Early results of species distribution modelling

Description:

This repository provides early access to species distribution models prepared in HiQBioDiv project. Modelling was performed within a presence-background framework using maximum entropy analysis. The analytical workflow is illustrated in 00_ModellingWorkflow.png, where the iterative nature can be seen: every model is evaluated statistically (00_StatisticalEvaluation.png), as well as by species experts. If a model was of low quality, showed strong artefacts or unreasonable ecological associations with ecogeographical variables, another version with a reduced set of environmental predictors was created. The habitat suitability maps issued here are considered by experts to be usefull for guiding fieldwork to increase the number of presence locations (for rare species), or to be directly usable in nature conservation and management planning in most cases.

Species-specific results are stored in archives of wider taxonomic groups. Each species has its own directory named after its abbreviated scientific name and the version of the model. The statistical evaluation metrics for each final model are pooled together in the files 01_ModelStatistics.xlsx, 01_ModelStatistics_Results.csv and 01_ModelStatistics_README.csv (the last two are software-independent duplicates of the contents of the first file). These files also provide information to link the scientific name of a taxon with its abbreviation used in directory and file naming. 

Modelling was generally performed withholding a spatially independent test set (ca. 25% of both presence and background locations, both within the same spatial folds) before fitting models with four spatial-block cross-validation. However, in some cases, this was not possible due to the low number of observations. This is indicated in the statistical evaluation tables (01_ModelStatistics.xlsx, 01_ModelStatistics_Results.csv, and 01_ModelStatistics_README.csv).

Files:

Each archive contains species-specific directories with five files:

• HSmap_[taxon code].tif: cloglog-transformed {maxnet} habitat suitability map.
• thresholds_[taxon code].csv: table of the most popular thresholds used to stretch or binarise habitat suitability maps (HSmap_*).
• HSstretch_[taxonomic group]_[taxon code and model version].tif: habitat suitability maps version stretched so that "maximum training sensitivity and specificity" threshold is at 0.5, while keeping 0 and 1 stable.
• PicMargResp_[taxon code].png: marginalised for arithmetic mean response curves illustrating the effect of each ecogeographical variable on the habitat suitability when all the other variables have their arithmetic mean values. The y-axis shows the predicted habitat suitability (cloglog-transformed), while the x-axis shows the standardised (centred and scaled) value of the ecogeographical variable. Dots at a value of 1 on the y-axis illustrate presence locations in the environment (on the x-axis), while dots at a value of 0 illustrate background locations used in modelling.
• PicVarImpVIFs_[taxon code].png: illustrates the permutation-importance (mean and standard deviation from 99 permutations, on the x-axis) of the ecogeographical variables (first y-axis, left side), and their variance inflation factor values (second y-axis, right side).

Geographic projection:

The habitat suitability maps are provided in LKS-92 / Latvia TM (default for all layers in HiQBioDiv).

Interactive viewing:

Habitat suitability maps (HSstretch_*) can also be interactively viewed.