Mapping potential soil salinization using rule based object-oriented image analysis

[摘要] Soil salinization is a world wide environmental problem affecting plant growth andagricultural yields. Remote sensing has been used as a tool to detect and/or manage soilsalinity. Object-oriented image analysis is a relatively new image analysis techniquewhich allows analysis at different hierarchical scales, the use of relationships betweenobjects and contextual information in the classification process, and the ability to create arule based classification procedure. The Lower Orange River in South Africa is a regionof successful irrigation farming along the river floodplain but also with the potential riskof soil salinization. This research attempted to detect and map areas of potential high soilsalinity using digital aerial photography and digital elevation models.Image orthorectification was conducted on the digital aerial photographs. The radiometricvariances between photographs made radiometric calibration of the photographsnecessary. Radiometric calibration on the photographs was conducted using Landsat 7satellite images as radiometric correction values, and image segmentation as thecorrection units for the photographs.After radiometric calibration, object-oriented analysis could be conducted on one analysisregion and the developed rule bases applied to the other regions without the need foradjusting parameters. A rule based hierarchical classification was developed to detectvegetation stress from the photographs as well as salinity potential terrain features fromthe digital elevation models. These rule bases were applied to all analysis blocks.The detected potential high salinity indicators were analyzed spatially with field collectedsoil data in order to assess the capability of the classifications to detect actual salinization,as well as to assess which indicators were the best indicators of salinity potential.Vegetation stress was not a good indicator of salinity as many other indicators could alsocause vegetation stress. Terrain indicators such as depressions in the landscape at a microscale were the best indicators of potential soil salinization.

[发布机构] Stellenbosch University