[A-AS04] Machine Learning Techniques in Weather, Climate, Hydrology and Disease Predictions

[AAS04-08] Global seawater desalination plants prediction using species distribution models

*Zhipin Ai¹, Naota Hanasaki¹, Fumiko Ishihama¹ (1.National Institute for Environmental Studies)

Keywords:seawater desalination plant, generalized linear model, generalized additive model, random forest, support vector machine

Desalinated seawater gradually becomes a vital source of freshwater supply for coastal water scarcity regions. Predicting global geographical distribution of seawater desalination plants is crucial for integrated global water resources assessment. Here, we investigated the possibility of applying different species distribution models (SDMs) (widely used in the field of biology and ecology) to predict the spatial distribution of seawater desalination plants (by 2014) globally for the first time. Four statistical models (generalized linear model, generalized additive model, random forest, and support vector machine) were trained and tested with the cross-validation method at a spatial resolution of 0.5 degree. Results showed that random forest was found to have the best performance in terms of both AUC value and the correlation coefficient. To combine the unique feature of each individual prediction, we finally made an ensemble prediction map by averaging each individual prediction with consideration of the thresholds that determined by maximizing the sum of sensitivity and specificity. We also discussed the prediction uncertainties from the view of AUC, threshold values, and the number of training plants.