Modelling drought-tolerant Sorghum bicolor distributions in Eastern Indonesia using machine learning approaches

Abstract

Abstract. Utomo SW, Wibowo AA, Lestari F. 2025. Modelling drought-tolerant Sorghum bicolor distributions in Eastern Indonesia using machine learning approaches. Intl J Trop Drylands 9: 99-110. Tropical arid ecosystems require alternative species that are drought-tolerant. Sorghum bicolor (L.) Moench has been considered an alternative, drought-tolerant species. Despite its resistance to drought, information on the potential distribution of sorghum is very limited. This information is very important, especially in arid eastern Indonesia, where sorghum is nominated as an alternative species to sustain food security. This study aims to model the potential distribution of S. bicolor using machine learning (random forest), geoclimate (Bioclim), and statistical methods (GAM/GLM) on five arid islands, including Lombok, Sumba, Sumbawa, Flores, and the Timor Islands, Indonesia. Area Under Curve (AUC) was used to evaluate model performance. In general, all the models confirm that Timor, followed by Sumbawa and the Flores Islands, have large, suitable areas for sorghum. It is estimated that up to 99.71% of arid island ecosystems in eastern Indonesia are suitable for sorghum. The geoclimate and machine learning models generated the highest values for AUC in comparison to statistical methods in which the Domain model is 0.962, SVM is 0.903, and both GLM and GAM are 0.894. It is important for plant cultivation planning to consider species distribution modeling and not rely on any single modeling method. The plant cultivation should evaluate the performance of all available models for their crops and area of interest, and select the best representative methods to develop an accurate and representative sorghum crop distribution model.