Fuel mapping is a key activity for forest fire risk management. It is based on remote sensing images processing methods. Spatial patches of fuel types are complex and highly heterogeneous spatial entities. Complexity of fuel types, in relation to their remote sensing-based mapping problem, is classified in four sorts (Borgniet 2009): purely spectral complexity, ie. complexity of the relationship between fuel types and spectral signatures, spatial heterogeneity of fuel types spectral signatures, spatial horizontal structures of fuel types, and fuel types vertical structure heterogeneity. Some methods are developed to solve each kinds of complexity: pixel based spectral methods, texture analysis based methods, object based methods and 3D analysis methods. On an operational point of view, most of the methods are mixed. But it is not possible to propose one unique method able to produce a fuel map valid in any context with the same parameters. Proposed methods are context dependent and might be complementary in order to solve the global problem of fuel mapping in a given geographical and ecological context. They are usually implemented in one specific software environment. This lead us we choose an open knowledge based system, opposed to a closed processing solution. The system is aimed at helping the user to build a global successive processing approach that we call a “demarche”, in order to better respond to his needs (fig 1). Conceptual specification of the system is based on the model integration paradigm, in which methods are represented by models. In a first stage, the coupled DEV'S formal system (Ziegler 1999) s to conceptually specify coherent demarches. In a second stage, semantic integration is aimed at solving semantic heterogeneity between models to be integrated at a conceptual level. It specifies the semantic relationships between concepts handled by the models to be integrated. If semantics (i.e. list of concepts) handled by the models are different, integration will require the specification of models for models integration (Maillé 08): such models specify the relationship between concepts of the initial models. Finally, syntax integration is aimed at solving heterogeneity of representation terms of information handled by the models to be integrated. It permits models interoperability which allows proper functioning of the resulting model, without referring to its semantic consistence (Müller 08). Syntax integration might be specified at different abstraction levels: organisational level (architecture), logical level (data models, communication protocols, etc.), physical levels (networks), etc. The specified tool architecture includes a knowledge database of methods and resources, and an expert system for methods selection in relation to the user needs and constraints specification. It is a distributed system, where the different resources, either data or processing systems, are distributed on a network of "nodes". Although the database is unique, it is also distributed on the nodes. Selected methods can then be organized into demarches by the user. An executive engine is designed to execute the different methods of the demarche in their respective computer environment, through mediating wrappers. A research prototype called “Fuel Mapping Methods Integration Platform” (FMMIP) was developed.