Predicting Vector-Borne Disease Risk using Earth Observation and Machine Learning: A Case Study in northern Italy

Floodings exemplify the interconnection between climate change, environmental exposures, and human health. They are often characterized by the presence of stagnant water, which makes the habitat particularly favourable for the proliferation of vectors of arboviruses in during their reproductivity seasons. This poses significant threats to public health, because the geographical expansion of these vectors is responsible of an increase of the diffusion of imported infectious diseases such as dengue and chikungunya, together with other arbovirosis like West Nile, Usutu, Toscana virus infections and tick-borne encephalitis, which are endemic in Italy. This diffusion requires proactive monitoring and mitigation strategies. The monitoring of the distribution of these vectors is usually performed by installing attractive traps in the territory. However, the sites of these traps cannot be uniformly distributed over the territory. Therefore, it is useful to support them with other warning methods to identify areas with the ideal characteristics of ecological niches for these insects and thus at risk of becoming outbreaks for arbovirosis. 

The EASTERN project focuses on both direct and indirect consequences of flooding, by exploiting Earth Observation (EO) and meteorological data to implement Machine Learning (ML) models able to predict flood-related risks. One of the project’s use cases is dedicated to the implementation of ML-based predictive tools to identify areas suitable for vector proliferation, using meteorological parameters and satellite imagery.  

The meteorological parameters considered are humidity, temperature, wind speed and rain, which are known in literature as correlated with vector spreading. From optical imagery (Sentinel-2 constellation) ecological indexes like Normalized Difference Vegetation Index (NDVI) and Normalized Difference Moisture Index (NDMI) are retrieved. Entomological data were collected by IPLA S.p.A. The species of mosquitos that have been considered are Aedes caspius and Culex pipiens. Around 50 trap sites located in the Piedmont region have been monitored every two weeks from June to October. Data used for model training are referred to years from 2017 to 2023. 

The amount of collected mosquitos for each species has been divided into classes. Separated predictive models have been trained for each species. The dataset is highly unbalanced. Since most of the collected data have values proximal to 0 and only few sites collect up to thousands of vectors, the effect of the imbalance has need neutralized. For both species, temperature, NDMI, NDVI, wind speed and humidity are the predictors with the highest feature importance for this model. 

The synergy between satellite imagery, meteorological data and ML models, can be considered a promising tool to monitor vectors’ populations and assess associated health risks, enabling targeted interventions and strategic placement of monitoring traps. Our approach addresses the gaps in traditional monitoring methods, particularly in data-limited regions, and will be useful to provide risk maps and early warnings in case of flooding, crucial for informed decision-making. 
 
EASTERN project received funding from Cascade funding calls of NODES Program, supported by MUR - M4C2 1.5 of PNRR funded by the EU - NextGenerationEU (Grant ECS00000036)