Requesting help regarding some aspects of Maxent

[Radhika Jha]

Hello all,

I have recently submitted a manuscript for publication using Maxent 3.4.1 and environmental data from WorldClim. I have received the following queries from the reviewers. I would be very grateful if someone could help me out with these.

(1) The reviewer recommended using Boyce Index to evaluate model performance in addition to AUC. I have the outputs from Maxent 3.4.1, what would be the best way to proceed with calculation of boyce index with this? A detailed methodology would be very helpful.

(2) The reviewer queried about whether a static or a dynamic approach was used regarding the bioclimatic variables (dependent on a month or quarter of the year).  I understand that using data from worldclim constitutes the dynamic approach, however, I am not very clear on this. I would be helpful if someone could shed some light on this or share some literature.

(3) Is the future climate data bias corrected, using delta change mapping or QQ mapping? Again, I am not very clear on these topics, some guidance or literature would be greatly appreciated.

Thank you for sparing your time.

Regards,

Radhika

[Rogério]

Hello Radhika,

1 - The package enmSdm have a simple and easy way to calculate CBI:

You will need to extract the presences and background from your predict model and create to files: pres and contrast:

contBoyce( pres,  contrast, na.rm = TRUE).

The other 2 topics i'm not familiar.

Regards,

Rogério.

[Bede-Fazekas Ákos]

Dear Radhika,

I have answer to your second question, since the terms "static approach" and "dynamic approach" were defined by me. You can find the details in the following paper: https://doi.org/10.1111/2041-210X.13488
AFAIK, all the publicly available global datasets (such as WorldClim) contains future bioclimatic variables calculated by the dynamic approach, even though it might result comparing apple with orange if the specific quarter/month differ between the reference period and the future period. It might increase prediction uncertainty, produce prediction artefacts and transferability issues. There are three solutions to deal with this problem:
1) do not use those bioclimatic variables that are the most problematic. Most of all, do not use combined variables (i.e. precipitation of warmest/coldest month/quarter; temperature of wettest/driest month/quarter).
2) calculate the shift of the specific months/quarters in your study area, define a threshold for the shift (e.g. 2 months) that you think is too much for your species to cope with by its phenotypic plasticity, and do not use those bioclimatic variables as predictors in your model that are based on a month/quarter exceeding the threshold in your study area
3) calculate the bioclimatic variables from the monthly temperature/precipitation data based on the static approch. R script is available here: https://github.com/bfakos/bioclimatic_variables

Regarding questions no 1 and 3, I agree with Rogério, several packages are available in R that can calculate Boyce index. Future WorldClim data are, AFAIK, bias corrected, please read the paper of Fick and Hijmans about WorldClim. Never use non-bias-corrected future climate data for prediction with an SDM trained on observed climate dataset. If you have only non-bias-corrected future data and you are not able to do bias correction, than you must train your model on the climate model's prediction to the reference period and you must not interpret the predictor values along the response curve.

HTH,
Ákos
_______
Ákos Bede-Fazekas
Centre for Ecological Research, Hungary

--
You received this message because you are subscribed to the Google Groups "Maxent" group.
To unsubscribe from this group and stop receiving emails from it, send an email to maxent+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/maxent/08ffcea6-6d46-47e4-baec-f15637ea3d40n%40googlegroups.com.