USING LEGACY SOIL DATA FOR STANDARDIZING PREDICTIONS OF TOPSOIL CLAY CONTENT OBTAINED FROM VNIR/SWIR HYPERSPECTRAL AIRBORNE IMAGES
- 1IRD, UMR LISAH (INRA-IRD-SupAgro), F-34060 Montpellier, France
- 2Department of Soil Science and Soil Protection, Czech University of Life Scieneces, 165 21 Prague, Czech Republic
- 3INRA, UMR LISAH (INRA-IRD-SupAgro), F-34060 Montpellier, France
Keywords: VNIR/SWIR spectroscopy; Airborne remote sensing; Clay content mapping; Legacy soil data, Spectral indexes; Mediterranean context
Abstract. Mapping of topsoil properties using Visible, Near-Infrared and Short Wave Infrared (VNIR/SWIR) hyperspectral imagery requires large sets of ground measurements for calibrating the models that estimate soil properties. To avoid collecting such expensive data, we proposed a procedure including two steps that involves only legacy soil data that were collected over and?or around the study site: 1) estimation of a soil property using a spectral index of the literature and 2) standardisation of the estimated soil property using legacy soil data. This approach was tested for mapping clay contents in a Mediterranean region in which VNIR/SWIR AISA-DUAL hyperspectral airborne data were acquired. The spectral index was the one proposed by Levin et al (2007) using the spectral bands at 2209, 2133 and 2225 nm. Two legacy soil databases were tested as inputs of the procedure: the Focused-Legacy database composed of 67 soil samples collected in 2000 over the study area, and the No-Focused-Legacy database composed of 64 soil samples collected between 1973 and 1979 around but outside of the study area. The results were compared with those obtained from 120 soil samples collected over the study area during the hyperspectral airborne data acquisition, which were considered as a reference.
Our results showed that: 1) the spectral index with no further standardisation offered predictions with high accuracy in term of coefficient of correlation r (0.71), but also high bias (−414 g/kg) and SEP (439 g/kg), 2) the standardisation using both legacy soil databases allowed an increase of accuracy (r = 0.76) and a reduction of bias and SEP and 3) a better standardisation was obtained by using the Focused-Legacy database rather than the No-Focused-Legacy database. Finally, the clay predicted map obtained with standardisation using the Focused-Legacy database showed pedologically-significant soil spatial structures with clear short-scale variations of topsoil clay contents in specific areas.
This study, associated with the coming availability of a next generation of hyperspectral VNIR/SWIR satellite data for the entire globe, paves the way for inexpensive methods for delivering high resolution soil properties maps.