxent interface are much higher than in the predictions from the model obtained though R. – M.G. Jun 14 '23 at 18:13

0 Answers0