Landscape Predictors of Tick-Borne Encephalitis in Latvia: Land Cover,Land Use,and Land Ownership

Abstract

Although the presence of tick-borne encephalitis (TBE) virus circulating in tick populations depends on large-scale patterns of climate, and the local density of infected ticks depends on the abundance of mammalian hosts, the risk of human infection depends on the access and use by human populations of tick-infested habitats, particularly forests, at the landscape level. We investigated the incidence of reported TBE cases in rural parishes (i.e., municipalities) in Latvia. The following major characteristics of parishes were considered: whether their environment is suitable for tick and tick-host populations (depending on land cover); whether the local human population is likely to enter the forest on a regular base (depending on land use); and whether the spatial distributions of these two aspects are likely to intersect, through access rules (as a function of land ownership). The results indicated that all three aspects are important in explaining and predicting the spatial distribution of TBE cases in the rural areas of Latvia. The concept of landscape is here given new depth by consideration of its physical structure, its use by human populations, and its accessibility as modulated by ownership.

Introduction

H uman infection with tick-borne encephalitis (TBE) virus occurs widely, but focally, throughout much of Europe. Ixodes ricinus and Ixodes persulcatus ticks are the sources of TBE infection, acting as both reservoir hosts and vectors. The circulation of TBE virus in tick populations depends on large-scale climate patterns (Randolph 2000), and the local density of infected ticks depends on the abundance of their mammalian hosts [large mammals to feed adult ticks (Moràn Cadenas et al. 2007) and rodents to transmit the virus (Labuda and Randolph 1999)]. The risk of human infection partly depends on fine-scale factors such as use of, management practices in, and public access to forests, where ticks live predominantly. This is especially true as different demographic and socioeconomic groups use forest resources for a variety of reasons, such as collecting wood or wild foods and recreation. These factors and the spatial distribution and nature of human land use determine the spatial distribution and incidence of infection.

The abrupt and massive upsurge of TBE that was observed in the Central and Eastern European countries after the end of the Soviet rule indicates that enzootic cycles of the virus are well established in this region, with the increase in human infection evidently triggered by environmental and socioeconomic changes related to political transition (Šumilo et al. 2007, 2008a, Randolph 2008). A spatial analysis of TBE incidence in the 85 counties of Estonia, Latvia, and Lithuania for the period 1993–1998 identified a significant negative relationship with the percentage coverage of habitats classified as of medium suitability for ticks (urban green sites, pastures, and natural grassland), based on data from the CORINE Land Cover project (www.eea.europa.eu/themes/landuse/) (Šumilo et al. 2006). Habitats of high suitability were not significant predictors (although their extent was negatively correlated with that of habitats of medium suitability). These statistical results demand greater understanding of the interactions between the spatial distribution of different types of tick habitats and human contact with them, through use and access. Large areas of tick habitats (woodlands and forests) would be irrelevant to human infection if humans rarely enter them.

Here we present an investigation at the finer scale of rural parishes (i.e., municipalities) in Latvia, using land cover data from a variety of sources, to distinguish different land use patterns, including land ownership and forest management. The aim was to identify the risk factors for TBE infection among human populations, originating in both ecological and socioeconomic landscapes, and their spatial intersection. We investigated the incidence in rural parishes of the whole country and then focused on Vidzeme, the north-eastern region of Latvia, a more homogeneous rural area of the country with few large towns. This study design allows the scale-dependency of the spatial determinants of symptomatic TBE infection to be investigated.

Modeling vector-borne disease risk simply according to the habitat requirements of a vector, as determined by land cover attributes (e.g., Daniel et al. 1998, Glass et al. 2000, Jackson et al. 2006), is unlikely to include all factors necessary for understanding the relationship between environment and human disease risk (Beck et al. 2000, Curran et al. 2000). In our comprehensive consideration of landscapes, land cover and physical structure relate to the ecology of ticks and their hosts, land use to human exposure, and land ownership to the institutional control on the potential overlap between the two previous factors. The factors contributing to each of these elements in the case of rural Latvia (Fig. 1) are detailed below.

FIG. 1.

Hypothesized relationship between tick-borne encephalitis and land cover, land use, and their intersection.

Ticks and most of their hosts (rodents, ungulates, birds, etc.) thrive in forests that have thick undergrowth, because of the availability of sufficient humidity, cover, and browse (Randolph and Storey 1999, Süss 2003). The composition (proportion in the landscape) and configuration (arrangement) characterize the spatial structure of land cover types (Forman 1995). Forests are shaped by a range of management practices, such as felling and clear-cut management, which could affect human–tick contact. In Latvia, where logging is economically important (Latvian Forest Industry Federation 2006), the practice of natural or artificial forest regrowth could result in varying habitat quality for ticks and tick hosts. Hosts as well as ticks could also be favored when agricultural land is abandoned and perennial weeds and bushes take over, particularly near forests. The spatial structure of different land cover types also affects habitat suitability for ticks: ecotones between forests and open areas commonly have high tick densities (Daniel and Dusbàbek 1994), while the proximity and connectivity between highly suitable land cover classes determine the presence of larger wildlife hosts (Estrada-Pena 2003).

Socioeconomic factors such as income and education have been quantitatively related to the frequency of human visits to forests and exposure to tick bites (Šumilo et al. 2008b). The main reasons for people (other than forest workers) to enter forests include looking for alternative livelihood sources, mostly through the collection of wild foods, or recreation, which might also involve wild-food collection. People in both the lower and higher income brackets visit forests regularly, but people with lower income and education more frequently and more commonly with the purpose of collecting wild food. People in the middle and upper socioeconomic classes are more likely to visit the forest for recreation (SKDS 2001, Šumilo et al. 2008b). Therefore, people with low income or social support, or unemployed, who may be looking for alternative sources of food or income, are hereafter referred to as “pickers,” whereas those in employment, possibly with a higher level of education, are hereafter referred to as “hikers.” Residents of rural parishes where more intensive farming reduces the need to collect wild foods could be at lower risk of infection.

Land ownership and access rules determine whether distributions of ticks and human activities overlap in space and time. While forests managed by the State are legally open to anyone, access to forests managed by non-State owners can be restricted. The State also has a legal obligation to provide forested recreational areas around towns (Donis 2003), most of which are located in rural parishes.

Data and Methods

Data sources

TBE incidence was recorded per parish by the Latvian Public Health Agency. Parishes are the smallest administrative unit in Latvia (Fig. 2) (mean population: 1512 persons; standard deviation: 1225; mean area: 119.97 km²; standard deviation: 62.24). The analyses focused on rural parishes (n = 478, 84% of all municipalities), whose residents have close interactions with their surrounding environment and are likely to become infected close to home. For cases for which the place of infection was known, a higher proportion was acquired in the parish of residence for rural (69%) than for urban dwellers (24%) (Table 1). We focused on the period 1999–2003 to avoid the abrupt year-to-year changes in conditions of farming and socioeconomic status in the few years immediately following independence in 1991. TBE incidence was consistently high during this period, and lower in Vidzeme than the national average (Table 2). The 2000 population and housing census and the 2001 agricultural census were obtained from the Central Statistical Bureau, and the 2001 forest census from the State Forest Service (SFS). Land cover data were retrieved from two sources to allow analyses at different spatial scales: national CORINE 2000 data (100-m resolution; European Environmental Agency, www.eea.europa.eu/themes/landuse/clc-download), and Landsat images (4th of June 1992; 10th of June 2000) for the region of Vidzeme (Global Land Cover Facility; 30-m resolution). In May 2007, supportive information on agriculture and forestry practices was acquired by interviewing national, regional, and local representatives of the Ministry of Agriculture and the SFS.

FIG. 2.

Tick-borne encephalitis incidence in the rural parishes of Latvia, 1999–2003.

Table 1.

Tick-Borne Encephalitis Records by Parish for All Latvia During 1999–2003

	Whole country	Urban parishes	Rural parishes
Population (2000)	2,369,833	1,618,165	751,668
Total cases recorded by parish of residence 1999–2003	1707	907	800
TBE incidence (cases/100,000 population)	72.03	56.05	106.43
Total cases for which recorded parish of infection matches parish of residence 1999–2003	764	216	548
Proportion of cases contracted in parish of residence	0.45	0.24	0.69

TBE, tick-borne encephalitis.

Table 2.

Tick-Borne Encephalitis Incidence in Rural Parishes of Residence, 1999–2003

	Cases in whole country (cases/100,000)	Cases in Vidzeme (cases/100,000)
1999	137 (18.23)	24 (11.58)
2000	236 (31.40)	53 (25.58)
2001	163 (21.68)	35 (16.89)
2002	78 (10.38)	21 (10.14)
2003	186 (24.74)	37 (17.86)
Mean 1999–2003	160 (21.29)	34 (16.41)

Land cover classification

Landsat images were processed using maximum likelihood classification. The following land cover classes were identified: water; peat bogs and wetlands; coniferous forests and mixed deciduous forests; transitional areas with trees; mixed disturbed areas (bushy or grassy vegetation); agricultural areas, including meadows and cultivated areas; bare soils (including cultivated areas) and areas with sparse vegetation; clear-cuts. Clear-cuts could not be detected directly because their spectral signature was similar to that of transitional land. Instead, they were defined as any pixel classified as transitional or bare soil in 2000 that had been classified as forest in 1992. A majority filter was applied to increase the spatial consistency of the map and improve accuracy. The Landsat-based land cover classification is more detailed than the CORINE map because it identified clear-cuts, in accordance with the method aforementioned, and included more spatial detail in the complex rural landscapes.

Accuracy of the Landsat-based land cover classification was assessed based on a random sample of points stratified by class. Points were assigned to classes using a combination of image interpretation, field observations (May 2007), topographic maps, and aerial photographs. The global accuracy was 73%, and the Kappa index (Congalton 1991) was 0.67, indicating acceptable accuracy.

At the national level, the classes “broad-leaved forest,” “coniferous forest,” and “mixed forest” (CORINE classes 3.1.1, 3.1.2, and 3.1.3) were merged into one forest class for consistency in the comparison with Landsat-based class. The classes “transitional woodland-shrub” (3.2.4), “nonirrigated arable land” (2.1.1), “pastures” (2.3.1), and “complex cultivation pattern” (2.4.1) were also included in the analysis. There was no information on clear-cut.

Comparison of data sources and forest management practices of various users

Various data sources were available for land cover: censuses, CORINE, and Landsat. To evaluate the level of confidence that could be allocated to each source and which one was best to use, consistency between data from various sources was assessed. Data on similar factors were compared using the Spearman correlation coefficient, and Mann–Whitney U tests (MWUs). Forest management practices implemented by the State and by non-State users, as recorded by the SFS, were compared using MWUs.

Variables relevant to TBE incidence

Variables were selected in the available data sets to test explicitly the relationships between TBE incidence and various factors outlined in the conceptual model aforementioned. Variables representing land cover, use, and ownership on the spatial distribution of TBE are detailed here (Table 3).

Table 3.

Hypothesized Risk Factors for Tick-Borne Encephalitis in Latvia

Hypothesis group	Variables selected (data source)	Hypothesized sign of relationship with TBE incidence
Land cover	% Parish covered with forest^a	+
	% Forest regeneration that is natural regeneration^b	+
	Mean area of forest patches^a	+
	Mean shape index of forest patches^a	+
	% Parish covered with clear-cut^a	+
	Mixed and transitional vegetation around forests^a	+
	% Parish covered with mixed disturbed vegetation^a	+
	% Parish covered with transitional forest^a	+
	% Agricultural land under fallow^c	+
	% Agricultural land that is arable^c	−
Land use	Mean area per farm^c	−
	% Population having an economic activity^d	Hikers+/pickers−
	% Population receiving social assistance^d	Hikers−/pickers+
	% Population receiving a pension^d	Hikers−/pickers+
	% Population with higher education^d	Hikers+/pickers−
	% Population with <4 years of formal education^d	Hikers−/pickers+
Land ownership	% Forest area managed by the State/other managers^b	+/−
	% Parish area managed by the State/other managers^b	+/−

Land cover map.

State Forest Service.

Agricultural census.

Population census.

Land cover

Landscape composition was characterized using percentage occupancy of parish area. Forest configuration was measured using the mean area and shape index of patches in each parish; higher index values indicate more complex shapes relative to a square (for raster data). All patches on the CORINE map were used, and patches larger than 4 pixels (3600 m²) on the Landsat-based map were included. Logging activity in Vidzeme, measured as the ratio of clear-cut to forest on the Landsat-based map, indicated the cumulative state of the landscape after several years of felling. Because clear-cuts are not depicted on CORINE, national data on forest felling were taken from SFS records for 2000, which also indicated the proportion of natural regeneration that reflect different management practices. The percentage of transitional vegetation indicated agricultural land abandonment. The agricultural census recorded the percentages of agricultural land under fallow or in arable use. Percentages of transitional and agricultural land in 200-m buffers around forest edges measured the forest-related ecotones of potential importance to TBE incidence.

Land use

The two categories of forest users identified earlier, “pickers” and “hikers,” were differentiated using the following proxy variables: education, economic activity (persons whose main source of livelihood is a wage or any form of income from entrepreneurship, sale of production, or self-employment), and receipt of a pension or some form of financial assistance (see Introduction section). In addition, mean farm area within each parish was included on the hypothetical basis that larger farms would be related to higher local economic dynamism and lesser need to look for alternative livelihood sources.

Land ownership

The proportion of forest area and of the area of each parish used by either the State or private operators indicated accessibility of tick habitats to people.

Analysis of risk factors for TBE incidence in Latvian parishes

The relationship between the incidence of TBE and hypothesized explanatory factors was investigated using generalized linear models with a negative binomial distribution. First, regressions with one independent variable were performed. Significant factors (p < 0.05) were then introduced into multiple models, tested for colinearity, and nonsignificant variables were discarded. The models were offset for population per parish and number of years included (five). Residuals were tested for spatial autocorrelation using Moran's I index (Fortin and Dale 2005). Autoregressive models were also fitted (Augustin et al. 1996, Gumpertz et al. 1997) using the same procedure as described earlier but including the proportion of parishes where TBE is present among parishes within 20 km of each parish. This accounted for the possibility of people acquiring infection in neighboring parishes, if TBE was present there (Haining 2003). Models were evaluated using the pseudo-r ²; values between 0.2 and 0.4 indicate a good model fit.

Results

Comparisons between various land cover data sources

In Vidzeme, measures of agricultural area by the agricultural census and Landsat data were highly correlated (Spearman correlation index [SCI], p < 0.0001) and did not differ at the parish level (MWU, p > 0.05). Any small discrepancies are probably related to the thematic content of the Landsat-based map, for example, bare soils and settlements are included in the same class. For the whole of Latvia, the agricultural area as measured by the census was also strongly correlated (SCI, p < 0.0001) with CORINE classes “nonirrigated arable land” and “pastures” in rural parishes, but there were significant differences between the two datasets (MWU, p < 0.0001).

The SFS somewhat underestimates the area of forest compared with that mapped using Landsat or CORINE (MWU, p < 0.0001). This could be because, as cultivated areas decreased during and after the Soviet period, abandoned marginal land largely reverted to forest, but this change was not always recorded (Melluma 1994, Nikodemus et al. 2005). Correlations between forest measured by the SFS and by land cover maps, however, were high (SCI, p < 0.0001).

Based on these comparisons, as long as no information on land ownership was necessary, Landsat-based or CORINE data were used in preference to census data because the former also includes spatial details of landscape composition and configuration.

Forest ownership and management

On average, forest occupies 46% of parishes (Latvia, CORINE data). Non-State owners manage a larger fraction of forest in rural parishes, allow a greater degree of natural regeneration (MWU, p < 0.0001), and have a smaller proportion of coniferous forest (MWU, p < 0.0001) compared with State-managed forests (Table 4). Non-State managed forests were also more intensively felled in 2000.

Table 4.

Comparison of Management in State and Non-State Used Forests in Latvian Rural Parishes

	State forests, mean (standard deviation)	Forests managed by non-State users, mean (standard deviation)	Mann–Whitney U test (p-value)
Forest per parish (km²)	26.84 (31.34)	26.75 (19.73)	—
% Forest per parish	43.77 (24.90)	56.23 (24.90)	—
% Regeneration that is natural	47.90 (35.64)	69.71 (36.60)	−25189 (<0.0001)
% Coniferous	65.55 (19.12)	40.71 (16.90)	52143 (<0.0001)
% Area felled in 2000	2.94 (3.06)	6.28 (3.02)	−47389 (<0.0001)

Analysis of risk factors for TBE incidence in Latvian parishes

Regression analyses with one independent variable

Nationally, in rural parishes, TBE incidence was significantly positively related to the proportion of forest in parishes, mean areas of forest patches, and shape index of forest patches, and negatively to the percentage of natural regeneration (Table 5, columns 2 and 3). TBE incidence was higher in parishes with more transitional vegetation, including such vegetation around forest edges, and less agricultural land around forests. Forest felling and types of agricultural land (arable or fallow) had no apparent effect. Of all the proxy land use variables, only an increase in the fraction of the population with less than 4 years of education had a negative impact on TBE incidence. Parishes with a larger fraction of the forest or the parish territory under State management had a higher TBE incidence.

Table 5.

Results of Negative Binomial Regression with One Independent Variable for Tick-Borne Encephalitis Incidence in Rural Parishes, 1999–2003

	All Latvia		Vidzeme
	Estimate	Pr > Chi²	Estimate	Pr > Chi²
Land cover variables
% Parish covered with forest^a	0.0205	<0.0001	0.0057	0.5060
% Forest regeneration that is natural^b	−0.0061	0.0035	0.0099	0.0146
Mean area of forest patches (km²)^a	0.0000	0.0001	0.0003	0.5369
Mean shape index of forest^a	0.0101	0.0003	−0.2900	0.3653
% Forest felled in 2000^b	0.0175	0.5437	—	—
Ratio of clear-cut to (forest + clear-cut)^a	NA	NA	−0.0417	0.0757
% Mixed and transitional vegetation around forests^a	0.0147	0.0010	0.0314	0.0140
% Agricultural land (cultivated, meadows) around forests^a	−0.0069	0.0405	−0.0509	0.0015
% Parish with mixed disturbed vegetation^a	NA	NA	0.0287	0.0867
% Parish with transitional vegetation^a	0.0324	0.0160	0.0064	0.7523
% Agricultural land under fallow^c	0.0034	0.8298	−0.0362	0.2707
% Agricultural land that is arable^c	0.0017	0.5991	−0.0155	0.0073
Land use variables
Mean area per farm^c	0.0054	0.5437	−0.0373	0.0205
% Population with an economic activity^d	−0.002	0.8233	−0.0524	<0.0001
% Population receiving financial assistance^d	0.0054	0.6828	0.0523	0.3903
% Population receiving a pension^d	−0.0184	0.1560	0.0762	0.0003
% population with higher education^d	−0.0342	0.1582	−0.1884	<0.0001
% Population with <4 years of education^d	−0.0825	0.0005	0.0700	0.0391
Land ownership variables
% Parish under non-State forest^b	−0.0043	0.4776	−0.0190	0.0584
% Parish under State forest^b	0.0256	<0.0001	−0.0000	0.6959
% Forest owned or used by the State^b	0.0102	<0.0001	−0.0000	0.9930
Spatial structure of dependent variable
Proportion of neighboring parishes with TBE incidence	1.8080	<0.0001	1.5312	0.0037

Bold values indicate statistical significance.

CORINE or Landsat land cover map.

State Forest Service.

Agricultural census.

Population census.

In Vidzeme (Table 5, columns 4 and 5), TBE incidence was positively related to the percentages of natural forest regeneration and of transitional vegetation around forests, and negatively related to the percentages of agricultural land around forests and of agricultural land that is arable. TBE incidence was lower in parishes with larger farms, and where more people had an economic activity and higher education, but higher where more people received a pension or had less than 4 years of education.

Nonspatial multiple model

TBE incidence in rural parishes across Latvia (Table 6) was positively related to the percentages of forest cover, of forest felled in 2000, and of the parish area occupied by State forest, but negatively related to the percentage of the population with an economic activity and with less than 4 years of education. This model had a pseudo-r ² of 0.23, and the residuals were spatially autocorrelated (Moran's I, p < 0.01), suggesting that an autoregressive model (below) would be more appropriate.

Table 6.

Results of Multiple Nonspatial Negative Binomial Regression for Tick-Borne Encephalitis Incidence in Rural Parishes, 1999–2003

	All Latvia		Vidzeme
	Estimate	Pr > Chi²	Estimate	Pr > Chi²
% Forest^b	0.0179	<0.0001
% Forest felled in 2000^a	0.0763	0.0072
Ratio of clear-cut to (forest + clear-cut)^b			−0.0591	0.0053
% Agricultural land (cultivated, meadows) around forests^b			−0.0468	0.0047
% Agricultural land that is arable^c			−0.0173	0.0008
% Population with an economic activity^d	−0.0274	0.0051
% Population with higher education^d			−0.1516	0.0006
% Population with <4 years of education^d	−0.0806	0.0025
% Parish under State forest^a	0.0078	0.0019	−0.0139	0.0439
Pseudo-r ²	0.23			0.17

State Forest Service.

CORINE or Landsat land cover map.

Agricultural census.

Population census.

In Vidzeme (Table 6), TBE incidence was lower in parishes with higher ratio of clear-cuts, higher percentages of agricultural land around forests, of arable land, of the population with higher education, and of the parish area that is occupied by State forest. This latter factor was the only direct contradiction of the national pattern. This model had a pseudo-r ² of 0.17 and the residuals were not spatially autocorrelated (Moran's I, p > 0.15), but to compare results at national and regional scales, an autoregressive model was also fitted.

Autoregressive multiple model

The presence of TBE in neighboring parishes was a highly significant positive predictor only for the nation-wide model (Table 7). Nationally, the positive relationship of TBE incidence with the percentage of forest felled in 2000 and of parish area occupied by State forest persisted, as did the negative effect of economic activity. The percentages of forest and of people with less than 4 years of education were no longer significant. In Vidzeme, the negative effects of percentages of arable land and of people with higher education were confirmed, and a new positive effect of the percentage of people receiving a pension was added. Residuals were spatially autocorrelated for the national model (Moran's I, p < 0.01), where the model fit was improved (pseudo-r ² = 0.35), but not for the regional model (Moran's I, p > 0.15).

Table 7.

Results of Multiple Spatial Negative Binomial Regression for Tick-Borne Encephalitis Incidence in Rural Parishes, 1999–2003

	All Latvia		Vidzeme
	Estimate	Pr > Chi²	Estimate	Pr > Chi²
% Forest felled in 2000^a	0.0642	0.0181
% Agricultural land that is arable^b			−0.0143	0.0066
% Population with an economic activity^c	−0.0236	0.0053
% Population with higher education^c			−0.1330	0.0081
% Population receiving a pension^c			0.0544	0.0193
% Parish under State forest^a	0.0256	<0.0001
Proportion of neighboring parishes with TBE incidence	0.2376	<0.0001	0.1651	0.2400
Pseudo-r ²	0.35		0.13

State Forest Service.

Agricultural census.

Population census.

The significant risk factors identified by the four multiple models are summarized in Table 8.

Table 8.

Summarized Results of the Four Multiple Models at the National Level and for Vidzeme Region

	Nonspatial multiple model		Spatial multiple model
National level	% Forest	+
	% Forest felled	+	% Forest felled	+
	% Economic activity	−	% Economic activity	−
	% <4 years of education	−
	% State forest	+	% State forest	+
			Neighboring parishes with TBE incidence	+
	Pseudo-r ² = 0.23		Pseudo-r ² = 0.35
Vidzeme region	Ratio of clear-cut	−
	% Agricultural land around forest	−
	% Arable land	−	% Arable land	−
	% Higher education	−	% Higher education	−
	% State forest	−
			% Pension	+
			Neighboring parishes with TBE	NS
	Pseudo-r ² = 0.17		Pseudo-r ² = 0.13

Plus and minus signs indicate significant positive and negative relationships, respectively.

Discussion

The regression analyses confirm that land cover, land use, and land ownership variables were related to TBE incidence in rural parishes of Latvia during the period 1999–2003. Decomposing the landscape into ecological constraints, human activities, and institutional rules of access proved useful in understanding the fine-scale spatial distribution of TBE.

Land cover variables were important. First, landscape composition was important: greater overall area, larger patch size of forest, and known habitat for ticks and their hosts were related to higher TBE incidence. Second, landscape configuration was significant: the positive effect of a higher mean shape index of forest patches may be related to greater perimeters where ecotones are known to favor tick abundance (Daniel and Dusbàbek 1994); conversely, TBE incidence was lower not only where there were relatively large areas of unfavorable land covers, such as arable land, but also where forests were surrounded by more agricultural land.

Some land cover effects are less obvious. The area of forest felled in 2000 was positively related to TBE incidence at the national level, but the relative area of clear-cut in Vidzeme was negatively related to TBE incidence. These two variables, however, were not correlated: the former reflects a specific activity in the forest for 1 year (forest workers may have contributed to the higher TBE incidence), whereas the latter reflects cumulative effects over several years, possibly leaving the forest less suitable for ticks and/or human visitors.

Land use was also significant. TBE incidence was positively related to variables hypothesized to indicate “pickers” (<4 years of education, receiving a pension), but negatively related to variables more likely to indicate “hikers” (economic activity, higher education). The exception was the negative impact of low education at the national level, but only a very small fraction of the population, presumably amongst the oldest, would not have completed the mandatory 8 years of education. “Pickers” were more common in Vidzeme, fitting the other features of this area that indicate the likelihood of a high proportion of this risk group. These results are consistent with other studies of socioeconomic conditions affecting relevant behavior (SKDS 2001), which appeared to play a major role in the upsurge of TBE incidence in the early 1990s (Bormane et al. 2004, Šumilo et al. 2008a) and subsequent decrease after 1998 (Šumilo et al. 2008b), that is, these conditions have now been shown to be related to both spatial and temporal variation in TBE risk.

Finally, land ownership, controlling the intersection between suitable tick habitats (land cover) and human agents (land use), proved to be important, but its effect is not entirely straightforward. At the national level, the significant positive relationship between TBE incidence and State-owned forests to which people have access reflects the obvious need for human access to tick habitats if infection is to occur. Along ecotones, however, where the influence of forest extends beyond its boundaries as indicated by the significance of agriculture or transitional areas surrounding forests, the distinction between State and non-State forests could break down. Further, because forest ownership is associated with different forest management practices (Table 4), the significant land ownership variables may reflect either access to forests, and/or the impact of management on habitat quality for ticks and tick-hosts and on the attractiveness for human activities. Clear-cutting could adversely affect the presence of ticks, rodents, and forest food, and landscape attractiveness for hikers, which could explain its negative effect on TBE risk in Vidzeme (Table 6). Such confounding effects of management versus access could explain the inconsistent impact of natural regeneration on TBE incidence at the national and regional levels in the models with one independent variable (Table 5).

Land cover maps based on CORINE and Landsat images gave largely consistent results, but the latter gave bespoke, specific information with a high degree of spatial detail, such as clear-cuts, which was identified as a significant factor. Some important information was partly lost in CORINE; mixed classes such as “complex cultivation pattern,” for example, included areas of both agriculture and natural vegetation, and the spatial arrangement of these two landscape elements proved to be a significant factor for TBE incidence. Such considerations also help to resolve the paradox of the earlier study (Šumilo et al. 2006), in which TBE incidence in the Baltic counties was most significantly, but negatively, related to the proportion of marginally suitable tick habitat, based on an aggregated version of CORINE. The more precise distinction here between the wide range of vegetated land covers included in such habitats, especially between arable and abandoned fields, now confirms the negative impact of agricultural land on TBE incidence. The model with one dependant variable also identified more intensive agriculture, as indicated by larger mean farm size, as less favorable for TBE, perhaps, as suggested earlier, because of the lesser need to harvest wild food. The loss of such thematic detail in CORINE was offset by its ease of use and greater spatial extent, so both CORINE and Landsat have complementary values for this sort of analysis.

This study reveals the benefits of both fine-scale (parish) analysis and comparison between national and regional analyses. A major advantage of focusing on rural parishes was being able to establish a firm link between people and place. Although most people in Latvia live in towns (50% of the population live in the seven major cities, none of which are in Vidzeme), the results of this study are relevant to the whole population. Urban dwellers cannot be directly related to specific risk areas through their residence, but they enter similar areas through outdoor recreational activities and are subject to the same land-related risk factors.

Once the spatial autocorrelation of TBE incidence was taken into account, the multiple models gave a much better fit at the national level than within Vidzeme, indicating that additional, local scale variables might be necessary to understand regional heterogeneity fully. The relative uniformity in seasonal climate across Latvia (Šumilo et al. 2007) suggests a minimal role for it, despite its evident importance across Europe (Randolph 2000), but its integration into the land-related aspects considered here is clearly the next step. Public health considerations, such as differential access to health care and vaccination, might also prove to be a significant determinant.

In conclusion, complex combinations of land cover, land use, and land ownership all contribute to understanding the spatial distribution of TBE in Latvia. Some forest features, such as felling and clear-cuts, are highlighted for the first time but require further investigation to define their precise role. The results emphasize the importance of humans in determining the degree of exposure to zoonoses and therefore the realized infection incidence, including the spatial aspects.

Footnotes

Acknowledgments

The data for this study were supplied by the Public Health Agency, the Central Statistical Bureau, and the SFS in Latvia. This study was funded by EU Grant GOCE-2003-010284 EDEN; it has been catalogued by the EDEN Steering Committee as EDEN133 (www.eden-fp6project.net/). The help of Jānis Ozolinš from the SFS in Riga, and Helma Jirgena and her colleagues from the Ministry of Agriculture is gratefully acknowledged. The authors also thank Gert Olsson for comments on an earlier version of this article.

Disclaimer

The contents of this publication are the sole responsibility of the authors and do not necessarily reflect the views of the European Commission.

Disclosure Statement

No competing financial interests exist.

References

Augustin

, Mugglestone

, Buckland

. An autologistic model for the spatial distribution of wildlife. J Appl Ecol, 1996; 33:339–347.

Beck

, Lobitz

, Wood

. Remote sensing and human health: new sensors and new opportunities. Emerg Infect Dis, 2000; 6:217–227.

Bormane

, Zeltina

, Lucenko

, Mavcutko

. Tick-borne encephalitis—pathogen, vectors and epidemiological situation in Latvia 2002–2003. Acta Univ Latviensis Biol, 2004; 676:27–37.

Congalton

. A review of assessing the accuracy of classification of remotely sensed data. Remote Sens Environ, 1991; 37:35–46.

Curran

, Atkinson

, Foody

, Milton

et al. Linking remote sensing, land cover and disease. Adv Parasitol, 2000; 47:37–80.

Daniel

, Dusbàbek

. Sonenshine

, Mather

. Micrometeorological and microhabitat factors affecting maintenance and dissemination of tick-borne diseases in the environment. Ecological Dynamics of Tick-Borne Zoonoses. New York: Oxford University Press, 1994; 464.

Daniel

, Kolàr

, Zeman

, Pavelka

, Sádlo

. Predictive map of Ixodes ricinus high-incidence habitats and a tick-borne encephalitis risk assessment using satellite data. Exp Appl Acarol, 1998; 22:417–433.

Donis

. Designating a greenbelt around the city of Riga, Latvia. Urban Forestry Urban Green, 2003; 2:31–39.

Estrada-Pena

. The relationship between habitat topology, critical scales of connectivity and tick abundance Ixodes ricinus in a heterogeneous landscape in Northern Spain. Ecography, 2003; 26:661–671.

10.

Forman

. Land Mosaics: The Ecology of Landscapes and Regions. Cambridge: Cambridge University Press, 1995.

11.

Fortin

, Dale

. Spatial Analysis: A Guide for Ecologists. Cambridge: Cambridge University Press, 2005.

12.

Glass

, Cheek

, Patz

, Shields

. Using remotely sensed data to identify areas at risk for Hantavirus pulmonary syndrome. Emerg Infect Dis, 2000; 6:238–247.

13.

Gumpertz

, Graham

, Ristaino

. Autologistic model of spatial pattern of phytophthora epidemic in bell pepper: effects of soil variables on disease presence. J Agric Biol Environ Stat, 1997; 2:131–156.

14.

Haining

. Spatial Data Analysis: Theory and Practice. Cambridge: Cambridge University Press, 2003.

15.

Jackson

, Hilborn

, Thomas

. Towards landscape design guidelines for reducing Lyme disease risk. Int J Epidemiol, 2006; 35:315–322.

16.

Labuda

, Randolph

. Survival strategy of tick-borne encephalitis virus: cellular basis and environmental determinants. Zentr Bakteriol, 1999; 289:513–524.

17.

Latvian Forest Industry Federation. Forest Sector in Latvia 2006. Riga: Latvian Forest Industry, 2006.

18.

Melluma

. Metamorphoses of Latvian landscapes during fifty years of Soviet rule. GeoJournal, 1994; 33:55–62.

19.

Moràn Cadenas

, Rais

, Humair

, Douet

, Gern

. Identification of host bloodmeal source and Borrelia Sensu Lato in field-collected Ixodes ricinus ticks in Chaumont (Switzerland) J Med Entomol, 2007; 44:1109–1117.

20.

Nikodemus

, Bell

, Grine

, Liepins

. The impact of economic, social and political factors on the landscape structure of the Vidzeme Uplands in Latvia. Landsc Urban Plann, 2005; 70:57–67.

21.

Randolph

. Ticks and tick-borne disease systems in space and from space. Adv Parasitol, 2000; 47:217–243.

22.

Randolph

. Tick-borne encephalitis incidence in Central and Eastern Europe: consequences of political transition. Microbes Infect, 2008; 10:209–216.

23.

Randolph

, Storey

. Impact of microclimate on immature tick–rodent host interactions (Acari:Ixodidae): implications for parasite transmission. J Med Entomol, 1999; 36:741–748.

24.

SKDS, Marketing and Public Opinion Research Centre. Investigation into the Factors Hindering Immunisations of the Population Against Diphtheria and Tick-Borne Encephalitis. Riga: SKDS, 2001.

25.

Šumilo

, Asokliene

, Avsic-Zupanc

, Bormane

. Behavioural responses to perceived risk of tick-borne encephalitis: vaccination and avoidance in the Baltics and Slovenia. Vaccine, 2008a. 26:2580–2588.

26.

Šumilo

, Asokliene

, Bormane

, Vasilenko

et al. Climate change cannot explain the upsurge of tick-borne encephalitis in the Baltics. PLoS One, 2007; 2:e500.

27.

Šumilo

, Bormane

, Asokliene

, Vasilenko

. Socio-economic factors in the differential upsurge of tick-borne encephalitis in Central and Eastern Europe. Rev Med Virol, 2008b. 18:81–95.

28.

Šumilo

, Bormane

, Asokliene

, Lucenko

et al. Tick-borne encephalitis in the Baltic States: identifying risk factors in space and time. Int J Med Microbiol, 2006; 296:76–79.

29.

Süss

. Epidemiology and ecology of TBE relevant to the production of effective vaccines. Vaccine, 2003; 21:S1/19–S1/35.