GLOBE Projects

GLOBE Side Navigation

A Random Forest Analysis of Remote Sensing Driven Mosquito Habitat Prediction in West Africa

Student(s):Ruchi Bondre, Ryan Chan, Calista Huang, Andrew Liu, Neil Sangra
Grade Level:Secondary School (grades 9-12, ages 14-18)
GLOBE Educator(s):Cassie Soeffing
Contributors:Dr. Russanne Low, SME, IGES. Peder Nelson, SME, OSU. Andrew Clark, SME, IGES. Dr. Erika Podest, SME, NASA JPL
Report Type(s):International Virtual Science Symposium Report, Mission Mosquito Report
Protocols:Land Cover Classification, Earth As a System, Mosquitoes
Presentation Video: View Video
Presentation Poster: View Document
Language(s):English
Date Submitted:01/18/2024
Mosquito vector-borne diseases, such as Dengue, West Nile Virus, Malaria, and Zika, pose significant global health risks for upwards of 3.9 billion people. An essential component to limiting the spread of mosquito vector-borne disease and assessing disease risk is the prediction of mosquito abundance. Given the severity of mosquito-borne diseases, there is a need for intelligent and automated mosquito abundance forecasting models. Such models would empower government and healthcare authorities to proactively address the mosquito threat and establish long-term disease prevention strategies. This study proposes the implementation of random forest models to predict mosquito larvae abundance in West Africa, suitable for forecasting future mosquito vector-borne disease outbreaks. Our models leverage remote sensing satellite data to extract features including normalized difference vegetation index (NDVI), average rainfall, temperature, humidity, and sporadically recorded GLOBE Mosquito Habitat Mapper (MHM) citizen science data to develop accurate predictions of mosquito population densities. We performed a comparative analysis of random forest classifiers and random forest regressors for the prediction of mosquito larvae counts as categories or numerical values, and determined both models to offer practical benefits for real-world implementation in mosquito habitat forecasting. The outcomes of our research indicate that random forest classifiers exhibit strong viability for predicting mosquito habitats and larvae abundance, achieving an accuracy of over 85%. Whether applied to a classification task or regression task, our work demonstrates the ability of random forest machine learning models to effectively identify correlations between environmental variables and mosquito population characteristics to predict mosquito abundance with high accuracy. In doing so, our research underscores the utility of remote sensing data and machine learning models for real-world mosquito threat management. Moreover, our results provide valuable insights for future research to address mosquito-borne disease prevention by targeting other areas or developing mosquito surveillance systems. Keywords: remote sensing, random forest, mosquito abundance, NASA, GLOBE, NDVI, citizen science, mosquito habitat, mosquito-borne disease



Comments