Merge pull request #5 from alexsla/master

New stuff in info

modelling_info.Rmd

@@ -35,15 +35,15 @@ To model species presence-only data with statistical models a random sample of b
 
 For each species we created a spatial kernel density estimate around its occurrence records, projected onto a raster masked to only include countries in which the species is known to occur. We then generated 1000 background samples for each species by randomly sampling from the masked raster, with the sampling probability for each cell weighted by the spatial kernel density estimate.
 
-**PLOT TBA**
+![](methodology.png)
 
 ### Species distribution modelling
 
 There are only few records per species, so we fitted a logistic binary regressions using a Hierarchical Generalized Additive Model (**HGAM**; Pedersen et al. 2019) with Point Process weighting (Fithian & Hastie 2013) to increase the predictive power of the model.
 
-We downloaded global layers of 19 bioclimatic variables representing the mean, extremes, and variation in temperature and precipitation from WorldClim v2.1 (Fick et al. 2017), and a layer of the Landsat Enhanced Vegetation Index (EVI; Masek et al. 2006, Vermote et al. 2016). We then used Principal Coordinate Analysis (PCA) to convert the layers into orthogonal Principal Components (PCs) to reduce collinearity in the predictor variables. We used the 4 first PCs, which collectively explain **XX%** of the variation in the bioclimatic variables, as predictor variables for the HGAM model.
+We downloaded global layers of 19 bioclimatic variables representing the mean, extremes, and variation in temperature and precipitation from WorldClim v2.1 (Fick et al. 2017), and a layer of the Landsat Enhanced Vegetation Index (EVI; Masek et al. 2006, Vermote et al. 2016). We then used Principal Coordinate Analysis (PCA) to convert the layers into orthogonal Principal Components (PCs) to reduce collinearity in the predictor variables. We used the 4 first PCs, which collectively explain 91.8% of the variation in the bioclimatic variables, as predictor variables for the HGAM model.
 
-The model was trained on a random sample of 80% of the data, and model performance was evaluated by making predictions on the remaining 20% and calculating the area under the receiver operating characteristic curve (AUC). The model achieved good predictive performance, with an AUC score of **XX**.
+The model was trained on a random sample of 80% of the data, and model performance was evaluated by making predictions on the remaining 20% and calculating the area under the receiver operating characteristic curve (AUC). The model achieved good predictive performance, with an AUC score of 0.85.
 
 We then used to model to make predictions for all cells in Australia and projected the resultant habitat suitability scores onto a map of Australia.