Chevalier et al. 20 conducted a serological study on horses in five ecologically contrasted regions of the Senegal River basin (Senegal) to assess WNV transmission. Blood samples were taken from 367 horses from the five regions and screened by ELISA for anti-WNV IgM and IgG. Positive samples were then confirmed by seroneutralization. The seroprevalence rate was 85% overall but it varied significantly between sites. To assess whether environmental conditions could explain these differences, a land cover map was derived from two satellite images from the dry and wet seasons, and the surface covered by each land-cover type was calculated for each study area. Environmental data were analysed using principal components analysis and generalized linear mixed models.

Vanwambeke et al. 21 investigated the landscape-level factors influencing TBE incidence in rural parishes of Latvia, distinguishing between land cover, use and ownership. Land cover was used to depict the ecological suitability of the landscape for ticks and their hosts. Landscape composition and configuration were extracted from land cover maps. Land use represented human exposure to ticks, mostly through visits to forests. It was measured using proxy variables extracted from agricultural and household censuses. Land ownership of forests represented the accessibility to vector habitats for the public. Data were analysed using non-spatial and spatial negative binomial regression models.

A spatially-explicit R₀ model was developed for canine leishmaniasis in the French Pyrennées 9. An important variable for such a model is the density of the vector, which was estimated continuously in space using multivariate regression models. Based on extensive field trapping of sandflies, and using landscape composition and configuration at a medium spatial resolution and remotely sensed climate-related factors at a coarse spatial resolution, the abundance of sandflies was predicted throughout the study area. This was then used as an input for the calculation of a spatially-explicit R₀.

The impact of landscape variables on the transmission of RVF in small ruminants was investigated in the semi-arid region of Barkedji, Ferlo, Senegal 22. The relationship between landscape features, derived from a very high spatial resolution image, and serologic incidence was analysed statistically using a mixed-effect logistic regression model. A model of mosquito population dynamics was also developed. This model, based on current knowledge on the biology of the two RVF vectors species, Aedes vexans and Culex poicilippes, takes into account the main events of the mosquito life cycle and climatic fluctuations. Simulations of daily spatial and temporal variations in the area of temporary ponds around a village in Senegal relied on the Tropical Rainfall Measuring Mission (TRMM) rainfall product 1819. Mosquito population dynamics was simulated based on variations in water level and surface.

Based on the seasonal distribution of mosquito and bird populations, simulations of introduction, amplification and emergence of WNV under different realistic scenarios were produced and compared with seroprevalence measured in horse and bird populations 23.

The link between environmental features and PUUV prevalence in bank vole population was investigated in Belgium. Linard et al. 24 explored the relationship between environmental variables and host abundance, PUUV prevalence in the host, and human cases of nephropathia epidemica. Statistical analyses were carried out on 17 broadleaf forest sites. To understand causal pathways between environment and disease risk, the study distinguished between environmental factors related to the abundance of hosts, such as land-surface attributes, landscape configuration, and climate, and factors that may favour virus survival in the environment, such as climate and soil attributes. A national scale model explained the spatial distribution of PUUV human infections at the municipality level 25.

The impact of fine-grained landscape patterns on the exposure of people to LB infection was also investigated in Belgium. A combination of factors linked to the vector and host populations, landscape attributes, and socio-economic factors were included in a negative binomial regression to explain the number of LB cases per municipality 25.

A larval index for the main potential vector of malaria in the Camargue area, Anopheles hyrcanus, was defined as the probability of observing its larvae in a given site at least once over a year. It was mapped by associating in a statistical model environmental indices that were derived from high spatial resolution imagery and entomological field data (Figure 1) 26. Linard et al. 14 developed a spatially-explicit multi-agent simulation representing the spatio-temporal dynamics of interactions between the agents that could influence malaria transmission in the Camargue: people, mosquitoes, animal hosts and landscape. This model integrates movements and behaviours of people and vectors, and factors influencing transmission risk spatially. The model allowed testing potential drivers of malaria re-emergence such as changes in biological attributes of vectors, agricultural practices, land use, tourism activities and climate. Scenarios of possible futures varied the value of exogenous variables (e.g., tourist population visiting the area), initial conditions (e.g., land cover, in response to changes in land use policies or market forces) or parameters (e.g., level of protection of visitors against mosquito bites) 27.

This general proposition applies to all our studies (Table 2). The distribution, density, behaviour, and population dynamics of arthropod vectors and their non-human hosts are partially controlled by landscape features. The spatial distribution of vectors and the level of transmission is thus influenced by the environment 2829. This was illustrated by studies testing the relationship between disease occurrence and environmental features 3031. The case of WNV in Senegal illustrates this link, and the subsequent propositions provide details on the causal links between landscapes and diseases (Figure 2). In the serological study in the Senegal River basin, Chevalier et al. 20 found that IgG seroprevalence rate in horses was decreasing with proximity to seawater, flooded banks and salted mudflats. These landscape features acted as protective factors as they are highly unfavourable to the presence of Culex mosquitoes, the main WNV vectors. In Senegal, environmental constraints on vector distribution are more limitative for WNV transmission than bird distribution. Similar results were observed in the Camargue region, France and in Iowa, also with significant relationships between landscape features and seroprevalence of WNV 3233, with different risk and protective factors though. This highlights the need for a landscape scale analysis of infections, especially when there are multiple possible hosts and vectors species.

Numerous studies aimed at understanding associations between land cover and disease risk focus on the presence of critical habitats for vectors or reservoir hosts 34353637383940. In most cases, the surface area of those habitats was used to quantify landscape characteristics, ignoring their spatial configuration. Yet, more complex and fragmented landscapes are associated with more ecotones (i.e., transition areas between two adjacent ecosystems), which increase the likelihood of contacts between species associated with various habitats 41. Moreover, fragmented landscapes provide more habitats for edge species, and a greater diversity of resources. For example, ecotones between forests and open areas commonly have high tick densities 42, as well as higher incidence of infection 43. The incidence of LB was significantly associated with the importance of land cover edges between forest and herbaceous land cover types in the US 4445. Allan et al. 46 found an increasing density of infected tick nymphs with decreasing size of forest patch. In Belgium, Linard et al. 25 showed that the probability of LB infection was higher in areas with a large interface between settlements and forests in peri-urban areas. In the multivariate statistical analysis of TBE incidence in rural parishes of Latvia 21, human cases of TBE were associated with the mean area of forest patches, their shape, and the proportion of mixed and transitional vegetation cover types around forests. Conversely, TBE incidence was lower not only where there were relatively large areas of unfavourable land covers for ticks, such as arable land, but also where forests were surrounded by agricultural land. Similar results were found in the statistical model predicting the abundance of the sandfly Phlebotomus ariasi in the French Pyrenees 9. It was best predicted with landscape descriptors measured in a 1000-m buffer, including Shannon's landscape diversity index (i.e., a measure of the diversity and balance in landscape composition), the proportion of forests, and the mean area of forest patches. Others studies on different vector-borne and zoonotic diseases highlighted the importance of habitats configurations for infection transmission. For example, Pradier et al. 32 showed a significant association between WNV's level of circulation in Southern France and a landscape metrics measuring the degree of interweaving of land cover classes. Graham et al. 31 also demonstrated that habitat form was related to the prevalence in China of human alveolar echinococcosis caused by a helminth, the tapeworm Echinococcus multilocularis.

The proximity of vector and host habitats may not result in a high level of risk if these critical habitats are not connected spatially by landscape features favourable to the circulation of vectors and/or hosts. Spatial diffusion of vectors is particularly crucial in the case of mosquito-borne diseases. At night, when female mosquitoes generally feed, most of the hosts (e.g., humans, cattle) are immobile. The ability of female vectors to spread from their breeding sites to hosts increases host/vector contacts. Landscape features largely control these movements 47. The study of RVF in the semi-arid region of the Ferlo in Senegal showed that ruminant herds living around temporary water bodies were at greater risk of RVF if they were located close to ponds surrounded by vegetation 22. A landscape closure index - representing the proportion of surface around each pond covered by vegetation such as dense forest and shrub savannah - was positively correlated with higher serologic incidence. The presence of dense vegetation around water bodies favours the spread of mosquitoes from the pond where they breed to the nearby ruminant herds (Figure 3). The importance of connectivity between habitats for the spread of mosquitoes was also demonstrated for Culex species in Southern France 48. The connectivity between forest patches may also influence rodent populations and therefore the transmission of rodent-borne diseases such as hantavirus. Linard et al. 24 found that the spatial distribution of bank voles was different during epidemic and non-epidemic years. The number of bank voles captured was higher in more isolated forest patches during the non-epidemic year, whereas it was higher in less isolated patches during the epidemic year. Well-connected patches have more chances to be recolonized after local extinctions 49. Habitat connectivity could also influence the virus occurrence in hosts by controlling movements of individuals and thus contact rate between infected and susceptible rodents. Langlois et al. 50 observed that hantavirus incidence in deer mice was higher in landscapes with a higher level of fragmentation of the preferred habitat. In Western Africa, Guerrini et al. 51 showed that the riverine forest fragmentation level is a critical factor to determine the habitat of riverine tsetse species, vectors of animal trypanosomosis.

Great numbers of hosts and vectors species are potentially involved in the transmission of WNF, making its epidemiology complex. One should better understand underlying processes accounting for observed patterns of WNF when correlations between land use/cover and disease prevalence have been established 52. In the Camargue region, we showed that epidemic processes of introduction, amplification and emergence of WNV were related to specific associations of hosts and vectors species in time and space. Based on maps of seasonal distributions of mosquito and bird populations, simulations of introduction, amplification and emergence of WNV were compared with seroprevalence measured in horse and bird populations 23. Introduction of the virus by migratory birds explained accurately the observed spatial patterns of WNV transmission, whereas overwintering of WNV in mosquitoes did not. Cx. modestus was identified as the main amplifier of WNV, which is consistent with competence studies 53. In the Camargue, the virus was only isolated in sparrows and magpies 54. Yet, other competent bird species - or all bird species - may contribute to WNV amplification. The bird community composition did not seem to play a major role in the amplification of WNV in the Camargue (no "dilution effect"), unlike what was observed in the New World 525556. The final risk map, based on landscape attributes, synthesizes the different processes leading to WNV emergence in horses and identifies risk areas requiring veterinary surveillance.

The transmission of vector-borne diseases requires a direct contact - i.e., a bite - between an infectious vector and a susceptible host. By contrast, zoonoses such as PUUV can be transmitted directly, by physical contact between infected and susceptible hosts, or indirectly, with the environment as an intermediate. Actually, the virus can be shed in the environment via rodent excretions, and transmitted to humans in aerosols 5758. The environmental conditions controlling the direct and indirect transmission paths differ. In a study of the link between environment and PUUV prevalence in bank voles in Belgium, Linard et al. 25 showed that PUUV prevalence among bank voles is more linked to variables favouring the survival of the virus in the environment, and thus indirect transmission. In particular, low winter temperatures were strongly linked to high prevalence among bank voles, and high soil moisture was associated with high numbers of nephropathia epidemica cases among humans. The risk of transmission to humans is therefore not only determined by host abundance, but also by the indirect transmission path, largely controlled by climatic factors and soil characteristics influencing virus survival in the environment. Another illustration is provided by the differences in dynamics of two pathogens vectored by the same species, Borrelia spp. and TBE virus. Transmission to ticks related to systemic and non-systemic host infections (respectively, infection of many and specific parts of the body) has been described for both pathogens. Yet, transmission through co-feeding of ticks in non-systemically infected hosts may be more important to sustain the transmission of TBE 5960. As a result, spatial variations of TBE infection risk are controlled by more restrictive ecological conditions, as closer interactions between infected competent hosts and ticks are required compared to LB. The TBE transmission cycle is therefore more fragile 6162. The transmission via the environment can also play a role in the transmission of avian influenza viruses: contaminated water supplies can constitute a reservoir that maintains the virus in the environment and be a source of contamination for wild or domestic birds 63.

The emergence of a disease in a particular region is associated with multiple macro-level changes such as a shift in political and economic regimes that influence people's livelihood strategies and therefore their interaction with natural ecosystems 6465, rapid conversion of natural habitats, and urbanization. In some cases, variations in climate may act synergistically with these socio-economic changes, even though the relative magnitude of climate change impacts remains contentious 66. Macro-economic and climatic conditions create favourable or unfavourable background conditions for the development of pathogens, vectors, and hosts. The actual realisation of the transmission cycle and transmission to humans depends on the overlap of the preferred habitats of competent vectors and infected hosts, and on the entry of humans into these infected areas. Therefore, the fine-grained spatial heterogeneity in disease emergence is determined by natural and cultural landscape attributes that act as fine-scale spatial determinants of the multiple factors controlling the transmission cycle. For example, while an acceleration of warming of spring temperatures in the Baltic region played some role in the dramatic increase in TBE incidence 67, it did not account for the high spatial heterogeneity of incidence at an infra-national scale. Landscape factors related to the ecology of the disease and to human activities better explain this spatial heterogeneity 21. In Western Europe, bank voles are known to display cyclic abundance peaks, which significantly increase PUUV infection among rodents and humans. These peaks are related to high tree seed production, triggered by high summer and autumn temperatures the preceding years 6869. While such macro-scale factors influence PUUV incidence at the regional scale, landscape characteristics related to rodent habitats and human land use determine the local scale spatial heterogeneity of transmission. Spatial variations in sandfly abundance in the French Pyrenees was also explained by a combination of coarse and fine resolution remotely-sensed metrics that are associated with, respectively, regional climatic trends and local-scale landscape variables reflecting land use 9. In (disease) ecology, the relevance of the scale factor has been widely demonstrated 70 and yet it remains poorly explored in empirical studies of landscape epidemiology.

In the semi-arid region of the Ferlo, Senegal, temporary ponds - i.e., that are flooded after the first rains and remain dry for most of the dry season - constitute a favourable habitat for RVF vectors, Ae. vexans and Cx. poicilippes. Modelling spatial and temporal variations in water level and surface of temporary ponds around a village in Senegal allowed simulating mosquito population dynamics and host distribution 1819. Results showed that rainfall drives Culex and Aedes populations and therefore the risk of circulation of RVF virus. Depending on the frequency of rainfall events, high densities of Culex and Aedes may occur simultaneously. This co-occurrence of species leads to a higher risk of transmission, Ae. vexans and Cx. poicilipes acting as, respectively, initiators and amplifiers of RVF virus circulation. Thus, in the Ferlo region, a combination of rainfall frequency and amount identifies years at risk for RVF. Landscape features also influence the spread distance of mosquitoes away from their breeding sites 19. In East Africa, heavy rainfall events were associated with RVF outbreaks 71. Such events lead to major increase in mosquito vector populations 72. The flooded areas that are potential mosquito breeding sites ("dambos") constitute the main risk areas 73. Meteorological conditions can also influence exposure of humans 74.

Land cover is defined by the attributes of the earth's land surface and immediate subsurface, including biota, soil, topography, surface and groundwater, and human structures. Land use is defined by the purposes for which humans exploit the land cover. While a focus on land cover helps understanding the presence of vectors and hosts, a focus on land use identifies which places people visit for specific activities, at what time of the day and of the year, and at what frequency. The attractivity of various places for a given activity depends on their attributes such as accessibility and value for that activity - e.g., presence of recreational features, easily accessible forests with amenities such as good trails. Our studies on human cases of PUUV and LB in Belgium showed that the spatial distributions of the two diseases were very dissimilar 25. PUUV was mostly prevalent in forested, rural and low income municipalities as transmission to humans was mostly associated with hunting and forestry work. By contrast, LB was mostly found in forested, peri-urban and wealthy municipalities, as infection was mostly associated with gardening and recreational activities in forests. Thus, it is less land cover (i.e., the presence of broad-leaved forests) than land use (associated with income level and type of settlements) that controls the spatial distribution of human cases of these two diseases. Agents involved in different land uses - e.g., a tourist and a local farmer - also have varying levels of awareness of the risk of infection transmission and rate of adoption of preventive measures. Socio-economic factors such as income and education have been quantitatively related to frequency of human visits to forests and exposure to tick bites 75. In Latvia, the main reasons for people to enter forests included looking for alternative livelihood sources or recreation. People with low income and education visit forests more commonly with the purpose of collecting wild food, while more wealthy and educated people are more likely to visit forests for recreation 75. In our study of TBE incidence in rural parishes of Latvia 21, human cases of TBE were positively related to variables indicating wild food pickers and negatively related to variables associated with hikers. Different socio-economic groups are represented in various proportions in different regions of a country. In the Camargue, the scenarios tested using a multi-agent simulation showed that water management practices in rice fields influence the rate of contact between people and potential malaria vectors 27. According to the flooding date of rice fields, the maximum abundance of vectors and people can be reached simultaneously, resulting in high contact rates. In a study on West Nile Virus disease risk in the US, Winters et al. 76 emphasized the importance of taking into account human activities to assess risk of vector exposure in montane areas heavily used for recreation in the summer.

Land ownership and access rules determine whether distributions of ticks and human activities overlap in space and time. In most countries, State forests are more accessible to the public than privately-owned forests. In the statistical analysis of TBE incidence in rural parishes of Latvia 21, land ownership proved to be an important explanatory variable, with a positive relationship between TBE incidence and the presence of State-owned forests. Because forest ownership is associated with different forest management practices, land ownership may reflect not just access to forests but also differences in forest management practices. The latter have an impact on habitat quality for ticks and tick-hosts, and on the attractiveness of forests for human activities. For example, forest age, use of plantation and sanitary cutting may affect the presence of ticks, rodents, and forest food. While issues of land ownership have been widely studied in conservation science 77, they have been ignored in landscape epidemiology.

Humans can adopt a range of preventive measures to decrease the risk of contact with vectors or pathogens. The rate of adoption of these measures depends on the availability of such preventive measures, the nuisance of the vector, the risk of infection and its level of perception by susceptible agents. For transmission to occur, one needs a combination of a conducive environment, which is created by an ecosystem favourable to interactions between pathogens, vectors and their animal hosts, and risk behaviours by human hosts that lead to a greater incidence among a human population. For example, the emergence of TBE in the Baltics was associated with a complex combination of abiotic, biotic and human factors 6475. Similarly, the spatial distributions of PUUV and LB infections in Belgium were related to human activities and behaviours, among other factors 25. In an area of Sweden endemic for TBE and LB, differences were found in adoption of preventive measures between permanent and part-time residents and between genders, suggesting differing risk perception 78. Dengue infection in northern Thailand was explained by variables related to the ecology of Aedes, the mosquito vector, and by risk and protective behaviours 7980. One of the greatest challenges in landscape epidemiology is to better integrate human behaviours.