Hydrobasin level 7 units offer superior predictive accuracy compared to traditional administrative borders when predicting malaria cases in Loreto, Peru, according to a retrospective ecological study. Across multiple modeling approaches, spatial units designated as hydrobasin level 7 consistently demonstrated superior predictive accuracy and produced the lowest prediction errors for regional malaria cases. Authored by Antony Barja, Edson J. Ascencio, Bryan Fernandez-Camacho, Luis C. Revilla, Jazmin Qquellon, and Gabriel Carrasco-Escobar, the study suggests that the choice of spatial unit affects predictive accuracy, with hydrobasin level 7 producing the lowest prediction errors.
Within the analyzed dataset, Plasmodium vivax accounted for 79.48 percent of all cases, while Plasmodium falciparum represented 20.52 percent. Researchers derived environmental and meteorological variables from satellite data and incorporated them into Zero-Inflated Negative Binomial (ZINB) regression models and machine learning algorithms, including Random Forest, Support Vector Machine, and XGBoost. Predictive performance across these spatial aggregation levels was evaluated using Root Mean Squared Error (RMSE) and Akaike Information Criteria (AIC).
For Plasmodium falciparum, hydrobasin level 7 consistently produced the lowest prediction errors across all modeling approaches. Among the machine learning models, XGBoost achieved a Root Mean Squared Error of 15.79, with a 95 percent confidence interval of 10.42 to 21.03. In comparison, the corresponding ZINB-PCA model at hydrobasin level 7 achieved an RMSE of 39.19 (95% CI: 17.09 to 62.24). Shifting to traditional administrative borders or alternative hydrological boundaries resulted in substantially higher prediction errors: district-based ZINB-PCA models yielded an RMSE of 50.89 (95% CI: 23.65 to 77.87), while ANA-based ZINB-PCA models produced an RMSE of 59.22 (95% CI: 33.26 to 82.21).
A similar pattern was observed for Plasmodium vivax, with hydrobasin level 7 again demonstrating superior predictive accuracy. For this species, XGBoost achieved an RMSE of 38.44 (95% CI: 26.90 to 49.19), whereas the corresponding ZINB-PCA model at hydrobasin level 7 achieved an RMSE of 63.87 (95% CI: 46.38 to 80.95). In contrast, models based on alternative spatial units generated markedly higher errors under the ZINB-PCA framework: district-based models produced an RMSE of 102.81 (95% CI: 73.15 to 131.77), hydrobasin level 6 produced an RMSE of 175.87 (95% CI: 116.08 to 232.61), and ANA models produced an RMSE of 263.78 (95% CI: 144.98 to 389.87).
The authors emphasized that the associations captured by these modeling frameworks represent predictive relationships rather than causal effects. Integrating hydrologically defined boundaries into surveillance systems may enhance malaria control strategies in environmentally complex regions.
Subject of Research: Predictive accuracy of watershed-based spatial units versus administrative boundaries for malaria surveillance in the Peruvian Amazon
Article Title: Comparing watershed-based spatial units and administrative boundaries for malaria prediction in the Peruvian Amazon
Article References: Barja, A., Ascencio, E. J., Fernandez-Camacho, B., Revilla, L. C., Qquellon, J., & Carrasco-Escobar, G. (2026). Comparing watershed-based spatial units and administrative boundaries for malaria prediction in the Peruvian Amazon. PLOS Global Public Health, 6(11), e0007553. https://doi.org/10.1371/journal.pgph.0007553
Image Credits: AI Generated
DOI: 10.1371/journal.pgph.0007553
Keywords: Malaria, Peruvian Amazon, Loreto, Plasmodium vivax, Plasmodium falciparum, Hydrobasins, Watershed-based spatial units, Administrative boundaries, Disease surveillance, Retrospective ecological analysis, Zero-Inflated Negative Binomial regression, Machine learning, Random Forest, Support Vector Machine, XGBoost, Root Mean Squared Error, Akaike Information Criteria, Satellite environmental data
News Source: Ophelia Keating. (November 11, 2026). Hydrobasin Level 7 Produces Lowest Malaria Prediction Errors in Loreto, Peru. Scienmag.



