Photo credit  | 

Forecasting cholera outbreaks from space

Blog  |  23 August, 2021

Few people associate climate impacts with health and well-being, but scientists are trying to improve understanding of how changes in climate are already affecting human health.

Amy Marie Campbell is a PhD Researcher at National Oceanography Centre with a research background in Earth Observation and environmental science. She spent a year working with the European Space Agency’s Climate Office exploring how Earth Observation (EO) data could be utilised for climate-health applications. She currently researches how climate change will drive the future epidemiology of water-borne diseases at the University of Southampton and CEFAS. Here she shares the results of a recent model that uses satellite data to forecast cholera outbreaks in coastal India.

When most people think of climate change, it is likely they will picture ice caps melting or extreme weather events. Not many people associate the impacts of climate change with health and well-being.

The reality is that the Earth’s changing climate is also affecting our health. The health of our planet and of its people are intrinsically interlaced and so, as the world changes, so too does the dynamic of climate-sensitive diseases.

Scientists are now trying to improve understanding of how changes in climate are already, and will in the future, impact upon human health, including threats from pathogens, viruses and diseases. This is where space comes in – satellites provide us with a range of environmental datasets to allow us to start drawing connections between outbreaks of climate-sensitive infectious disease, such as cholera, and climate anomalies.

Nearly 3 million cases of cholera every year

Around 1.3 billion people are currently at risk of cholera, with an estimated 2.9 million cases annually. On the densely populated coast of India, home to approximately 200 million people, cholera is endemic. This means there is a long history of cholera outbreaks, appearing year after year, and seasonally, which  suggests that there is more at play than just human-to-human transmission.

Studies have discovered that certain environmental variables drive the coastal distribution and seasonal dynamics of the bacteria that causes cholera (Vibrio cholerae). The bacteria exist naturally in coastal waters, however cholera outbreaks are associated with optimum water temperatures and salinities, and the presence of plankton (which the bacteria can latch on to, increasing survival). While it wouldn’t be feasible to collect water samples around the whole of the Indian coast to survey the dynamics of cholera bacteria, satellites monitoring the Earth are providing high-resolution global data on environmental variables that drive the bacteria, daily.

Working with Marie-Fanny Racault, from the Plymouth Marine Laboratory, we set out to explore whether we could forecast cholera outbreaks on the Indian coast, using environmental data, machine learning (ML) and artificial intelligence (AI). We created a model trained on historic environmental data, including satellite data, and previous cholera outbreak data.

The model was able to detect 89.5% of outbreaks in the test dataset. We also found that chlorophyll-a concentration – a pigment marker for phytoplankton presence – as well as salinity and temperature, were the strongest predictors of cholera outbreaks in our dataset.

Historical climate time series data – ESA’s Climate Change Initiative

To predict future environmental cholera outbreaks, we first looked backwards. We need strong retrospective data of environmental trends and anomalies that could potentially be associated with previous cholera outbreaks. These can then be used as training data which is fed into AI algorithms to establish possible relationships between certain environmental conditions and outbreaks.

The datasets that facilitated this analysis and the subsequent predictions were time series data of Essential Climate Variables (ECVs). Coordinated by the Global Climate Observing System (GCOS), ECVs are defined as key indicators that provide the empirical evidence needed to understand climate change and can include terrestrial variables (e.g. Land Temperature, Soil Moisture, Glaciers) and oceanic variables (e.g. Sea Ice, Salinity). The European Space Agency Climate Office, as part of the Climate Change Initiative (ESA-CCI), utilises data from archived satellite missions (including Earth Explorers and the Copernicus Sentinel constellation) to generate continuous, global data records for these key climate components, to be used for further analysis.

ESA’s CCI program follows strict standards that all datasets comply to, allowing them to be used in combination seamlessly. This is critical for an application such as disease forecasting, as the cholera bacteria have complex interactions with a wide range of environmental variables, that also interact amongst themselves, requiring a range of variables for effective analysis.

Figure 1: A confusion matrix showing the results of the model when run on the unseen test dataset. 221 of the 248 outbreaks were correctly identified (true positives), and 2559 non-outbreaks were correctly identified (true negatives). There were 26 false negatives (outbreaks missed) and only 1 false positive (false alarm).

Satellite missions providing capabilities for climate-driven disease forecasting

As mentioned, the CCI data is generated from archived satellite data from Earth Explorers and Copernicus satellites. However, continuous data collection at high-resolution and high-repeat frequency is needed to allow sustained analysis and to top-up these climate variable time series. Such datasets will allow near-real time analysis for future forecasting.

Missions that will facilitate future generations of forecasting information for cholera outbreaks include the recently launched Sentinel-6 altimeter, for topographic coastal measurements, and the upcoming Copernicus high priority candidate missions including CHIME (for coastal mapping), LSTM (improved land surface temperature) and CIMR (sea surface temperature and salinity).

Paramount to our model development is the need for sustained measurements of salinity, plankton presence (chlorophyll-a concentration) and land surface temperature particularly, as these were the variables found to be the best predictors of cholera outbreaks in this application.

Figure 2: Bar graphs showing the model performance score when individual variables were used. (LST= land surface temperature, ChlorA= chlorophyll a concentration, SSS= sea surface salinity, SLA= sea level anomaly and SM= soil moisture). The highest sensitivity and F1 scores were seen for land surface temperature, followed by chlorophyll a concentration and sea surface salinity.

Next steps in climate-driven disease forecasting

Forecasting climate-driven diseases will become increasingly important as environmental anomalies that lead to outbreak risks increase in frequency and severity. Another avenue we are exploring is how environmental factors may be driving genomic evolution of pathogens.

Our study generally focuses on environmental conditions that are suitable for transmission and survival however we need to explore how pathogens themselves adapt to climate change, as this may allow certain disease-causing strains to emerge in new places around the world. This has been seen recently with the seafood poisoning bacteria Vibrio parahaemolyticus, which causes human gastrointestinal illness following ingestion of this pathogen in raw or undercooked seafood. Usually confined to warmer waters, this species of seafood-poisoning bacteria has now been found at high latitudes as north as the Baltic Sea, with particular variants emerging around the globe. Understanding the mechanisms behind this, such as whether certain variants will be more resilient to climate change or respond opportunistically, will help us mitigate the threat of future waterborne disease outbreaks.

Research paper

Campbell, A.M.; Racault, M.-F.; Goult, S.; Laurenson, A. Cholera Risk: A Machine Learning Approach Applied to Essential Climate Variables. Int. J. Environ. Res. Public Health 2020, 17, 9378. doi: 10.3390/ijerph17249378

About Amy Marie Campbell