Potential corridors and barriers for plague spread in central Asia
© Wilschut et al.; licensee BioMed Central Ltd. 2013
Received: 2 July 2013
Accepted: 23 October 2013
Published: 31 October 2013
Plague (Yersinia pestis infection) is a vector-borne disease which caused millions of human deaths in the Middle Ages. The hosts of plague are mostly rodents, and the disease is spread by the fleas that feed on them. Currently, the disease still circulates amongst sylvatic rodent populations all over the world, including great gerbil (Rhombomys opimus) populations in Central Asia. Great gerbils are social desert rodents that live in family groups in burrows, which are visible on satellite images. In great gerbil populations an abundance threshold exists, above which plague can spread causing epizootics. The spatial distribution of the host species is thought to influence the plague dynamics, such as the direction of plague spread, however no detailed analysis exists on the possible functional or structural corridors and barriers that are present in this population and landscape. This study aims to fill that gap.
Three 20 by 20 km areas with known great gerbil burrow distributions were used to analyse the spatial distribution of the burrows. Object-based image analysis was used to map the landscape at several scales, and was linked to the burrow maps. A novel object-based method was developed – the mean neighbour absolute burrow density difference (MNABDD) – to identify the optimal scale and evaluate the efficacy of using landscape objects as opposed to square cells. Multiple regression using raster maps was used to identify the landscape-ecological variables that explain burrow density best. Functional corridors and barriers were mapped using burrow density thresholds. Cumulative resistance of the burrow distribution to potential disease spread was evaluated using cost distance analysis. A 46-year plague surveillance dataset was used to evaluate whether plague spread was radially symmetric.
The burrow distribution was found to be non-random and negatively correlated with Greenness, especially in the floodplain areas. Corridors and barriers showed a mostly NWSE alignment, suggesting easier spreading along this axis. This was confirmed by the analysis of the plague data.
Plague spread had a predominantly NWSE direction, which is likely due to the NWSE alignment of corridors and barriers in the burrow distribution and the landscape. This finding may improve predictions of plague in the future and emphasizes the importance of including landscape analysis in wildlife disease studies.
Spread of pathogens through a population of hosts can be enhanced or restricted by the landscape. The landscape offers a ‘matrix’ of more and less suitable habitat  and therefore can influence both the abundance and movements of hosts and vectors [2–4]. Suitable habitat areas for hosts and/or vectors may lead to a higher contact rate of hosts  and vectors, and these areas therefore may act as corridors for pathogen transmission through a larger landscape. On the other hand, unsuitable habitat may prevent transmission because it cannot be crossed by hosts or vectors, and thus such areas may act as barriers for pathogen transmission . Corridors and barriers can be present at various spatial scales, depending on the movement and migration patterns of the host and vector. The influence of the landscape or the spatial distribution of the host or vector population on infectious disease dynamics has been shown for several diseases, such as those caused by hantavirus [7, 8], Lyme disease  and plague [10–12]. In this paper, we focus on plague.
Plague is a vector-borne zoonotic disease that is caused by the bacterium Yersinia pestis. The bacterium has a broad host range, but mostly infects rodents , and is spread from host to host by fleas. In Eastern Kazakhstan, the presence of this pathogen has been monitored in great gerbils (Rhombomys opimus), social rodents that live in burrows, for decades. It has been found that outbreaks of plague occur only when a host and flea abundance threshold is exceeded [14, 15], but the direct influence of the landscape on spatial spread and the dynamics of this disease has not previously been investigated.
This study investigates the role of the landscape and the host distribution on the occurrence of plague in the PreBalkhash focus in Eastern Kazakhstan. In this area, the burrows of the great gerbil were recently mapped with considerable success across several landscape types using semi-automated object-based classification . These burrow maps will be used to analyse burrow distribution patterns as a proxy for patterns in a network of great gerbil contacts for the spread of plague. The landscape will be analyzed using characteristics derived from satellite images and a digital elevation model.
We distinguish functional and structural corridors and barriers. Functional barriers are areas where, due to the spatial distribution of hosts, a disease is expected to spread with greater difficulty or not at all. In the PreBalkhash focus, these are areas where relatively few or no great gerbil burrows are present. Functional corridors are areas where, due to the spatial distribution of hosts, a disease is expected to spread more easily. In the PreBalkhash focus, these are areas where the great gerbil burrow density is relatively high. Structural barriers, then, are landscape features that are difficult or impossible for great gerbils to cross. Structural corridors are not identified in this study, but are defined as landscape features where movement of individuals is enhanced (such as roads in cases where humans are the transmitting hosts).
Is the distribution of burrows random, regular or aggregated? At what scale is any structure most pronounced?
Which landscape-ecological variables best explain burrow density?
Can we identify corridors and barriers in the burrow distribution and landscape? If so, what is their spatial configuration and how will this potentially influence the direction and speed of plague spread through the landscape, assuming burrow densities correlate with the speed of plague spread?
Using 46 years of historical epidemiological data, was the observed spread of plague radially symmetric?
Background and study area
The plague system
Great gerbils live in family groups usually consisting of one male, one or more females and their offspring . While foraging, great gerbils usually stay close to their burrow entrances . When gerbils disperse, defined as gerbils permanently moving from their natal colony to another colony during the first year of their life, they move larger distances: dispersal distances for males and females are on average 350 m and 150 m, respectively .
The main vectors of Yersinia pestis are fleas of the genus Xenopsylla. The abundance of the fleas varies with an approximately two-year delay with great gerbil population fluctuations. In addition, flea abundance is influenced by the microclimate inside burrows, which is influenced by the local weather conditions. For example, flea survival decreases when temperatures are high and air humidity is low .
Like many current plague foci, the PreBalkhash focus is located in a semiarid area . At the landscape level, the study area (Figure 1) can be described by three main landforms: the active river, floodplains and dunes. Overall, vegetation cover is low and has a decreasing trend towards Lake Balkhash. Soils are sandy, with variable clay and low organic matter content.
The vegetation cover on the floodplains ranges from totally bare to abundant shrub vegetation. Vegetation cover is higher near to abandoned river branches (Figure 2), most likely due to subsurface flow and hence higher moisture availability. In the dunes, variations in vegetation cover are most pronounced. Vegetation ranges from very sparse short grass vegetation, to more abundant long grass and small shrub vegetation.
The burrows of the great gerbil
The burrows of the great gerbil are omnipresent in the landscape (Figure 2). On the surface of a burrow that is presently or has recently been occupied, no or relatively little vegetation is present, because this has been removed by the great gerbils. Burrows are typically round in relatively flat areas and more elongated in the dunes and have a diameter of 20-60 m . Their large size and bare surface enable the recognition of burrows on satellite images. Burrows have multiple entrances that lead to a network of subterranean tunnels and chambers. Depending on the stability of the soil, a burrow can extend to three meters in depth . Deeper burrows will have a more stable temperature inside the chambers and hence are likely to increase survival of the gerbils . Shallow burrows may provide a barrier to fleas, as fleas are less prone to survive extreme temperatures and low humidity .
Landsat 7 ETM + images (28.5 m resolution) from 2000 and 2001 were used for the creation of multi-scale landscape objects, together with a 90 m resolution Digital Elevation Model (DEM) acquired by the Shuttle Radar Topography Mission (SRTM). Landsat images were converted to top-of-atmosphere reflectance .
In a previous study, burrows of the great gerbil were mapped in two areas covered by high-resolution satellite images . The 2.5 m pixels in the images were grouped into objects, at a scale such that they coincided with the burrows. The burrow objects were then separated from non-burrow objects by a Random Forest classifier, using spectral and neighbour signatures. The end product of that study has a mean overall accuracy of 90%. For the present study, the polygon-objects representing burrow systems were converted into point data files, using the coordinates of the centroid of the polygon as burrow locations.
The plague dataset consists of biannual data collected by the Kazakh Scientific Center for Quarantine and Zoonotic Disease (formerly Anti-Plague Institute) between 1949 and 1995 in about 90 so-called Primary Squares (PSQs). PSQs are plague monitoring units (measuring 18.5 km by 19.6 km) that were and are still used by the anti-plague services in Kazakhstan. Every spring and autumn great gerbils and their fleas were caught in a number of PSQs and tested for infection with Yersinia pestis. The plague dataset thus shows presence or absence of plague at different locations in the PreBalkhash focus through the years. For more detailed information on the collection of the data set we refer to a previous study .
In total, the data set used consists of 3493 data points, distributed over the 90 PSQs. Of those data points, 1750 were collected in spring and 1743 in autumn. In 8.8% of the data points, plague was detected.
Selecting research areas
Cover of floodplain and dunes in the three research areas
Characterising burrow patterns
To characterise the burrow distribution patterns, nearest neighbourhood statistics were calculated for the burrow patterns in each research area using the statistical environment R . Also, variograms  were created to get insight in the spatial correlation in the abundance of burrows, by first rasterizing the burrow maps into burrow-density maps, using cells of 100 m, 400 m and 1000 m, respectively. The range was also calculated for each variogram, using the R package gstat.
For each research area, 100 random dummy burrow patterns with an equal number of burrows as in the real burrow pattern were generated using the R package spatstat, to be able to compare completely random spatial point patterns with the burrow distributions.
Analysing the spatial burrow density pattern using landscape objects
To investigate whether there is a spatial structure in the burrow distribution and whether there is a spatial scale at which the spatial structure is most pronounced, the density of burrows across the landscape was evaluated using multi-scale object-based image analysis (MSOBIA). MSOBIA has been suggested for analysing the relationship between landscape and animal abundance [29, 30], but, at least until now, has never been used with real animal abundance and distribution data. In the present study, the burrow maps make it possible to test whether landscape objects coincide with the patterns in the burrow maps.
Creating landscape object layers
The MSOBIA was performed using Landsat ETM + satellite images and an SRTM elevation model. Three layers were extracted from the images: Tasselled Cap  soil Brightness (Brightness), Tasselled Cap Greenness (Greenness), and the standard deviation in the SRTM elevation (SRTM-SD). Brightness is a proxy for soil colour and hence soil type, Greenness is a proxy for the vegetation cover, and SRTM-SD is a proxy for the variation in the local topography. Together they capture the major variation in the landscape.
The three layers were segmented into objects representing the landscape patches by grouping neighbouring pixels based on a homogeneity criterion. At a high homogeneity criterion value, i.e. the pixels are required to be quite similar in order to be grouped, many small objects are created, whereas at a low homogeneity criterion value, larger objects are obtained. Segmentation was started at low homogeneity with a mean object size of ~17 km2, and was then repeated step-wise with increasing homogeneity. In the final layer with most objects, the mean object size was approximately 200 m2. In total 20 landscape object layers were produced. Segmentation level was plotted against mean object area for each research area, to compare the landscape homogeneity in the three areas. Also, mean values for Greenness, Brightness and SRTM-SD were calculated.
Calculating burrow density differences within landscape objects layers
m = number of objects
n = number of neighbouring objects to object i
BDi = burrow density of object i
BDij = burrow density of jth object neigbouring object i.
This variable will show relatively high values when landscape objects coincide with burrow density patterns, and low values when they do not coincide. Also, when the burrow pattern itself is random, instead of spatially structured, the value of MNABDD will be low. MNABDD was calculated as follows. First, for all the objects in the landscape object layers a burrow density value was calculated. To prevent extreme density values, objects at the edge of the research areas smaller than 0.025 km2 were removed.
The landscape object layers with their associated burrow density values were then used to calculate, for every object, the absolute neighbour burrow density difference (NABDD), i.e. the difference in burrow density between an object and its neighbouring objects. After this, MNABDD was calculated by taking the mean of the NABDDs for every layer.
Finally, to evaluate whether the burrow patterns in the three research areas are spatially structured rather than randomly distributed, MNABDD was calculated using the 100 random dummy point patterns. For every research area, using the landscape object layers and the square cell layers, the confidence intervals of MNABDD using the 100 random point patterns were calculated.
Relating burrow density to landscape-ecological variables
To investigate which landscape-ecological variables best explain burrow density, multiple linear regression was performed. Ideally, in cases where burrow density can be explained very well using the landscape-ecological variables, the model can be used to predict burrow density for areas where no burrow maps are available. This could be very helpful for epidemiological models.
Multiple linear regression was performed using the values from the rasterized burrow maps with a cell size that was well within the variogram range, and also small enough to capture local variations in burrow density. Four explanatory variables were used: Greenness, Brightness, SRTM-SD and SRTM. Second order interactions were also included. Spatial autocorrelation was accounted for in the model ; that is, the model fitted to data was a linear mixed-effects model with an exponential correlation structure with a nugget effect (function lme, in R package nlme). We used the AIC , a measure of the relative quality of a model, to determine the relative support for each model. This was first done for the three research areas combined. The model with the lowest value of AIC was chosen in case ΔAIC was greater than 2. A similar model using the same combination of variables was then also applied for the three areas separately, so that differences between the three areas could be evaluated.
Mapping corridors and barriers
To detect the functional corridors and barriers in the burrow distribution, burrow densities were investigated using the landscape object layers. To determine thresholds for the corridors and barriers, the mean and standard deviation of the burrow densities in the three research areas (using the 400 m rasters) were extracted. The threshold for the functional corridors was set at mean burrow density plus one standard deviation. The threshold for the functional barriers was set at mean burrow density minus one standard deviation.
Then the functional corridors and barriers were mapped in each research area, using only areas larger than ~700 m × 700 m (twice the average great gerbil migration distance ).
The structural barriers were mapped by identifying the rivers, canals and irrigated agricultural areas in the area. This is because great gerbils do not likely cross them , and the infected fleas on them are inclined to leave the gerbil when they go into the water.
Analysing the spatial configuration of functional corridors and barriers
The procedure was repeated using a homogeneous burrow density map as input, with values equal to the mean burrow density of the three research areas. The resulting mean cost distance value was used to normalize the cumulative resistance axis.
Analysing the direction of plague spread in the plague data set (1949–1995)
If all conditions were equal in all directions, i.e. if great gerbil densities and abundances were equal and no structural barriers or corridors existed, plague spread can be assumed to have been on average radially symmetric. The spread of plague might however be influenced (in) directly by the landscape.
As several gaps in the data exist, the variables describing directional spread were aggregated two by two. In effect, two nearly identical models were fitted to the data: one with the two variables describing NWSE + WE (ax 1) and the perpendicular NESW + NS (ax 2); and one with variables describing the other two axes. The two models predict the probability that plague will be present in a given PSQ i at time t (yi,t = 1) given four explanatory variables: the presence/absence of plague in the same square six months earlier (yi,t-1); the proportion of neighbouring squares on one of the four combined axes where plague was present six months earlier (nax 1i,t-1); the proportion of neighbouring squares on the perpendicular axis where plague was present six months earlier (nax 2i,t-1); and the great gerbil abundance in the square, averaged over the past two years (g t,i ). The models were used to test whether plague spread was significantly stronger along the NWSE + NS, NESW + NS, NWSE + WE or NESW + WE axes (see spatial example in Figure 5). Estimates of the magnitude of spread along these axes were also obtained.
Burrow network characteristics
The variograms (Additional file 2: Figure S1) show that variation in burrow density is highest in area West and lowest in area East. Furthermore, they show that the range, or in other words the length at which there is spatial structure in the data, varies from ~2.4 km in area North and ~3 km in area East, to ~5.2 km in area West.
Burrow density distribution at increasing scales
For all three areas, the burrow patterns (black lines) show significantly higher values than the random patterns (grey). This is the case when the values are calculated using the square cell layers (top row), as well as using landscape objects layers (middle row). This indicates that there is spatial structure in the burrow distributions in all three areas.
To evaluate the effects of using landscape objects, it is useful to compare the results from the square cell layers with the landscape object layers. For area North, the analysis using the landscape objects gives significantly higher values from layers with mean object sizes of ~3 km2 and higher. This suggests that the landscape objects better represent the structure in the burrow density compared to square cells in this area. For area East this is the case as for layers with mean object sizes between ~4 km2 and 9 km2. For area West the square cells seem to describe the data better than the landscape objects, although this is only significant for layers with a mean object size greater than ~5 km2.
The influence of scale on the MNABDD can be evaluated by looking at the variation of MNABDD with increasing object size. In area West, the difference with the mean values of the 100 random point patterns is largest at scale with a mean patch area of 3.95 km2. In area North, this difference is still increasing at 10 km2. In area East, the pattern is comparable with the pattern in area West and the maximum occurs at ~6 km2. In other words, at these scales the burrow patterns differ most from random point patterns.
Relation burrow density and landscape-ecological variables
Results of the multiple regression, for the three areas combined (top row) and the areas separately
Intercepts, coefficients and p-values multiple regression
The model with the lowest AIC (ΔAIC was >2) uses Greenness, Brightness and SRTM-SD and the interaction term Greenness:SRTM-SD (marginal R2 is 0.1 and conditional R2 is 0.2; for calculation is referred to ). When the spatial autocorrelation was not taken into account, all variables had a significant contribution.
The interaction term is positive, which indicates that on the floodplain (i.e. low SRTM-SD) the negative correlation of Greenness with burrow density is larger than in the dunes. The same variables were used to create models for the three separate areas (Table 2). Other models were also tested, but the models that used the same variables as for the combined model gave for all three areas the lowest AIC. Greenness shows in all areas the lowest p-values. The SRTM-SD has highest p-values, however if the interaction term was removed, SRTM-SD showed lower p-values than Brightness.
Corridors and barriers
The structural barriers (shown in blue in Figure 1) show that the Ili River forms a large barrier.
Spatial configuration functional corridors and barriers
Predominant directions of plague spread
To test along which axis most plague spread occurred using the 1949–1995 plague presence/absence data, two tests were carried out: one testing whether plague spread occurred more along the NWSE + WE axes or along the NESW + NS axes, and the other whether plague spread occurred more along the NWSE + NS axes or along the NESW + WE axes (Figure 5). 1423 data points, i.e. seasons in which presence or absence of plague was recorded in one of the PSQs, were available for the former test; 1372 data points were used for the latter. The number of data points is different for the two tests, because the data points where less than two of the neighbours in each direction had been sampled for plague were omitted.
Parameter estimates for the directional spread of plague model
NWSE + WE vs. NESW + NS
NWSE + NS vs NESW + WE
NWSE + WE
NESW + NS
NWSE + NS
NESW + WE
β estimates (±s.e.)
1.83 (± 0.5)
0.30 (± 0.5)
1.06 (± 0.59)
1.32 (± 0.56)
In this study, the distribution of great gerbil burrows was examined in three areas located in a plague focus in Eastern Kazakhstan. Multi-scale object analysis indicated that the burrow distribution differed significantly from random. This result complies with other studies  and field observations suggesting that some landscapes are more suitable than others for the great gerbil. The spatial structure of the burrow pattern is explained best using the variables Greenness, Brightness, SRTM-SD and the interaction term SRTM-SD:Greenness. Higher Greenness values result in lower burrow density values, especially in the floodplain areas (i.e. when SRTM-SD is low). In the dunes (i.e. when SRTM-SD is high), Greenness has a less negative correlation with burrow density. A possible explanation for this is that great gerbils might prefer short vegetation over dense shrubs, something which has been shown for the Mongolian gerbil . In the floodplain areas, high Greenness values likely correspond to the presence of tall and dense shrubs. In the dunes however, higher Greenness values correspond to dense grass cover.
The multiple regression also showed that a large part of the variation in burrow density cannot be explained by the three landscape-ecological variables. One variable that might further explain the variation in burrow density is the height of the ground water table . Which other variables need to be considered is yet uncertain. Some uncertainty arises from the burrow maps themselves, as they have a mean overall accuracy of 90%. Generally, the burrow maps are more accurate in areas with little vegetation and few takirs . Apart from spatial heterogeneity in the great gerbil distribution arising from heterogeneity in the landscape, it can also arise from the great gerbil’s social and behavioural interactions, or from predator-host interactions . Although the great gerbil distribution in the study area was found to be non-uniform, compared to spatial distributions of other animals, the great gerbil distribution is likely relatively homogeneous. It is useful to compare spatial structures of host populations, as this can give insight in how different spatial host distributions can lead to different disease dynamics. However, few studies exist that examine the spatial structure of a population in detail. Nevertheless, some comparisons are possible. For example, prairie dogs, which are hosts of plague and live in desert areas as well, have an aggregated distribution; they live in so called “prairie dog towns”. These towns range from 0.5 to 100 ha, within which prairie dog densities also vary . In between these colonies, no prairie dogs are present.
MNABDD was developed in this study, because it made it possible to evaluate the effect of using landscape objects. Another advantage of the method used here is that it possible to view the spatial pattern of the NABDD (neighbour absolute burrow density difference) within the layers, since it is calculated per object or square cell. This makes it possible to examine whether there are certain areas where the burrow differences between objects are larger than in other areas. For example, in area North, the largest NABDD values occur in the mid-south of the area, where there is an alternation of dunes and floodplain.
In this study, corridors and barriers were mapped in the plague focus in eastern Kazakhstan. In theory, barriers can influence disease spread and persistence in the following ways. Barriers influence the connectivity between subpopulations in a metapopulation, and hence will influence the persistence of the disease, and also the speed and direction of disease spread [41, 42]. The permeability of the barriers, i.e. whether they are completely impermeable or only partially, determines the contact rates between the subpopulations on opposite sides of the barriers, and hence can influence the pattern of disease spread . This makes it insightful to map barriers in a disease focus [42, 43]. The structural barriers that were identified in the research area comprised canals, rivers and lakes. Although these barriers are likely not 100% impermeable to great gerbils, they do likely form a considerable barrier to plague . On the other hand, disease spread can also be faster alongside rivers, which has for example been shown for rabies . However, it has not been ruled out that that birds or larger mammals occasionally spread plague, in which case these structural barriers could be overcome .
The functional barriers were identified using the burrow distribution. These barriers were most prominent in area East, where there are large areas with dense shrubs present. In area West, the amount of barriers may be underestimated in the areas where many takirs are present. In area North the barriers coincide mostly with the dunes. Currently, no empirical data exist to define a burrow density threshold value. However, models such as the flea-density-threshold model  predict occupied burrow density threshold values below which no invasion of plague can occur. However, as the occupancy varies over space and time these modelled values cannot be translated into a single threshold burrow density value. The value we used to determine the barriers was therefore based on the statistical distribution of the burrow density data. Besides the barriers identified in this study, there may be other features acting as barriers. For example, areas where fleas cannot survive, for instance due to too high temperatures or a too low moisture content of the soil, also could act as barriers.
Apart from barriers, functional corridors using the burrow density were also mapped. No structural corridors were mapped, since no data or information existed making it possible to map them. In theory, corridors will increase the connectivity between subpopulations  and hence possibly influence the speed and direction of disease spread , and the persistence of disease . In this study the functional corridors were defined as areas with high burrow densities. The underlying assumption for this is that transport of fleas between burrow systems occurs more often in high-density areas, because when distances between occupied burrows are shorter, it is more likely that fleas will be exchanged and hence plague will be transmitted. Possible other corridors for pathogen transmission could be areas where the conditions are optimal for fleas. The spatial arrangement of the corridors showed that in general the alignment of the corridors was NWSE. This was confirmed by the cost distance analysis, which showed that in all areas the NWSE direction had the highest burrow densities. An explanation for this is the geography of the landscape: the abandoned Ili River branches have a WE and NWSE direction as a result of the local topography (see Figures 1 and 2). Also the dunes are most often aligned along this axis (Figure 2), which is a result of the dominant wind directions. Both the abandoned branches as well as the dunes have lower burrow densities and hence these features are less permeable, which promotes movement along these features instead of crossing them.
The spatial arrangement of the corridors and barriers raised the question whether plague spread is radially symmetric in the plague focus, or whether there exists a predominant axis for plague spread. That the geography of a landscape and the host population structure contributes to the direction and velocity of disease spread has been shown in several studies [39, 47, 48]. In this study it was shown that corridors and barriers at the PSQ scale have a predominantly NWSE direction. In the PreBalkhash focus as a whole, the larger landscape features also have a predominantly NWSE direction, as Figure 1 and landscape maps of the PreBalkhash  show. Thus as well at a local as at a landscape scale, there are many features with a NWSE alignment. In the plague dataset, where plague spread was examined on a PSQ to PSQ scale, the predominant direction was found to be NWSE + WE. A plausible explanation for this is the NWSE structure in the spatial distribution of the burrows and the landscape. Although no detailed statistically analysed data on the direction of great gerbil movements yet exist, it has been suggested  that great gerbils – for example in the dunes – prefer to move from burrow to burrow on the sides of the dunes, as this is easier than crossing the dunes and also likely a safer route in terms of predation. When great gerbils are infected with plague, this would make plague spread in these directions (NW to SE and vice versa) likely. The same can apply for example to takirs that are aligned along the NWSE axis. Directional plague spread can thus be a result of gerbils moving along corridors, or in between and along barriers. It is currently yet unclear which contributes most to the directional spread of plague; for that detailed movement data of great gerbils are needed.
The plague spread model showed that spread is also more dominant along the WE axis. This is not as obviously related to the alignment of the landscape. Although some of the abandoned rivers have a WNW alignment rather than a NW alignment, overall there are less landscape features that can explain the more dominant plague spread on the WE axes. Another unanswered question is to which extent plague movement equates to gerbil movement. Plague spread is partly a stochastic process, but will depend on the behaviour of the fleas, as well as on the movement patterns of the gerbils. It is not yet clear where and when transmission of plague occurs most: is that in empty burrows visited by great gerbils from neighbouring burrows so that fleas can jump on susceptible great gerbil hosts, or does transmission occur mostly when young great gerbils migrate - along a barrier or corridor perhaps - to start a new family? Although this is yet unclear, the finding that the landscape plays a role in the spatial dynamics of plague in great gerbil populations in Central Asia underlines the value of including spatial landscape analysis in wildlife disease studies. Therefore, it is useful to include environmental variables, such as the topography, in existing plague models.
In this study the landscape and burrow distribution in a plague focus in Eastern Kazakhstan were studied extensively. This was done to gain insight in the possible (in)direct influence of the landscape on the spread of plague. Plague data from 1949–1995 were used to determine whether a predominant direction of plague spread exists in this area.
The great gerbil burrow network is spatially structured and not randomly distributed. The landscape likely influences great gerbil burrow density structure. Corridors and barriers showed a NWSE alignment. This axis was also found to be predominant in the 46-year plague dataset. This makes it plausible that the landscape and the burrow network influences the direction of plague spread. This finding improves the current knowledge of the spatiotemporal dynamics of plague in Central Asia and may improve the prediction of plague in this area.
This project was part of the Wellcome Trust project (090213/z/09/z) ‘Advances in the prediction of plague outbreaks in Central Asia’ and of the VICI project 918.56.620 ‘Population dynamics and control of infectious diseases: modelling interaction at different scales’ funded by the Netherlands Research Organisation. The SPOT images were obtained through the ISIS programme of CNES and SPOTIMAGE in ISIS projects 395 and 476 ‘Plague Surveillance’, Copyright CNES. Finally, we would like to thank P.C. Wilson for his help with the calculations and N. Wanders for his enthusiasm to help with R.
- Forman RTT, Godron M: Landscape Ecology. 1986, Canada: John Wiley & Sons, IncGoogle Scholar
- Apps CD, Mclellan BN, Woods JG, Proctor MF: Estimating grizzly bear distribution and abundance relative to habitat and human influence. J Wildl Manage. 2004, 68 (1): 138-152. 10.2193/0022-541X(2004)068[0138:EGBDAA]2.0.CO;2.View ArticleGoogle Scholar
- Delattre P, Giraudoux P, Baudry J, Quéré JP, Fichet E: Effect of landscape structure on Common Vole (Microtus arvalis) distribution and abundance at several space scales. Landscape Ecol. 1996, 11 (5): 279-288. 10.1007/BF02059855.View ArticleGoogle Scholar
- Pardini R, de Souza SM, Braga-Neto R, Metzger JP: The role of forest structure, fragment size and corridors in maintaining small mammal abundance and diversity in an Atlantic forest landscape. Biol Conserv. 2005, 124 (2): 253-266. 10.1016/j.biocon.2005.01.033.View ArticleGoogle Scholar
- McCallum H, Barlow N, Hone J: How should pathogen transmission be modelled?. Trends Ecol Evol. 2001, 16 (6): 295-300. 10.1016/S0169-5347(01)02144-9.PubMedView ArticleGoogle Scholar
- Wheeler D, Waller L: Mountains, valleys, and rivers: the transmission of raccoon rabies over a heterogeneous landscape. J Agric Biol Environ Stat. 2008, 13 (4): 388-406. 10.1198/108571108X383483.PubMedPubMed CentralView ArticleGoogle Scholar
- Langlois J, Fahrig L, Merriam G, Artsob H: Landscape structure influences continental distribution of hantavirus in deer mice. Landscape Ecol. 2001, 16 (3): 255-266. 10.1023/A:1011148316537.View ArticleGoogle Scholar
- Guivier E, Galan M, Chaval Y, Xuéreb A, Ribas Salvador A, Poulle M, Voutilainen L, Henttonen H, Charbonnel N, Cosson JF: Landscape genetics highlights the role of bank vole metapopulation dynamics in the epidemiology of Puumala hantavirus. Mol Ecol. 2011, 20 (17): 3569-3583.PubMedGoogle Scholar
- Barrios JM, Verstraeten WW, Maes P, Clement J, Aerts JM, Farifteh J, Lagrou K, Van Ranst M, Coppin P: Remotely sensed vegetation moisture as explanatory variable of Lyme borreliosis incidence. Int J Appl Earth Observation Geoinformation. 2012, 18: 1-12.View ArticleGoogle Scholar
- MacMillan K, Enscore RE, Ogen-Odoi A, Borchert JN, Babi N, Amatre G, Atiku LA, Mead PS, Gage KL, Eisen RJ: Landscape and residential variables associated with plague-endemic villages in the west Nile region of Uganda. Am J Trop Med Hyg. 2011, 84 (3): 435-442. 10.4269/ajtmh.2011.10-0571.PubMedPubMed CentralView ArticleGoogle Scholar
- Keeling MJ, Gilligan CA: Metapopulation dynamics of bubonic plague. Nature. 2000, 407 (6806): 903-906. 10.1038/35038073.PubMedView ArticleGoogle Scholar
- Neerinckx S, Peterson A, Gulinck H, Deckers J, Leirs H: Geographic distribution and ecological niche of plague in sub-Saharan Africa. Int J Health Geogr. 2008, 7 (1): 54-10.1186/1476-072X-7-54.PubMedPubMed CentralView ArticleGoogle Scholar
- Gage KL, Kosoy MY: Natural history of plague: perspectives from more than a century of research*. Annu Rev Entomol. 2005, 50 (1): 505-528. 10.1146/annurev.ento.50.071803.130337.PubMedView ArticleGoogle Scholar
- Davis S, Begon M, De Bruyn L, Ageyev VS, Klassovskiy NL, Pole SB, Viljugrein H, Stenseth NC, Leirs H: Predictive thresholds for plague in Kazakhstan. Science. 2004, 304 (5671): 736-738. 10.1126/science.1095854.PubMedView ArticleGoogle Scholar
- Reijniers J, Davis S, Begon M, Heesterbeek JA, Ageyev VS, Leirs H: A curve of thresholds governs plague epizootics in central Asia. Ecol Lett. 2012, 15: 554-560. 10.1111/j.1461-0248.2012.01767.x.PubMedView ArticleGoogle Scholar
- Wilschut LI, Addink EA, Heesterbeek JAP, Dubyanskiy VM, Davis SA, Laudisoit A, Begon M, Burdelov LA, Atshabar BB, de Jong SM: Mapping the distribution of the main host for plague in a complex landscape in Kazakhstan: an object-based approach using SPOT-5 XS, landsat 7 ETM+, SRTM and multiple random forests. Int J Appl Earth Observation Geoinformation. 2013, 23: 81-94.View ArticleGoogle Scholar
- Atshabar BB, Burdelov LA, Ageyev VS, Aukabirov SA, Dubyanskiy VM, Grazhdanov AK, Zhumadiolova ZB, Izbanova UA, Kuznetsov AN, Kunitsa TN, Lukhnova LJ, Malakhov DV, Meka-Mechenko VG, Meka-Mechenko TV, Nekrassova LE, Nurmakhanov TI, Pazylov EK, Pole SB, Prihodko DE, Rybakova MA, Sadovskaja VP, Sapozhnikov VI, Syzdykov MS, Mamadaliev SM, Orynbaev MO, Aburaimov EO KMM: Atlas of Bacterial and Virus Zoonotic Infections Distribution in Kazakhstan. 2010, Almaty, Kazakhstan: M.Aikimbayev´s KSCQZDGoogle Scholar
- Randall JA, Rogovin K, Parker PG, Eimes JA: Flexible social structure of a desert rodent, Rhombomys opimus: philopatry, kinship, and ecological constraints. Behav Ecol. 2005, 16 (6): 961-973. 10.1093/beheco/ari078.View ArticleGoogle Scholar
- Randall JA, Rogovin KA, Shier DM: Antipredator behavior of a social desert rodent: footdrumming and alarm calling in the great gerbil, Rhombomys opiums. Behav Ecol Sociobiol. 2000, 48 (2): 110-118. 10.1007/s002650000199.View ArticleGoogle Scholar
- Rapoport LP, Melnichuck EA, Orlova LM, Nuriev KK: Comparative analysis of the flea fauna and its epizootic importance in the deserts of Southern Kazakhstan. Entomol Rev. 2010, 90 (8): 1003-1013. 10.1134/S0013873810080051.View ArticleGoogle Scholar
- Ray C, Collinge SK: Potential effects of a keystone species on the dynamics of sylvatic plague. Disease ecology: community structure and pathogen dynamics. Edited by: Collinge SK, Ray C. 2006, Oxford, UK: Oxford University Press, 204-218.Google Scholar
- Laity J: Deserts and desert environments. 2008, UK: Wiley-BlackwellGoogle Scholar
- Naumov NP, Lobachev VS: Ecology of desert rodents of the U.S.S.R. (Jerboas and Gerbils). Rodents in desert environments. Volume 28. Edited by: Prakash I, Ghosh PK. 1975, The Hague, The Netherlands: Dr. W. Junk Publishers, 549-598.Google Scholar
- Dubyanskiy MA: Types of settlements of the great gerbil and their epizootic significance in the Aral karakum desert. (Original Russian title: типы поселений большой песчанки и их эпизоотологическое значение в приаральских каракумах). Зоол журн (J Zool). 1963, 42 (1): 103-113.Google Scholar
- Chander G, Markham BL, Helder DL: Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors. Remote Sens Environ. 2009, 113 (5): 893-903. 10.1016/j.rse.2009.01.007.View ArticleGoogle Scholar
- Heier L, Storvik GO, Davis SA, Viljugrein H, Ageyev VS, Klassovskaya E, Stenseth NC: Emergence, spread, persistence and fade-out of sylvatic plague in Kazakhstan. Proc R Soc B Biol Sci. 2011, 278: 2915-2923. 10.1098/rspb.2010.2614.View ArticleGoogle Scholar
- R Development Core Team: R: A Language and Environment for Statistical Computing. 2011, Vienna, Austria: R Foundation for Statistical ComputingGoogle Scholar
- Burrough PA, McDonnell RA: Principles of Geographical Information Systems. 1998, Oxford: Oxford University PressGoogle Scholar
- Burnett C, Blaschke T: A multi-scale segmentation/object relationship modelling methodology for landscape analysis. Ecol Model. 2003, 168 (3): 233-249. 10.1016/S0304-3800(03)00139-X.View ArticleGoogle Scholar
- Ehlers M: Proceedings of the Proc. SPIE 4545, Remote Sensing for Environmental Monitoring, GIS Applications, and Geology, Multi-Scale Image Analysis for Ecological Monitoring of Heterogeneous, Small Structured Landscapes: September 17, 2001. 2001, Toulouse: SPIEGoogle Scholar
- Crist EP, Cicone RC: A physically-based transformation of thematic mapper data–-the TM tasseled Cap. IEE Trans Geoscience Remote Sensing. 1984, GE-22 (3): 256-263.View ArticleGoogle Scholar
- Dormann CF, McPherson JM, Araújo MB, Bivand R, Bolliger J, Carl G, Davies RG, Hirzel A, Jetz W, Daniel Kissling W, Kühn I, Ohlemüller R, Peres-Neto PR, Reineking B, Schröder B, Schurr FM, Wilson R: Methods to account for spatial autocorrelation in the analysis of species distributional data: a review. Ecography. 2007, 30 (5): 609-628. 10.1111/j.2007.0906-7590.05171.x.View ArticleGoogle Scholar
- Akaike H: A new look at the statistical model identification. IEEE Trans Automatic Control. 1974, 19: 716-723. 10.1109/TAC.1974.1100705.View ArticleGoogle Scholar
- Lavrovskiy AA, Pavlov AN, Barkalova AG, Ivanov BC, Trofimov VI: On the extension of the boundaries of the area of the great gerbil (Rhombomys opimus) and its next move to the right bank of the Urals (original Russian title: О расширении границ ареала большой песчанки (Rhombomys opimus) и очередном ее переходе на правый берег Yрала). Зоол журн (J Zool). 1978, 57 (2): 306-308.Google Scholar
- ESRI: Using ArcGISTM Spatial Analyst. 2001, US: RedlandsGoogle Scholar
- Nakagawa S, Schielzeth H: A general and simple method for obtaining R2 from generalized linear mixed-effects models. Methods Ecol Evol. 2013, 4 (2): 133-142. 10.1111/j.2041-210x.2012.00261.x.View ArticleGoogle Scholar
- Liu W, Wan X, Zhong W: Population dynamics of the Mongolian gerbils: seasonal patterns and interactions among density, reproduction and climate. J Arid Environ. 2007, 68 (3): 383-397. 10.1016/j.jaridenv.2006.07.002.View ArticleGoogle Scholar
- Leontjeva MN: The range of the great gerbil and groundwater (original Russian title: ареал большой песчанки и грунтовые воды). Зоол журн (J Zool). 1968, 47 (9): 1422-1425.Google Scholar
- Real LA, Biek R: Spatial dynamics and genetics of infectious diseases on heterogeneous landscapes. J R Soc Interface. 2007, 4 (16): 935-948. 10.1098/rsif.2007.1041.PubMedPubMed CentralView ArticleGoogle Scholar
- Johnson WC, Collinge SK: Landscape effects on black-tailed prairie dog colonies. Biol Conserv. 2004, 115 (3): 487-497. 10.1016/S0006-3207(03)00165-4.View ArticleGoogle Scholar
- Jesse M, Heesterbeek H: Divide and conquer? persistence of infectious agents in spatial metapopulations of hosts. J Theor Biol. 2011, 275 (1): 12-20. 10.1016/j.jtbi.2011.01.032.PubMedView ArticleGoogle Scholar
- Smith DL, Lucey B, Waller LA, Childs JE, Real LA: Predicting the spatial dynamics of rabies epidemics on heterogeneous landscapes. Proc Natl Acad Sci. 2002, 99 (6): 3668-3672. 10.1073/pnas.042400799.PubMedPubMed CentralView ArticleGoogle Scholar
- Ostfeld RS, Glass GE, Keesing F: Spatial epidemiology: an emerging (or re-emerging) discipline. Trends Ecol Evol. 2005, 20 (6): 328-336. 10.1016/j.tree.2005.03.009. SPECIAL ISSUE: BUMPER BOOK REVIEWPubMedView ArticleGoogle Scholar
- Russell CA, Smith DL, Waller LA, Childs JE, Real LA: A priori prediction of disease invasion dynamics in a novel environment. Proc R Soc Lond Ser B Biol Sci. 2004, 271 (1534): 21-25. 10.1098/rspb.2003.2559.View ArticleGoogle Scholar
- Beier P, Noss RF: Do habitat corridors provide connectivity?. Conserv Biol. 1998, 12 (6): 1241-1252. 10.1046/j.1523-1739.1998.98036.x.View ArticleGoogle Scholar
- McCallum H, Dobson A: Disease, habitat fragmentation and conservation. Proc R Soc Lond Ser B Biol Sci. 2002, 269 (1504): 2041-2049. 10.1098/rspb.2002.2079.View ArticleGoogle Scholar
- Hendrickx G, Gilbert M, Staubach C, Elbers A, Mintiens K, Gerbier G, Ducheyne E: A wind density model to quantify the airborne spread of culicoides species during north-western Europe bluetongue epidemic, 2006. Prev Vet Med. 2008, 87 (1): 162-181.PubMedView ArticleGoogle Scholar
- Grenfell B, Harwood J: (Meta)population dynamics of infectious diseases. Trends Ecol Evol. 1997, 12 (10): 395-399. 10.1016/S0169-5347(97)01174-9.PubMedView ArticleGoogle Scholar
- Burdelov AS, Bondar EP, Zhuravljova VI: Mobility of great gerbils and its epizootical value in continuous settlements in the northern desert (Southern PreBalkhash) (Original Russian title: Подвижность больших песчанок и ее эпизоотологическое значение в сплошных поселениях в северной пустыне (Южное Прибалхашье)). Зоол журн (J Zool). 1964, 43 (1): 115-124.Google Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.