Sorghum bicolor (L.) Moench, called sweet sorghum, is a drought-resistant and heat-tolerant plant used for ethanol bioenergy production, and is able to reduce the competition between growing crops for energy vs. growing crops for food. Quantitatively mapping the marginal lands of sweet sorghum is essential for the development of sorghum-based fuel ethanol production. However, knowledge of the contemporary marginal lands of sweet sorghum remains incomplete, and usually relies on sample data or is evaluated at a national or regional scale based on established rules. In this study, a novel method was demonstrated for mapping the global marginal lands of sweet sorghum based on a machine learning model. The total amount of global marginal lands suitable for sweet sorghum is 4802.21 million hectares. The model was applied to training and validation samples, and achieved high predictive performance, with the area under the receiver operating characteristic (ROC) curve (AUC) values of 0.984 and 0.978, respectively. In addition, the results illustrate that maximum annual temperature contributes more than do other variables to the predicted distribution of sweet sorghum and has a contribution rate of 40.2%.
Information-optimal Abstaining for Reliable Classification of Building Functions
