Is there any manual guide to perform multicollinearity test between environmental (19 Bioclimatic) variables for MaxEnt modelling? Using all variables without removing the correlated one affects the model performance (Accuracy)? i have known about doing through SDM tool in ARCGIS but have not go the exact idea. i would be grateful if someone suggest me.
Dear Neelam,
To my understanding, because of its algorithm benefit, MaxEnt can model with input all 19 bioclimatic variables, but it is not obviously a good way of estimating the model of distribution with predictors with severe multi-linearity problems.
Instead, a alternative way to deal with predictor sets is that simply using PCA-derived variables (see De Marco and Nóbrega, 2018 ), or you can remove multicollinearity among the 19 variables by deleted highly correlated factors based one of the following criteria:
(a) the variance inflation factor (VIF) value of each factor less than 10 (Kubota et al. 2015);
(b) the pairwise Pearson correlation coefficients |r| < 0.7 (Dormann et al. 2013).
BTW, you can find what I did when dealing with the model preditors from the following two publications.
Article Climate change jointly with migration ability affect future ...
Article Potential distributions of seven sympatric sclerophyllous oa...
Best wishes,
Ziyan Liao
[1] De Marco P Júnior, Nóbrega CC (2018) Evaluating collinearity effects on species distribution models: An approach based on virtual species simulation. PLOS ONE 13(9): e0202403. https://doi.org/10.1371/journal.pone.0202403
[2] Kubota, Yasuhiro, Takayuki Shiono, and Buntarou Kusumoto. 2015. “Role of Climate and Geohistorical Factors in Driving Plant Richness Patterns and Endemicity on the East Asian Continental Islands.” Ecography 38(6):639–48.
[3] Dormann, Carsten F., Jane Elith, Sven Bacher, Carsten Buchmann, Gudrun Carl, Gabriel Carré, Jaime R. Garcí. Marquéz, Bernd Gruber, Bruno Lafourcade, Pedro J. Leitão, Tamara Münkemüller, Colin Mcclean, Patrick E. Osborne, Björn Reineking, Boris Schröder, Andrew K. Skidmore, Damaris Zurell, and Sven Lautenbach. 2013. “Collinearity: A Review of Methods to Deal with It and a Simulation Study Evaluating Their Performance.” Ecography 36(1):027–046.
Dear Neelam Pandey
The manual method is through pairwise Pearson correlation coefficients |r| < 0.7 spss --- analysis -- correlation --- bivariate
PCA is great but not advised for future predictions or spatial transfer. I suggest to think a little about the organism and choose only a few relevant constraints. Later and depending on the organism, I would add predictors such as soil pH, heat load index, distance to water and vrm (sappington).
Ziyan Liao I am really grateful to you for explaining this to me. can you help me with this question? Let us suppose we chose 9 variables after minimizing redundancy of explanatory bioclimatic variables by performing above mentioned methods; pairwise Pearson correlation coefficients or VIF or PCA (removing the highly correlated ones) will these same 9 variables applicable for all the species for the particular region (Country) or it might vary for different species?
Neelam Pandey Better to check multicollinearity among the inital predictors for each species. I can send you the R code for this analysis by email if you like.
Ziyan Liao
Rainer Ferdinand Wunderlich suppose we chose some variables after minimizing redundancy of explanatory bioclimatic variables; will these same variables applicable for modelling all the species for the particular region (Country) because 19 bioclimatic variables will be same for that region or it might vary for different species?
- Badges
- Science topic