Filtering multi-collinear predictor variables from multi-resolution rasters of WorldClim 2.1 for Ecological Niche Modeling in Indonesian context

Abstract

Abstract. Pradhan P, Setyawan AD. 2021. Filtering multi-collinear predictor variables from multi-resolution rasters of WorldClim 2.1 for Ecological Niche Modeling in Indonesian context. Asian J For 5: 111-122. WorldClim is one of the popular environmental datasets which hosts multi-resolution interpolated gridded climate raster surfaces and derived bioclimatic variables for both the immediate past, present and future scenarios. Bioclimatic variables along with other environmental factors like solar radiation, wind speed, water vapour pressure etc. have been used as primary set of explanatory variables for mapping and spatial modeling of many biological processes, including defining environmental niche of a species and identifying potential areas for its distribution through machine learning methods like Ecological Niche Modeling or Species Distribution Modeling or Habitat Suitability Modeling. However, the interpolated explanatory datasets are known to cause over-fitting of the models mainly due to multi-collinearity or redundancy within the variables. In the present study, 58 bioclimatic and environmental variables of Indonesian extent extracted from WorldClim 2.1 are screened to investigate the presence of multi-collinearity or redundancy. From the total 3364 variable pairs per raster resolution, 174 variable pairs were known to be affected by multicollinearity, from which temperature related bioclimatic variables, water vapour pressure and elevation associated variables were highly notable. For all the raster resolutions, bioclimatic variable 2, 3, 4, 15, 18 and 19, as well as slope, aspect, solar radiation for January, April, May, September, wind speed for August and November were found to be non-collinear. While, solar radiation for March and July were found to be non-collinear for 30s, 2.5m and 5m raster resolutions; Wind speed of July was non-collinear for 30s and 2.5m; Solar radiation for February and June were non-collinear for 10m; water vapour pressure for August for 2.5m and wind speed for January was non-collinear for 30s raster resolutions. The results of this study might serve as a convenient reference for investigators of the region for selection of bioclimatic and other environmental variables for conducting ecological niche modeling studies.