Eye in the Sky: Predicting Outbreaks from Orbit

Cholera is in its seventh pandemic and due to the magnitude of this problem (7 million people in 50 different countries have so far been affected since the 60s), it is important to be able to forecast outbreaks. This would allow aid to be readied before an outbreak even occurred. However, despite greater advances in our understanding of the disease our ability to predict outbreaks remains limited. Remote sensing perhaps provides the best way of achieving this goal. Many advances have been made since the publication of Lobitz et al. 2000 (see Sam's recent blog post) and a strong link has been established between chlorophyll levels (so plankton blooms) and incidence of cholera. However, recent models have become mired in detail by adding variables such as population immunity, which are hard to quantify over large scales. While ignoring the significance of plankton variability over time and space, more important when providing broad scale predications.

Jutla et al. took an exclusively macro-scale approach in order to construct models linking coastal chlorophyll levels (measured using data from the SeaWiFS satellite) and river discharge into the Bay of Bengal to predict spring outbreaks in the Bengal Delta, Bangladesh. Innovatively, as data for river discharge cannot be remote sensed (yet), they used air temperature for the Himalayas instead. Warming in this region is negatively correlated with spring cholera outbreaks as it signals elevated river discharge, which flushes plankton away from the river delta. After running a series of different models over on an eleven year data set they found that models which predicted cholera outbreaks two months in advance worked best (accuracy of 82%). This model effectively links autumn chlorophyll level and river discharge to cholera outbreaks the next spring. The authors believe that the reason for this is that a large phytoplankton bloom and increase in zooplankton in autumn (indicated by elevated chlorophyll) leads to a large pool of in winter CDOM after collapse and degradation. CDOM or coloured dissolved organic matter, is strongly correlated with autumn chlorophyll levels and spring cholera levels. This could allow Vibrio cholerae to persist and multiply as indicated by laboratory evidence. Low levels of river discharge during this period would allow CDOM and V. cholerae to infiltrate estuaries and cause an outbreak after winter had ended.

Additionally there is an autumn cholera outbreak that remarkably has a positive relationship with spring outbreak of that year. The authors presumed this is because cholera introduced into estuaries during spring is being spread across the landscape by monsoon floods, contaminating water and sanitation infrastructure. This allows a reasonable prediction of the autumn outbreak to be made from the spring outbreak, although the addition of summer sea surface temperature foes improve accuracy. Amazingly, linking the two models allows an outbreak to be predicted a year in advance. This is because the chance of a spring outbreak, calculated from the previous year's autumn chlorophyll level is used to predict that following year’s autumn outbreak chance.

Overall, I found this paper an excellent example of the use of macroscale approach to tackle problems of a microscopic origin. However, the authors admit these are not without caveats. For example: the societal role of transmission is not being factored in and the comparatively short data set used. Despite this, I think that the output of these models could be fed into models for specific local areas which could incorporate more assumptions and would improve predictions. 

Reference: Jutla, A.S., Akanda, A.S. and Islam, S. (2013). A framework for predicting endemic cholera using satellite derived environmental determinants. Environmental Modelling & Software, 47, 148-158.