Cloud-based applications for accessing satellite Earth observations to support malaria early warning

Climate change significantly impacts human health, particularly infectious diseases like malaria. Variations in temperature and precipitation affect mosquito vectors and parasite transmission. Satellite Earth observations provide valuable environmental data for studying these relationships and developing early warning systems. However, accessing, processing, and analyzing these large datasets requires significant expertise, computational resources, and reliable internet access, especially in low- and middle-income countries. This paper focuses on developing a user-friendly tool to overcome these challenges and facilitate the use of satellite data for malaria forecasting in Ethiopia. The study highlights the need for readily accessible climate data to support national-level malaria early warning systems in Ethiopia and other similar regions. Existing barriers to data access include the lack of expertise in satellite data handling, limited computational resources, and unreliable internet access. Cloud computing, specifically Google Earth Engine (GEE), offers a solution by providing access to geospatial datasets, processing power, and software as virtualized resources over the internet. GEE facilitates data access, processing, and summarization, making it ideal for users with limited resources in low-bandwidth environments.

Existing literature extensively documents the impacts of climate change on human health, specifically focusing on heat-related illnesses, infectious diseases, and waterborne diseases. Research emphasizes the strong correlation between climate variations and the transmission cycles of many infectious diseases, including malaria. Satellite remote sensing has emerged as a powerful tool for studying environmental health risks, providing valuable data on vegetation, surface water, soil moisture, temperature, and precipitation. Studies have effectively used satellite data for various applications, such as mapping heat waves, monitoring air pollution, and delineating populations at risk of disease. However, the challenges of data access, processing, and analysis remain significant hurdles, particularly in resource-limited settings. Previous work highlighted the need for user-friendly tools to overcome these barriers and facilitate the integration of satellite data into public health applications.

The study developed REACH (Retrieving Environmental Analytics for Climate and Health), a GEE-based application designed to automate data processing and provide summarized data for malaria early warning in Ethiopia. REACH uses data from three sources: Global Precipitation Mission (GPM) for precipitation, MODIS Terra for land surface temperature (LST) and spectral indices (NDVI, SAVI, EVI, NDWI), and the Ethiopian Mapping Agency for woreda boundaries. The application processes the data to generate daily summaries of environmental variables for each woreda in Ethiopia. Cloudy pixels are removed, and the number of cloud-free pixels is reported. The data are then summarized as zonal means for each woreda and downloaded as CSV files. Three different implementations were developed: a JavaScript API implementation using the GEE code editor, a standalone Earth Engine App, and a Python package. The JavaScript API implementation provides a user interface for selecting dates and downloading data. The Earth Engine App offers a publicly accessible web application with a similar user interface, but with limitations on large data requests. The Python package allows for automated data processing and downloading, controlled by a Python script. The data generated by REACH are used as inputs to the EPIDEMIA malaria early warning system, which combines them with malaria surveillance data to generate forecasts of future malaria burden.

REACH successfully addressed the challenges of accessing and processing large satellite datasets in Ethiopia. The application simplified data access by providing daily summaries of key environmental variables for each woreda, significantly reducing data volume. The three implementations of REACH—JavaScript API, Earth Engine App, and Python package—each offered unique strengths and weaknesses. The JavaScript API offered flexibility but required a GEE account and some programming knowledge. The Earth Engine App provided greater accessibility but faced limitations with large data requests. The Python package allowed for automation but increased system complexity. The study demonstrated the feasibility of using a cloud-based approach for providing timely access to climate data to support malaria forecasting. The use of REACH within the EPIDEMIA system in the Amhara region of Ethiopia resulted in timely data access and forecast generation, meeting the goal of obtaining data and generating a forecast within one hour. The generated data were harmonized with malaria surveillance data and incorporated into routine reports, which included district-level charts displaying climate and malaria patterns and forecasts. Maps visualized malaria incidence and environmental variables, highlighting areas with high malaria burden and unusual climate anomalies. The study generated a 20-year archive of historical woreda-level data summaries (2002-2021), significantly enhancing the capabilities of malaria early warning systems.

The success of REACH in the Amhara region demonstrates the effectiveness of a cloud-based approach to overcoming barriers to environmental data access for malaria early warning. The application's ability to provide readily usable data at the district level significantly improved the operational efficiency of the EPIDEMIA system. The choice of implementation (JavaScript API, Earth Engine App, or Python package) depends on the specific needs and technical capabilities of the users. While the Earth Engine App provided the most sustainable approach for routine data access in the Ethiopian public health agencies, the other implementations offer flexibility for researchers and developers. Cloud-based approaches, such as that employed by REACH, are highly effective for providing timely access to climate data at regional and national levels, contributing significantly to improving climate-sensitive disease forecasting.

The REACH application successfully provided a user-friendly and efficient method for accessing and processing satellite-derived environmental data to support malaria early warning in Ethiopia. The cloud-based approach proved highly effective in overcoming the limitations of internet connectivity and computational resources. The development of three different implementations caters to diverse user needs and technical capabilities. The successful pilot implementation highlights the potential for scaling up such cloud-based systems for climate-sensitive disease forecasting in other regions. Future research could focus on expanding the application to incorporate additional environmental datasets and refining the data processing algorithms to further improve accuracy and efficiency.

The primary limitation of the study was the reliance on satellite data, which can be affected by cloud cover, resulting in missing data. While the study addressed this by removing cloudy pixels and reporting the number of cloud-free pixels, this still results in incomplete data. Future improvements could incorporate techniques for data imputation to minimize the impact of missing data. The study focused primarily on the Amhara region of Ethiopia. Further research is needed to evaluate the generalizability of the findings to other regions with varying climatic and epidemiological characteristics. The success of the system also depends on the reliability of internet connectivity, which can be a significant challenge in some regions of Ethiopia.