Skip to content

README.md

Latest commit

 

History

History
54 lines (38 loc) · 3.31 KB

README.md

File metadata and controls

54 lines (38 loc) · 3.31 KB
Raw

Rethinking Spatial Units for Malaria Prediction in Loreto, Peru: The Use of Watershed-Based Boundaries in the Amazon Rainforest

Description

Malaria remains a major public health threat in the Peruvian Amazon, where Loreto accounts for more than 76% of national cases. Most spatial modelling studies rely on administrative boundaries (districts) that follow arbitrary political divisions, poorly reflecting the ecological dynamics that drive transmission. This study evaluates whether watershed-based spatial units (hydrobasins) outperform traditional administrative boundaries for predicting Plasmodium vivax and Plasmodium falciparum cases in Loreto from 2009 to 2018. Four boundary types are compared: administrative districts and three hydrological levels (HydroSHEDS L6, L7, and ANA watersheds). Modelling combines Zero-Inflated Negative Binomial (ZINB) regression with three machine learning algorithms — Random Forest, Support Vector Machine, and XGBoost — using PCA-derived climatic, hydrological, and vegetation predictors from satellite imagery.

Key Results

Hydrobasin level 7 (78 units) consistently achieved the lowest prediction error across all models and both malaria species.

P. falciparum — RMSE by boundary and model

Boundary XGBoost Random Forest SVM ZINB
Hydrobasin L7 15.79 17.21 19.42 24.09
Districts 17.10 17.32 26.26 36.99
Hydrobasin ANA 33.15 38.00 45.58 138.61
Hydrobasin L6 43.06 39.96 56.80 90.67

P. vivax — RMSE by boundary and model

Boundary XGBoost Random Forest SVM ZINB
Hydrobasin L7 38.44 40.98 46.24 77.08
Districts 50.76 51.84 76.27 97.34
Hydrobasin ANA 68.99 79.10 95.32 215.61
Hydrobasin L6 118.93 122.48 148.68 229.05

For P. falciparum, hydrobasin L7 reduced RMSE by 24–68% relative to district and ANA boundaries. For P. vivax, reductions ranged from 28–74%. Spatiotemporal cross-validation (2009–2014 train / 2015–2018 test) confirmed these patterns, with lowest temporal RMSE at hydrobasin L7 for both species (28.67 and 73.85 respectively).

Spatial distribution of residuals

Residual values (observed − predicted cases per 100 inhabitants) mapped across boundary types and models. Negative values indicate overestimation; positive values indicate underestimation.

Residual map

  • Hydrobasin L7 shows the most spatially balanced residuals for both species, with moderate over/underestimation distributed across the region
  • ANA and District boundaries exhibit large overestimation zones (dark red) concentrated in the northwest, particularly for P. vivax across all three models
  • Hydrobasin L6 shows predominantly mild residuals but loses spatial detail due to its low number of units (N = 13)
  • P. vivax residuals are consistently larger in magnitude than P. falciparum across all boundaries, consistent with its higher prediction error in RMSE tables

Data Availability

Malaria surveillance data from the Peruvian Ministry of Health are not publicly available. Access to anonymised data may be requested subject to institutional approval. Contact: imtavh.innovalab@oficinas-upch.pe