Digital Mapping of Soil Drainage Classes Using Multitemporal RADARSAT-1 and ASTER Images and Soil Survey Data

Mohamed Abou Niang,Monique Bernier,Isabelle Perron,Michel Nolin

doi:10.1155/2012/430347

Abstract

Discriminant analysis classification (DAC) and decision tree classifiers (DTC) were used for digital mapping of soil drainage in the Bras-d’Henri watershed (QC, Canada) using earth observation data (RADARSAT-1 and ASTER) and soil survey dataset. Firstly, a forward stepwise selection was applied to each land use type identified by ASTER image in order to derive an optimal subset of soil drainage class predictors. The classification models were then applied to these subsets for each land use and merged to obtain a digital soil drainage map for the whole watershed. The DTC method provided better classification accuracies (29 to 92%) than the DAC method (33 to 79%) according to the land use type. A similarity measure (S) was used to compare the best digital soil drainage map (DTC) to the conventional soil drainage map. Medium to high similarities (0.6≤S<0.9) were observed for 83% (187 km2) of the study area while 3% of the study area showed very good agreement (S≥0.9). Few soil polygons showed very weak similarities (S<0.3). This study demonstrates the efficiency of combining radar and optical remote sensing data with a representative soil dataset for producing digital maps of soil drainage.

Highlights

Much of Canada’s agricultural land has been mapped at reconnaissance or semidetailed scales
This study has demonstrated that earth observation data can be effectively used for soil drainage mapping at regional scale
This paper presents an approach to predict soil drainage classes using selected raster data derived from RADARSAT-1 and spectral indices extracted from ASTER image

Summary

Introduction

Much of Canada’s agricultural land has been mapped at reconnaissance (scale 1 : 125 000) or semidetailed scales (scale 1 : 50 000 or 1 : 63 000). Conventional soil maps are generally created using a polygonbased approach, where different soils on the landscape are represented as polygons with discrete borders Using this approach, representative soil attributes for a soil are usually assigned to a whole polygon and result in maps that contain abrupt transitions between neighbouring polygons [2]. Representative soil attributes for a soil are usually assigned to a whole polygon and result in maps that contain abrupt transitions between neighbouring polygons [2] This approach often sufficiently serves the needs for representing the spatial distribution of soils over a landscape, it is not well suited for quantitative environmental modeling.

Methods

Results

Conclusion