Airborne hyperspectral imaging of spatial soil organic carbon heterogeneity at the field-scale

Abstract

A detailed knowledge on the heterogeneity of the soil organic carbon (SOC) content in agricultural soils is required to support applications such as precision agriculture and soil C monitoring. Imaging spectroscopy in the visible (VIS) and near-infrared (NIR) region has proven to be highly sensitive to organic soil components and can efficiently provide data with high spatial resolution. The objectives of our study were (i) to test the suitability of airborne hyperspectral imaging for the characterisation of the spatial heterogeneity of the SOC content at the field-scale, (ii) to investigate the impact of various soil surface conditions (roughness, vegetation) on SOC prediction and (iii) to produce SOC maps for arable fields on a pixel-wise basis. The soil reflectance was recorded by the aircraft-mounted hyperspectral sensor HyMap (450–2500 nm) on test sites with the following varying soil surface conditions: bare soil, fine seed-bed; ploughed, bare soil; volunteer crops; straw residues. A partial least squares regression (PLSR) was performed for data analysis. Our results reveal an accurate prediction of the SOC content at a comparatively small concentration range (8.3 to 18.5 g SOC kg^− 1) on long-term uniformly cultivated fields. Site-specific characteristics influenced the calibration models; highest prediction accuracy was performed over a bare, fine soil (RMSEP = 0.76 g SOC kg^− 1; RPD = 2.08). A generated pixel-wise map (8 m × 8 m) allows the detection of small-scale spatial variability of SOC content and comparatively more realistic than an interpolated map. Thus, airborne hyperspectral imaging constitutes a substantial progress compared to point observations and facilitates well-directed applications in precision agriculture.

Introduction

Soil organic carbon (SOC) is a particularly important soil property affecting other soil parameters as well as crop growth. The spatial distribution of the SOC content is often heterogeneous within agricultural fields. A detailed knowledge on the variability of SOC at the field-scale is required to support applications such as precision agriculture (Gebbers and Adamchuk, 2010). Precision agriculture aims to match both agricultural input and practices to the spatial and temporal variability of soils, crops, pests and weeds within a field, instead of the uniform and suboptimal management of an entire field. The spatial distribution of the SOC content within a field influences various crop stand parameters such as crop nutrient status or crop yield and the related amount and distribution of fertilizer needed as well as the behaviour of pesticides in the soil (Ladoni et al., 2010, Patzold et al., 2008, Wauchope et al., 2002). In addition, SOC is closely related to soil quality, as it performs as an indicator of soil erosion and degradation (De Gryze et al., 2008). Depending on land-use and management, agricultural soils can be a major source or sink for carbon (Lal, 2010, Sleutel et al., 2007). A high sampling density is necessary to produce SOC maps with an adequate spatial resolution, as required in precision agriculture or for soil C monitoring (Chang et al., 2001, Kerry et al., 2010, Stevens et al., 2006). However, conventional soil analyses are often both laborious and expensive. Imaging spectroscopy in the visible (VIS) and near-infrared (NIR) region is a more efficient, rapid and less expensive technology which provides data with high spatial and temporal resolution (Gomez et al., 2008, Stevens et al., 2010).

VIS, NIR and also mid-infrared (MIR) techniques are reported to be highly sensitive to the organic components of a soil (Viscarra Rossel et al., 2006). Spectral signatures related to various components of soil organic matter (SOM) generally occur in the MIR range (2500–25,000 nm), but their overtones can be found in the VIS/NIR (400–1200 nm) and the shortwave-infrared (SWIR; 1200–2500 nm) ranges (Shepherd and Walsh, 2002). Several studies reveal that on the average MIR outperforms VIS/NIR, because MIR spectra consist of more defined peaks and thus are often described as performing better in estimating the SOC content (Chang et al., 2001, Ladoni et al., 2010, Patzold et al., 2008, Reeves, 2010, Stevens et al., 2006, Viscarra Rossel et al., 2006). Due to its increasing use and wide acceptance, MIR spectroscopy (MIRS) may be considered as a reference method, even if it's not yet an established laboratory standard. However, MIRS is to be used preferentially under laboratory conditions, while NIR spectroscopy (NIRS) can be easily applied in the field either as portable or as airborne sensors.

NIRS has been used for several years for the assessment of SOC (Dalal and Henry, 1986, Henderson et al., 1992). Ladoni et al. (2010) reviewed the satisfying ability to predict SOC via NIRS under laboratory conditions. However, the choice of spectroscopic techniques used in field studies depends mainly on the accuracy of prediction, the cost of the technology and the amount of sample preparation required (Viscarra Rossel et al., 2006). To enhance the efficacy of spectroscopic techniques for soil mapping and monitoring, airborne sensor applications are becoming more popular, though they remain in the testing phases (Patzold et al., 2008, Selige et al., 2006, Stevens et al., 2006, Stevens et al., 2008, Stevens et al., 2010). According to DeTar et al. (2008), soil properties can be accurately detected using airborne hyperspectral imaging due to its high spatial resolution and supplying a complete spectrum of data for every pixel location. Several authors successfully mapped SOC on large agricultural fields and regions via airborne hyperspectral sensors using multivariate calibration statistics (Selige et al., 2006, Stevens et al., 2006, Stevens et al., 2008, Stevens et al., 2010).

Multivariate calibrations, such as a partial least squares regression (PLSR), allow for a quantitative determination of several soil characteristics from spectral signatures (VIS, NIR, SWIR and MIR) and are a common tool used to derive soil properties from hyperspectral data (Selige et al., 2006, Stevens et al., 2010, Viscarra Rossel et al., 2006). However, airborne hyperspectral imaging still has several limitations, as reviewed by Ben-Dor et al. (2009) and Cécillon et al. (2009). These limitations can be atmospheric absorptions or sensor-based characteristics such as a low signal-to-noise ratio and a limiting spatial resolution, or can occur due to spatial and temporal soil surface conditions, such as variable moisture content, soil surface roughness, or green manure or crop residue covers. In consequence, local calibration outperforms regional calibration (Stevens et al., 2010). The airborne HyMap sensor has shown itself to be an appropriate tool used to monitor SOC at a regional scale (Selige et al., 2006). As recently reported by Ben-Dor et al. (2009), Cécillon et al. (2009) and Ladoni et al. (2010), more experience is required to accurately predict SOC via airborne hyperspectral imaging, as only few studies dealt with this topic by now. Former research focused primarily on the detection of SOC at large regional scales for entire landscapes. However, SOC can considerably vary within few metres. For effective site-specific management, which is especially required for precision agriculture, a detailed knowledge of the small-scale variability of SOC within individual agricultural fields is an essential requirement.

The objectives of this study were (i) to test the suitability of airborne hyperspectral imaging (HyMap sensor) for the characterisation of the spatial heterogeneity of the SOC content at the field-scale including (ii) investigations concerning different soil surface conditions (roughness, vegetation) and (iii) to produce SOC maps on a pixel-wise basis as required for precision agriculture.