A Mapping Review on Urban Landscape Factors of Dengue Retrieved from Earth Observation Data, GIS Techniques, and Survey Questionnaires

To date, there is no effective treatment to cure dengue fever, a mosquito-borne disease which has a major impact on human populations in tropical and sub-tropical regions. Although the characteristics of dengue infection are well known, factors associated with landscape are highly scale dependent in time and space, and therefore difficult to monitor. We propose here a mapping review based on 78 articles that study the relationships between landscape factors and urban dengue cases considering household, neighborhood and administrative levels. Landscape factors were retrieved from survey questionnaires, Geographic Information Systems (GIS), and remote sensing (RS) techniques. We structured these into groups composed of land cover, land use, and housing type and characteristics, as well as subgroups referring to construction material, urban typology, and infrastructure level. We mapped the co-occurrence networks associated with these factors, and analyzed their relevance according to a three-valued interpretation (positive, negative, non significant). From a methodological perspective, coupling RS and GIS techniques with field surveys including entomological observations should be systematically considered, as none digital land use or land cover variables appears to be an univocal determinant of dengue occurrences. Remote sensing urban mapping is however of interest to provide a geographical frame to distribute human population and movement in relation to their activities in the city, and as spatialized input variables for epidemiological and entomological models.

parameters that define vectorial capacity, through habitat suitability, socio-ecological processes and local temperature variations such as urban heat islands (UHI) [24]. However, the impacts of landscape structure on epidemiological processes have been largely neglected in the past [25], and there is still a need for a spatialized integrated approach at various spatial scales [20,24], to combine methods from epidemiology, ecology, statistics and geographic information sciences [25][26][27].
Over the last twenty-five years, advancement in spatial epidemiology has been largely driven by the use of Geographical Information Systems (GIS) and georeferencing data systems [28,29]. In the case of vector-borne diseases, it may also include remote sensing techniques, which present a high-potential in disease risk mapping and environmental contextualizing [30][31][32][33], but probably still remains underutilised [34,35]. Remote sensing uses the notion of a proxy, that is a measurable variable which represents an indirect measure of an impractical physical variable that cannot be measured directly [35]. In the case of vector-borne diseases, entomological data surveys are often costly, labor-intensive and remain scarce [13,36]. Therefore, authors often use the proxies of mosquito breeding or resting sites based on the vector-knowledge reviewed in the literature [17,37]. Despite a more systematic use of GIS and the implementation of spatial statistical methods, the availability of health data and appropriate exposure data often remain limiting factors [38]. National passive notification systems present high variability in the standard of data and metadata storage, which highlights the importance of local knowledge through seroprevalence survey and questionnaire-based responses that can help to add clarity in uncertain regions [39].
We propose here a mapping review to create an inventory and identify the most relevant landscape factors potentially involved in dengue transmission in urban contexts from different data sources. Mapping reviews enable the contextualization of in-depth systematic literature reviews within broader literature and identification of gaps in the evidence base [40]. Mapping reviews share common purposes with scoping reviews, such as examining how research is conducted and structured on a certain topic, the identification of available evidence and the investigation of knowledge gaps [41,42], but provide a systematic map representation to categorize the included articles. Taking an interdisciplinary view, we propose a systematic search of articles into the literature to: (i) identify the landscape factors according to various sources and geographical units of production; (ii) map co-occurrence networks associated with the landscape factors, in order to identify the potential underlying structure of fields; (iii) evaluate qualitatively the respective importance of the above for the mapping of the dengue risk.

Systematic Search of Articles
This systematic review used the guidelines presented in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [43]. The methodology is summarized in Figure 1 and the detailed steps are presented hereafter. Data at the identification and the screening process steps were extracted by two independent researchers (RM and ZL), and discrepancies were resolved concordantly. The searches were performed in four on-line bibliographic databases, from inception to 31 December 2019: 4. Scopus: e.g., e.g., Asia Pacific Journal of Public Health, BMC Infectious Diseases, Epidemiology and Infection, Geocarto International, etc.; and considered either "all fields" (including bibliography references) or only "title-keywords-abstract" according to the database query form, and limited to the type "journal article". The logical structure of the queries was based on the following formula: (i) dengue AND (urba* OR cit*) AND ("land use" OR "land cover" OR landscape OR dwelling OR habitation) The character * being the classical symbol for regular expressions, corresponding to any character or group of characters, for example, urba* refers to the words urban, urbanization, and so forth. No constraints on the study period and language were imposed in the search queries. All search records from the four on-line databases were then combined together [n = 2342], using the free and open-source reference management software Zotero (https://www.zotero.org/). In addition, a search in Google Scholar R  iv. Considers a sound method to characterize the relationship between landscape and dengue 1. Publication meta-data : id number assigned by alphabetic order, first author, date, title, journal name, Appendix A1. 2. Geographical context of the study : country, study area, geographical unit of spatial analysis; Appendix A1. 3. Epidemiological descriptors : time span, source, diagnostic method, serotype, number of dengue cases, qualitative spatial variations, vector involved; A2 4. Landscape evaluated factors : source (remote sensing, GIS, survey questionnaire), data type group and subgroup, factors, type of proxy ; Appendix A3. 5. Dengue cases-landscape relationship : statistical method used to assess the relationship , three-valued interpretation (positive, negative, non significative) of the relationship. Appendix A3. 6. Risk of bias assessment table, which includes a checklist on key features based on a four-valued choice (yes, no, partial, ?: can't tell), and a overall appraisal of the level of contributive information respect to the topic "dengue--relationship characterization" (from 1: high to 4: poor). Supplementary material.

Identification
Screening Information extraction Eligibility criteria Google scholar ii. Considers spatialized dengue occurrences associated to the geographical units of the city iii. Considers at least one landscape factor associated to the geographical units of the city Systematic queries from inception to 2019 [n =2 614] dengue AND (urba* OR cit*) AND ("land use" OR "land cover" OR "landscape" OR "dwelling" OR "habitation")" Manually review of all titles and abstracts after duplicate removal [n=2 303] Manually  Minimum quality threshold associated to the thematic criteria Figure 1. Stages of systematic search to retrieve included article to our four criteria, following the PRISMA statement [43].

Screening, Selection Criteria, Risk of Bias, and Contribution of the Articles
A systematic checking of the titles and abstracts was conducted in order to select only the peer-reviewed original research articles dealing with the relationships between landscape factors and dengue cases, leading to [n = 234] at the end of the screening step, excluding those deemed irrelevant to the topic. Based on a full text reading, screened studies at the previous step were included if: (i) they consider geographical units within a city; (ii) they included spatialized dengue cases, collected by passive notification systems or by serological surveys; (iii) they identified and characterized the influence of landscape factors on dengue occurrences in an urban context; (iv) they described the explicit relationships between landscape classes and dengue data.
In contrast, studies that: (i) consider rural areas, or include large part uncovered by urban areas; (ii) do not consider dengue occurrences, but solely Aedes mosquitoes as proxy of dengue presence; (iii) do not include any explicit landscape feature, for example, solely consider meteorological variables (temperature, wind speed etc.) or socio-economic variables (income, status etc.); (iv) do not bring any evidence or information on the used models to perform the relationship between dengue occurrences and landscape features; were excluded, which finally resulted in [n = 78] articles included in the review, at the end of the eligibility step. A total of 156 articles were discarded at the end the screening stage based on criteria 1 (does not consider an urban geographical unit of a city, [n = 36]), criteria 2 (does not consider spatialized dengue cases [n = 26]), criteria 3 (does not consider at least one landscape factor, [n = 31]), criteria 4 (does not perform a relationship between dengue and landscape, [n = 49]), or based on an insufficiently described methodology ([n = 13]). We considered landscape factors in a "broad" definition, centering around a virus perspective: vectors and humans are hosts, and their respective trajectories lead to a complex interaction, which facilitate or hamper the virus circulation. Therefore, we considered entomological variables and human densities or movements as dynamic features of the landscape. On the other hand, we limited our definition of landscape factors to physical variables, and discarded direct references to socio-economic data, as level of income, per capita gross domestic product (GDP), or unsatisfied basic needs. We have in the first place considered a "Built City", i.e. a city as a physical entity, or the area devoted to primarily urban uses [44]. Such definition is in line with the global urban mapping approaches, and automatic extraction of built-up area [45][46][47]. As a proxy of human presence and Aedes habitats, urban areas within a city reflect a "certain density" of buildings, which threshold varies according to the geographical context and authors definition, out of the scope of this paper. We did not have either considered the question of city size, an issue of considerable significance in urban and regional analysis.
Various methods exist to appraise the quality of studies included in a review, and assess the corresponding risk of bias. These methods differ greatly in applicability across study designs, and approaches: e.g., scale vs checklist, presence/absence of summary score etc. [48]. During the screening stage, we performed a first "minimum quality threshold associated to the thematic criteria" (Figure 1) in order to discard articles were the data set or the methodological descriptions remain unclear. At the eligible stage, we included a checklist on key features of the 78 included articles based on a four-valued choice ("yes", "no", "partial", "can't tell") to characterize (i) the completeness of the epidemiological and the entomological dataset (ii) the degree of maturity of the methods to produce the landscape factors (iii) the characterization of the dengue-Landscape relationship. We also provide an overall appraisal of the level of contributive information respect to the topic "dengue-relationship characterization" (from 1: high to 4: poor). These information are available in a table format as Supplementary Materials.
Our entire bibliographic database, structured according to the PRISMA steps, may be consulted at the following web address: https://www.zotero.org/groups/2159925/article-review_dengue_ landscape/items/collectionKey/. By browsing the Zotero folders, readers could see the different results obtained through the systematic requests on the one-line databases, and by picking one particular article in the "non eligible" folder, readers could visualize the reason associated to the inclusion/exclusion decision in the note section (right window in the online application).

Structuring of the Information Extracted from the Included Articles
We referenced the included articles by an identification (id) number assigned alphabetically from 1 to 78, which corresponded to reference numbers [135] (Ali et al., 2003) to [212] (Zellweger et al., 2017) in the bibliography section (please refer to the appendix for a full description). We manually extracted the information concerning the data, the methods, and the main results to build three analysis tables, according to the following categories (please refer to the appendix section for exhaustive tables): (i) the geographical context: country, study area (city), geographical unit of spatial analysis (Table 1 and Appendix A); (ii) the epidemiological descriptors: start and end years of an outbreak or survey, dengue data type (incidence, prevalence, case number), medical analysis to confirm the diagnosis (clinical signs, laboratory analysis), number of dengue cases (and incidence rate when available), spatial variation and pattern(s) observed, vector species involved ( Table 2 and Appendix A); (iii) the landscape factors: data source according to three subcategories: remote sensing images (sensor name), Geographic Information System (GIS) layers, and survey questionnaires. We also extrapolated the type of proxy associated (i.e., the element of the transmission cycle represented, for example, "exposure to Aedes bite"), and the type of data (e.g., land use or housing type and characteristics) according to a two-level classification, called data group and sub-group, respectively (Table 3 and Appendix A); (iv) the search of a relationship between urban determinants and dengue cases: type of statistical and spatial methods used to quantify the relationship between dengue cases and environmental determinants, interpretation of the relationship through a three-valued index: positive (+), negative (−), or non-significant (NS) ( Table 3 and Appendix A).  Based on the information extracted from the geographical context and the epidemiological information, we mapped the cities corresponding to the 78 study sites (QGIS LTR 3.4). We distinguished the types of epidemiological data according to their sources: passive surveillance system, or serological studies (incidence or prevalence). We also mapped the techniques employed to produce the information related to landscape factors: survey questionnaire, GIS data, and remote sensing imagery.

Co-Word Analysis through Self-Defined Tags Co-Occurrences
To understand how landscapes factors are produced and those that could be critical in urban dengue transmission, we adapted a method derived from bibliometric visualization techniques ( Figure 2). Such approaches are based on the mapping of a network, which represents the degree of keyword co-occurrence of predefined article descriptors, like co-authors, or tags. Co-word networks may help to identify the conceptual structure, that uncovers links between concepts through term co-occurrence. Promising implementations of such literature analysis tool have been recently developed ( [49,50], NAILS, bibliometrix). To perform this network mapping, here we used VOSviewer software (V1.6.11), a tool for constructing and visualizing bibliometric networks [51], and already used to perform review analysis ( [33], e.g., Remote Sensing in Human Health). To map the structure associated with the landscape factor production, we exported the bibliographic references according to three categories: remote sensing images, GIS data, and survey questionnaire. From the bibliometric manager (Zotero 5.0.73), we chose a standardized tag format developed by Research Information Systems (RIS), compatible with VOSviewer and the module create map based on bibliographic data. To map the networks, we chose Co-occurrences with Keywords as units of analysis, associated with the full counting method. Here, keywords refer to self-defined tags, identified by the authors of this review, and associated with landscape factors, structuring terms, and a three-valued interpretation associated with the dengue-landscape relationship (positive, negative, or non-significant) ( Figure 2). We defined the minimum number of occurrences as 1, in order to map the entire landscape factor network. Here, a node is associated with a tag (or keyword), with an edge representing a link of co-occurrence between two tags. To map the networks associated with the nature of the relationships between the landscape factors and the observed dengue cases, we adopted the same approach for each of the four defined spatial units: household, neighborhood, small-administrative, large-administrative (including city-level ( Figure 2). As VOSviewer is mainly designed to visualize large maps containing thousands of items, it could have been challenging to read the full-network, so we added a post-treatment step, in order to make some items more readable by modifying the character font (Inkscape, version 0.92.4).
Survey questionnaires and census data originate from socio-geographical approaches, while entomological observations are part of medical entomology. As these were mainly collected during household investigation, they were associated it with survey questionnaires in the data structure representation, as part of socio-ecological surveys.  Figure 2. Method used to map the co-occurrence relationship between the self-defined tags, here keywords, for each of the articles. Keywords are specific self-defined tags, which may here refer to: landscape factors (e.g., "Urban Heat Island"), structuring terms (in bold, e.g., "Urban typology" or "large administrative-level"), or nature of the relationship (in color, e.g., "positive"). We added a tag, called Nb (number), which helps to identify the id number of the included article (here 3 of [n = 78]).

Geographical and Epidemiological Contexts
Temporality and location of the included articles ( Figure 3): • Collectively, these review articles propose a broad spatial sampling of the inter-tropical belt, traditionally associated with dengue occurrences [2], and consider dengue cases observed over a thirty seven year time-span, between 1982 and 2019 ( Figure 3).

Epidemiological characteristics of the included articles:
• The dengue virus can cause a large range of symptoms, ranging from an asymptomatic form, which includes the vast majority of infections, and may be associated with various degrees of infection: dengue fever (DF), dengue hemorrhagic fever (DHF) to the potentially fatal dengue shock syndrome (DSS) [52]. Generally, most articles refer to dengue cases that include a broad interpretation of the disease expression, especially fever (DF). Twelve studies in the method section refer explicitly to DHF cases (ids: 7,12,17,25,31,38,49,59,60,65,75,68), and two to DSS (id: 31, 65). In Indonesia for example, only DHF cases are mandatorily reported (id: 49); • We identified 23 articles based on serological surveys performed by the authors (ids: 2, 7, 8, 9, 20, 22, 26, 28, 30, 34, 35, 43, 48, 49, 52, 55, 61, 67, 70, 71, 73, 75, and 77). In such approaches, based on fieldwork, household location is used to spatially identify the dengue cases. Fifty-five other articles were based on passive notification of cases collected by local and national health agencies. Such databases may collect the patient address or refer to an administrative division to locate the cases, without further information on a potential place of transmission (ids: 15,16,19,23,32,35,57,64,66,78 Almost all of the 78 publications included articles which confirmed a highly non-uniform spatial distribution in the urban context, regardless of the spatial scale of analysis. Global or focal cluster detection are commonly based on global/local Moran's index to detect the presence of overdispersion based on autocorrelation analysis [53], and is based on either a sliding circular window (cylinder, if the time dimension is considered), or consider each spatial unit towards contiguous neighbor units (ids: 10,16,17,18,38,46,54,58,65,78). Its value comprises between [-1,+1], and reflects the assumptions about the spatial phenomenon in question to detect negative or positive spatial auto-correlation. In the articles of this review, a local Moran's index often highlights the presence of a spatial correlation at fine scale. Various articles identify clusters (ids: 1, 3, 10, 16,17,18,24,31,36,37,38,39,46,51,53,58,63,65,70,71,74,78), hotspots (ids: 10,19,50,56,59) and coldspots (id: 10, 50). In one study (id: 42), the authors tested several structures of spatially explicit Bayesian models in order to estimate the relative risk (RR) of dengue.

Entomological consideration in the included articles:
• The majority of the articles only mention the implication of the Aedes vector in the introduction and/or the discussion sections, and exclude entomological consideration in the method or in the data acquisition. Nineteen articles performed entomological observations of: Aedes aegypti The potential heterogeneous nature of the spatial dispersion of mosquito density has been analysed in some studies (in relation withe dengue occurrences), through, notably (i) the intensity of larvae-positive breeding sites by properties inspected in each block, unsing the kernel estimator method (id: 5), parameterized with a flight distance of 280 m which is associated with the Aedes aegypti female [54], (ii) the extrapolation by ordinary kriging of entomological indicators associated with the four life stages of Ae. aegypti: (absolute) number of A. aegypti eggs in the block, and number of positive buildings for Ae. aegypti larvae-pupae and adults in the block, divided by the number of buildings surveyed in the block (id: 6).

Epidemiological data Passive
Sero-incidence  We indicate the type of sources (serological surveys or passive notification system) and the temporal range associated with the dengue data. Bottom: localization and characteristics of the landscape data sets of the 78 articles of the review. We indicate the type of sources: questionnaire surveys, GIS, Remote sensing data, and the availability of entomological data (*).

Production of the Landscape Factors Associated to Dengue Cases
Type of approaches: We identified five approaches that led to the production of landscape characteristics (Figures 3 and 4): (i) Survey questionnaire, including census data; (ii) in situ entomological observation; (iii) Geographical Information system (GIS) data; (iv) Topographical measurements; (v) Remote sensing data (RS data), originated from satellite images.
Data sources network considering all approaches: The graphical representation of the data sources network, considering all type of data, highlights the strong polarization between "survey questionnaire" and "remote sensing images" (Figure 4): • "RS images" are strongly connected to the "land cover" properties of the environment, while "survey questionnaire" is strongly connected to "housing characteristics", "housing type", "construction material" and "entomological observation". "GIS data" sources are both connected to "remote sensing images" and "survey questionnaire", highlighting its interface position as a bridge between human geography approaches and digital geography (e.g., [55]); • "GIS data" connect well to the "land use" characteristics of the environment, the "infrastructure level" and the "typology" of the urban area. It is noteworthy that the node "Aedes aegypti mention" is at the centre of the network, which shows that entomologist information relative to the 78 included studies, centred on observed dengue cases, are coming from a knowledge base of the mosquitoes rather than direct observations. Entomological observations concerning Aedes aegypti and albopictus, considered together or separately, belong to the "survey questionnaire" cluster, while Aedes aegypti and Ae. albopictus mentions belong to "remote sensing image" or "GIS data" clusters ( Figure 4); • Considering the publication year associated with the data source ( Figure 4), it is noteworthy that "survey questionnaire" and "entomological observations" are associated with the oldest publications, and "remote sensing" and "GIS data" with the most recent. However, the "remote sensing images" cluster is associated with the 2000-2015 period satellite missions (Landsat 5-7, MODIS, IKONOS, ALOS), and not to the most recent ones (e.g., Sentinel missions, except for id: 41). Satellite imagery and GIS data have been used to complete and contextualize some survey questionnaires in multi-sources studies, e.g., Google Earth images used for photo-interpretation (ids: 20, 57), normalized difference vegetation index (NDVI) index and urban characteristics (id: 50), or GIS data used to localize entomological observations (ids: 24,58) or altitude associated with the mosquitoes' environment (ids: 21, 34, 44, 64); • By jointly using remote sensing and GIS data sources, some authors were able to describe both land use and land cover properties of the study area, e.g., vegetation index and urbanization level (id: 10), road network density and aging infrastructure (id: 14), bare soil detection and building type (id: 19), urban typology ("Urban Park") and vegetation cover through NDVI index (id: 29), "urban village" and NDVI index (id: 51).
Data sources network considering remote sensing images: By mapping the structure of data from the "remote sensing images" source ( Figure 5), we observe a strong structuring around the "land cover" properties of the landscape, mainly retrieved by the MODIS (500 m), ASTER (30 m), and Landsat 5 TM, 7 (30 m) moderate and high resolution sensors: • "Land cover" is characterized by: "Land use" is thematically associated with "urban typology" and refers to the buildings function, e.g., residential, commercial, religious, industrial, or temporary construction (ids: 10,19,20,57). Some authors define a local spatial index associated with the degree of urbanization and infrastructure of the area, e.g. the "percentage of urban villages" (ids: 10, 53), the percentage of "village area with vegetation" (id: 71), or the "quality of neighborhood" (id: 32).
Data sources network considering GIS: "GIS data" sources are initially collected from various sources such as digitised maps, geocoded census data, or in situ observations. The network shows a strong connection with the "land use" properties of the environment ( Figure 5). Urban landscape is characterized through: • "urban typology" associated with (i) urban morphology with construction height, e.g. "high or low-rise housing" (id: 58), (ii) building function, e.g., "tire repair shops" (id: 18) (ii) area functions, e.g., "residential/commercial/recreation" areas (ids: 19, 23, 57), "informal settlement" areas (id: 23, 51), "Park" (id: 29) "cemeteries" (id: 18); • "infrastructure level", e.g., proximity to the hospitals (id: 1), water network connection (ids: 15,18,23), canal and ditches (id: 15), "road density" or "parks area"(ids: 10, 18, 37, 50, 51); • "housing type", e.g., connections between houses. Some authors also considered topographic data, like shade or altitude, which influence the Aedes presence; • GIS Land cover data indicates the presence of water areas and wetland (id: 16), and cropland (id: 16, 29); • "Human presence" is characterized by geocoded density (id: 7); Data sources network considering survey questionnaires: In the context of this mapping review, "survey questionnaires" associated with census data constitute the largest data sources for landscape characterization associated with dengue cases (Figure 6), and inform at household-level according to:        Figure 6. Co-occurrence network mapping of the self-defined keywords related to the article using survey questionnaires to produce the landscape factors. We indicate in orange those factors that could also be produced through remote sensing techniques.

Proxies According to the Geographical Units of Spatial Analysis
Of the articles in this review, all the relationships between dengue occurrence and landscape features were based on aggregated data at a given geographic level. Relationships were not identified for individual dengue cases, except in id 22 (human mobility patterns of recently DENV-infected subjects). Since we considered data from survey questionnaires, a large number of relationships were identified at fine scale household-level, where the authors mainly considered the influence of house type and characteristics in the dengue transmission process, and exposure to Aedes bites by including entomological observations (ids: 1,4,8,12,13,20,25,26,34,35,48,52,55,60,68,75,77). Urban administrative divisions were often considered because (i) they represented the legal unit of dengue cases reports (ii) other datasets, such as demographic or socio-economic data, were aggregated and available at the same levels. Generally, the authors considered the smallest local administrative level, but we noticed a large diversity in the 78 articles in the names of organizations and the denomination of national administrative units: "Districts" (ids: 3, 32, 33, 36, 65), "Li" (id: 15), "BSA" (id: 16), Locality (id: 19), "Barrangay" (id: 23), "Cantones" (id: 44), "Municipios" (id: 62), "Colonies" (id: 63), "Villages" (id: 74), "health sectors" (ids: 27, 69) and "national census tracts" (ids: 11,17,38,46). Five authors proposed a study considering the whole city (ids: 21, 41, 50, 64, 66) or very populated areas (id: 60). Various authors aggregated the data at the neighborhood level, considering dengue diffusion at fine scale linked with Aedes flight, or human density and proximity to Aedes presence (ids: 5,6,7,9,14,18,24,28,49,54,56,57,58,59,61,67,70,71,73,78). According to individual authors justifications, we interpreted the choice of a landscape factor, considered at a given geographical unit of analysis, by its link to one or several mechanisms involved in the dengue transmission process ( Table 4): 1. ecological factors favorable to Aedes presence and development through direct entomological observations, or elements of the landscape favoring the presence of breeding-resting sites; 2. probabilities of human exposure to Aedes bites at household-level through small-scale proxies associated to the housing type or its characteristics; 3. probabilities of human-vector encounter considered at neighborhood, small and large administrative levels; 4. virus conservation and diffusion through human mobility. Table 4. Landscape factors interpreted as proxies of different processes involved in dengue transmission according to the geographical level of data aggregation.

Landscape Factors
Proxies of Geographic Level Housing characteristics: Animal water pans, Households with water supply, regular water supply, water containers, sewage system, garbage collection

Statistical Models
To quantify the relationships between urban landscape factors and dengue cases, the authors adopted methodologies based on statistical and spatial analysis fields, classically employed in spatial epidemiology or disease risks geography [38]. Correlation is commonly used to quantify the direction and strength of the relationship, through Pearson and Spearman (ranking) correlation coefficients (ids: 1,24,29,31,33,42,44,53,56,60,61,62,64,65,67,69,76). The odds ratio, which quantifies the strength of the association between two events is also often used (ids: 13,20,25,26,27,34,48,68). Ecological regression analysis was used to estimate a relationship equation between "dengue cases" and one or more independent "landscape-based predictors" at a given area-level, underlying several assumptions on the data distribution and its associated errors, such as independence between observed cases. Assuming a Gaussian conditional distribution of the dependent variable in respect to the predictors, several studies considered simple, multiple, or generalized linear models (ids: 17,45,47,62,66). Based on a Bernoulli conditional distribution of the categorical outcome variable in respect to its predictors, most of the authors used logistic and multivariate logistic regression models to estimate the probabilities of a dengue infection (ids: 2,9,13,18,22,26,39,41,43,49,70,71,75,77). To introduce non-linearity terms due to the spatial dependence of the predictors, some authors considered the generalized additive model (GAM) (ids: 6,10,28,50,51). To adapt the model to local contexts, some authors used the geographically weighted regression method (GWR), which takes non-stationary variables into consideration and models the local relationships between predictors and dengue cases (ids: 14,17,32,53,54). Two studies considered a generalized linear mixed model (GLMM, id: 8, 29), a model that, in addition to the fixed effect, includes a random effect for which the hypothesis of independence of observations is no longer assumed [36].

Mapping of Relationships at Household-Level
Except for the use of air conditioning, which could appear as a protective factor (ids: 52, 55), the housing characteristics considered in the included articles generally presented non-significant relationships with dengue cases (Figure 7): e.g., the number of windows in a house, the distinction between "public" or "private" multi-storey flats, floor of principal entry, the use of water containers, or the housing size. Screens on windows might appear to be a protective factor in some cases (ids: 26,43,55,70,73), but the association with dengue cases was also observed as statistically non-significant (ids: 4,13,20,30,65), and positively associated (id: 35), which might reveals the high density of Aedes or vector-borne disease in the area. No clear relationship was generally associated with construction materials: e.g., wood can appear as non-significant (ids: 26, 55), positively (ids: 70, 73) or negatively (id: 71) associated to dengue cases according to the study. Concrete, stone, or brick do not appear to be protective factors (ids: 55,65,70,71,78). Entomological observations are generally positively associated with the presence of dengue: direct Aedes observations of adults, pupae, larvae, or infested and discarded containers (id: 1, 25, 34, 60). Aedes aegypti is much more cited than Ae. albopictus in the included articles. In the domestic environment of a house, the presence of shaded and vegetated areas, and the lack of street drainage appear as exposure factors (ids: 26, 30).

Mapping of Relationships at Neighborhood Level
At the neighborhood level, it is possible to define an urban typology associated with an area, by considering the housing type and the building functions ( Figure 8). This led the authors to propose various urban ecotypes, and to consider the residential, commercial, or social function of a construction, after taking into consideration transportation or ecological aspects like density of roads or vegetation. Despite the difficulty in comparing authors' self-definitions, the mix of residential and highly frequented areas, associated with multi-scale human mobility (e.g., road network density, ids: 14, 37), with vegetation in the surrounding areas generally show the strongest associations to dengue occurrences (ids: 10,14,19,28,35,37,51,57). Considered separately as individual proxies, urban functions are generally not significant (ids: 18, 35). Slum-like or informal settlement areas may be positively associated with the presence of dengue (ids: 14, 28, 51, 53, 73), but not systematically (ids: 3,49). Well structured urban areas, defined by a "quality index", may have protective effects (id: 32). The height of buildings could have an influence: low-rise buildings may be more exposed than high-rise buildings (ids: 49,58). Few articles considered human density directly as a proxy at neighborhood level, and it appears non significant or positively related to dengue cases (ids: 7, 26, 35). Entomological observations are fewer than at household-level, and may show significant (e.g., with Aedes house index) or non-significant relationships (e.g., with Aedes eggs, larvae, and pupae abundance, or Breteau index, defined as the number of positive containers per 100 houses inspected).

Relationships at Administrative Units
The authors considered a small administrative level to integrate data from institutional sources at fine scale (Figures 9 and 10). A co-occurrence network shows some similarities with the neighborhood level, highlighting the role of human density through residential area mapping (ids: 16,19), and the importance of mixed areas, characterized by coming and going of people with some hot spots or a context favorable to the persistence of Aedes: urban villages (id: 10), deprived areas with medium-high density (id: 38,44,63), residential areas with commercial and industrial areas (id: 23), or informal settlement areas (id: 23). With regard to infrastructure level, it is useful to consider waste management and the state of the sewage networks (ids: 15,27,65), as well as road structure and density (ids: 10). The orientation of a street, the presence of empty houses, or the use of gutter rain are urban characteristics that could play a role in maintaining Aedes (id: 27,74). Building height is also a variable of interest (id: 46). Some authors have information on human mobility, generally significantly associated with dengue cases, which highlights the usefulness of estimating human fluxes (ids: 11,22,77). Historical epidemiological data are scarce, but allow for the study of dengue urban patterns over time, and are especially significant when associated to DEN serotypes (id: 35). Entomological observations are not aggregated or available at the level of administrative units. The presence and density of the Aedes mosquitoes are addressed through prior knowledge on vector bio-ecology and remotely-sensed environmental data: (i) the classical index NDVI is used as a proxy of the vegetation, and is positively associated to dengue cases in two of the three studies (ids: 10, 42, 50), (ii) urban surface temperature was not significant (id: 42). At larger administrative levels, authors considered the influence of altitude, which is negatively correlated to dengue occurrences (ids: 21,34,44,64). This result illustrates the influence of the temperature gradient on Aedes ecology. Human mobility is also correlated with dengue cases (id: 20,22). Vegetation also seems positively associated with dengue occurrence (id: 36), although NDVI is associated with a negative relationship to dengue in two cases (id: 3, 45), which could be due to a decrease in residential surfaces in respect to vegetation surfaces.

Methodological Considerations
The expansion of evidence-based practice across scientific disciplines has led to an increasing variety of review types. We chose a mapping review, which enables the contextualization of in-depth systematic literature reviews within broader literature and identification of gaps in the evidence base [40]. The network, based on calculating the barycenter of the structured textual information, is aimed at proposing a coherent synthesis in a graphical way. The forms of the network graph are however quite dependent on the way information is sorted, structured and grouped. Our work is limited to a broad descriptive and qualitative level, and thus may oversimplify the considerable variations (heterogeneity) between studies and their findings [40]. Mapping reviews do not usually include a quality assessment process to preselect the articles, which could limit considerably the quality of the information and analyses produced. To provide an assessment of the risk of bias, we proposed here a simple checklist on key features based of metadata completeness, and an overall appraisal of the level of contributive information respect to the topic "dengue-relationship characterization" (Supplementary Materials). In addition, we did not include conference papers, which could contain some relevant information at the front-line of the research. We focused on urban areas, but rural areas could contribute at least as much to the dissemination of dengue fever as cities [56]. In a context of significant increase of dengue publications over time [57], our study highlights that specific research on spatial epidemiology, like dengue landscape factors, is not at the front line compared to virology, biochemistry or molecular biology research areas. Surprisingly, we did not find any articles which follow our inclusion criteria related to other Aedes-borne diseases, like Zika and Chikungunya when we swap dengue to one of them. These can be relativized by the recent character of the massive outbreaks associated to the Zika flavivirus [58,59]. We found only one study concerning Africa, which might be due to (i) many other competing public health problems (e.g., malaria or Ebola) and limited resources [60], which cause a lack of diagnostic testing and systematic surveillance [61] and (ii) a less suitable environment for dengue [62], with potential differences in terms of vector efficiency and viral infectivity between Africa and other dengue-endemic regions [63]. However, depending on location, rapidly increasing urbanisation, and/or higher temperatures and increased rainfall could increase dengue incidence in the following decades [62,63]. In general, only one article mentioned a given landscape factor, which prevented us from performing a more in depth meta-analysis, and limited us to the present qualitative analysis.

Limitations Associated with Epidemiological and Entomological Data
Through this review, we noted that passive notification cases, reported by official health systems, and dengue serostatus surveys, performed by research teams, can show two different realities of dengue occurrences, relativizing in this way the comparison between the factors proposed in the types of studies. Passive case notification datasets present strong identified biases due to (i) the absence of asymptomatic cases (ii) the absence of symptomatic cases when patients do not consult because of, particularly, the distance to health centers, or their cultural habits, and (iii) misdiagnose based on insufficient medical evidence. On the other hand, intra-urban dengue seroprevalence surveys are based on a sampling strategy where assumptions and representativeness may be inaccurate, and could limit interpretation: lack of demonstrable spatial variation between self-defined areas (id: 8), complexity to define an appropriate urban ecosystem (id: 35), relative influence of contextual indicators versus individuals (id: 48), and limitation to school children population (ids: 49, 67). Unknown socio-demographic drivers, the retrospective nature of questionnaires, and associated recall bias are other issues that should be mentioned (id: 49).
Four distinct serotypes of DENV have been identified, and infection from one serotype confers protective immunity against that serotype but not against other serotypes [64]. Acquired immunity may therefore introduce a bias in any dengue pattern study. From that perspective, historical studies of dengue epidemics can provide valuable information. However, such data are scarce, and few studies have performed both IgM and IgG analysis in the correct time window. Early tests (up to day 7) using Reverse Transcription Polymerase Chain Reaction (RT-PCR) should be preferred because their specificity is much higher than serology, but only one study has performed a Plaque Reduction and Neutral Test (PRNT) to distinguish between dengue serotypes (id: 36). In one study, two time-periods have been considered to distinguish potential infections by DENV-1 and DENV-2 (id: 16).
Underreporting in dengue surveillance systems has been identified in various studies [65][66][67] demonstrate, through a systematic review, that a large proportion of the data from any affected population has not been captured through passive routine reporting-misdiagnosis or subclinical cases, non-users of health services, users of private versus traditional sectors, or certain age groups. In high endemic settings, however, if the dengue cases are geographically representative and laboratory confirmed, dengue data may be representative, to some extent, and possibly corrected by calculating an expansion factor. Improvements in dengue reporting could come from improvement in indicators/alert signals, laboratory support, motivation strategies, shifts in dengue serotypes or genotype surveillance, and data forms/entry/electronic-based reporting [66].
Dengue cases were rarely associated with entomological data, probably due to the difficulty in obtaining these data in a cost-effective way. Except for household-level studies, mosquitoes were generally considered from prior knowledge, and not from in situ observations. Aedes were sometimes considered as composed of a unique species, without differentiating albopictus from aegypti despite their different ecological behaviours. This point could however be relativized because of the remarkable ecological plasticity of both species, especially to urban settings [10,11].

The Difficulty in Defining a Geographical Unit of Spatial Analysis
The first requirement in performing a relationship between dengue cases and environmental determinants is the geolocation of the cases. Most of the selected studies do not go into detail on that point, except when an automated procedure has been implemented (id: 42). Generally, a hypothesis is made after dengue cases have been located at a patient's home address as the transmission may have occurred at home or in the vicinity of the household. Aedes aegypti and Ae. albopictus are day time biting mosquitoes, which implies to consider human commuting pattern. Such hypothesis might be strengthened when considering an age stratification, as the mobility of elderly persons or young children mobility can be limited for example (ids: 17,70). If the dengue cases are located within a given area, the probability of the transmission may increase up to a threshold distance, but it might become more difficult to identify the correct environmental determinants associated with the transmission. These proximity-hypotheses are consistent with local, density dependent transmission as key sources of viral diversity, and with home location being the focal point of transmission [68]. Using geolocated genotype and serotype data, Salje et al. [68] showed that in Bangkok (Thailand), dengue cases came from the same transmission chain for (i) 60% of cases living in less that 200 meters apart, and (ii) 3% of cases separated by 1 to 5 kilometers. At distances closer to 200 meters from a case, the authors estimated the effective number of chains of transmission to be 1.7, and that this number rises by a factor of 7 for each 10-fold increase. As in the large majority of ecological-related issues ( [69], Modifable Area Unit Problem), the choice of an appropriate spatial unit to associate a relationship between dengue cases and their risk factors has a strong influence on effective analysis. We identified various type of infra-urban areas of spatial analysis in the 78 included articles (e.g., buffers around the infected households, census tracts, health regions, small and large administrative areas), which varied according to authors' choices, data sources and availability. Dengue cases and landscape factors are often aggregated to an administrative level or census tracts to perform comparisons with socio-economic or demographic datasets. When considering an administrative area, there is a risk of disruption with dengue transmission mechanisms as it does not represent a spatial homogeneous area for vector ecology or the human exposures to Aedes bites. According to the specific objectives and time period of the study, the use of an administrative unit as an analysis area could be justified [70], but the inevitable simplification that occurs when attempting to model real-world phenomena should be considered and systematically discussed, independent of the type of spatial units or chosen methods [38].

Highlights and Perspectives to Improve the Frame of Urban Dengue-Landscape Relationships Studies
Our purpose was originally to identify studies based on remote sensing techniques to produce landscape factors, so we opened our search to all kinds of information sources, including survey questionnaire and GIS data. Such strategy is guided by the consideration of a holistic conceptual risk and vulnerability framework [71], to allow for the identification of new factors that would be potentially achievable by using remote sensing techniques. The main purpose was to identify what makes a given landscape "pathogenic" or not, in respect to dengue transmission [72]. We privileged a "Built City" approach, i.e. a city as a physical entity, [44], to avoid direct socio-economic considerations in landscape factors. Discursive links between dengue and poverty may have contributed to an inappropriate transfer of globally dominant dengue control strategies to non-poor local environment [73]. From this perspective, the quantification of human exposure to Aedes bites through salivary antibody-based biomarkers may be a promising method for estimating the influence of the bio-physical environment on human-Aedes contact [74]. Only two articles used landscape metrics to explore the impact of more in-depth ecological characteristics of an urban landscape on dengue transmission (ids: 57, 69). Landscape metrics have been separately applied to malaria transmission for assessing the influence of landscape factors relative to exposure risk [75,76]. The representativeness of sampling strategies during intra-urban dengue seroprevalence surveys may be improved by the use of GIS and remote sensing techniques ( [77], e.g., urban environmental clustering and Aedes density); ( [78], e.g., Urban typology) and help to objectify the choice of geographical units ( [70], e.g., criteria of intra-unit homogeneity, areal and population size, compactness); ( [71,79], e.g., Concept of integrated geons). Public health services could also benefit from original visualization techniques to map metrics or indexes related to dengue vectors or occurrences ( [80], e.g., Ring mapping).
Id 22 highlighted the importance of human movement, and time spent in places at various scale in human exposition and DENV spreading. Id 37 showed that high-density road network is an important factor to the direction and scale of dengue epidemic, and that the dengue cases were mainly concentrated in the vicinity of narrow roads. Id 63 insisted on the "forest fire" signature of DENV epidemiology in the context of Dehli (India), while id 61 refers to a "silent epidemic in a complex urban area" in the context of Salvador (Brazil), where "high rates of transmission were observed in all studied areas, from the highest to the lowest socio-economic status." Many authors referred to the necessity of an improvement in the individual geolocalisation capacity to estimate human mobility patterns, since an "importation of infected individuals into a frequented area could lead to a local foci of infection included with a low Aedes density". Id 12 considered that "dengue transmission occurs, not at a fixed entomologic figure/quantity but rather at a variable level based on numerous factors including seroprevalence, mosquito density and climate." Entomological indices may be good proxy of DENV occurrences at household-level (ids: 4, 34, 68, 75), but seem less significant when aggregated at coarser resolutions (ids: 6, 26, 28, 59), or when considering only larvae (id: 5). Some important data relative to vector borne diseases are exclusively accessible by field survey, e.g., type of material construction or screens on windows, but their knowledge do not seem so critical in the case of Aedes borne disease (ids: 4, 13, 70, 73). Many survey questionnaires based studies confirmed the large inadequacy of remote sensing techniques to properly identify potential dengue risk factors in link with Aedes habitats, characterized by a fine or micro-scale level: empty houses, sewage system, garbage system, street drainage, water pumps, water containers, open sewers, tyres, water puddle, ditches, cans (ids: 8,9,17,18,33,65,68,74). However, remote sensing techniques should be now in capacity to provide more than land cover information, and could help to systematically inform on land use and urban typology, without the need of a questionnaire, as (i) proxies of human presence and activity, or as (ii) macro-scale hotspot proxies of Aedes habitats e.g., cemeteries (id: 17), construction site (id: 36), vegetation height (ids: 26, 73), shade (ids: 26, 73), or roof shape (id: 54). Based on sound statistical machine learning, such complex urban typology could be labeled from space at neighborhood or small administrative level: informal settlement areas (ids: 23, 28, 49), urban villages (52), quality of neighborhood index (ids: 32, 52), or multiple association of urban functions (ids: 18,19,23,35,57), especially if completed by building height (ids: 58,46,75). Such improvement could help to explicit the multiscale geographical framework where DENV transmission occurs as a result of a multifactorial process. At the same time, remote sensing products could help to guide the questionnaire during the field survey, while GIS provide the framework to combine all spatialized information and performs geo-analysis (id: 10). Although remotely sensed radiometric measures like NDVI or LST could provide conflicting conclusions (ids: 3, 10, 42, 44, 50, 69), their use in a sound methodological framework could be of some interest, especially when available at higher resolutions. Digital archiving in GIS context of geocoded and confirmed dengue cases should help to easily inform on historical dengue risks areas (id: 35). Such digital layers could provide an interesting proxy of dengue transmission patterns when DENV-serotype is known.
As was apparent during this review, we were not able to identify a set of land cover and land use classes unequivocally related to dengue risk factors. This is consistent with the fact that reliable predictors for dengue have not yet been established in the literature [36], and the Aedes presence and density are not sufficient to determine dengue epidemics [13], which justifies the scope of this review, centering on dengue cases. DENV transmission is complex, and the relationship between vector density and risk is not static nor adequately characterized through periodic entomological surveillance [81]. However, even if Aedes indicators serve as surrogates of true exposure [81], vector control will remain the primary prevention strategy in most dengue endemic settings [1], including when an effective dengue virus (DENV) vaccine would become commercially available [18]. To better target surveillance programs, effective control of Aedes could benefit from available evidence-based guidance by considering an Integrated Aedes Management framework ( [82,83], IAM).
Some specific factors are unachievable using remote sensing techniques due to their limited spatial dimension and should continue to be acquired by field and entomological surveys, e.g., decimetric spatial resolution for breeding sites or for gutter rain, or because they are hidden from the sensor perspective. However, building detection remains a central task as it allows human presence and density to be identified, and is constrained geographically to the urban area. Building environment, e.g., vegetation or water areas, is also of interest since it could influence Aedes ecology or human activities. Building function, e.g., residential or commercial, can give important information about human activities and human presence related to time. Road and transport networks may also constraint Aedes and DEN virus diffusion, and can be related to patterns of human commuting. Land use data related to human movement and places visit frequency should help in reducing the difficulty of acquiring detailed knowledge about "the non-random nature of encounters" [8]. In this way, urban mapping, particularly by including land use, could provide the geographical context in which, with adequate parameters that compensate for missing information, dengue-related processes could be modelled ( [36], Review on modeling tools for dengue risk mapping; [84][85][86], Getis-Ord Gi in GIS context; [87][88][89], Spatial Mechanistic Modeling of Aedes Mosquito Vectors; [90], Spatial agent-based simulation model of the dengue vector Aedes; [91], Environmental hazard index mapping methodology of Aedes aegypti; [92], Modeling Dengue vector population using remotely sensed data and machine learning; [93], Comparison of stochastic and deterministic frameworks in dengue modelling).
To improve surveillance and monitor of dengue occurrences and Aedes mosquitoes, intercomparison model projects could help to identify the most general and efficient models considering various geographical contexts and data set: ( [94], e.g., Airborne spread of foot-and-mouth disease -Model intercomparison; https://www.theia-land.fr/en/anisette-tracking-mosquitoes-thatcarry-disease/, e.g., Inter-Site Analysis: Evaluation of Remote Sensing as a predictive tool for the surveillance and control of diseases caused by mosquito, and future impacts of climate and/or land use changes may also be considered; [95], e.g., Malaria and climate; [17,23,96], e.g., Urbanization). Review of literature are also needed to update the ever-increasing output of scientific publications, and lead to new synthetic insights ( [97]; [10], e.g., Determinants of Aedes Mosquito Habitat for Risk Mapping, [98], e.g., New frontiers for environmental epidemiology in a changing world, [99], e.g., Current challenges for dengue; [100], e.g., Mosquito-Borne Diseases: Advances in Modelling Climate-Change Impacts; [101], e.g., A 10 years view of scientific literature on Aedes aegypti; [102], e.g., Satellite Earth Observation Data in Epidemiological Modeling).
The potential of satellite images and remote sensing techniques should continue to be explored. As mentioned in this review, the images used often corresponded to old missions or end-of-life satellite sensors, and methodologies should consider more state-of-the-art-approaches: • the native pixel resolutions were often aggregated at a coarser resolution during the mapping production ( Figure 11). Recent satellite missions should bring greater possibilities to fit spatial resolution and temporal windows over urban areas two studies have exploited the thermal sensors from Landsat-TM and MODIS instruments, and used them to retrieve land surface temperature (LST) parameters (ids: 3,19). This is particularly useful to detect urban heat islands that could indicate improved conditions for Aedes viability and dengue virus replication, due to the potentially amplified higher temperatures (typically greater than 30 • C), and resulting in a reduction of the extrinsic incubation period from 12-14 days to 7 days ( [111], id: 3). New thermal sensors with higher spatial resolution may promote consideration of thermal sensors, such as the CNES-TRISHNA mission [112,113], even if methodological issues remain: that is, hotspot effects, separation of temperature and emissivity parameter. • dengue is often spread in tropical or subtropical regions, where the presence of clouds and cloud shadows result in missing data in optical images. Synthetic aperture radar SAR images could penetrate such barriers and might be combined with optical sensors for overcoming this issue. Such an approach to optical and SAR fusion has been applied in the studies of malaria [114,115]; • very high resolution imagery may be more suitable for extracting the direct dengue-related landscape factors, such as (i) the type of vegetation near human settlements [104,116] (ii) the footprint of built-up areas [46,117], and (iii) land use types, such as slum areas [118,119]; • from high-resolution built-up area detection, population growth estimation due to urbanization could be assessed, improving the estimation of census and incidence rates [120,121]. In this regard, only one article proposed a proxy for a spatially-corrected population density by digitizing and excluding inhabited areas (id: 24). To improve the population density assessment, cities should be considered in their verticality and volume, through the use of a digital height model, potentially generated from unmanned or satellite remote sensing stereo imagery [122][123][124]; • although we did not consider meteorological factors here, surface air temperature or soil moisture, traditionally measured by in situ weather stations, could be derived from satellite passive microwave radiometry [102,125].
The temporal dimension remains largely absent in the spatio-temporal relationship studies of this review. Populations commute, as well as mosquitoes. If a decrease in mean distance between dengue cases may generally correlates with activity, and could lead up to an outbreak, a decrease in temporal distance between dengue cases may increase geographic spread of the disease [126]. Landscape changes associated with human mobility, like transportation infrastructure changes, may create favorable conditions for the establishment of dengue virus [127]. However, relationship investigations are usually done under a stationary analysis scheme, and the mapping of dengue patterns often ignore "temporal kinetics" (id: 32). A complementary approach to this static view should be to consider human mobility in relation to Aedes-bites exposure, and not only to mosquito dispersal associated with its flight, as this former could affect significantly the spread of infection [128]. Adams and Kapan [129] enhanced the fact that hubs and reservoirs of dengue infection can be places people visit frequently but briefly. Authors from id 74 found that most of the space-time distances of non-commuting dengue cases clustered within 100 m and one week, whereas commuting cases clustered within 2 to 4 km and one to five weeks. Human commuting patterns may be estimated through the use of GPS data-logger (id: 22) [130] or regularly logged cellphone tracking data [131], which could be in the next decade generalized in the so-called Smart City model ( [132] Real Time Health Monitoring, [133] Smart Health care Internet of Things and Aedes monitoring, [134] Geospatial artificial intelligence).   Figure 11. Comparison between pixel size (x axis, in log scale) and typical dimension of geographical area used to perform relationships with dengue cases (Y axis, in qualitative dimension).

Conclusions
We propose here a mapping review which focuses on the landscape factors potentially related to urban dengue transmission. By analysing the 78 included articles that satisfied these criteria, we found that the landscape mapping linked to human dengue infection was mainly guided by (i) vector ecology-based considerations through vegetation and water surface mapping and (ii) human presence and activities deducted from the settlement typology.
We extracted each of the specific landscape features that have been assessed in the context of DENV transmission. We proposed a systematic three-valued interpretation of the relationships performed between each landscape factors and dengue occurrences, and provided a representation in a graphical way according to the considered spatial scale of the studies. Even if some characteristics appear essential, as human density and movement pattern, or the presence of a minimum vegetation in the surrounding, considering only one landscape factor at a time should be avoided, as we highlighted the complexity of the "pathogenic landscape" associated to dengue transmission. In a broad and simplified approach, relevant landscape is characterized by a mix of residential and highly frequented areas, associated to multi-scale human mobility, with an entomological thresholds that can be low. From a remote sensing perspective, there is a need to identify land uses more than solely land covers to characterize more complex urban environment: informal settlement, building typology, transportation network, and consider the vertical dimension of the city. Up to now, these kinds of information have been more often retrieved from costly and time-consuming survey questionnaires than from automatic remotely-sensed approaches. To provide a realistic geographical context in dengue modelling and to take into account the complexity and the multi-factorial nature of DENV transmission in tropical environments, remote sensing approaches need to be promoted through the use of recent HR and VHR sensors such as, Copernicus (Sentinel) or Orfeo (Pleiades) programs, a combination of optical, including stereo, and RADAR approaches, and state-of-the-art image processing algorithms, including deep learning techniques when possible. A strengthening of relations between environmental epidemiology and urban mapping communities should help to standardize the mapping of the urban typology of interest, and therefore enable better assessment of the influence on dengue transmission.
As integrated approach combining remote sensing, GIS, and field survey preferable when possible, since health data and entomological observation availability and quality would probably remain the main limiting factors if landscape and urban typology mapping, including human movement pattern, continue to improve. Due to the silent characteristics of DENV presence within the city, dengue control still requires above all an active search and an early detection of new cases, including serotype detection, associated to an entomological control at fine scale involving both citizen and health agencies.

Acknowledgments:
The authors would like to thank the members of the work-groups from the Environment, Societies and Health Risks inter-disciplinary (ESoR, http://www.espace-dev.fr/index.php?option=com_content& view=article&id=37&Itemid=193), and from the ANalyse Inter-Site CNES-TOSCA project (ANISETTE, https:// anisette.cirad.fr/), particularly Eric Daudé and Renaud Misslin, for the constructive discussions that considerably enriched the paper. We also thank the administrative and technical supports.

Conflicts of Interest:
The authors declare no conflict of interest. The founders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

Appendix A. Raw Descriptive Tables of the 58 Included Articles
Appendix A.1. Identification and Localization Table of the 58 Included Articles   Table A1. Extraction of the publication meta-data (first author, date of publication, title, name of the journal), and description of the geographical contexts (country, city, geographical unit) of the 78 included studies.               Table A3. Data extracted from the 78 articles on the landscape factor production (type of source), on the landscape factor classification according to groups and subgroups, and on the dengue-landscape relationship (three-valued interpretation: +, −, or NS, and statistical method performed).