GIS and Satellite Imagery Track Encephalitis Infection
By Kevin P. Corbley

Epidemiology is a complex science that involves tracking the occurrence, distribution, and source of infectious disease. Because many illnesses are caused by viruses transmitted by organisms such as insects, ticks, rodents, and humans, epidemiologists focus their studies on the environmental, cultural, and societal conditions that favor proliferation and interaction of these carriers.
    Researchers refer to these disease transmitters as vectors because they can carry an infectious agent from one animal to another. This adds yet another variable to the convoluted relationship between the source of disease and its victim. Despite the many variables involved in the spread of disease, however, there is one constant--nearly every link in the infection chain has a spatial component.
    This makes GIS one of the most valuable tools ever offered to epidemiologists. Its ability to handle large volumes of data, track multiple carriers, detect mathematical patterns in disease distribution, and find relationships between vectors and the environment is assisting epidemiologists in pinpointing more quickly and efficiently the factors responsible for spreading disease.
    The Division of Vector-Borne Infectious Diseases at the Centers for Disease Control (CDC) in Fort Collins, Colo., has engaged GIS technology in its fight to prevent ailments transmitted by insects and ticks. Specifically, it is using GIS, satellite imaging, and a variety of field data collection techniques to zero in on sources of mosquito-borne encephalitis, a dangerous group of diseases found worldwide.
    Although it is still too early in the project to yield definitive answers regarding encephalitis, CDC has made significant progress in determining which GIS capabilities are most useful in tracking vector-borne illnesses.
    "The abilities to view data sets at different scales and to integrate multiple types of data have offered the greatest benefits to disease researchers," says Dr. Chet Moore, a research entomologist at CDC's Division of Vector-Borne Infectious Diseases in Fort Collins.
    These and other GIS capabilities make their most critical contributions helping CDC researchers to devise and test new hypotheses on the vector-virus-victim relationship.

GIS and Encephalitis
The CDC laboratory in Colorado bought its first GIS software and hardware in 1993 after late summer flooding in the US Midwest created ideal conditions for mosquitoes and, possibly, an encephalitis outbreak. This prompted Congress to allocate extra funding for CDC to track and control the disease in case it flared up.
    The Division of Vector-Borne Infectious Diseases purchased TNTmips software developed by MicroImages, Inc., of Lincoln, Nebraska. They chose this product because it already was widely used among other federal agencies and it offered both raster image processing and vector-based GIS functionality in one package.
    "GIS is not our mission; it's simply a tool," says Moore. "We don't have GIS technicians here, so we selected TNTmips because it is easy for our scientists to learn to use."
    Since then, GIS processes have routinely been applied by CDC researchers to detect patterns between disease and human population. Encephalitis was a natural choice for application of the technology because it has a complex infection cycle involving wild animals, arthropods (ticks and mosquitoes), and humans.
    Encephalitis is a viral infection that afflicts the central nervous system in humans and causes an inflammation of the brain. In its mildest form, symptoms include headache, stiff neck, fever, nausea, and vomiting. If left untreated, severe disease can result in partial or total paralysis or even death.
    CDC entomologists in Colorado included two different mosquito-borne encephalitis viruses in their GIS studies--LaCrosse and St. Louis encephalitis. Although they cause similar disease in humans, these viruses are transmitted by different vectors in separate regions of the country, making for an interesting comparison with GIS tools.

LaCrosse: Finding the Right Scale
Between 75 and 150 cases of LaCrosse encephalitis are reported in the US each year, and more than half occur in West Virginia. Interestingly, nearly all reports of the illness in that state come from its southern counties.
    "We are using GIS to determine whether the increase in LaCrosse in certain areas is actually due to an increase in the disease there, or merely the result of enhanced awareness and reporting procedures," says Esther Brannon, a surveillance nurse and epidemiologist with the Bureau of Public Health in Charleston, W.Va.
    The LaCrosse virus is normally carried by squirrels and chipmunks and transmitted to humans by the eastern treehole mosquito, Aedes trisariatus. The insects are infected when they feed on the blood of those animals. To make matters even more complicated, infected female mosquitoes are able to pass the virus to their offspring through the eggs. This information gave the researchers a starting point for tracking down clues in the state.
    To begin analysis on the state level, West Virginia furnished CDC with historical statistics on cases per county. Researchers compiled these statistics in simple Excel files and downloaded them to TNTmips for display on a digital map of West Virginia with delineated county boundaries. This clearly showed a much higher occurrence rate in the southern half of the state.
    Next, the researchers added density and location data for mosquito species caught in live traps across the state. Oddly, an overlay of these two data sets showed no direct correlation between Aedes trisariatus habitat and the disease. In fact, the tree hole species was found throughout West Virginia, even where there were no reports of illness.
    As a result, researchers hypothesized that some environmental condition played a larger role than thought in the spread of LaCrosse encephalitis. "There are many variables in the forest related to this disease, and we are relying on GIS to help us sort them out," says Brannon.
    In an attempt to correlate the disease with a specific habitat or environment in a specific county, CDC obtained Multispectral Scanner (MSS) imagery acquired by the US Landsat satellites over Nicholas County, W.Va. With 80-meter spatial resolution, MSS data is considered moderate-scale imagery and deemed well suited for identifying specific habitats related to land cover.
    Working in TNTmips, researchers applied a standard unsupervised classification to the MSS data, dividing the state into broad classes based on dominant land cover in each area. Once again, however, no correlation was found between any land cover classes and occurrence of the LaCrosse strain or presence of the mosquito.   
     At this point, the researchers began to consider the scale of their data and combined county and state level data. In GIS, scale is often thought of only in terms of surface feature detail, but ecologists view scale much differently. In ecology, finding the right scale can reveal complex environmental patterns that are otherwise invisible.
    With this in mind, CDC decided to examine less-detailed, broader-scale imagery of the entire state, reasoning that ecological mechanisms involving mosquitoes, small animals, and humans probably take place on the macro rather than micro level. In 1998, they obtained 1-kilometer (km) resolution imagery collected by the AVHRR sensor aboard US meteorological satellites. Images acquired during various times of the year were selected with the thought that seasonal changes in vegetation might be a good indicator of specific types of forest habitat.
    Working in TNTmips, CDC again ran unsupervised classifications on the imagery using the normalized difference vegetation index, or NDVI, which provides a measure of "greenness" or plant leaf growth. Researchers immediately noticed certain land cover classes repeatedly coinciding with cases of the disease.
    Because the classes were generated by an unsupervised classification algorithm, CDC scientists had to determine what land cover was represented in each. The US Forest Service's Forest Inventory Analysis database provided a digital forest inventory in vector point format which was overlaid on the classification vectors. The suspect classes were identified as having large expanses of hardwood forests.
    "At this point, we are looking at the issue of total acreage or percent cover, rather than specific kinds of trees because it's difficult to identify tree type in 1km data," says Moore, adding that the USGS has joined in this phase of the project.
    Project participants will follow up on this important clue in future phases of the research. The Colorado researchers plan to team with West Virginia health officials and USGS GIS and remote sensing specialists to gather additional information about the suspected relationship between hardwood forests and LaCrosse virus. They will try to determine how forested areas in the positive counties differ from similar forested areas in other parts of the state.
    "Among other things, they will be looking for specific tree conditions that might bring together the LaCrosse vectors such as mosquitoes and the natural vertebrate hosts, chipmunks and squirrels, or possibly mosquitoes and humans," says Moore. "The intersection of vector, virus, and vertebrate host is necessary for transmission. When humans enter areas where all three factors are found, they can inadvertently become a part of the cycle."
    Because the positive pattern appeared in multi-season NDVI imagery, the researchers suspect they are seeing the effect of different times of beginning and ending of leaf development. These patterns will differ for evergreens, deciduous forests, agricultural lands, and urban developments.
    "Most likely, we will find a combination of factors such as forestation, animal population, rainfall, and climate creating the right conditions for transmission," says Brannon.
    CDC researchers in collaboration with USGS researchers, are analyzing smaller-scale imagery from the Landsat TM satellites for the Eastern and Midwestern US, and perhaps Space Shuttle photos of the entire East Coast, to see if any other macro-level ecological mechanisms can be detected with GIS and image processing.

St. Louis: Examining Multiple Data
In Grand Junction, Colo., where CDC researchers recently began studying St. Louis encephalitis, the GIS challenges differ from those in West Virginia. Although a greater variety of data is immediately available in Grand Junction, the scientists have less base knowledge on the disease there. Therefore, in these early phases, they rely on the GIS to determine which data is most significant.
    Researchers already know that the St. Louis transmission cycle involves birds and mosquitoes. Two different mosquito species, Culex tarsalis and Culex pipiens, are common vectors, and each has a distinctly different ecology.
    "We are not sure if one or both species of mosquitoes serves as the vector here," says Steve DeFeyter, environmental health director in the Mesa County (Colo.) Health Department. "The last outbreak of St. Louis encephalitis in humans was reported here in 1985, so we have to find other indicators of the virus besides human illness."
    The Health Department established sentinel chicken flocks at several sites throughout the county in 1995. Blood is drawn from the chickens every two weeks during the summer and tested for encephalitis antibodies to the St. Louis encephalitis virus. In 1995, about 15 percent of the chickens tested positive, indicating the virus was certainly being transmitted by mosquitoes.
    When CDC teamed with Mesa County in 1997, they immediately set up mosquito traps at about 12 locations in the Grand Valley to identify and count the key species every week or month. CDC researchers tracked the results in TNTmips and used a GIS modeling function called surface fitting to map the density of mosquitoes in three dimensions.
    "In this process, the location of the traps provided the x and y coordinates, while the number of a particular species in the trap served as the z value," explains Moore. "This 3D view was overlaid on a map of the county, giving a vivid depiction of where each species was most numerous--presumably where larval habitats were located."
    With these maps generated throughout the summer, the researchers can study the movement of the insects across the county as they mature from their larval to adult stages and begin to disperse. By overlaying this 3D view on the chicken flock maps, researchers hope to detect a timing correlation between infection of chickens and the appearance of one of the two known vector species. So far, no relationship has been established.
    In the next phases of the project, CDC and Mesa County will begin collecting human blood samples from hospitals and test them for the presence of St. Louis antibodies. They suspect the disease may still be striking children in cases too mild to warrant a visit to the doctor, but the antibodies remain in the bloodstream for years. If so, the scientists will know humans are being infected.
    Several other data sets will also be collected. For instance, CDC has already begun counting and testing sparrows, starlings, and song birds that are known to be common hosts of the St. Louis encephalitis virus. Their nesting and feeding habitats will be carefully mapped and compared with mosquito locations. In addition, Mesa County has a wealth of spatial data, including aerial photographs and Landsat images which will provide the land cover data needed to introduce environmental factors to the study.
    "There are just so many variables here with spatial components that we couldn't possibly view and study them all at once without the digital mapping capability of GIS," says DeFeyter.
    Ultimately, these raster and vector data sets will be layered atop one another in TNTmips so the CDC scientists can isolate areas where two or more variables intersect and possibly transmit the disease.
    "We will carefully watch riparian habitats along the Colorado River because this is where song birds nest, mosquitoes lay eggs, and humans canoe, camp, picnic, and fish during the summer," says Moore.
    Many other intersection points exist and will be studied to determine if any coincide with cases of the disease in humans. Once the researchers understand how and where St. Louis encephalitis is transmitted to its victim, they will be better prepared to prevent its spread.

Reporting Back to States
As mentioned, these studies are still in early phases, but they already are providing disease researchers with clues to the carriers and conditions that facilitate the spread of encephalitis. CDC does not normally become directly involved with disease control operations but rather it reports findings back to state and local health departments. State and local health and vector control agencies take appropriate action based on information provided by CDC and their own surveillance activities.
    "Typically, the state and local agencies use this information to educate their citizens on how to avoid mosquito habitats or reduce the likelihood of being bitten in those environments," says Moore. "They also notify health care workers to watch for the disease and report any cases."
    Ultimately, CDC researchers expect that the GIS will enable state authorities to pinpoint exactly where the larval habitats of each vector species are so it can be eradicated at that point rather than waiting until adult mosquitoes have emerged and become involved in the transmission cycle. In the future, states and counties will use GIS-generated habitat information to target specific areas for application of control measures such as water management or pesticide sprays.

About the Author
Kevin Corbley is a freelance writer and consultant specializing in remote sensing, GIS, and GPS. He is located in Denver and may be reached by telephone at 303-722-0312, or by E-mail at [email protected]

Back