Recognition and mapping of soil salinization in arid environment with hyperspectral data

Abstract

Hyperspectral imagery of airborne imaging spectrometer (Pushbroom Hyperspectral Imager (PHI)) was acquired over KeLaMaYi, which situated in arid region of northwestern China. In situ hyperspectral data obtained with FieldSpec® HandHeld spectrometer (ASD) simultaneously were analyzed for recognition of soil salinization. Some types of transformation were applied to the reflectance data of 60 soil samples, which preprocessed with a simple smoothing followed by band merging. A comparative study among these methods was made to ascertain their applicability for recognition accuracies. After multivariate analysis between ion concentration and reflectance data or their derivatives, a best statistical model was then extracted to predict the soil salinity and PH. Using this prediction model, subpixel classification applied to the corrected imagery helped to yield quantitative maps of soil salinity and PH. Such maps contributed to suggesting soil distribution and aggregation, estimating the spatial controls of salinization, and consequently, helping to plan soil improvement and soil conservation schemes. Detailed spectral information in the spectral range of the reflected electromagnetic solar energy (i.e., visible (VIS), near-infrared (NIR), and short-wave infrared (SWIR)) can be acquired by recent imaging spectrometry systems. They allow the identification of a large range of land surface materials that cannot be identified with broadband, low spectral resolution satellite remote sensing systems (2). The presence of salts at the terrain surface can be detected from hyperspectral data either directly on bare soils, with salt efflorescence and crust, or indirectly through vegetation type and growth as these are controlled or affected by salinity (3). The capabilities of hyperspectral imagery for salinity mapping have been recently investigated. Reference (4) mapped saline areas characterized by salt scalds, halophytic vegetation, and soils with varying salinity degrees and types. Spectral features in the visible and near-infrared parts of the spectrum, related to combined water in hydrated evaporite minerals, allowed saline soils to be mapped by the HyMap scanner. Further tests showed that the subtle hydrate absorption features at 980 and 1170 nm are crucial for correct mapping of salt-affected soils. Reference (5), using the hyperspectral DAIS-7915 sensor, concluded that the Visible and Near-Infrared Analysis (VNIRA) approach is a feasible tool to produce quantitative soil surface maps of organic matter, soil field moisture and soil salinity. The objective of our study, based on multivariate statistical analysis of field-derived hyperspectral spectral data of soil samples and hyperspectral imagery mapping, was to investigate the potential of high resolution reflectance spectra to access soil salinization level and spatial distribution in an inland arid environment of northwestern China. In this paper, we describe the spectral capability in the 400 to 900nm range of spectra of selected soils in KeLaMaYi. These are used to develop a strategy for the use of field-derived spectra for the mapping of salinized soil from Pushbroom Hyperspectral Imager (PHI) airborne scanner imagery.