Earth observation for public health: Biodiversity change and emerging disease surveillance.

One Health is an emerging concept in the health sciences that approaches human, animal and environmental health from a single framework. This policy approach is grounded in the knowledge that approximately 70 percent of emerging diseases in humans originate from other species, and that this species crossover is precipitated by stresses to environmental systems such as habitat change and biodiversity loss. Remote sensing tools apply well to this approach due to the multitude of variables that can be measured across borders in real-time. This paper explores the challenges and opportunities of using satellite remote sensing to monitor biodiversity loss in real time, with a goal of predictive surveillance for emerging disease events. Key findings include that (1) certain emerging disease events are preceded by biodiversity changes that can be observed from space; (2) refining quantitative assessments of biodiversity loss is a critical next step; and (3) biodiversity loss as observed from space merits inclusion in emerging disease surveillance programs as a complement to in situ and epidemiological surveillance data.

Whoever wishes to investigate medicine properly, should proceed thus: in the first place to consider the seasons of the year, and what effects each of them produces for they are not at all alike, but differ much from themselves in regard to their changes. Then the winds, the hot and the cold, especially such as are common to all countries, and then such as are peculiar to each locality. We must also consider the qualities of the waters, for as they differ from one another in taste and weight, so also do they differ much in their qualities.

Introduction: context and need for One Health

In the aftermath of the 2014 outbreak of Ebola hemorrhagic fever in West Africa, during which over 11,000 lives were lost and considerable economic and social costs were incurred [1], much of the criticism directed at the international system focused on the isolated and reactive nature of the response [2]. The Ebola virus is an example of an emerging infectious disease (EID), a pathogenic agent appearing in a population for the first time or posing a novel risk in a population [3]. Other contemporary examples include HIV (human immunodeficiency virus), SARS (severe acute respiratory syndrome) and the Zika virus.

It is the novelty of these threats relative to public health preparedness that heightens the risk environment associated with them. The 2003 SARS outbreak, for example, is estimated to have cost over $30 billion in economic losses, largely due to disruptions in economic markets, travel and preventative health measures made necessary by the unknown and therefore unpredictable factors of the pathogenic agent [4].

Despite advances in global medical infrastructure, the number of EID events per year has roughly quadrupled since 1940 (accounting for better surveillance) [5]. Approximately 70 percent of these pathogens are zoonotic, which means they originate from non-human hosts [5]. David Heymann, a leading scholar in the field of emerging diseases, suggested in a forum on emerging infections hosted by the National Academy of Sciences that the “windows of opportunity” to combat the EID rise are closing in the face of rapid environmental and societal change [3]. This paper proposes the use of space-based Earth observation, as a complement to robust in situ monitoring and epidemiological surveillance, as a timely and necessary interdisciplinary approach to getting ahead of the EID curve. Specifically, we suggest that real-time monitoring for biodiversity loss via remote sensing may be a critical component of the surveillance effort to predict the next emerging disease outbreak.

Recent advancements and opportunities

Hippocrates, writing in 400 BCE, noted a need to consider the effects of environmental conditions on human health. In a return to this theme, the public health community has directed much attention to the importance of interdisciplinary activity in emerging disease surveillance and response. Organized under the One Health theme, this work has been characterized by diversified Emergency Operations Centers (EOCs) and broader dissemination of environmental, animal and human health data by practitioners and researchers in the field. Examples include the EMPRES (Emergency Prevention System) database, a joint Food and Agriculture (FAO) and World Health Organization (WHO) venture that has been implemented in the global surveillance of highly pathogenic avian influenza, as well as the Predict pandemic surveillance program piloted by the United States Agency for International Development (USAID) and the One Health Institute at the University of California, Davis [6,7].

Despite these advancements, emerging disease surveillance is still largely reactive in its scientific and policy approaches. The One Health framework, supported by advances in data storage, processing and computing, offers an opportunity to anticipate disease outbreaks with sufficient time for intervention and mitigation. Surveillance tools must generate ongoing, real-time information, which is why long-term and unpredictable risk factors (discussed below) often remain the subject of research journals and do not permeate into national and international EOCs. An integrated, remote-sensing based platform to provide early warning of EID events could ameliorate the distance between long-term trends and real-time risk.

The geospatial dimensions of disease

To understand the role that Earth Observation tools might play in disease surveillance, one must consider the geospatial dimensions of disease and the current methodology utilized to track global outbreaks. The field of epidemiology is intimately linked with spatial analysis since its genesis: in 1854, as cholera decimated the Soho district of London, a physician by the name of John Snow famously began tracking the cases as they emerged on a map of the city (Fig. 1 ) [8]. Geographic clusters of individual cases quickly emerged along unique corridors of the city's streets. This methodology yielded a critical insight: the clusters occurred along shared routes of water distribution [8]. The geospatial patterns, which existed neither in the lab nor in the isolated clinical context, extended the Cartesian relationship of disease transmission into another dimension. Patterns portend predictability, which offers public health officials an opportunity to intervene.

Fig. 1

Snow's Cholera map demonstrates geospatial connections between disease and environment [8].

The prevalence of cartography in disease sleuthing has grown considerably since Snow's initial investigation, ranging from explaining outbreaks to the public to mission-critical geospatial analysis of ongoing epidemics. Visualizing the broader context of disease has become critical in a time of growing, rapidly moving human populations interacting more frequently with a dynamic biosphere. Given this historical context, the field is well-suited for the remote sensing program proposed here.

Contemporary Emergency Operation Centers, such as those maintained by the World Health Organization and Centers for Disease Control and Prevention (CDC), use GIS as standard practice in monitoring outbreaks [9]. Data are often limited to case counts and demographic information (not unlike Snow's original Cholera maps); real-time environmental measurements are difficult to integrate and thus limit the predictability of such platforms. Real-time alert systems, such as the healthmap.org project (Fig. 2 ), have expanded the scope of disease mapping by integrating a broader range of data sources through advanced text mining through newspaper reports and social media [10].

Fig. 2

Healthmap.org project uses text mining to create real-time global health alerts [10].

This is an important step in reducing the disparities in reporting and monitoring events that often occur between developing and developed economies. Considering that environmental risk factors affect developing countries at a greater rate [11], such advancements in surveillance play a critical role in better responding to the EID rise.

Because the public health community has a high degree of familiarity with geographic information systems, and the environmental data captured by space-based remote sensing is readily applicable to highly visual, real-time models, we suggest that the proposed platform is a natural fit for the needs and context of the global health community.

The role of biodiversity

Several variables have been implicated in the alarming rise of EID events, including increased international travel; the proliferation of antibiotic misuse; and the rapid growth of urban environments corresponding with human population increase [5]. Here, the growing EID event curve is considered in the context of global biodiversity loss, which has also been associated with the emergence of novel pathogens in human populations [12] and has arguably received less attention than the factors above.

Mechanisms of action

In a review of the past several decades of biodiversity research, Wardle et al. conclude that while global species loss continues at an alarming rate, “the impact of biodiversity on any single ecosystem process is non-linear and saturating, such that change accelerates as biodiversity loss increases” [13]. These ecosystem processes include human health, which is at greater risk to EIDs due to the non-linear impact of biodiversity loss. Pongsiri and colleagues [12] discuss three principal connections between biodiversity loss and emerging infectious disease rates: the dilution effect, changing encounter rates and the spread of nonindigenous vectors.

The dilution effect refers to an overall increase in disease reservoir competency that makes transmission of a disease to humans more likely [14]. Zoonotic diseases originating from wildlife are more likely to crossover to human populations after experiencing a loss in diversity because of the associated loss of non-competent disease hosts.

An experimental example is provided in the case of Puumala hantavirus, a zoonotic RNA virus that causes hemorrhagic fever in humans. The disease reservoir of this hantavirus is the bank vole, a common mammal that thrives in human environments (it is therefore more likely to remain in nature after other species have been removed due to human activity) [15]. Researchers have demonstrated that declines in non-host species (those hosts that cannot further transmit the pathogen) corresponded with an increase in both bank vole and human infections of Puumala hantavirus [15].

Similar work has suggested the same mechanism in the emergence of Lyme disease in new areas of the United States: while blacklegged ticks commonly attach themselves to both the white-footed mouse and the Virginia opossum, the Virginia opossum is both better at grooming off and killing ticks and is less likely to be infected with Lyme disease and subsequently further propagate the disease in the environment [16].

Biodiversity loss has affected these organisms in different ways, with the white-footed mouse demonstrating greater resilience; this ecological disruption has been proposed as one of the driving forces for the rapid spread of tick-borne illness in the United States [16]. Fewer non-competent hosts in the system, such as the Virginia opossum, lead to more favorable outcomes for pathogens spread by ticks.

The encounter rate among hosts is a complementary driving force behind the biodiversity-EID relationship [12]. As species diversity is lost, competent disease hosts are more likely to come into contact with each other rather than members of another species that would not be able to further propagate the disease. Experimental studies of Sin Nombre virus (an emerging hantavirus in the southwestern United States) have suggested that the competent disease host, deer mice, come into contact with each other much more frequently in less biologically diverse environments and are therefore more susceptible to infection [17]. This heightens the human risk of contracting Sin Nombre virus in the region.

Finally, nonindigenous disease vectors are more likely to succeed in environments that have experienced biodiversity decline [12]. Entomologists have noted the expanded range of important mosquito vectors including Aedes aegypti, responsible for spreading dengue fever, chikungunya, Zika fever and yellow fever [18]. In a less diverse ecosystem, these vectors are less susceptible to predation, which otherwise might be facilitated by diverse bird, bat and insect populations [12].

Additional considerations: scope and limitations of theory

The body of experimental evidence on zoonotic spillover suggests that while no single mechanism is universally causal, disruption of ecosystems and loss of species diversity has been clearly and consistently associated with the emergence of novel pathogens [16]. However, this statement must be made carefully, as infectious disease ecology is often pathogen-specific. Indeed, there is some disagreement as to which of these factors serves as the principal driver of EID events [[19], [20], [21]].

Moreover, there is some confusion as to whether a relationship exists between ecological composition and infectious disease agents in general, beyond those which meet the criteria of emerging zoonotic pathogens with origins in wildlife. It is critical to note that such an extrapolation would be beyond the state of what the current data support. For example, associations between ecological change and the propagation of malaria have been shown to be inconsistent, and in some cases biodiversity seems to be associated with increases in human infection rates [20]. Importantly, this example is of an endemic or established pathogen, the ecology of which no longer behaves in accordance with the body of evidence presented here. The need to distinguish between these two distinct states of infectious agents in the context of ecological change is discussed in depth by Hosseini et al. [22], who suggest that the differences between the terms “hazard” and “risk” may play a role in misunderstandings. In the case of EIDs with origins in wildlife, diverse ecosystems present a hazard of high pathogen diversity; this paper is concerned with measuring the risk of an outbreak. While drivers of endemic infectious disease, especially vector-borne pathogens such as malaria, have shown promise as targets of remote-sensing based surveillance programs, these efforts must be considered separately from the approach described here as they involve different assumptions of both hazard and risk.

Importantly, this paper considers only those infectious disease events with zoonotic origins in wildlife, which have a more consistent response to changes in local ecological change [23]. Furthermore, we propose adoption of a methodology which is mechanism agnostic: this is to say that the target of measurement in high-risk areas should be ecological disruption in general rather than a measurement specific to, for example, the dilution effect.

We suggest that observing real-time disruptions in the ecological composition of areas at risk for this subset of EIDs at the interface of human activity can be used to forecast pathogen emergence and support mitigation efforts. Furthermore, it is likely that these efforts can also support further examination of the mechanisms that promote zoonotic spillover, which in turn can be used to improve upon surveillance platforms. To begin to understand the role of remote sensing, it is critical to understand how emergence-promoting changes can be quantified at scale.

Measuring change quantitatively

Biodiversity is, at its simplest, a measure of the variation of living organisms in an ecosystem (current work in the field also considers function) [24]; alpha, gamma and beta diversity describe different general classifications of biodiversity [25]. diversity is a measurement of species diversity in a local area, diversity is the overall biodiversity in a region (alpha diversity multiplied by the number of sites), while diversity reflects the difference between local sites (gamma/alpha; the higher the value, the more unique the composition of each local alpha relative to each other) [25].

Biodiversity measurements typically begin at the alpha level, and these measurements can be characterized quantitatively using the Shannon index. The Shannon index (H’) is a well-known tool in the biological sciences that allows standardized comparisons of alpha diversity [26]. Mathematically, the index is calculated by equation (1):where s represents the number of species in the sample area and p represents the ratio of individuals, i, of a species divided by all individuals, N, of all species [26]. As the Shannon outputs for a unit area (the hectare is common) increase, the biodiversity of that area is considered to be higher [27].

The unitless index has an upper limit of 5 and a minimum of 0; outputs are difficult to compare from one type of ecosystem to another, but do provide a reference point of change over time for the same unit area [27]. Additional indices have been developed to quantify biodiversity, each with advantages and disadvantages over the Shannon index, but due to the familiarity of the Shannon index across fields and its long-term use, it is most readily applicable to this study.

Roche and colleagues [28] modeled the effects of biodiversity changes on an ecosystem using the Shannon index, and demonstrated that as the Shannon values for an ecosystem approached 5, the proportion of infected individuals decreased. Reductions in host diversity (approaching 0) conversely increased the proportion of infected individuals. This provides strong theoretical support for the experimental observations noted by Pongsiri and others above, and also suggests that areas with high indigenous biodiversity are most at risk for EID-promoting changes.

The earth observation opportunity

The opportunities for Earth observation technologies, in particular satellite-based remote sensing, in predictive health surveillance are many and varied. The wide range of environmental conditions that can be monitored at a low cost and consistent time scale add significant value to models of risk and prevalence for maladies attributable to environmental factors (a classification that includes approximately one-quarter of the global disease burden, according to the Üstün and Corvalán of the WHO) [29].

At the recent One Earth – One Health workshop in Montreal, Canada (June, 2017) researchers and practitioners highlighted several examples of pre-existing and developing projects [30]. Well-developed activities include air and water monitoring (examples include asthma and harmful algal bloom risk predictions, respectively), as well as modeling changes in vector dynamics responsible for malaria.

The application of remote sensing to understanding biodiversity changes which affect health is a nascent, developing technique. Why has the development of surveillance tools in this area not kept pace with advances in water and air quality, among others? This is in part because the relationship between rapid biodiversity loss and EID increases has been demonstrated relatively recently. However, monitoring biodiversity through SRS has also presented challenges unique among the presented health applications.

It should come as no surprise that biodiversity cannot be taken as a direct measurement from space. Rather, measurements of key variables are taken that can be calibrated to biodiversity estimates using field observations. The Group on Earth Observations – Biodiversity Observation Network (GEO BON) has been organizing much of the effort to do so. To this end, the group has constructed a range of Essential Biodiversity Variables (EBVs) to sufficiently capture the complexity of biodiversity in a unit area [31].

GEO BON has identified the following EBVs as readily able to be sensed remotely via currently operating satellites: fractional cover, forest cover, land cover, fraction of absorbed photosynthetically active radiation, leaf area index, phytoplankton, phenology, soil moisture, fire disturbance, and inundation [31].

These variables have been constructed in the context of ecology at large, and therefore can be refined for the specific context of disease monitoring. For contexts of interest (terrestrial areas with high Shannon indices) the fraction of absorbed photosynthetically active radiation, phenology and inundation are especially relevant. In addition, several EBVs have been proposed for future programs when more accurate measurements are available: species occurrence; specific leaf area; taxonomic diversity; vegetation height; and above-ground biomass [31].

One important advancement in the field is the application and experimental refinement of the spectral variation hypothesis, which states that spectral heterogeneity in an image reflects environmental and taxonomic heterogeneity in the unit area sampled (if an image yields a greater number of spectral signatures, the effective biodiversity of that region must be greater, according to the hypothesis) [32].

Groundbreaking work by the Carnegie Airborne Observatory has demonstrated the ability to correlate Shannon values in a sampled area using airborne spectroscopy in lowland Amazonia [33]. While this work does not apply well to the needs of a real-time, cost-effective and long-term surveillance program, it suggests that upcoming measurements from the HyspIRI (Hyperspectral Infrared Imager) and DESIS (Deutsches Zentrum für Luft-und Raumfahrt Earth Sensing Imaging Spectrometer on the ISS) instruments may well be of interest in future space-based biodiversity monitoring [34,35].

Currently, the standard remote-sensing measurement technique for observing changes in biodiversity composition is the Normalized Difference Vegetation Index (NDVI). This index is calculated by comparing differences in reflected light in the near-infrared range relative to reflected visible light [36]. Mathematically, this is calculated by equation (2):or the near-infrared radiation minus visible radiation divided by near-infrared radiation plus visible radiation (unitless output values range from −1 to 1, where vegetative output increases from 0 to 1 and values less than zero reflect aqueous areas). Vegetation characteristically reflects near-infrared light from 700 to 1100 nm and characteristically absorbs light in the 400–700 nm range [36].

It is important to note that although NDVI is readily applicable for applications such as agriculture or monitoring land-use change, it is not sensitive to species-level changes in a unit area (one can imagine, for example, an increase in NDVI due to the proliferation of an invasive plant that actually decreases biodiversity in a given pixel while concurrently demonstrating greater vegetative output).

A warming global climate further complicates efforts to use NDVI as an SRS EBV, as vegetation increases are experienced even in some areas that have undergone rapid land-use change and subsequent loss of taxa-level diversity [36]. However, due to its standard use in the field and the high quality of available data (beginning in 1972 with the deployment of the Landsat program), NDVI serves as a useful first step in beginning to monitor the types of biological change that can portend emerging disease events [36].

Due to the complexity of biodiversity measurement, a combination of satellite remote sensing, in-situ measurements, air-borne observations, modeling and expert opinion will be necessary components of a long-term, real-time monitoring program.

Toward a predictive emerging disease surveillance platform

The utility of a remote sensing-based public health surveillance program is dependent on the degree to which the produced information is timely and actionable. Biodiversity monitoring presents a readily available and valuable component of such a system. While the GEO BON group has demonstrated that biodiversity surveillance cannot function independently of in situ data collection and expert evaluation, it has also advocated for remote sensing as a critical part of a comprehensive monitoring platform [31]. A remote-sensing based program can operate around the clock and algorithmically fill the gaps between in-situ monitoring, which brings the possibility of a real-time alert system closer to reality.

A useful analog exists in the global land-use monitoring program of Hansen et al. (2014). In partnership with the Google Earth Engine team, Hansen and his colleagues created the Global Forest Watch program, an open-source platform that provides real-time FORMA (FORest Monitoring for Action) deforestation alerts [37]. The algorithm utilizes NDVI data from MODIS (Moderate Resolution Imaging Spectroradiometer, aboard the NASA Terra satellite), precipitation data and historical information on each pixel in a given image (to account for seasonal variation). A user-friendly interface is freely available at globalforestwatch.org, providing actionable, real-time information to stakeholders concerned with rapid changes in land use [37].

While the Global Forest Watch algorithm does not directly measure the biodiversity changes that portend emerging disease events, it provides critical information about habitat change that suggests species loss. Recent advances in mapping the relative risk of emerging disease threats by Jones et al. (2008) enable prioritization of the monitoring program proposed here [5]. In fact, it seems that remote sensing is well-suited to respond to one of the primary conclusions of the authors, where they found that:

Global resources to counter disease emergence are poorly allocated, with the majority of the scientific and surveillance effort focused on countries from where the next important EID event is least likely to originate. (Jones et al., 2008) [5].

Remote sensing allows countries with fewer resources to understand what ecological changes are occurring within their borders, and also allows for a global picture to be constructed, without country-level errors in reporting.

We suggest that, in a similar approach to Hansen et al., remote sensing can be utilized to detect statistically significant biodiversity changes and create timely alerts for policymakers. By focusing on the high-risk areas identified by Jones, in-situ validation and testing can be conducted on a cost-effective basis. This open-source platform can be utilized by One Health practitioners to consider environmental areas of concern, with possibilities for mitigation in areas undergoing the rapid species loss that is dangerous to human health.

Case studies

The need to include biodiversity change in a health surveillance platform can be demonstrated by the following case studies, which highlight: (a) the relationship between environmental change and emerging disease outbreaks; (b) the ability to detect such environmental changes, both in real time and over the long term, through remote sensing; and (c) the significant cost of delayed detection of emerging outbreaks.

Cases of Eastern Afromontane

In 2007, a novel Ebolavirus caused an epidemic of Ebola hemorrhagic fever in Uganda's Bundibugyo District (this virus is distinct from, although similar to, the Ebolavirus of the 2014 epidemic in West Africa mentioned in the introduction to this paper) [38]. Though less deadly than similar Ebola virus outbreaks, the disease caused considerable loss: among the fifty-six cases involved in the outbreak, approximately 40% resulted in fatality [38].

The suggested reservoir of the Bundibugyo ebolavirus is the fruit bat [39]. This is consistent with recent research that suggests bats harbor a disproportionately high number of zoonotic viruses than other mammals [40]. The changes in host competency caused by rapid biodiversity loss are therefore highly relevant to bat populations, which can migrate great distances after land-use change events.

Indeed, the Bundibugyo outbreak appears to have been precipitated by a rapid biodiversity shift, combined with a rapidly growing human population in the same area, that increased the risk of the emerging disease event. The Bundibugyo District is on the western border of Uganda abutting the Democratic Republic of Congo (DRC). The district is bounded by Ntoroko to the north and east, Kabarole to the south and the North-Kivu region of the Democratic Republic of the Congo to the west. The district is in close proximity to Semuliki National Park, Rwenzori National Park, Semuliki Wildlife Reserve, Kibale National Park, and the Virunga National Park.

Importantly, the district is situated in a global biodiversity hotspot, the Eastern Afromontane, as defined by Conservation International. This hotspot, a thin arc curving through East Africa, contains approximately 10,856 unique species comprised of the taxonomic groups amphibians (229); birds (1,299); fishes (893); mammals (490); plants (7,598); and reptiles (347) [41].

Approximately one-third of these species are endemic to the region [41]. This biodiversity coexists in the region with a dense human population: Bundibugyo maintains a population density of 309 persons/km2 (261,700 persons in total), compared with 207 persons/km2 in Uganda overall (41.49 million persons in total) [42].

Remote-sensed data, based on historical NDVI changes at the single-pixel level, demonstrate that Bundibugyo lost 1307 ha of forest cover from the year 2001 (the earliest available year of coverage) to 2007 (the emergence date of the Bundibugyo ebolavirus) [37]. While this change in forest cover is significant, it does not provide a full picture of the region. The neighboring DRC region of Kivu lost a remarkable 457,357 ha of forest cover during the same period (Fig. 3 ) [37].

Fig. 3

NDVI loss (in pink) from 2001 to 2007 in region west of Bundibugyo, Uganda outbreak demonstrates significant vegetation loss, suggesting biodiversity loss and ecosystem disruption. Generated by the authors using Google Earth Engine and Hansen et al. (2014) algorithm for detecting statistically significant NDVI changes at the pixel level. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

This change in land cover has been driven by economic, political and social factors. Bundibugyo has developed into one of the largest cocoa-producing regions in the country in recent years [43], while the neighboring DRC region of Kivu is experiencing human migration due to conflict [44].

Although in-situ measurements would have been difficult during a period of conflict, the remotely sensed data demonstrate that considerable biodiversity loss likely occurred during the forest cover clearing in the years preceding the Bundibugyo outbreak.

This relationship is further suggested by a 2012 novel viral infection in a remote Ugandan village, approximately 150 km from Bundibugyo. In 2012, a novel paramyxovirus called Sosuga virus appeared in Uganda. The infected individual was a wildlife biologist who had been sampling bats to the northwest of Bundibugyo [45]. Importantly, the regional ecosystem disruption that preceded the Bundibugyo ebolavirus outbreak continued apace in advance of the Sosuga EID event: the district in which the pathogen was discovered (Kibaale, approximately 100 km from Bundibugyo) experienced 32,173 ha of loss from 2007 to 2012 [37]. Again, bats were implicated in the pathogen's emergence [45], suggesting that the ecosystem disruption described by Pongsiri and colleagues may have played a role.

While no causality can be inferred from this analysis, the theoretical and epidemiological evidence for the role of biodiversity loss in the aforementioned outbreaks in the Eastern Afromontane is strengthened by space-based observations. Better experimental data could further elucidate the role species loss played in precipitating the emergence of novel pathogens in this area.

For example, recent work [46,47] has highlighted a critical ecological factor in the emergence of Ebola virus disease in Central Africa: the degree of tropical forest intactness and fragmentation. Here, we consider ecosystem disruption in general. However, both [46] and [47] present findings that suggest disruptions are more likely to promote outbreak events when they incur in such a way as to fragment previously intact ecosystems of a certain minimum diameter. These findings are important for the remote sensing community, because adoption and refinement of EBVs can facilitate real-time measurement of tropical forest fragmentation to a more robust degree than previous methodologies. Specifically, fraction of absorbed photosynthetically active radiation and leaf area index can supplement the methodologies utilized in these studies to broaden our scientific understanding of how ecosystem change promotes spillover events. Does loss of specific keystone species in tropical forest ecosystems, such as tree species that provide food sources to bat populations, play an outsized role in promoting zoonotic spillover, or are outbreaks more likely irrespective of the nature of the disruption? Does clearing of tree species of a certain age or density matter? Further characterization with EBVs can provide important insight into these questions.

Cases of Sundaland

In 2006, two similar orthoreoviruses emerged in Malaysia: the Melaka virus and the Kampar virus. Symptoms of infected individuals were similar to influenza-like illness, with infection of the respiratory tract, a high fever and severe aches noted by local physicians [48].

As in the case of the novel virus outbreak in Bundibugyo, epidemiological and serological data suggest that fruit bats likely serve as the natural reservoir for the Kampar and Melaka viruses and are therefore involved in transmission to humans [48]. (Bats were also implicated in Malaysia's most famous disease outbreak, the 1999 Nipah virus, during which the genus Pteropus infected pigs with a Henipavirus which in turn caused fatalities in over 100 of 300 human cases and resulted in approximately 500,000 euthanized pigs [49].) The abundance of bats in the region, and their suitability as viral reservoirs as described by Olival and colleagues [40], poses a considerable health risk as regional biodiversity changes.

Each virus is named based on its area of origin: Melaka is a region on the southwest of peninsular Malaysia, while Perak is a larger region on the west coast on the northern side of Kuala Lumpur [48]. Peninsular Malaysia is a component of the Sundaland biodiversity hotspot in Southeast Asia (so designated by Conservation International), and is home to over 8000 species of plants; 200 unique mammalian species (including, importantly, 81 bat species); and some 600 bird species [37].

Remote-sensing data yields important insights into the environmental conditions of the surrounding area in the five years leading up to the Kampar and Melaka virus emergence events. Peninsular Malaysia as a whole lost a considerable 905,579 ha of forest cover, while the districts of interest, Melaka and Perak, lost 14,654 and 93,362 ha respectively [37]. Experimental work in the field has demonstrated H’ values of greater than 4 distributed in a non-uniform pattern throughout Peninsular Malaysia, suggesting that biodiversity loss would have been considerable during the clearing events (Fig. 4 ) [50].

Fig. 4

NDVI loss (in pink) from 2001 to 2006 in Peninsular Malaysia near Malacca. Generated by the authors using Google Earth Engine and Hansen et al. (2014) algorithm for detecting statistically significant NDVI changes at the pixel level. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

At the national level, urban development and agriculture, particularly palm oil production, have driven land use change and anthropogenic forest clearing [33]. At the district level, a more complex picture emerges: while Perak faces threats to natural systems from agricultural and urban development (reflected by its higher hectare loss compared with Melaka), Melaka is experiencing a tourism boom and is incentivized by this to maintain its natural systems (though it likely means greater interaction between a diverse human population and the biodiversity of the region) [51]. In fact, Melaka experiences roughly 15 million tourists per year, including approximately 500,000 medical tourists [51]. This increases the risk of transporting an emerging disease from the region to the global population, making surveillance efforts in the region particularly valuable over the long term.

Discussion of presented cases

The cases presented suggest that the biodiversity changes that heighten EID risk can be viewed from space, even with NDVI-based methods. The health community should take note of several critical features of the Hansen et al. algorithm used here, which provides (1) real-time risk (2) based on a historical trend that (3) can be used to implement better policy for the long term. While causality should not be inferred, we do make the case that remote sensing is a powerful tool for detecting the ecological changes that can lead to novel outbreaks.

The Bundibugo and Sosuga cases were precipitated by highly localized and significant ecosystem disruption both within Uganda and in the DRC. A key feature of remote sensing is the ability to observe environmental changes beyond national borders, which would appear to have been beneficial to public health officials in Uganda prior to the outbreak. While the ecosystem disruption in Peninsular Malaysia was not as isolated, extreme habitat loss was consistently observed throughout the region. Combined with an urbanizing population, the risk of this ecosystem disruption to human health appears to continue to be high.

Additionally, these cases complement recent work on emerging disease threats that has suggested bats as an important mammalian host of novel pathogens. Observing disruptions of bat habitats might therefore be a critical component of future space-based surveillance efforts.

However, much work needs to be done to refine the temporal and spatial relationship between ecosystem change, human societal activity and disease outbreak. Importantly, spectral heterogeneity analysis and direct measurement of EBVs can greatly improve on the time series, spatial resolution and accuracy of currently available methods, while field testing will be critical to calibrate new techniques that can better account for species-level loss.

The cases presented here provide insight into opportunities for improvement at each level of the data, as well as how these data layers might function together in a surveillance system. At the remote sensing level, inclusion of additional EBVs can supplement current techniques to understand both ecosystem intactness and fragmentation beyond simply forest cover. As discussed, fraction of absorbed photosynthetically active radiation and leaf area index are readily available candidates, while vegetation height and taxonomic diversity are appropriate targets for forthcoming LiDAR and hyperspectral instruments. This highlights the importance of airborne observations: as new instruments become available, data from short-duration airborne instruments will play a critical role in calibration. Finally, field-level data will continue to be the foundation of future surveillance efforts. Understanding the sensitivity and extent of ecosystem change, as well as human factors, necessary to promote an outbreak requires precise geographic identification of the index or original human case in an outbreak event. Often, this information is provided at the city or district level, which complicates the system described here. Close partnership with the public health community, including considering the ways in which index cases might be more easily reported and logged, will be critical in future work.

Analysis and future directions

Early warning of biodiversity disruption could be an important predictor of EID events, and thus offers an opportunity to mitigate their initial spread in vulnerable populations. Advances in remote sensing and computational ability suggest that observing this disruption at a spatial and temporal resolution sufficient to provide this early warning might soon be possible. Applying satellite-based EBVs to characterize ecosystem intactness and fragmentation in real time should be the focus of future work on this topic.

Due to the exponential nature of epidemics, the time gained from an early warning system could prevent unnecessary mortality and produce considerable economic savings. The highly visual nature of space-based remote sensing data and its ready integration with epidemiological information via the cloud-computed Google Earth Engine will facilitate user-friendliness to practitioners operating in Emergency Operation Centers, or otherwise monitoring EID events.

Currently, our team is developing an open-sourced platform to analyze ecosystem change in areas at high risk of EID events in real time. The proposed systems architecture of this surveillance tool is described in supplementary figure 1, which describes a spatial reduction approach: beginning at the level of the geographic boundaries of pathogen risk, moderate resolution imagery can be used to define areas of ecosystem intactness, similar to [37] and expanded upon in Refs. [46] and [47]. SRS EBVs are used to identify statistically significant changes in this intactness, or fragmentation, in real time. From here, higher resolution and airborne imagery can be tasked using a “tip and cue” strategy to validate or invalidate alerts. Finally, positive identifications can be reported to partners in the field for manual analysis. This system is in its early stages, and we plan to incorporate system design suggestions from a wide variety of partners during the development phase.

We intend to use this platform to further understand the relationship between specific EBVs and zoonotic spillover. These results will be shared with the public health community, which has a strong interest in real-time characterization of the drivers of outbreak events. This platform can also inform future conservation work by adding human health effects to the equations considered by policymakers in determining biodiversity preservation schemes. By considering human and environmental health concurrently, tradeoffs between development and conservation can be more readily navigated in a rapidly changing world. Successful implementation may extend to the monitoring of livestock and wildlife health, which are also sensitive to changes in biodiversity.

Limitations of the proposed platform include the necessity of in situ monitoring and validation, which creates challenges for resource-limited areas, as well as the nascent use of SRS EBVs and their application to health outcomes. For successful adoption by the health community, errors in predictive capability must be maintained at a practical level.

Biodiversity loss is not currently included in standard EID surveillance programs because of the cost and difficulty of long-term, real-time measurement. The results of the presented cases, combined with empirical and experimental evidence of the EID-biodiversity relationship and forthcoming advances in space-based spectral imaging suggest that biodiversity now can and should be included in such systems.