Extended Data Fig. 2: Performance of machine learning models used to predict plant sodium (Na) concentrations across sub-Saharan Africa.
From: Sodium constraints on megaherbivore communities in Africa
Root mean squared error (RMSE) and R2 model performance metrics of the buffered leave-one-out cross validation (BLOOCV) for (a, b) median and (c, d) 95th percentile foliar Na concentration prediction models. The black circles above the first graph represents the average number of training points removed from within each buffer distance. CUBIST = cubist model; GBM = Gradient boosted model; NNET = neural network model; RF = random forest model; SPATIAL = kernel smoothing model; SSA = sea salt aerosol model; SVM = support vector machine model. e) Spatial mapping of distance to the nearest training point across sub-Saharan Africa. Contour lines represent 300 km intervals. Map elements created with Natural Earth (https://www.naturalearthdata.com).
