Abstract
Rural tropical regions face escalating threats from zoonotic AIV and dengue virus but lack sewered infrastructure for conventional wastewater surveillance. We implemented surface water-based surveillance (SWBS) in peri-urban Dhaka (Bangladesh) and Ruili (China) from July to November 2023 and coupled it with machine learning-enhanced digital epidemiology. Reverse transcription quantitative PCR (RT-qPCR) was employed to detect the M gene of AIV and to subtype H1, H5, H7, H9, and H10 in surface water. Wild bird feces (n = 40) were collected within 3 km of positive sites to source-track AIV. For the dengue virus, a serogroup-specific RT-qPCR assay targeting the CprM gene was used. Genomic sequencing of AIV and dengue virus was performed to elucidate phylogenetic relationships with local clinical strains. Clinical data related to dengue fever were also collected for correlation analysis. Meanwhile, 13 dengue-related keyword search volumes were harvested daily from Google, Bing and Baidu for four cities to reveal the relationship between dengue epidemics and the web search index. AIV H5 was detected in Dhaka city from week 38, peaking at week 39, while dengue virus was persistently detected from week 29 to week 45, aligning with clinical trends. Time-series cross-correlation analysis revealed that variations in surface water viral load led clinical case reports by approximately two weeks (max CCF = 0.572 at lag -2). In Ruili city, dengue virus was detected from week 32 to week 44. To sharpen sensitivity, 383 weekly web search series for 13 dengue keywords from four countries were screened; random-forest and XGBoost models retained five symptom queries that generated a composite index explaining 79% of variance in dengue RNA levels in an independent Ruili test set (n = 24) and reduced superfluous sampling by 35%. Phylogenetic analysis verified identity between water-derived and patient-derived DENV-2, confirming local transmission. The study demonstrates that AIV SWBS is optimally integrated with wild bird sampling for source attribution, whereas dengue SWBS achieves maximal efficiency when combined with real-time web search monitoring, providing tailored, low-cost early-warning modules for resource-constrained tropical settings.