Improved classification of soil salinity by decision tree on remotely sensed images

Abstract

Soil salinity, caused by natural or human-induced processes, is not only a major cause of soil degradation but also a major environmental hazard all over the world. This results in increasing impact on crop yields and agricultural production in both dry and irrigated areas due to poor land and water management.Multi-temporal optical and microwave remote sensing can significantly contribute to detecting spatial-temporal changes of salt-related surface features. The study area is located in the west of Jilin Province, Northeast China, which is one of most important saline-alkalized areas in semi-arid and arid area in North China. Decision tree classifiers are used to improve the classification of soil salinity on Landsat Thematic Mapper (TM) images in later autumn of 1996. The Kauth-Thomas (K-T) transformation was performed after TM image preprocessing including image registration, mosaic and resizing for the study area. Then the first component of KT transformation, TM 6 imagery (thermal infrared imagery), and NDVI (Normalized Difference Vegetation Index) from TM 4 and TM 3 images, were density-sliced respectively to establish suitable feature classes of soil salinity as the decision nodes. Thus, the classification of soil salinity was improved using decision trees based on these feature classes.Compared with the conventional maximum likelihood classification, this method is more effective to distinguish soil salinity from mixed residential and sand areas in the west of Jilin Province, China.

Assessment of remote sensing-based classification methods for change detection of salt-affected areas (Biskra area, Algeria)

“…41,42,80 DTC techniques, either manual or automatic, have been used successfully for a wide range of classification problems, but only recently tested in detail by the remote sensing community. 33,36,37,39,40,81 Several studies have compared DTC methods with other classifiers. Otukei and Blaschke 40 compared an automated decision tree (using data mining approaches for calculating thresholds), maximum likelihood, and support vector machine-based techniques for land cover change assessment using Landsat TM and ETM+ data and found decision tree-based methods performed better than others.…”

“…Because it is easily interpreted, it is not a "black box," like the neural network, the hidden workings of which are concealed from view. 33,36 When the analyzed data are too complex in nature, data mining DTCs, or other algorithms employed for training of automated DTCs, are inappropriate for determining decision thresholds, thus manual decision trees are "safer." Traditionally, the thresholds are obtained using the knowledge provided by experts who employ their expert knowledge to assess and create the decision boundaries.…”

“…36,40,41,42 Decision tree classification (DTC) is one such technique, found to be very effective for the remote sensing community for land cover and land use (LULC) classification and salt-affected areas mapping, among a vast gamma of environmental applications. 12,32,33,35,39,40 Consequently, in this study, we employ a customized decision tree classifier that copes with the aforementioned issues, managing to delineate correctly 12 land cover classes and differentiate between two types of salt-affected areas: highly and moderately saline. We employ new indices in the decision tree classifier, 43 which aim to cope with the issues risen by the study areas, in terms of well-known spectral confusion of highly reflective land features in semidesert/desert areas.…”

“…Among these, salt features classification issues also have been reported, either with supervised or unsupervised classifications, or with spectral mixture methods, mainly regarding spectral confusion of salt features with other land ones, especially urban fabric, bare soil, or carbonaterich lithology. 11,12,19,31,32 To overcome the aforementioned issues, which are related not only to spectral confusion and misclassification but also to the difficult access to field data, two different classification schemes are approached in this study: (1) a DTC 12,[33][34][35][36][37] and (2) IsoDATA classification applied to the principal components analysis (PC/PCA) of Knepper ratios.…”

Diachronic analysis of salt-affected areas using remote sensing techniques: the case study of Biskra area, Algeria

“…Knepper (1989) proposed specific band ratios for the delineation of hydroxyl -bearing minerals, hydrated sulphates and carbonates, vegetation and iron oxides and hydroxides, namely the 5/7:3/1:3/4 red-green-blue (RGB) combination, used mainly for geological remote sensing mapping [8], but the novelty of the current study is that it is employed for salt features identification. Thus, two different classification schemes are proposed in this study: 1) a decision tree analysis (DTA) [9] [10,11], [3] based on spectral analysis, band transformation and expert-knowledge and 2) the analysis of the Principal Components (PCA) of Knepper ratios, adequate in the fast, spectral-based delineation of mineral components. In the Biskra area, ground truth data was difficult to achieve in the correct amount needed for such studies, thus available ancillary data were used as basis throughout the study phases.…”