Outbreaks of many infectious diseases, including malaria, are often linked to climate variations and land use changes that affect the life cycles of vectors, hosts, and pathogens. Early warning of the timing and locations of epidemics can facilitate more effective targeting of resources for disease prevention, control, and treatment. However, these predictions must be accurate to ensure that potential outbreaks are not missed and resources are not wasted responding to predicted outbreaks that do not occur. Our research involves the development of novel informatics systems to acquire and harmonize environmental and epidemiological data from multiple sources, and the analysis of these data to identify the environmental drivers of malaria outbreaks and determine the most effective forecasting approaches.

EPIDEMIA System Development

The Epidemic Prognosis Incorporating Disease and Environmental Monitoring for Integrated Assessment (EPIDEMIA) system was developed to link large environmental datasets obtained via satellite remote sensing with malaria surveillance to support outbreak detection and forecasting. A Public Health Interface enables uploading of epidemiological data, downloading of harmonized epidemiological and environmental data, and access to forecasting reports via a password-protected interface. Remote Sensing Data Acquisition and Processing is carried out by the EASTWeb software, described in the next section. The Data Integration subsystem harmonizes multiple environmental and epidemiological datasets, and the Modeling subsystem uses these data to calibrate forecasting models and predict future outbreaks. The Reporting subsystem automatically generates formatted reports that display the predictions using risk maps and control charts. The Public Health Interface uses the PHP programming language for the back end; HTML, Cascading Style Sheets, JavaScript, and JQuery for the front end; and AJAX to handle asynchronous message calls. Database management is implemented in MySQL and data integration, modeling, and reporting are carried out using the R environmental for statistical computing.

Conceptual diagram of information flow through the EPIDEMIA system (Merkord et al. 2017).

EASTWeb Software Development

EASTWeb is an open-source software application that automatically connects to remote sensing data archives and acquires, processes, and summarizes selected datasets for a user-specified geographic location and time period. After building a historical database, EASTWeb continues to search for and acquire new data as they become available in the online archives. Data summaries are stored in a PostgreSQL database in a format that can be easily linked to epidemiological datasets for analysis and forecasting. The software is programmed using JAVA for overall system control and SWT for user interface development. Spatial processing is carried out using the GDAL open source geospatial library. EASTWeb currently supports a variety of plugins that provide access to MODIS land surface temperature, spectral indices derived from MODIS BRDF-adjusted reflectance, and satellite-based precipitation measurements from TRMM and GPM. A more detailed technical description of EASTWeb is provided by Liu et al. (2015).

User Interface for EASTWeb Software Version 2.0.

Epidemiological Studies

Our research addresses fundamental questions about the environmental determinants of malaria in the highlands of Ethiopia. There was significant spatial synchrony in the temporal patterns of malaria epidemics from 2001-2009, suggesting that at least a portion of the interannual variability in malaria risk is driven by large-scale climatic anomalies. (Wimberly et al. 2012a). A time series analysis of monthly variability in malaria cases found lagged associations with satellite-derived metrics of temperature and moisture availability ranging from 1-3 months (Midekisa et al. 2012). Malaria outbreaks during the peak September-December epidemic season were also associated with longer-term effects of precipitation and temperature anomalies from the early peak season in May-June (Midekisa et al. 2015). Satellite imagery and digital elevation models were used to map landscape characteristics hypothesized to influence malaria risk, and the geographic distribution of wetlands was found to be an important factor influencing the spatial pattern of malaria cases throughout the region (Midekisa et al., 2014).

Mike Wimberly and Gabriel Senay conducting field work in Ethiopia.

Disease Forecasting

To produce forecasts, we apply data assimilation techniques that continually validate and update the models as new data are collected. Currently, statistical time-series models are implemented using a state-space framework with the Kalman filter for model updating. Early Detection Models are used to estimate thresholds for outbreak detection based on seasonality and interannual trends. Early Warning Models are used to predict malaria risk in future weeks based on lagged responses to temperature, precipitation, and vegetation anomalies. Because these models are implemented in the operational EPIDEMIA system, we are able to continually assess the accuracy and usefulness of the forecasts by comparing them to new observations. Current research efforts are focused on identifying geographic zones with similar patterns of malaria outbreaks, determining how and why environmental determinants of malaria vary across these zones, and comparing the accuracy of different modeling approaches.

Malaria forecast for Abargelie Woreda, Waghimira Zone, for week 39 of 2016.