A Two-Steps Remote Sensing and Machine Learning Framework for Predicting Malaria Prone Areas by Mapping Anopheles Larval Habitats Publisher



Rezvan H ; Valadan Zoej MJ ; Hassanpour G ; Youssefi F ; Hanafibojd AA ; Ghaderpour E
Source: Earth Systems and Environment Published:2025

Abstract

Malaria is a serious and life-threatening infectious disease transmitted by female Anopheles mosquitoes. It poses a significant global health and security concern. Preventing the disease is far more cost-effective than attempting to control it after outbreaks have occurred. Since the Plasmodium parasite develops within Anopheles mosquitoes, areas where these vectors thrive are critical high-risk regions for public health. Therefore, mapping and predicting larval habitats are crucial for developing effective early warning systems and vector control, which facilitate timely epidemic detection. Previous early warning systems that relied on remote sensing faced difficulties in achieving comparable accuracy and coverage across the study region. To address this challenge efficiently, the present study proposes a two-stage monitoring approach. Firstly, medium-resolution optical and synthetic aperture radar (SAR) data were utilized to identify key areas across a broad region, focusing on residential zones, their 2 km buffers, and temporary and permanent water bodies. The second stage utilized high-resolution data to pinpoint mosquito habitats within the province’s highest-risk areas. A key contribution of this research is the ranking of indicators using the Analytic Hierarchy Process (AHP) model, which assesses key factors such as depression locations, soil type, precipitation, temperature, and a novel feature based on mosquito flight altitude. Among these, depression mapping emerged as the most influential factor in habitat formation. Risk maps were created using fuzzy, MaxEnt, multilayer perceptron (MLP), and random forest (RF) models, with accuracy validated through Google Earth-derived test points and field visits. The fuzzy model demonstrated the highest accuracy, correctly predicting all habitat presence points, while MLP achieved the highest area under the curve score of 94%. The findings offer valuable insights for health authorities, enabling the development of precise vector control strategies to mitigate disease transmission and minimize public health impacts. © The Author(s) 2025.