A Machine Learning-Based Dynamic SST Index for Long-Lead Malaria Prediction in the Peruvian Amazon.

Malaria imposes a major health burden in the Peruvian Amazon, and its early warning is essential for effective disease prevention. The tropical sea surface temperature (SST) variability, fundamentally shaping the global weather patterns, may also alter malaria transmission and potentially improve its long-lead predictability. In this study, we propose a machine learning-based methodology that leverages comprehensive tropical SST variability for malaria prediction in the Peruvian Amazon. First, we demonstrate that significant correlations broadly exist between tropical SST anomalies and Peruvian malaria occurrence across different seasons and time lags, confirming the potential predictability from the tropical ocean. Then, we apply the self-organizing map to synthesize the spatiotemporally varying SST-malaria relationship and identify a unique dynamic SST index for Peruvian malaria. The dynamic SST index provides better performance (higher correlation coefficients and lower root mean square errors) in the generalized linear model, compared to the traditional El Niño-Southern Oscillation (ENSO) index, with lead times exceeding 3 months. Furthermore, the dynamic SST index captures the evolution of the ENSO life cycle from its precursor climate mode (Pacific Meridional Mode) and appears to influence Peruvian malaria by altering the local near-surface air temperature and specific humidity. Such underlying mechanisms provide the physically plausible basis for the long-lead predictability of Peruvian malaria using a machine learning-based remote predictor. Last but not least, we provide open-source code for broad applications in linking tropical SST variability and vector-borne disease transmission, or other climate-sensitive socioeconomic issues.

Duke Scholars

Author Shineng Hu Earth and Climate Sciences

Author Mark Janko Duke Global Health Institute