Soil aggregate stability (AS) is controlled by an array of elementary soil properties such as soil organic carbon (SOC), texture and extractable metal oxides [1
]. Agricultural management practices and topographic positions also influence its dynamics, thus making AS a dynamic property that evolves with space and time [3
]. Decrease of AS in croplands not only hinders agricultural production through the control over surface crusting and seedling emergence [5
], but also increases the risks of soil degradation, thereby compromising the physical protection of SOC by intact soil structure [6
] and amplifying the erosion-induced nutrient and pollutant transfer from soils to surface water [7
]. Among all influencing factors, SOC is frequently reported to be an essential one that positively controls the dynamics of AS [8
]. Meersmans et al. [11
] reported a decreasing trend in cropland SOC content during a 50-year period, which increased the risk for soil structural degradation, especially in areas where SOC concentrations were close to 1% or lower. The Voluntary Guidelines for Sustainable Soil Management published by the Food and Agriculture Organization of the United Nations [12
] also identified the increase of SOC stock as an important measure to combat soil degradation. In light of the current sustainable soil management efforts aiming to mitigate climate change (e.g. “4 per 1000” initiative [13
]), numerous studies have shown the positive impact of crop residue retention and conservation tillage on AS in croplands [10
]. To better understand the underlying mechanisms leading to the dynamics of SOC and AS, and to target areas that are vulnerable to soil degradation and nutrient losses, it is necessary to monitor SOC and AS over large spatial and long temporal scales.
The importance of accounting for the spatiotemporal variability of AS is also recognized in soil erosion assessments. As an indicator of soil erodibility to reflect a soil’s inherent resistance to external erosive forces [16
], AS is often used as an input into spatially distributed soil erosion models to investigate erosion-induced soil redistribution patterns [17
]. Performance assessments of such models pointed to the gap between satisfactory spatial predictions of soil erosion and insufficient representation of the spatial variability of soil erodibility [18
]. Due to the lack of large-scale, high-resolution AS data, current modelling practices often assume soil erodibility (as indicated by AS) as a static parameter, instead of treating it as a spatiotemporally dynamic input. Among limited number of studies that assessed the spatial variability of AS, Mohanmmadi and Motaghian [19
] showed the large spatial variation of AS with a range value of ca. 3 km from variogram analysis; Annabi et al. [20
] also demonstrated the spatial structure of AS in a geologically diverse region and pointed out the need to capture small-scale variability of AS by increasing the sampling density. However, with wet-sieving remaining to be the conventional method to measure AS in those studies, it is unrealistic to analyze a large number of samples required to assess the large-scale variability of AS at a fine spatial resolution. This highlights the necessity to develop new methods that allow efficient quantification of AS at large scales and fine resolution.
Recent developments of hyperspectral remote sensing techniques have shown the potential to map key soil properties, such as topsoil SOC and soil texture [21
], both of which are important determinants of AS. In particular, the Airborne Prism Experiment (APEX) [22
] sensor offers high spectral and spatial resolution images and has been proven to be capable of predicting SOC at field to catchment scales [23
]. Other airborne hyperspectral imaging applications in digital soil mapping include developing a spectral index to correction for soil moisture effects [24
], and analysis of spatial organizations of mapped soil properties (e.g. CaCO3
, iron and cation exchange capacity) [25
]. Apart from using hyperspectral images to predict and characterize soil properties, spatial patterns of erosion and deposition could also be characterized by developing classification methods with spectrally-predicted elementary soil properties as inputs [26
], and by matching soil properties of different soil horizons emerging at the surface with different soil erosion and deposition stages [27
As of yet, few studies used hyperspectral remote sensing images to directly map secondary soil physical properties across large scales, while laboratory-based hyperspectral data have already been extensively explored to predict properties, such as AS, that are related to known soil chromophores [28
]. For instance, soil mean weight diameter (MWD), a lumped index commonly used to express AS, and different aggregate size fractions were successfully predicted using laboratory visible-infrared (Vis-NIR) spectroscopy [29
]. The authors attributed the good model performance to the close correlation between AS and SOC, as the wavelengths that are known contributors to the prediction of SOC were also found to be significant in the MWD prediction. This warrants further investigations on whether the successful application of hyperspectral imagery to SOC mapping could be transferred to the mapping of AS.
The objective of this study was to develop a method for large-scale, high-resolution AS mapping for the investigation into the spatial dynamics of AS. To this end, we aim to test the capability of APEX hyperspectral imagery to predict AS across an agricultural region in Belgium at a 2×2 m spatial resolution. The approach used in this study began with extracting bare soil fields based on pre-defined spectral indexes that are representative of bare soils. Then, partial least squares regression (PLSR) models were established using the hyperspectral data extracted from APEX images. Finally, the AS prediction model was evaluated against an independent validation dataset and an AS map of the study area was produced. Such a map will not only provide detailed information on field-level soil degradation status for the sake of precision agriculture, but at the same time allow assessments of spatial variation of AS at multiple scales.