Intégration de données multisources de télédétection et de données morphométriques pour la cartographie des formations meubles région de Cochabamba en Bolivie

Abstract

In this study, we develop a new method for mapping surficial deposits based on the modeling of the synergistic relationship between objects on the surface identifiable from remote sensing data, morphometric information derived from a DEM and geoscience ancillary data. The performance of SPOT-4 HRVIR, Landsat-7 ETM+, and RADARSAT-1 S4 and S5 mode SAR satellite data are compared in order to identify and map the surface indicators of surficial deposits. Furthermore, morphometric data such as altitude, slope, slope orientation, slope curvature and the potential moisture index extracted from the DEM are used to define the ground topographical characteristics. Supervised classifications were carried out from single source images and multisource image combinations using image fusion techniques including RVB, ACP and HIS. Different spectral indices including NDVI, TSAVI, RI and IF, as well as texture indices (average, standard deviation, entropy, contrast) are also compared. The different layers of information obtained are regrouped into categories of variables in relation to the vegetation cover, soils, the textural organization of the landscape and the topography. Results show that the supervised classification using the maximum likelihood method based on the RVB fusion of HRVIR and SAR S4 data provides the best classification rate of the land cover estimated at 96%. The morphometric data derived from the DEM are integrated with the results of the spectral indices, the texture indices and the land cover map in a linear equation within a discriminant analysis. Based on this equation, we were able to model the synergy between the surficial deposits indicators which allowed their identification and mapping. Based solely on the binary units derived from the land cover variables, the model is capable of identifying and classifying surficial deposits in the study area with a global accuracy of 70%. The addition of spectral indices relating to vegetation and soil to the preceding information increases the global classification precision to 79%. This improvement in the results confirms the importance of the two categories of spectral indices in providing information on the density of the vegetation cover, the state of growth of the vegetation and the soil spectral characteristics. The addition of texture indices added to the previous information increases the classification accuracy to 81%. Finally, the addition of topographical information to the parameters of the previous step provides a further improvement in the global classification rate from 81% to 88%, a further increase of 7%. Validation of the final results of the model applied to the entire study area, in comparison with ground truth data and geological maps, gives a global classification rate of 88%"--Résumé abrégé par UMI.