Automatic Interpretation of Quickbird Imagery for Digital Soil Mapping, North Caspian Region, Russia

Abstract

The digital analysis of a Quickbird image has been conducted to develop a procedure for automatic interpretation of soils within the north Caspian Depression. The soil cover in the study area has a spotted pattern and consists of contrasting soils (in terms of the humus content, salinity, pH, etc.): chernozem-like soils, light chestnut soils, and solonetzes (sodic soils). Multispectral data from the Quickbird satellite (September 13, 2006) of 2.4 m spatial resolution were used. Computer-based image analysis was conducted using the ILWIS Open GIS software (ITC, the Netherlands) and STATISTICA 6.0. This work represents the results of image interpretation for rangeland subjected to low grazing pressure. The ground truth data were collected in 2002–2004 and 2007. The original DN values (pixel brightness) of different soil types in the near-infrared (NIR), red, green, and blue bands were analyzed and NDVI values were calculated. Two methods to analyze DN values were used: descriptive statistics and discriminant analysis. The spectra showed that the brightness in the NIR band and the NDVI values are the most informative indices to discriminate soils. The indices were put into the image classification by threshold values obtained from descriptive statistics and by classification functions obtained from discriminant analysis. Both methods of image classification gave similar self-test and cross validation results, with an accuracy of classification of about 80%. Chernozem-like soils were best discriminated. Light chestnut soils and solonetzes were not delineated as well by automatic methods. The approach developed in this study can be used to map regions with contrasting soils changing within short distances, provided that the average soil area is 2–3 times more than a pixel area on the image (e.g., for solonetzic complexes of semiarid regions or for some cryogenic complexes in the tundra zone).