Éléments méthodologiques de diagnostic paysager utilisant les systèmes d'information géographique

Although landscape is not a recent notion, its place in environmental management is now of major concern. Unprecedented land transformations have become a source of worry in terms of impacts on the landscape. The dynamics are still poorly understood and need to be studied in detail to be able to design consistent planning strategies. Today's policy is largely reactive and calls for prospective, as well as participative, methods and tools at different scales. Information Technologies (IT) and more particularly Geographic Information Systems (GIS) are fast-expanding domains proposing more and more better-suited solutions for problems concerning our environment. They generally provide means to capitalize on the different land data sources, whether they are spatial or not. Though their use is still limited in the context of landscape analysis, they have a huge potential that needs to be developed. Therefore, the current study is concerned with landscape research as well as GIS development. The contribution of GIS is assessed with respect to landscape management and both methodological and technological adaptations are proposed in return. One of the most important question concerns the integration of the qualitative dimension of landscape evaluation in fundamentally quantitative systems. The Jura mountains and more specifically the Marchairuz region in the higher part of the Jura vaudois constitute the study region. The main result of the work consists in the proposal of a participative and multiscale landscape assessment method, relying on the specific use of information technologies. Landscape is considered to be the visible part of the land, that the local community perceive and consider. Different forms of appreciation are modelled within a system of spatial indicators. It is used on a small scale to identify and locate landscape impacts coming from land transformations, in order to compare them with public perception. On a larger scale, indicators are used to translate social preferences in space, from the definition of landscape representation models. The assessment method relies on visibility analysis functions, improved with visual correction factors and content analysis operators. Finally, the method proposes an objective way to put together subjective landscape perceptions and thus offers an interesting basis for negotiation. The relevant approach is defined by examining its integration potential in existing management processes. Additionally, the pros and cons of the method and the tools developed here are described and some final perspectives are drawn.

Amphibian conservation in human shaped environments

Flavio Zanini

Global biodiversity is experiencing a worrying decline. Habitats destruction, associated to their degradation and fragmentation are among the greatest causes. Amphibians are particularly interesting because they are more threatened and decline more rapidly than either birds or mammals. In this context, the objective of our research is to improve some methodological approaches and offer practical scientific bases for decision making in landscape management and amphibian conservation. Our study focuses on fragmented Swiss landscapes. We developed a method that uses land-cover data and expert knowledge to enable a spatially explicit assessment of 1) the temporal changes in the nature conservation value of the landscape and 2) the rehabilitation potential of the landscape. We applied this Geographical Information Systems (GIS) based approach in the Swiss Rhone plain and we used the years 1900 as the reference state. The method constitutes a helpful tool for communication, decision-making and biological conservation management in landscape planning. Effective and optimal species management strategies can only be formulated after relationships between species distribution and environmental factors have been identified. Concerning amphibians, several approaches exist but they generally suffer from two limitations: 1) the spatial autocorrelation (i.e. the dependency between two observations) in data is rarely explicitly analyzed, even if it may affect the accuracy of species-habitat relationships models. We showed how this spatial autocorrelation can be measured and included in logistic models with the example of the agile frog (Rana dalmatina) in north-eastern Switzerland. We used the Moran's I and the autologistic model (i.e. a logistic model including a measure of the spatial arrangement of the response variables). We found that if spatial autocorrelation is not considered, then conclusions on species-habitat relationships can be incorrect. 2) The effect of landscape on amphibian occurrence in ponds is often assumed to be equal in every direction (isotropic). However, barriers and inhospitable surfaces may reduce movement patterns and the area around ponds accessible to species. This implies that the ideal circular area has in reality a shape depending on the surrounding landscape. We developed a method to determine the effect of habitat variables on amphibian species distribution, considering physical barriers in their movement around ponds. We studied two amphibian species: the common toad (Bufo bufo) and the common frog (Rana temporaria) in the Rhone plain. We demonstrated that reducing the boundaries of circular area following barriers, allowed to compute landscape predictors which better explained species distribution. These results suggested that the proposed approach is more pertinent than the traditional circular buffers analysis. Our results stress the necessity to consider barriers and ecological corridors in species distribution models in order to avoid incorrect inferences. Species distribution models are usually established for a single region. It is generally unknown whether the identified relationships between species distribution and environmental variables can be directly transferred to another geographical area. We examined landscape-level habitat relationships for six amphibian species by measuring correlations with their presence in 655 ponds of five different regions. We analyzed several models by using the information-theoretic approach and the Akaike Information criterion (AIC). For five out of six species, the best models predict that site occupation probability depends on region. Our results suggest that caution is needed when predictions of species occurrence and species management strategies are done using models built in other geographic regions. We also observed that connectivity was generally more explicative than landscape variables. In addition, we found that the spatial scale at which habitat affected species occurrence varied from pond to several km around ponds. Management strategies for amphibian conservation should be conducted taking into account the geographic context, connectivity of ponds and habitat characteristics at multiple spatial scales. Finally, we demonstrated that the landscape, separating patches in metapopulation models, has to be considered in order to avoid incorrect conclusions on population viability analyses. We explored how patch occupancy is sensitive to Euclidean (shortest) versus a landscape-based distance (least-cost). We found: 1) from a theoretical standpoint, that inter-patch landscape affects patch occupancy; 2) from a practical and conservation standpoints, which patches should be considered in priority for landscape management. The approach was illustrated in the case of two metapopulations of the Yellow-bellied Toad in the Rhone plain. We applied successfully the developed practical approaches to the case of several amphibian species, but they can doubtlessly be extended to any species functioning on a spatially defined patch basis (e.g. pond, nesting place, den …), structured as a metapopulation and affected by landscape structure during movement. By improving and combining spatially explicit approaches, we are more likely to provide wildlife managers with tools for valuable decision making.

EPFL2006

Dispersal modelling

Séverine Vuilleumier Varisco

In human-dominated landscapes, populations and species extinctions are directly related to habitat destruction and fragmentation. To provide genetic diversity as well as population viability, individual exchanges among isolated populations must be maintained. Therefore, animal dispersal processes in fragmented landscape become an important topic for ecologists, and ecological networks planning has become one of the major challenges for landscape planners. Identification of habitat patches as well as assessment of the effect of ecological networks is badly needed. Since little information on the effect of landscape heterogeneities on animal dispersal is available, simulation models are being developed. As dispersal pattern and success strongly depend on the spatial context, species' interactions with landscapes, species behaviour and species ability to disperse, these models must be able to simulate them explicitly. This research work therefore aims first at developing methods and models that allow realistic animal dispersal simulations in fragmented landscapes. Second it aims at evaluating the effect of landscape heterogeneities and animal behaviour on dispersal and on species persistence. Additionally, the ability of such a model to estimate gene flow is analysed. To carry on this research, the following fields have been explored: landscape ecology, metapopulation dynamic, animal behaviour, genetics, Geographical Information Systems, modelling approaches and programming. A method, based on properties provided by Geographical Information Systems software, is first proposed to generate ecological networks by simulating animal dispersal according to animal movement constraints induced by human infrastructures. The resulting maps provide a spatial identification of ecological networks, corridors and conflicting areas. This model has proved to be a useful and straightforward tool for landscape planning, even if this model, similar to other present-day models used in dispersal simulations, presents numerous technical and scientific limitations. To improve models for animal dispersal, a feature-oriented landscape model associated with an expert system has been developed. Its conceptualisation, its formalism, its data structure and its object-oriented design implementation provide a very accurate representation of landscape features and simulation of complex interactions between model entities (individuals and landscape features) based on simple rules. It allows the spatial identification of simulated processes. The ability of this model to incorporate states, relations and transition rules between entities makes it applicable to simulate large ranges of dispersal processes according to specific behaviour and/or landscape uses. To analyse the influence of landscape heterogeneities and species behaviour on dispersal and their incidence on metapopulation dynamics, the proposed feature-oriented model has been coupled with an animal model. The latter assigns different cognitive and dispersal abilities to individuals. Based on simulations according to three movement strategies (corresponding to the cognitive abilities of the simulated species), two measures evaluate the effect of cognitive abilities on dispersal: the colonisation probability between habitat patches and the ecological distance (due to landscape heterogeneities). These measures give an estimation of metapopulation structures (the habitat patches belonging to the metapopulation) and metapopulation dynamics induced by the landscape heterogeneities (for example, the habitat patches which release individuals). The complexity of dispersal processes, considering species behaviours and dispersal abilities, can therefore be reproduced and analysed at different levels. This application has shown the importance of animal behaviour on metapopulation dynamics and structure. Since tracking animals and providing sufficient data remain difficult, calibration and validation procedures of dispersal models are difficult to perform. One approach proposed here is to measure one of the consequences of dispersal: genetic differentiation among populations. Geographical distances are in general used to explain a part of the genetic differentiations. But as our fundamental assumption states that landscape heterogeneities and spatial arrangements of landscape features may strongly affect dispersal successes, genetic distance between populations must be better explained by the estimate of a model which considers these factors. We have tested this assumption with the greater white-toothed shrew (Crocidura russula). Scenarios considering various behaviours and dispersal abilities of C. russula have been performed. Relating measures of genetic, geographical and ecological distances (the latter emerge from scenario simulation results) highlights the model capability to reproduce dispersal of C. russula by explaining a greater part of the genetic differentiation than that explained by the geographical distances. This application has not only pointed out the ability of the model to quantify connectivity between habitat patches but also the difficulty to relate gene dispersal and individual dispersal.

EPFL2003

Multi-scale landscape genomic models

Kevin Leempoel

Environmental heterogeneity is one of the main actors of biodiversity and species adaptation as it exerts a selective pressure on observable characteristics of living organisms. Consequently, local adaptation favours certain genetic variants and, by doing so, leaves a footprint in the genetic heritage across populations. The identification of these adaptive genetic variations is the main objective of landscape genomics and allows, among other things, to study the role of specific regions of the genome in evolutionary processes. Landscape genomics studies also provide essential information for species conservation and for the prediction of migrations due to environmental changes. To identify these adaptations to the environment, it is necessary to define a study area where the populations are sampled. However, defining the scale of the study area is not a trivial task. In fact, whether the work is carried out at a local or at a broad scale determines the relevance of environmental factors and the type of signature of selection that will be observed. In addition, the concept of scale in ecology takes into account not only the extent of the study area but also the pattern and density of the geographic distribution of observations, and the spatial resolution of predictors (environmental variables), which is intrinsically linked to the extent. However, a priori indications about the relevance of any resolution over another are rare in the literature and it is therefore essential to question this issue. In this thesis, we propose a multi-scale landscape genomic framework to identify signatures of adaptation to the environment. This multidisciplinary framework lies at the interface between geographic information systems, spatial analysis, environmental modelling, population genetics and computer science. Specifically, we focus on the relevance of variables derived from Digital Elevation Models (DEMs) and on the application of multi-scale analysis aiming to detect signatures of selection. We applied this analytical framework to three case studies, comprising four species: Biscutella laevigata sampled at a local scale, Plantago major at a regional scale, sheep and goats at a large scale. In particular, the case of B. laevigata allowed us to evaluate the role of topographic features based on Very High Resolution DEMs and to include DEM-derived variables as predictors in association models to study the adaptation of species to their local environment. On the other hand, the case of Moroccan sheep and goats permitted to include for the first time whole genome sequence data within landscape genomic models. The results revealed several important findings. We showed that micro-climate variability is highly dependent on topographic factors at a local scale and that therefore, DEMs are relevant for understanding species adaptation to a mountainous environment. We also demonstrated that it is essential to consider the scale of spatial representativeness by assessing DEM-derived variables at various spatial resolutions. Indeed, two out of three case studies showed that the models involving topographic variables were sensitive to changes in resolution. In summary, we used several landscape genetics approaches to understand the role of environmental factors in the local adaptation of various species. Our findings mainly provide an important contribution to the understanding and use of scale in landscape genomics, also useful in landscape ecology.

EPFL2015