日本地球惑星科学連合2021年大会

講演情報

[E] 口頭発表

セッション記号 A (大気水圏科学) » A-AS 大気科学・気象学・大気環境

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

2021年6月4日(金) 13:45 〜 15:15 Ch.10 (Zoom会場10)

コンビーナ:Jayanthi Venkata Ratnam(Application Laboratory, JAMSTEC)、Rajib Maity(Indian Institute of Technology Kharagpur)、Behera Swadhin(Climate Variation Predictability and Applicability Research Group, Application Laboratory, JAMSTEC, 3173-25 Showa-machi, Yokohama 236-0001)、土井 威志(JAMSTEC)、座長:Swadhin Behera(Climate Variation Predictability and Applicability Research Group, Application Laboratory, JAMSTEC, 3173-25 Showa-machi, Yokohama 236-0001)、土井 威志(JAMSTEC)、Venkata Ratnam Jayanthi(Application Laboratory, JAMSTEC)

14:30 〜 14:45

[AAS04-04] An early warning malaria prediction system based on climate predictors using machine learning

*Patrick Martineau1,2、Swadhin Behera1、Venkata Ratnam Jayanthi1、Masami Nonaka1、Qavanisi E. Mabunda3、Philip Kruger3、Noboru Minakawa4 (1.Japan Agency for Marine-Earth Science and Technology, Yokohama, Japan、2.Research Center for Advanced Science and Technology, The University of Tokyo, Japan、3.Malaria Control Programme, Limpopo Department of Health, Tzaneen, South Africa、4.Institute of Tropical Medicine, Nagasaki University, Japan)

キーワード:Malaria prediction, Machine learning, Climate variability

Malaria outbreaks remain a serious issue in sub-Saharan African countries and early warning predictions of malaria incidence have the potential to reduce the burden of malaria through the planning of prevention interventions. Among useful predictors, both local weather variability (precipitation, temperature) and global modes of climate variability (sea surface temperature variability over the tropical Pacific and Indian oceans) have been previously identified. Based on these predictors, an early warning malaria prediction framework utilizing machine learning techniques is developed. To produce forecasts of malaria incidence with lead times ranging from 3 months to two years, a set of machine-learning classifiers are trained on 23 years of historical global climate and South African malaria incidence data. Through hindcast experiments, we find that the prediction accuracy is adequate during the austral summer when malaria frequency is largest, with an accuracy of 80% up to a year ahead. Among the identified predictors, sea surface temperature variability associated with the El Niño–Southern Oscillation and the Indian Ocean Subtropical Dipole are the most important sources of predictability of malaria incidence.