Integrating Earth Observations to Support Malaria Risk Monitoring in the Amazon

EarthzineEarth Observation, EO for Health, Original

An enhanced EO-derived information system can be applied to operational risk monitoring and early warning in the Amazon.

B.F. Zaitchik1, B.J. Feingold2, D. Valle3, W.K. Pan2

1 Earth and Planetary Sciences, Johns Hopkins University

2Nicholas School of the Environment and the Duke Global Health Initiative, Duke University

3University Program in Ecology, Duke University

Malaria is a vector-borne disease that is endemic in more than 100 countries with more than 3.3 billion people at risk. It is a leading cause of morbidity and mortality worldwide: In 2010, there were an estimated 216 million cases, globally, and 655,000 deaths [1]. In the Americas, more than 95 percent of all reported malaria cases occur in the Amazon [2]. The absolute number of cases in the Amazon has declined in the past decade, due largely to coordinated regional efforts such as ‰ÛÏRoll Back Malaria‰Û and the ‰ÛÏAmazon Malaria Initiative,‰Û which began in 2000. However, following a large-scale flood in 2012 affecting the northwestern Amazon Basin [3], malaria cases have increased by 7,000 in Colombia between 2012 and 2013, and 23,000 in Peru between 2011 and2013 [4], [5]. The disease continues to affect many vulnerable populations throughout the Amazon and it is imperative to identify and validate tools to monitor risk in highly malaria prone regions.

Malaria is the result of infection by Plasmodium parasites, which are transmitted by female mosquitoes of Anopheles species. Both the ecology and transmission dynamics of malaria in the Amazon are complex functions of human activity and the natural environment. While Earth observations (EO) offer a powerful tool to improve characterization of malaria risk, the effective application of EO to the problem depends on adequate characterization of these human and environmental processes. Satellites, after all, do not detect parasites or mosquitoes. They can,however, inform estimates of risk factors forf mosquito presence and malaria transmission, but only insomuch as the spatial and temporal relationships between biophysical conditions, disease ecology, and human exposure are understood.

In a recent NASA Applied Sciences pilot study, we applied EO to assess mosquito distribution and malaria risk on the Peruvian Amazon frontier. The study made use of a diverse suite of EO to investigate malaria ecology and transmission risk factors at two distinct scales: at the local scale, where land cover conditions and hydrology are thought to influence the presence and density of vector mosquitoes; and at a regional/administrative scale, where regional landscape characteristics, demographic patterns, and hydroclimatic variability are thought to be predictive of overall transmission risk.

Figure 1: Study area for local scale analysis, located in the Amazon frontier region south of the city of Iquitos. Red dots indicate mosquito collection sites and black lines are roads. Colors are supervised land cover classification based on 2001 Landsat imagery: water (blue), forest (dark green), grassland and crops (light green), and secondary forest and scrub (yellow). Image Credit: Ben Zaitchik / USGS Landsat Program.

Figure 1: Study area for local scale analysis, located in the Amazon frontier region south of the city of Iquitos. Red dots indicate mosquito collection sites and black lines are roads. Colors are supervised land cover classification based on 2001 Landsat imagery: water (blue), forest (dark green), grassland and crops (light green), and secondary forest and scrub (yellow). Image Credit: Ben Zaitchik / USGS Landsat Program.

For local scale analysis, we applied EO in two ways. First, building on previous work [6], we applied Landsat imagery to map deforestation and reforestation patterns in an area near the city of Iquitos that is known to have a high malaria burden (Figure 1). Second, we implemented aåÊrelatively high resolution (1kilometer) Land Data Assimilation System (LDAS) that combinesåÊEO of: a) meteorological variables, such as Tropical Rainfall Measurement Mission (TRMM) precipitation, b) land-surface parameters, including Moderate Resolution Imaging Spectrometer (MODIS) and Landsat-derived estimates of vegetation cover and land use, and c) time-varying surface properties such as soil moisture with an advanced land surface model that calculates the surface and subsurface water and energy balance at 15-minute time-steps.

The Landsat maps and LDAS hydrometeorological estimates were applied to predict the distribution of mosquito-breeding sites and of larval and adult mosquitoes of different species. Data on breeding sites and species-specific mosquito data were available from a previous highly detailed field survey of theåÊregion. Mosquito-species information is critically important, since only some Anopheles species are effective malaria vectors. Results of the EO-informed models are informative and not always intuitive. For example, hydrological and meteorological fields drawn from the LDAS indicate that the probability that mosquitoes will be found in a potential breeding site increases with the number of large water bodies surrounding the site but decreases with temperature, solar radiation and soil moisture, suggesting that mosquitoes react negatively when conditions are too warm or wet. And while the larva of all mosquito species reacted similarly to these hydrometeorological conditions and to land-use characteristics, the species diverged in their response to land disturbance as adults: Adults of many species prefer more pristine areas to cleared agricultural land, but the probability of finding Anopheles darlingi, the most effective malaria vector in the Amazon, increases as forest cover decreases. This suggests that land clearing might reduce the risk of getting a mosquito bite while increasing the risk of being bitten by a mosquito capable of transmitting malaria. Further complexities, including the interaction between land cover and the density of water bodies [7] are still under investigation.

At the administrative scale, we again used EO to characterize land cover and hydrometeorology, but for a longer time period and a significantly larger area (Figure 2). This necessitated some compromise in the resolution of the LDAS, which we implemented at 5-kilometer resolution for regional analysis, and in the nature of land cover information: We were not able to perform spatially complete and continuous Landsat analysis, so we relied on static Landsat-informed analyses of forest type and MODIS-derived estimates of land cover change through time. EO data were paired with 10 years of weekly malaria case reports from more than 300 health posts distributed across the Department of Loreto. More than 11 years of malaria surveillance data, preliminary analyses haveåÊindicated that malaria burden is a function of background ecological conditions—the frequency of flooding and general forest type—and that transmission risk responds to temporal variability in soil moisture and rainfall. We are continuing to refine these models for operational use, but even these preliminary results indicate that EO are capable of informing transmission risk estimates over space and time.

Figure 2: Examples of LDAS fields for the regional-scale malaria transmission risk analysis. The study was performed for the Department of Loreto, which covers approximately 370,000 square kilometers and has a population of about 900,000. Shown here are monthly average (A) net solar radiation, (B) precipitation, (C) surface temperature, and (D) volumetric near-surface soil moisture for January 2009, all at 5-kilometer spatial resolution. Image Credit: Ben Zaitchik / NASA Land Information System.

Figure 2: Examples of LDAS fields for the regional-scale malaria transmission risk analysis. The study was performed for the Department of Loreto, which covers approximately 370,000 square kilometers
and has a population of about 900,000. Shown here are monthly average (A) net solar radiation, (B) precipitation, (C) surface temperature, and (D) volumetric near-surface soil moisture for January 2009, all at 5-kilometer spatial resolution. Image Credit: Ben Zaitchik / NASA Land Information System.

To date, our experience applying EO to malaria risk monitoring in the Peruvian Amazon has provided useful information on malaria dynamics that have the potential to support intervention strategies. At the same time, the work has revealed several areas for future study. At local scale, we find ourselves limited by the resolution and reliability of Landsat-based land cover classification. The landscape mosaic on the Amazon frontier is not a simple checkerboard of forested and deforested land. There are diverse forest types, abandoned fields at various stages of regrowth, and lands disturbed by selective logging and artisanal mining activities that are not always easily detected at Landsat resolution. In ongoing work, we are attempting to improve our characterization of land disturbance at local scale through use of high texture-based classification algorithms to capture forest regrowth and incorporation of data from high-resolution imagery and additional Landsat scenes. At the administrative scale, we face the dual challenges of abstracting small scale processes to representative variables at large scale—for example, soil saturation atåÊ5-kilometer resolution as an indicator of potential for breeding sites—and of accounting for irregularities in malaria case data due to human behavioral factors that are difficult to model— for example, if people are more or less likely to travel to a health clinic at different times of year or in different districts.

At both scales, the application of EO to malaria studies is likely to benefit significantly from new and upcoming satellite missions. Landsat 8 is now providing regular imagery of the Amazon, providing a new influx of data at 30-meter resolution for forest mapping. The Global Precipitation Measurement Mission (GPM) and the Soil Moisture Active Passive (SMAP) satellite will both come on line in the near future and will provide higher resolution and more reliable precipitation and surface soil moisture estimates for use in LDAS. We intend to incorporate these new data streams into our analytic framework, raising prospects for an enhanced EO-derived information system that can be applied to operational risk monitoring and early warning in the Amazon.

References

[1] World Health Organization, ‰ÛÏWorld Malaria Report: 2011,‰Û WHO Global Malaria Program, Geneva, 2011.

[2] Pan American Health Organization, ‰ÛÏReport on the Situation of Malaria in the Americas,åÊ2009.‰Û PAHO, Washington DC, 2009.

[3] J. Espinoza et al., ‰ÛÏThe Major Floods in the Amazonas River and Tributaries (Western Amazon Basin) during the 1970‰ÛÒ2012 Period: A Focus on the 2012 Flood,‰Û J. Hydrometeor., vol. 14, pp. 1000‰ÛÒ1008, 2013.

[4] Peru Ministry of Health (MINSA), ‰ÛÏSala de Situacion de Salud, Peru. Sistema Nacional de Vigilancia Epidemiologica. D. G. d. Epidemiologia,‰Û Ministerio de Salud (MINSA), Lima, Peru, Semana Epidemiologica No 52 ‰ÛÒ 2013, Dec. 2013.

[5] Instituto Nacional de Salud de Colombia, ‰ÛÏBoletin Epidemiologico Semanal, Subdirreccion de Vigilancia y Control en Salud Publica,‰Û INSC, Bogota, Colombia, Semana epidemiologica numero 51 de 2013, Dec. 2013.

[6] A.Y. Vittor et al., ‰ÛÏThe effect of deforestation on the human biting rate of Anopheles darlingi, the primary vector of falciparum malaria in the Peruvian Amazon,‰Û American Journal of Tropical Medicine and Hygiene, vol. 74, pp. 3-11, 2006.

[7] D. Valle et al., ‰ÛÏAbundance of water bodies is critical to guide mosquito larval control interventions and predict risk of mosquito-borne diseases,‰Û Parasites & Vectors, vol. 6(1), ppåÊ179-180, 2013.